Div XUser Guide 521

Revision 5.2.1.
1
Contents
Contents
Index
Index
Introduction Psychovisual Enhancement
Welcome to the guide............................. 5 What is psychovisual enhancement?....56
The origins of DivX® video ...................... 5 The DCT, iDCT and
What is DivX? ......................................... 7 Human Visual System......................56
Why use DivX? ....................................... 8 The psychovisual
enhancement system .......................57
Quick Start Guide Experiencing PVE .................................58
Introducing VirtualDub .......................... 11 Fast psychovisual enhancement...........59
Your first DivX....................................... 12 Slow psychovisual enhancement..........60
Your first Multipass ............................... 17
Notes on audio ..................................... 19 Source Preprocessing
What is pre-processing? .......................62
Forward When to use pre-processing .................62
General concepts ................................. 24 Spatial processing.................................63
Frames ............................................ 24 Temporal filtering ..................................63
Macroblocks and motion ................. 24 Source pre-processing ..........................64
Intra-frames and Predicted-frames.. 26
Quantizers ....................................... 27 Crop and Resize
A key to the guide................................. 29 Cropping and resizing via the encoder .66
Why crop or resize? ..............................66
Bitrate Mode Crop ...............................................67
What is the bitrate?............................... 32 Resize ...............................................69
Bitrate calculation ............................ 33
The bitrate calculator ............................ 34 MPEG-4 Tools
1-Pass .............................................. 35 Bi-directional encoding
Multipass .............................................. 37 What is bi-directional encoding? ......72
Multipass, 1st pass .......................... 37 Adaptive bi-directional encoding ......73
Multipass log file .............................. 38 Multiple consecutive
Multipass, nth pass.......................... 40 adaptive B-frames.......................74
Bitrate modulation............................ 40 The ordering of frames.....................75
1-Pass, Quality-based .......................... 44 Bi-directional encoding.....................77
Quarter Pixel
Performance/Quality What is quarter-pixel? ......................79
Fastest .............................................. 48 Why use quarter-pixel? ....................80
Fast .............................................. 50 Quarter-pixel ....................................81
Standard .............................................. 51 Global motion compensation
The Rate-Distortion Algorithm .............. 52 What is global motion
Slow .............................................. 54 compensation? ...........................83
Global motion compensation ...........84
The
Guide
Contents
Index
Advanced
Scene-change threshold ...................... 86
Maximum key-frame interval ................ 88
Quantization type ................................. 89
Interlacing
What is interlacing? .............................. 92
Why de-interlace? ................................ 93
Maintaining interlacing ......................... 93
Source interlace ................................... 94
Video Buffer Verifier

What is the video buffer verifier?.......... 97
The video buffer verifier explained ....... 97
Video buffer verifier .............................. 98
Profiles
What is a profile?................................ 101
Why use profiles?............................... 101
Profiles ............................................ 102
DivX Certified™ Program

What is the DivX Certified Program? . 105
What’s in a profile?............................. 106
DivX certified requirements ................ 107
Electrokompressiongraph™
What is the EKG?............................... 109
How the EKG works ........................... 110
Using the EKG.................................... 111
DivX® Decoder
About the decoder .............................. 114
Post-processing.................................. 114
De-blocking ................................... 115
De-ringing...................................... 115
Automatic post-processing............ 115
Post-processing levels .................. 116
Film effect........................................... 116
Quality settings................................... 117
Smooth playback........................... 117
YUV Extended............................... 117
Overlay Extended.......................... 118
Double buffering ............................ 118
Disable logo................................... 118
Support generic MPEG4 ............... 118
Acknowledgements
Thanks and credits ............................. 120
Legal ............................................ 120
The
Guide
Introduction
The
Guide
Introduction
Introduction
Welcome to the guide,
Welcome to the guide
The origins of DivX video

Welcome to the official guide to DivX® 5.2!
With this guide we will demystify the black-art of DivX encoding - leading you
step-by-step through everything from creating your very first DivX video to
®
exploiting the many powerful advanced features that DivX has to offer.
This guide covers all of the features available in both the DivX and DivX Pro
encoders in detail. You will learn how to optimize video quality via the encoder
and where to use the EKG for even further enhancement.
We will discuss the DivX Certified™ Program and how you can create videos
compatible with the growing range of DivX Certified DVD players and other
portable devices that are set to take the consumer market by storm in 2004. We
will also show you how to configure the DivX decoder to improve the playback of
DivX videos on your personal computer.
Before we begin it is worth noting the outstanding pedigree of DivX video

technology. The next section transcribes over five years of DivX history!
The origins of DivX ® video

In 1999, French video engineer Jérôme Rota (better known by the alias “Gej”)
began experimenting with powerful new video technologies.
The “MS MPEG4” codec from Microsoft, a video compression technology then
new to market, was limited to storing video data inside their proprietary ASF
(advanced systems format) container file. Gej wanted to use a high-quality video
codec in combination with standard AVI (audio video interleave) containers,
which would enable nearly all video editing applications to benefit from new
compression.
The
Guide
Introduction
The origins of DivX video
By September, 1999 Gej had achieved his goal of using a high performance
codec in combination with the AVI container, and DivX ;-) (the emoticon a jibe at
Circuit City’s now-deceased protected DVD rental system) was unleashed upon
the Internet. Boasting compression ratios over five times superior to the
technologies used in conventional DVDs, DivX ;-) would allow entire movies to be
transferred to single compact discs or distributed over high speed connections in
®
a matter of hours rather than days.
DivX ;-) gained an overnight cult following amongst the Internet underground.
Within days of its release, it became possible to download entire feature-length,
DVD-quality videos for the first time, and DivX ;-) was well on its way to
becoming the MP3 of video.
But DivX ;-) was just the beginning. Intrigued by the possibilities of video
compression and not ready to retire and rest on his laurels, Gej quickly dropped
the winky face and began the long process of creating a new codec entirely from
scratch.
In May 2000 Gej co-founded DivXNetworks with CEO Jordan Greenhall, Director
of Product Management Joe Bezdek, R&D director Darrius Thompson and Vice
President of Operations Tay Nguyen.
In early 2001, DivXNetworks covertly launched ProjectMayo, a website that

would become the stage for creating OpenDivX - an open-source development
project that invited developers and video engineers worldwide to collaborate on a
free, open version of DivX. This would ultimately become the foundation of the
XVID project. Meanwhile, DivXNetworks secured approximately $5.6M in funding
from investors and began internally developing the future of DivX technology.
By August 2001, DivX 4 was born. A proprietary codec based on the international
MPEG-4 standard, DivX 4 was the first in a long line of DivXNetworks releases,
culminating in the commercial release of DivX 5.0 in March 2002.
Since that time DivX has gone from strength to strength - a patent-pending
technology that is constantly evolving in terms of the performance, quality, and
features. DivX 5.2 sees the ultimate culmination of this work.
The
Guide
What is DivX?
Introduction
What is DivX?
Firstly, let’s address what DivX is not. DivX is not a stand-alone encoding
application; that is when you install DivX you are not getting a program that will
convert your files to DivX format auomatically. For such an application, look to
Dr. DivX.
Instead, DivX is a video codec (COmpressor/DECompressor) accessible via

most video applications supporting either:
►Microsoft Video For Windows for applications running under Windows

►QuickTime for applications running under Mac OS
In short, this means DivX enables the export of high quality video from virtually
any software of your choosing. With the right software it becomes possible to
take any video source - be it live capture, DVD, DV, MPG, MP2, AVI or other -
and export them to DivX video.
The DivX 5 series encoder is an implementation of the MPEG-4 video standard,

supporting both simple profile and advanced simple profile encoding. Essentially
the Moving Pictures Expert Group creates standards and methods for achieving
various objectives, one of which is video compression. Defined within the
MPEG4 Video Standard are various profiles representing feature sets that a
codec complying with the standard must fully implement.
Because it is based upon the MPEG4 Video Standard, DivX naturally supersedes
both MPEG1 (as used for Video CD) and MPEG2 (the standard for DVD video
and Super Video CD) for low-bitrate encoding. To put this into perspective, DivX
5 achieves DVD-quality video at one tenth the size of a DVD, and because DivX
5 is fully standards compliant you can watch your DivX videos on a growing
range of DivX Certified hardware players*.
* Different MPEG4 hardware players tend to support the MPEG4 Video Standard to varying degrees. For
guaranteed playback capability and performance only DivX Certified players are recommended.
The
Guide
Why use DivX?
Introduction
Why use DivX?
There are many codecs available these days, so why choose DivX to export your
video?
►Quality
DivX is a professional grade video codec. The Pro version available to
consumers for less than $20 is exactly the same program licensed for
use in commercial production environments. While other MPEG-4
encoders exist, none include the extensive performance optimizations
and patent-pending algorithms that enable DivX to deliver the highest
possible video quality while maintaining full standards compliance.
►Certified video
DivX produces video guaranteed to play back flawlessly on a growing
range of DivX Certified™ devices. The DivX encoder includes profile
presets that let you encode for players with differing capabilities; from
the Handheld profile for palm-top computers to High-Definition profile
for players that can decode for HDTV. Simply match the logo shown in
the DivX encoder to the logo on your player. The DivX Certified
Program already includes over 30 companies, and DivXNetworks
rigorously tests every single certified device to guarantee its
performance.
►Home movies
With DivX you can export your home movies in unprecedented quality
then burn them to CD and share them with friends at next to no cost.
Not only that, but because DivX movies take up so little file space you
no longer have to worry about cutting down the running length and
missing out on all those golden moments. Use DivX now to create clips
you can be proud to show your relatives for years to come.
►Backup DVDs*
DVDs are expensive, and one little scratch is sometimes all it takes to
turn your favorite disc into a worthless coaster. By converting your
DVDs into DivX format you can make inexpensive CD backups that will
save your original discs from wear and tear. Let DivX help you to enjoy
your collection while protecting it from damage.
* It is legal in many countries to make a single backup for personal use only. Local laws may apply.
DivX is not intended as a tool for unlicensed copying.
The
Guide
Introduction
Why use DivX?
►Internet distribution
Video transferred over the Internet used to be small, jumpy, and blocky
with poor color definition. DivX removes the barriers, letting you quickly
and easily transfer high-quality video via the Internet. Using a regular
broadband connection it is possible to transfer a feature-length movie in
hours, and shorter clips in minutes.
The
Guide
Quick Start
Guide
The
10
Guide
Quick Start Guide
Quick Start Guide

Introducing VirtualDub
Introducing VirtualDub
Working with video requires a variety of tools to achieve the best possible results.
There are some processes for which dedicated-purpose tools can be invaluable
in the world of digital video. Most notable of these tools by far is VirtualDub.
VirtualDub is a free dubbing application that lets you cut, splice, dub, convert
between video formats, and apply filtering effects.. Although it lacks some
functionality typically associated with commercial video editing applications (easy
DVD conversion, programmable digital VCR, video transition effects, etc.), it is an
essential tool in anyone's arsenal: powerful, yet remarkably easy to use.
This Quick Start Guide will teach the basics of VirtualDub and walk you through
creating your very first DivX videos. For now, resist the urge to explore and
simply follow the instructions.
Download VirtualDub before we start. It’s a small download, open-source, and

entirely free (although the author does accept donations!).
http://virtualdub.sourceforge.net
Extract the .zip file to a new folder on your hard drive. No formal installation
procedure is required. This guide uses VirtualDub version 1.5.4, but the interface
of later versions should remain similar.
You will also need a pre-existing sample .avi or .mpg file to work with in this
guide. We recommend a short file taken from a reliable source (such as CD-
ROM) that is unlikely to be corrupt or otherwise damaged.
The
11
Guide
Your first DivX
Quick Start Guide

Your first DivX
To begin we will use VirtualDub to open an MPEG 1 video file and convert it to
DivX format. For the purposes of simplicity our video will not include an audio
track at this point.
1. Load VirtualDub by double-clicking on the VirtualDub

program icon.
VirtualDub Program Icon
In the VirtualDub main window

(from top to bottom): Program
menus, Video display area
(currently empty), Timeline and
seek control, Timeline control
buttons, Tool tips bar.
From left to right the timeline

control buttons are: Stop, Play
input, Play output, Go to start,
Backward, Forward, Go to end,
Previous key-frame, Next key-
VirtualDub Main Window frame, Find last scene change,
Find next scene change, Set
mark-in, Set mark-out.
2. From the File menu choose Open and

select your sample file.
The
12
Guide
Quick Start Guide
Your first DivX
3. From the Audio menu set No Audio. This instructs
VirtualDub that even if the source file contains audio
data we do not want to process it or copy it into our
output file.
4. From the Video menu set Full Processing Mode.

In this mode VirtualDub allows us both to
recompress the video (here we will be going from
MPEG1 to DivX) and also apply any filters should
we choose to do so.
Because we will not be applying filtering here

Normal Recompress might give us better
performance. Fast recompress gives the best
performance but may not work in all
circumstances.
For the present time leave the video mode set to Full Processing Mode.
5. Now we want to configure DivX as the

video compressor. Select the Compression
option from the Video menu to open the
compressors list.
Select DivX Pro(tm) 5.2

Codec and click the
Configure button to
bring up the encoder
configuration dialogue.
The
13
Guide
Quick Start Guide
Your first DivX
6. Congratulations!
If you made it this far, this is the first time

you will have seen the DivX encoder. At
this point we simply want to configure the
encoder with the default settings using
the Restore Defaults, button but you
might want to take this opportunity to
explore the encoder configuration
dialogue before continuing.
When you are ready, click Restore

Defaults and then OK.
All that remains is to instruct VirtualDub to

7. save the AVI file. When you do this
VirtualDub will begin processing the video
and writing the encoded data to file.
Enter a suitable filename and

click Save to commence
processing.
The
14
Guide
Quick Start Guide
Your first DivX
8. The VirtualDub Status window should
pop up and allow you to monitor the
encoding progress.
Here you see the progress bar, how

much video and audio data has been
encoded, and a projection of the final file
size. The projection is a very rough
estimate because DivX may choose to
vary the bitrate throughout the file.
If the estimated time is quite long and

you would like to work in other
applications while encoding set
Processing thread priority to lowest. This will make Windows share more
CPU time with other applications but will also result in longer encoding
times.
The Feedback
Window, first
introduced in DivX 5.1, lets
you monitor and manipulate
the inner-workings of the
DivX encoder in real-time.
You will see the feedback

window unless it is disabled
under the Settings button,
located in the bottom-left
corner of the encoder
configuration dialogue (see
illustration for step 6).
You might like to explore some of the display options shown here while
your video encodes, but for now avoid altering the Bitrate, pvLumaFlat, and
pvLumaTexture options in the bar at the very bottom of the window.
With regards to performance, the feedback window may reduce the rate at
which your video is encoded. To reduce overhead, increase the update
interval so that the feedback display is not updated every single frame.
The
15
Guide
Quick Start Guide
Your first DivX
The feedback and status windows will close automatically after encoding.
You should then be able to play your first DivX movie using the DivX Player
(bundled free with DivX 5.2), or any other media player.
If you would like The DivX Player to open your file by default as opposed to
any other player you can rename the AVI file after encoding so that the file
extension reads “.divx“ instead of “.avi”. The DivX Player automatically
registers the “.divx” file extension when it is installed.
The
16
Guide
Your first Multipass
Quick Start Guide

While 1-pass mode is fast and particularly suited to certain situations, Multipass
encoding offers more consistent quality by allowing the encoder to first analyze
the entire video before encoding it.
1. Load VirtualDub and configure it as you did in

Your first DivX steps 1 through 5 but this time
set the Variable bitrate mode to “Multipass, 1st
pass”.
When you save a video in Multipass, 1st pass

mode the encoder will generate a log file that
contains an analysis of the video but no actual
video output will be generated.
2. Save your AVI file using a suitable file

name. You don’t need to use a separate
file name for every pass that you
perform, so as before simply choose a
name that fits the video.
3. Without closing VirtualDub re-configure the

encoder by setting Variable bitrate mode to
“Multipass, nth pass” and ensure “Update
log file” is enabled.
When you save a video in Multipass, nth

pass mode the encoder will use the analysis
log from the previous pass to help it encode
the video with a consistent quality but still
targeting the bitrate you have specified.
The
17
Guide
Quick Start Guide
Because we selected Update log file, the encoder will record an analysis of
this nth pass with respect to the decisions it makes based on the previous
analysis from the log file. This lets you feed the analysis log from one nth
pass into successive nth passes and refine the encoding process toward
the optimal quality level.
4. Save the AVI file using the same file

name as you used for the 1st pass. It
is normal practice in Multipass
encoding to overwrite the output from
the previous pass every time you
save.
After completing your nth pass

encoding, play the AVI file to see the
output.
5. Without changing any settings try saving the AVI file one more time, again
choosing to overwrite the previous file. The encoder will use the log file we
updated in the last nth pass to create an even higher quality video.
Although you might be tempted to continue running nth pass after nth pass
there is of course a limit on the quality attainable at any given bitrate.
Typically 98-99% of the optimal quality will be realized after 3 passes (1st,
nth, nth) or less.
The
18
Guide
Notes on audio
Quick Start Guide

Notes on audio
DivX is a video codec. As such it does not handle audio in any way and we will
not cover the very broad topic of audio in detail. Most likely, you will want audio
with your DivX video, so we will briefly cover a few important points.
1. There are many audio formats and codecs that can be used in combination
with an AVI container, each with different properties and features. DivX AVI
files generally use Constant Bitrate MPEG1-Layer 3 Audio, or CBR MP3 for
short. CBR MP3 is the recommended audio format for DivX video and has
a number of benefits:
►MP3 audio is supported natively by most operating systems
►MP3 audio is supported by DivX Certified hardware players
►CBR MP3 audio is least likely to introduce de-synchronization between

audio and video.
The MP3 codec distributed with Windows is the Fraunhofer IIS MPEG
Layer-3 Codec (advanced). This codec is capable of decoding high-bitrate
MP3 format audio but is limited to low-bitrate encoding.
VirtualDub works only with audio codecs designed for Microsoft’s Audio
Compression Manager, or ACM, and thus you will need to install an ACM
MP3 encoder.
As a rough guide: 128 kbps is generally considered close to CD quality

when encoding to CBR MP3, but bitrates 112 kbps and up are generally
acceptable for stereo sound. Halve the bitrates for mono sound.
The
19
Guide
Quick Start Guide
Notes on audio
Audio contained in an AVI file must be correctly interleaved for stable
2. playback performance—particularly when content is to be stored on optical
media (such as CD-R or DVD-R), or other media with high seek times.
Consider this AVI container with one video and one audio stream:
AVI Container
Audio Video
This represents the worst possible situation with regards to interleaving.

During playback the system will need to seek back and forth within the file
continuously in order to load the audio associated with the video as time
progresses. If the viewer were to jump to a specific point in the video it
might be hard for the media player to quickly locate the corresponding point
in the audio stream, causing an undesirably high seek recovery time.
The
20
Guide
Quick Start Guide
Notes on audio
Now consider the corresponding correctly interleaved file:
AVI Container
Audio Video Audio Video Audio Video Audio Video Audio Video
Here audio chunks are interleaved within the file so they fall closely
alongside the related sequence of video. As playback progresses seeking
within the file is minimized, and in fact with appropriate buffering as
provided by most devices seeking will not even take place. This leads to
smoother playback performance. Because audio chunks are stored
alongside the corresponding video chunks if a viewer skips to a particular
point in the video the media player will likely locate the associated audio
and resume playback more promptly.
Using VirtualDub, you can access the Interleaving

options from the Audio menu.
Ensure “Enable audio/video interleaving” is enabled. The recommended

audio preload is the default 500 ms. The recommended audio interleave is
every 1 frames.
Note that the interleave options also present a method of adjusting audio/
video sync by adjusting the “Audio skew correction” value, either positively
or negatively in milliseconds.
For example: If the audio was running two seconds behind the video a
value of –2000 ms would correct for the delay. Conversely if the audio was
running one and a half seconds ahead of the video a value of 1500 ms
would be required.
The
21
Guide
Quick Start Guide
Notes on audio
VirtualDub contains two audio processing modes,
3. “Direct stream copy” and “Full processing mode”.
In direct stream copy mode no processing (other than

interleaving) is performed on the audio stream. If you
wanted to manipulate only the AVI file’s video stream,
configure the audio mode as Direct stream copy in
conjunction with “Source audio”. When you save the
new AVI file, the original audio stream is copied from
the source.
In Full processing mode the given audio stream is

decompressed and fully processed when saving, including the application
of any filters or compressors you have selected.
Both Direct stream copy and Full processing mode have exactly the same
meaning when applied to the video stream as opposed to the audio
stream.
Tip: Using the Audio skew correction (see note 2) in combination with
Direct Stream Copy mode for both audio and video streams allows you to
correct the audio/video sync of any pre-existing AVI file without requiring
recompression of either stream.
4. You can save a lot of time when performing Multipass encoding if you
enable audio processing only during the last pass you intend to save. It is
wasteful to process audio on every pass because the AVI file, including any
audio track, is overwritten each time.
The
22
Guide
Forward
The
23
Guide
Forward
Forward
General concepts, Frames,
General concepts
To understand the encoder features covered in the next section, we must first
Macroblocks and motion

cover the core concepts behind the MPEG-4 video standard on which DivX 5.2 is
based. Let’s begin with the way videos are composed.
Frames
At its most basic a
video is a series of
pictures shown one
after the other in
quick succession.
When the pictures
are played fast Frame 1 Frame 2 Frame 3
enough the image
appears to move. Each picture in a video is called a frame and the
speed at which they are shown is the frame rate, given in frames per
second, or fps.
If we were to slow a video down to such an extent that it were possible

to see individual frames it would become evident there is generally little
change in the picture between adjacent frames. Frames in sequence
would appear very similar to each other, objects in each consecutive
frame moving position only slightly. DivX takes advantage of this
behavior as a means to compress the data required to store a video.

The encoder divides a frame into a grid of blocks, called macroblocks,
and attempts to track the motion of each macroblock in the current
frame back to a matching area of picture in the previous frame. This
process is known as the motion search.
Frame 1 Frame 2 Frame 3
The
24
Guide
Forward
When the encoder finds a matching area for a block
in the previous frame its position is recorded by use
of a vector. A vector is simply a numerical
representation of direction and magnitude. For
example the vector (4, -7) might mean right 4 units,
down 7 units.
Vector (4,-7)
Because these vectors happen to represent the
movement of a macroblock they are called motion vectors. In the
illustrations the red arrows represent the motion vectors.
Storing the motion vector of each block from frame-to-frame allows

DivX to recreate much of each new frame from the last using far less
data than if each block were stored as a complete image. The process
is similar to cutting up one frame into little pieces, then maneuvering
them like a make-shift jigsaw puzzle using motion vectors in order to
recreate the next frame. Of course the resulting picture is not perfect
and DivX has to store some additional data to correct for the difference
(known as the residual). Even so, recreating the picture using motion
vectors and residual consumes far less storage than recording the
entire image for every single frame.
The process of tracking blocks back to matching areas in the previous

frame (known as the reference frame) is called the motion search, or
motion estimation. The frame recreated from a reference frame by
application of motion estimation is known as the compensated frame.
You can see the motion search

in action using the encoder
feedback window.
While encoding set Show picture to

Compensated frame and turn on
Show Motion Vectors.
The white lines overlaid on the

compensated frame represent motion
vectors returned by the motion search,
tracking each block back to a
matching area in the reference frame.
Where blocks are gray in the compensated frame, the motion search
failed to find a suitable match in the reference frame. These blocks will
be encoded as image data instead of motion vectors.
The
25
Guide
Forward
Intra-frames and Predicted-frames
Intra-frames and Predicted-frames
The DivX encoder uses three frame: Intra-frames, Predicted-frames and
Bi-directional-frames, commonly known as I, P and B-frames. B-frames
are discussed later in the guide as an advanced topic.
Intra-frames store all macroblocks as images rather than as motion

vectors. As noted previously, recording the image of a
block is the most expensive method in terms of
storage and hence I-frames are the most expensive
type of frame. Fewer I-frames in the video generally
translates to better compression and I-frames are
normally used by the encoder only when too few
blocks can be tracked from the reference frame by An intra-frame
the motion search algorithm.
I-frames serve a very important purpose: All the blocks in an I-frame are
stored as images, thus decoding an I frame reveals a complete picture
without dependency on reference frames. Therefore, I-frames are also
known as key-frames, and they are the only type of frame completely
independent of all others.
A predicted frame can contain forward-predicted blocks—blocks that

have been predicted via a motion vector from the previous frame by the
motion search. It is normally unlikely that all of the blocks in a P-frame
can be predicted, and where blocks can not be tracked from the
previous frame an intra-block is used in its place, similar to those found
in an I-frame. Because P-frames reconstruct much of the frame by
applying motion vectors to the previous frame (motion compensation)
they are far less expensive in terms of storage than I-frames.
One or more P-frames may follow an I-frame:
Frame 1 (I) Frame 2 (P) Frame 3 (P)
A greater ratio of P-frames to I-frames leads to a higher compression

ratio.
The
26
Guide
Forward
Quantizers
Quantizers
DivX uses a technique called quantization to control the accuracy of
the image data it stores. Quantizers are similar to the denominator in
an expression such as:
Data
Result = Quantizer
… and the co-efficient in the inverse quantization:
Data = Quantizer * Result
By maintaining some fixed Quantizer value (greater than one), it is

possible to reduce the value of Data using the first expression, and
then invert the process using the second expression.
However, after quantization the Result is normally rounded down to the

nearest whole number. This has some interesting consequences for
quantization with regards to image quality.
Examine the following two tables. The first quantizes a series of

number with a fixed quantizer of 3. The second quantizes the same
series of numbers but with a fixed quantizer of 5. Both tables round
down all of the results to the nearest whole number then perform the
inverse quantization and record the error from the original data series.
Quantizer = 3
Data series 7 14 21 28 35 42 49
Quantization data 2.33 4.66 7 9.33 11.66 14 16.33
Round-down result 2 4 7 9 11 14 16
Inverse quantization 6 12 21 27 33 42 48
Error 14% 14% 0% 4% 6% 0% 2%
Quantizer = 5
Data series 7 14 21 28 35 42 49
Quantization data 1.4 2.8 4.2 5.6 7 8.4 9.8
Round-down result 1 2 4 5 7 8 9
Inverse quantization 5 10 20 25 35 40 45
Error 29% 29% 5% 11% 0% 5% 8%
The
27
Guide
Forward
Quantizers
It soon becomes clear that higher 50
quantizers introduce larger errors 45
due to rounding before inverse 40

35
quantization; thus when encoding 30

25
lower frame quantizers lead to 20

15
higher quality images. 10
5
Plotting results from the tables Or i ginal Q=3 Q=5
with respect to the original series

illustrates lower quantizers recreating a more accurate signal.
Although lower quantizers give higher quality images they also produce
larger file sizes. This is because the range of results encompassed by
lower quantizers is larger than that of higher quantizers.
In the tables, results of data quantized at Q=3 could potentially include

any value of interval 3 (0, 3, 6, 9…). Data quantized at Q=5 has a
larger interval (0, 5, 10, 15…). Within a fixed range, e.g. 1 - 100, there
are fewer possible values for Q=5 than for Q=3 results. This leads to
better compression for Q=5, and higher quantizers in general.
Under most modes the DivX encoder will fully manage control of the
quantizer for you, although it is possible to specify a quantizer for
encoding if you want to.
The
28
Guide
A key to the guide
Forward
A key to the guide
From this point forward the guide will describe the features available in the DivX
encoder. Throughout the rest of the guide an easy to follow iconic key will be
used.
Features available only in the DivX Pro encoder will be marked
The following color-coded sections will appear where appropriate to a feature:
This indicates a feature is not compatible with the DivX Certified Program.
Use of the feature will mean your video will not be compatible with DivX
Certified devices.
This indicates that use of the feature may have implications for DivX Certified
devices.
This indicates there are performance tips associated with the feature.
This indicates that use of the feature may have implications upon other
options.
This indicates that there are usage notes associated with the feature.
The
29
Guide
Forward
A key to the guide
This indicates there are CLI parameters associated with the feature.
Additionally, at the beginning of each feature description you will see the Quick
Guide bar. QUICK GUIDE
Here the Quick Guide bar indicates that there are performance tips and usage
notes associated with the feature. All other icons are grayed out.
The
30
Guide
Bitrate mode
The
31
Guide
Bitrate mode
Bitrate mode
What is the bitrate?
In a computer the minimum unit of storage is one bit. A bit can represent two
values, 0 or 1, and hence computers use the binary number system (base 2) as
opposed to the denary system (base 10, 0-9) that humans use. When several
bits are combined they can be used to store more complex data, similar to
adding more digits in denary in order to store larger numbers.
For practical purposes the smallest unit of storage software normally works with
is actually one byte. A byte is composed of eight bits.
To avoid confusion when working with quantities of bits and bytes the following
tables should be adhered to:
Bits Bytes
One bit = Smallest unit One byte = 8 bits
One kilobit = 1,000 bits One kilobyte = 1,024 bytes
One megabit = 1,000 kilobits One megabyte = 1,024 kilobytes
or 1,000,000 bits or 1,048,576 bytes
or 8,388,608 bits
The bitrate describes how many thousands of bits per second on average the
encoder should aim to spend when encoding the video. Given a video of any
fixed duration encoding at a higher bitrate will lead to larger file sizes (and better
quality video), while conversely encoding at a lower bitrate will lead to smaller file
sizes (but lower quality video).
A key aim when encoding is to achieve a desired file size; after all the prime
reason for video compression is to reduce storage requirements. Given the
length of a video and a target file size it is possible to determine a suitable bitrate
in just five simple calculations.
The following example demonstrates step by step the manual bitrate calculation
for encoding 60 minutes of video with 128 kbps MP3 audio to fill a 700MB CD-R.
The algorithm is applicable to any video or target file size.
The
32
Guide
Bitrate mode
What is the bitrate? - Bitrate calculation
Bitrate calculation
60 minute video with 128 Kbps MP3 audio for 700MB CD-R
1. Calculate how many seconds long the video is

60 The number of minutes
x 60 The number of seconds in one minute
= 3,600 seconds duration
2. Calculate how many bits in total there are available

700 The number of megabytes
x 1,048,576 The number of bytes in one megabyte
x 8 The number of bits in one byte
= 5,872,025,600 bits available
3. Calculate how many bits will be consumed by audio

128 The number of kilobits per second for audio
x 1,000 The number of bits in a kilobit
x 3,600 The duration of the video in seconds from step 1
= 460,800,000 bits consumed by audio
4. Calculate the bits remaining for video

5,872,025,600 The number of bits available from step 2
- 460,800,000 The number of bits consumed by audio from step 3
= 5,411,225,600 bits remaining for video
5. Calculate the video bitrate

5,411,225,600 The number of bits remaining for video
/ 1,000 The number of bits in a kilobit
/ 3,600 The duration of the video in seconds from step 1
= 1,503 kbps video bitrate
Note: Considering the AVI container itself requires some bits we might use
a slightly lower bitrate to ensure hitting our target, e.g. 1,501 kbps
The
33
Guide
Bitrate mode
The bitrate calculated is the average bitrate for the video.
DivX may choose to vary the actual bitrate throughout the
video as it encodes, a technique known as variable
bitrate encoding.
Since not all sequences in a video are equally complex in Average bitrate
terms of image and motion, it is impossible to maintain a Actual bitrate
constant bitrate throughout the file without introducing

substantial quality differences between frames. Instead, DivX distributes the total
The bitrate calculator

bandwidth as appropriate with the goal of maintaining a consistent overall quality
that permits it to meet the average bitrate. It does this via the rate control, which
will be introduced shortly when multipass mode is covered in more detail.
The bitrate calculator
A bitrate calculator is included in the encoder GUI,

allowing you to enter a few simple statistics, similar to
those covered in the previous section, in order to
calculate your video bitrate.
Using the bitrate calculator is easy:
1. Enter your video duration. You can enter the

total play time of the video in hours, minutes and
seconds, or enter the number of frames and the
associated frame rate.
Tip: You can also enter the duration in seconds

only, by typing the total number of seconds into the secs box. The bitrate
calculator will work out the hours and minutes for you.
2. Select your target file size. Specify your target

either in megabytes, or the number of CD-R’s/
DVD-R’s you want to fill.
3. Select your audio bitrate. Specify either the

audio bitrate, or the file size, in kilobytes, of a
pre-prepared audio file you will be using.
4. Click Accept. The calculated bitrate will be set

automatically in the encoder configuration.
The
34
Guide
1-Pass
Bitrate mode
1-Pass
QUICK GUIDE
In 1-Pass mode the encoder will output a
working DivX video stream as it receives the
source video.
This is both the easiest and fastest method of encoding

DivX video while targeting an average bitrate, and is
particularly suited to video capture applications where it is
impossible to recreate the source video identically more
than once. However, 1-Pass mode is the least consistent
of all available modes in terms of video quality.
After running 1-Pass encoding your DivX file can be

viewed immediately.
During 1-Pass encoding it is

possible to modify the bitrate
in real-time by selecting the bitrate
option from the bottom panel of the
encoder feedback window.
This feature is likely to be of

particular interest when capturing
from a live source. For example, if
capturing from television you might
want to reduce the bitrate during
advertisements.
When you set the bitrate mode to 1-Pass you will not be able to use the
Bitrate modulation control or the EKG application.
The
35
Guide
Bitrate mode
1-Pass
1-Pass mode is particularly useful when you are capturing from a live source
and you desire control over the bitrate/file size.
When capturing from a live source and attempting to maintain a consistent

quality throughout the video you should use 1-Pass Quality Based mode.
When working from static sources (for example re-processing existing stored
video) you will achieve more consistent quality by using Multipass mode.
Bitrate mode is one of:
-bv1 <bitrate> 1-Pass

-b1q <quantizer> 1-Pass, quality-based
-bvn1 <bitrate> Multipass, 1st pass
-bvnn <bitrate> Multipass, nth pass
Bitrate is an integer value between 0 and 16,000 specifying the number of

kilobits per second.
Quantizer is floating point value between 1.0 and 31.0 specifying the fixed
quantizer to be used.
The
36
Guide
Multipass
Bitrate mode
Multipass
QUICK GUIDE
In Multipass mode you are required to run the
video through the encoder two or more times.
On every successive pass the video must be
identical to that which you used in your first pass.
As discussed in Bitrate mode—What is the bitrate?, DivX uses a rate control

algorithm to vary the bitrate throughout the video to achieve a consistent quality
while targeting the specified average bitrate. To do this effectively DivX has to be
able to examine the entire video in order to determine where bandwidth should
be allocated before it spends any bits at all.
Multipass, 1st pass

In 1-Pass mode the rate control strategy is not
optimal because the encoder has no foresight
into the content of the video. Therefore, to meet
the specified bitrate the rate control must make
extreme changes to the bitrate based on the
short term history as video is received.
Multipass mode is designed to avoid this

extremism by allowing the encoder to first
analyze the entire video before encoding it. This
analysis is performed in Multipass 1st Pass
mode, and during the first pass no actual video output is written (if you
attempt to play your file after running a 1st pass you will see no video).
In Multipass 1st pass mode the encoder writes its

analysis to the log file specified in the Multipass
encoding options area of the configuration dialogue.
Lets take a brief look at the log contents and what they mean:
##map version 8
nframes 5837
timescale 30000
passes 1
seq deltaT type total_bits motion_complexity texture_complexity modulation
0 0 I 7712 0.000000 0.251141 1.000000
2 2400 P 2952 0.035268 0.011100 1.000000
1 -1200 B 1144 0.015206 0.000000 1.000000
4 3600 P 6984 0.067981 0.069732 1.000000
3 -1200 B 2216 0.029374 0.000000 1.000000
6 3600 P 8160 0.079010 0.082583 1.000000
5 -1200 B 2264 0.029773 0.000745 1.000000
8 3600 P 14176 0.083506 0.294013 1.000000
The
37
Guide
Bitrate mode
Multipass
Multipass Log file
Log header
##map version <Value>
Value is the version number describing the format of the log file.
nframes <Value>
Value is the total number of frames listed in the log file.
Frame numbering begins from zero.
timescale <Value>
Value is some arbitrary number representing the number of intervals
in one second of video.
The timescale is used in combination with deltaT (see below).
passes <Value>
Value is the number of passes performed when the log was written.
Frame list
seq <Value>
Value is the sequence number of each frame.
The sequence numbers may appear to be out of order but in fact this
is because they are in display order. For example, in the section of
log shown the sequence reads 0,2,1 for a set of I,P,B frames
respectively. The order in which these frames must actually be
displayed is I,B,P—or 0,1,2 respectively.
deltaT <Value>
Value is the change in time (based upon the timescale value in the
header) between frames.
In the section of log shown the first three deltaT values are 0,2400,-
1200. Recalling that frames are listed in decoding order this actually
means the sequence is 0,1200,2400. With respect to the timescale
value (30000 units) this means the interval between frames (1200
units) is (1200/30000) = 0.04 seconds, and we can derive from this
frame-rate is 1/0.04=25 frames per second.
The
38
Guide
Bitrate mode
Multipass
type <Frame_type>
Frame_type is the type of each frame and can be one of ‘I’,’P’ or ‘B’,
representing intra-frame, predicted frame or bi-directional frame
respectively.
total_bits <Value>
Value is the total number of bits used by the encoder to encode the
frame.
motion_complexity <Value>
Value is a floating point number proportional to the number of bits
required to encode all motion data in the frame.
This will be zero only for intra-frames.
texture_complexity <Value>
Value is a floating point number proportional to the number of bits
required to encode the texture in the frame.
For intra-frames the value is associated with the bits required to

encode the image. For predicted and bi-directional frames the value
is associated with the bits required to encode the residual (difference
between source and motion-compensated frames).
modulation <Value>
Value is a floating point number representing the modulation value
set by the EKG application between passes.
Modulation acts as a co-efficient to the frame quantizer selected by

the rate control algorithm. By raising the modulation value for a frame
above one the quantizer will be increased resulting in lower quality
for that frame. Conversely, lowering the value below one will cause
the quantizer to be reduced thus raising the quality for the frame.
As an example, if the rate control were to determine quantizer 2

should be used but the modulation for the frame was 1.5 the result
would be a quantizer of 2 x 1.5 = 3.
Modulation should not be confused with the Bitrate modulation option

in the encoder configuration dialogue which is not directly related.
The
39
Guide
Bitrate mode
Multipass
Because DivX Pro includes the EKG tool for graphically displaying and
manipulating the contents of the log file it is rarely necessary to edit it
manually.
Multipass, nth pass

Once first pass has been completed the same
video must be encoded one or more times
using Multipass, nth pass mode. DivX video is
encoded and written to your DivX video file on
every nth pass.
In nth pass the encoder will read the log file

created by the previous pass and use the
analysis to form a more optimal rate control
strategy. This will lead to better bandwidth
distribution throughout the file, obtaining a
consistent quality while meeting the average
specified bitrate.
The previous pass need not necessarily be a

1st pass. During each nth pass results of the new optimized rate control
strategy can again be written back into the log file and used to further
refine the rate control strategy for a successive nth pass.
The log file will be updated during an nth pass if you have
selected the option to Update log file in the Multipass
encoding files area of the encoder configuration dialogue.
Although it is possible to run as many nth passes as you like, 98-99% of

the optimal quality is typically obtained in three passes or less (one 1st
pass followed by two nth passes).
Note that if you do not select to update the log file during nth pass then
running successive passes will not improve the consistency of video
quality because the rate control strategy will not be refined.
Bitrate modulation
By using the Bitrate modulation slider it is
possible to bias the distribution of bandwidth
towards frames with either low or high motion,
allowing you to adapt the rate control to better suit the particular type of
video being encoded.
The
40
Guide
Bitrate mode
Multipass
Moving the slider left of center, or
lowering the modulation value, biases
bandwidth in favor of high-motion frames
and consequently away from low-motion
frames. Moving the slider right of center, Motion complexity
or raising the modulation value, biases Unmodulated bitrate
Bitrate modulated in favor of high-motion
bandwidth in favor of low-motion frames Bitrate modulated in favor of low-motion
and consequently away from high-motion
frames.
When determining the most appropriate bitrate modulation it is

important to remember that quality is subjective. The optimal bitrate
modulation will vary from video to video and also upon the preference of
the viewer and their personal sensitivity to different types of artifacting.
Let us consider some examples where it is beneficial to use the bitrate

modulation control.
1. Consider the encoding of an action movie containing several

explosion scenes or other fast-paced action and imagine the
quality tending to degrade where there is significant motion in the
scene. Moving the bitrate modulation slider left of center towards
high motion will improve the quality consistency - the degree of
adjustment relative to the level of correction required.
2. Once again taking an action movie as an example, imagine

encoding at a very limited bitrate in order to create video suitable
for Internet distribution. It might be desirable to intentionally bias
bandwidth away from fast-motion scenes. Explosion scenes and
other short fast-motion sequences may consume so much of the
total available bandwidth that they actually starve the encoder of
free bits as it encodes the majority of the video, leading to low-
motion sequences that appear disproportionately degraded.
Biasing bandwidth towards low-motion will cause the fast-motion
scenes to appear at slightly lower quality, but the quality in the
remainder of the video may appear much improved.
3. Consider a scenic video in which the camera is panning across a

landscape and zooming in and out but pausing for periods to show
features in detail. In this situation we might like to bias bandwidth
so that low-motion scenes receive more bits than high-motion
scenes. In this way we could enforce that the highest quality is
afforded to scenes in the video where the camera focuses on
detailed features in the landscape.
The
41
Guide
Bitrate mode
Multipass
Fastest Performance/Quality mode is unavailable during Multipass encoding
because it does not perform the motion estimation necessary for the rate
control to work effectively.
If you intend to use Slow Performance/Quality mode you can substantially

decrease your encoding time by encoding all but the final pass using
Standard Performance/Quality.
Running a Standard pass followed by a Slow pass in place of two Slow

passes reduces encoding time by approximately 50%, while impacting so
negligibly on quality as to be virtually indistinguishable.
Disable audio processing on all but the last nth pass to reduce encoding time.
Use Multipass mode to obtain the optimal quality consistency when it is

possible to pass an identical source video through the encoder multiple
times.
When performing more than 2 passes (1st, nth) in order to be effective in

refining the rate control (and hence quality consistency) the option to Update
log file must be enabled during each nth pass.
Generally 98-99% of the optimal quality is obtained in three passes or less

(1st, nth, nth).
It is possible to use a 1-Pass, Quality-based pass in place of Multipass 1st

pass if you desire to perform the first pass based upon a fixed frame
quantizer. Ensure that Write log file is enabled. The XVID encoder performs
1st pass at Q=2. Using this technique it is also possible to capture a live
source via 1-Pass, Quality-based mode and then recompress it, proceeding
directly to nth pass.
The
42
Guide
Bitrate mode
Multipass


Log file:
-log <filename>
Filename is the log file name, fully qualified with a pre-existing path and
enclosed in parentheses. The default is “C:\DivX.log”.
The
43
Guide
1-Pass, Quality-based
Bitrate mode
QUICK GUIDE
In 1-Pass quality-based mode the encoder will
output a working DivX video stream as it
receives the source video, similar to in 1-Pass
mode except that you specify the quality rather than the
target bitrate.
This mode is only accessible when profiles have been

disabled from the Select Profile Wizard.
Quality is set by adjusting the Quantizer control. As

described in Forward—Quantizers, the quantizer is the
method DivX uses to control the
accuracy of the image data it
stores. Lower quantizers relate to
higher quality, higher quantizers to
lower quality.
By fixing the quantizer you guarantee a consistent quality throughout your entire
video file by ensuring the encoder stores the image data for each frame with a
consistent accuracy. However, when you fix the quantizer it is not possible to set
or predict the bitrate and hence file size because the encoder will always spend
as many bits as are required to encode each frame at the given quantizer.
Quality-based mode is ideal when you need to be assured of a fixed quality

regardless of file size, and particularly when capturing from live or analogue
sources that you intend to later re-compress.
Note that even when the quantizer is 1 and the display reads 100% quality the
encoded video will not be identical to the source. DivX is a lossy codec, and
there will always be some degradation from the source video regardless of the
combination of settings that you choose.
After running a 1-Pass quality-based encoding your DivX file can be viewed
immediately.
The
44
Guide
Bitrate mode
1-Pass, Quality-based mode is not explicitly unsupported by DivX Certified
devices, however the Video Buffer Verifier is disabled in this mode and there
is no guarantee that the video stream produced will not exceed the
capabilities of the certified device.
When 1-Pass, Quality-based mode is enabled any Video Buffer Verifier CLI
parameters are disregarded.
You cannot set a bitrate for 1-Pass, Quality-based mode.
Use 1-Pass, Quality-based mode when capturing from a live source and
attempting to maintain a consistent quality throughout the video without
consideration of the resulting file size.
It is impossible to control or predict the bitrate or file size under 1-Pass,

Quality-based mode, other than to know that encoding the same source at
higher quantizer will produce smaller files and vice versa.
1-Pass, Quality-based mode is particularly useful for capture when you intend
to later re-compress a video using Multipass mode. In fact, 1-Pass, Quality-
based mode can be used in place of Multipass, 1st pass mode so long as
Write log file is enabled. In this way it is possible to capture in 1-Pass,
Quality-based mode and proceed directly to Multipass, nth pass. When
making use of this technique take care to avoid overwriting your 1st pass file
containing the captured video with the Multi-pass, nth pass output
accidentally.
Under the same principle, 1-Pass, Quality-based mode allows you to perform
XVID-like Multipass encoding. XVID performs its first pass with Q=2.
The
45
Guide
Bitrate mode


Log file:
-log <filename>
Filename is the log file name, fully qualified with a pre-existing path and
enclosed in parentheses. The default is “C:\DivX.log”.
The
46
Guide
Performance/
Quality
The
47
Guide
Performance/Quality
Performance/Quality
Fastest
Fastest
QUICK GUIDE
The fastest performance/quality mode causes
the encoder to perform no motion search when
encoding. Because no motion estimation is done
all blocks will either be intra-blocks or
predicted with a null motion vector. In this
respect the encoded video will resemble an
MJPEG sequence.
When Fastest performance/quality mode is selected the encoder does not

respect the following options: Quarter-pixel, Global motion compensation, Bi-
directional encoding, Psychovisual enhancement, Scene change threshold.
The Maximum Keyframe Interval is limited to 60 frames when Fastest mode

is enabled. Entering any interval larger than 60 frames will cause the encoder
to default the Maximum Keyframe Interval to 10 frames.
Because fastest mode saves mainly intra-blocks it can consume an

excessive number of bits. Generally this means 1-Pass mode is unsuitable
when working with fastest mode and you should instead use only 1-Pass
quality-based mode, fixing the quantizer much the same as you might set the
compression factor when saving a JPEG image in a paint program.
If you choose to use Fastest mode in combination with 1-Pass mode the
encoder may fail to meet the average bitrate unless a very high average
bitrate is specified.
Fastest mode is unsuitable for use during multipass encoding because the
rate control algorithm used by the multipass system relies on accurate motion
complexity statistics to correctly distribute bandwidth throughout the video.
Therefore, multipass does not support Fastest mode.
Fastest mode is ideal for capture situations where encoding rate is critical
and bitrate is not a major concern - for example when you intend to later re-
compress the video using another mode.
The
48
Guide
Performance/Quality
Fastest
Performance/Quality:
-pq <mode>
Mode is one of:

1 Fastest
3 Fast
5 Standard
64 Slow
The
49
Guide
Fast
Performance/Quality
Fast
QUICK GUIDE
The Fast performance/quality mode causes the
encoder to perform a basic, performance-
optimized motion search when encoding.
Although the motion search implementation

is less complete than that of Standard
mode, it is a significant improvement on the
bit consumption of Fastest mode with
minimal impact on encoding rate.
Fast mode is ideal for capture situations where encoding rate is critical but
bitrate is also a concern. Capturing in Fast mode with a high bitrate will likely
produce higher quality output than capturing in Fastest mode, but raises the
performance requirements placed upon the CPU.
-pq <mode>
Mode is one of:

1 Fastest
3 Fast
5 Standard
64 Slow
The
50
Guide
Standard
Performance/Quality
Standard
QUICK GUIDE
The Standard performance/quality mode
enables the motion search algorithm and is
functionally equivalent to Slowest mode in
previous versions of DivX 5, but contains
algorithm enhancements that improve video
quality.
Standard mode respects the Maximum Keyframe Interval and Scene Change
Threshold settings (described later). Using these two settings, it is possible to
control the encoder’s decision process for frame-type selection with respect
to intra-frames.
Standard mode is the only mode supporting motion estimation compatible

with the QPel option.
-pq <mode>
Mode is one of:

1 Fastest
3 Fast
5 Standard
64 Slow
The
51
Guide
The Rate-Distortion Algorithm
Performance/Quality
The Rate-Distortion Algorithm
First introduced in DivX 5.1, the Rate-Distortion algorithm is enabled when the
Performance/Quality slider is set to Slow. The rate-distortion algorithm allows the
outcome of various decisions made by the encoder to be evaluated intelligently
with respect to bit spend against quality gain where previously simple algorithms
would govern the process.
Every decision made as a video is encoded has an impact on both the rate and
the distortion. The rate is a term used to describe the bits spent, distortion to
describe the change in some measure of quality based upon any particular
decision.
Decisions evaluated by the rate-distortion algorithm can include the coding of

blocks (intra-coded, predicted, or skipped), or which of several possible motion
vectors for a block yields the best balance of bits spent against quality.
Referencing the illustration opposite,

consider any decision the encoder might
make and imagine three possible
outcomes—A, B and C.
Quality
Under normal circumstances the encoder

will make the best choice for quality. In the
illustration this is decision C. However, the Decision A
Decision B
rate-distortion algorithm might choose Decision C
decision B instead since it offers
comparable quality at a reduced bit Bits
spend.
Notice that although decision B leads to

marginally lower than optimum quality
for this particular decision, overall
quality will rise because the encoder has
more free bits to spend.
Quality
The rate-distortion algorithm evaluates

Decision A
decisions based on a rate-distortion
Decision B curve, describing the change in
Decision C
distortion necessary for a rise in bit
spend to be considered a good decision.
Bits
The
52
Guide
Performance/Quality
The Rate-Distortion algorithm
Here points above the rate-distortion curve evaluate as good decisions, those
below the curve as bad decisions. The best decisions fall towards the upper-left
of the plot (higher quality and lower bit spend).
It follows that the encoder must derive this rate-distortion curve from somewhere.
The curve is approximated by some function inside the encoder accepting a
value that changes the shape of the curve. This value is Lambda, and you can
manipulate it via the Feedback window.
Begin encoding a video using Slow

performance/quality mode. From the
feedback window set Show picture to
Encoded frame and enable the Lambda
control in the bottom toolbar.
Pause encoding at any suitable frame and

vary the Lambda control, observing the
effect on image quality and the number of
bits spent for the frame. By altering Lambda
you are manipulating the shape of the rate-
distortion curve.
Increasing Lambda (dragging the slider

to the left) causes a sharper rise in the
rate-distortion curve, meaning larger
improvements are demanded for an
Quality
increase in bit spend.
Decreasing Lambda (dragging the slider

to the right) causes the rate-distortion Higher Lambda
curve to rise more slowly, meaning fewer Lower Lambda
improvements are acceptable for an
increase in bit spend. Bits
If the rate-distortion curve becomes too steep quality will be degraded because
the encoder will not be able to achieve the quality demanded of it for a given rise
in bit spend. If the rate-distortion curve becomes too flat quality will be degraded
because the encoder will waste a lot of bits for very marginal improvements,
reducing the number of free bits available overall.
Deriving the optimal value for Lambda is a very difficult process and under most
circumstances you should not over-ride the encoder default. The encoder will
normally vary Lambda automatically during encoding depending on
circumstances, over-riding instead fixes Lambda for the duration of one pass.
Lambda returns to its default mode of operation at the beginning of each pass.
The
53
Guide
Slow
Performance/Quality
Slow
QUICK GUIDE
Slow performance/quality mode enables the
rate-distortion algorithm, designed to
dramatically improve the video quality of low-
bitrate encoding.
The Rate-Distortion algorithm attempts to

best balance bit spend against quality gain by
intelligently evaluating all of the possible
outcomes of various decisions made by the encoder.
In Slow mode the encoder uses the new Rate-Distortion algorithm to make
frame type decisions based upon the best balance of bit spend and quality
gain. Because of this the Scene Change Threshold settings become
redundant and is ignored.
QPel is unsupported in Slow mode. If you have enabled both QPel and Slow
mode the performance/quality mode will automatically default to Standard
mode when encoding begins.
-pq <mode>
Mode is one of:

1 Fastest
3 Fast
5 Standard
64 Slow
The
54
Guide
Psychovisual
Enhancement
The
55
Guide
Psychovisual Enhancement
What is Psychovisual Enhancement?
What is Psychovisual Enhancement?
DivX is a lossy codec meaning after encoding the video will have been degraded
to some extent. In other words it will have lost some detail.
The DCT, iDCT and Human Visual System

DivX, like other lossy codecs (e.g. MP3, Vorbis etc.), use techniques based on
human perception and intelligently lose detail where it is least likely to be noticed
before degrading the remainder of the picture. This perceptual technique forms
the basis of psychovisual enhancement, letting DivX mask the artifacts inherent
in MPEG-4 video in areas of the picture where error is least perceivable.
To understand how psychovisual enhancement works, we must first explan the

method DivX uses to encode images. It is a complex subject and we will cover it
only briefly.
The DCT, iDCT and Human Visual System

Inside a computers memory, images are normally represented as a two-
dimensional array of pixels - the little colored squares that make up a complete
picture.
Storing the color value of every pixel is very expensive in terms of bits and thus
DivX uses a technique known as the Discrete Cosine Transformation to convert
this series of values into frequency information. The process of the DCT is an
advanced mathematical topic that will not be covered here. However, the main
feature of the DCT result is a set of co-efficients that represent the magnitude of
the frequencies composing the input series in
order of ascending frequency.
The
56
Guide
The DCT, iDCT and Human Visual System,
The human visual system is far less sensitive to high frequencies in an image
than it is to low frequencies. One perceptual technique DivX uses during lossy
compression is the reduction in accuracy of higher frequency co-efficients, saving
bits while causing the least perceivable quality degradation.
It is in fact these DCT co-efficients that are quantized by DivX when image data
is encoded (see Forward—Quantizers). Greater quantization of the DCT co-
efficients means fewer bits are ultimately required to store them, but a less
accurate image results when the inverse discrete cosine transformation (or iDCT)
is performed during decoding—the process of restoring the original series (image
data) from the DCT result. This error between source and encoded image is
known as quantization noise.
The psychovisual enhancement system

As described, image data is stored by DivX as a set of co-efficients linked to
different frequencies in the image. The psychovisual enhancement process
actually manipulates these co-efficients to produce two distinct effects:
1. In flat areas of the image co-efficients are manipulated so that fine details
are enhanced. If we were to encode in 1-Pass Quality Based mode (i.e. at
a fixed quantizer), this enhancement would naturally increase the bits
spent on flat areas of the image. However, at a fixed average, bitrate the
effect is actually that textured areas of the image will receive fewer bits
and hence when psycho-visual enhancements are enabled, artifacts will
be masked in textured areas of the image where they are least visible.
This works because of a balancing effect Original bits

Bits after PV
that is created - the encoder has a finite

number of bits to spend and if flat areas
are to consume more bits then textured
areas must consume fewer bits. Flat Textured
2. In strongly textured areas of the image, co-efficients are manipulated so

that fewer bits are spent and thus quality is slightly degraded. If we were to
encode in 1-Pass Quality Based mode (i.e. at a fixed quantizer), this
enhancement would naturally decrease the bits spent on heavily textured
areas of the image. However, at a fixed average bitrate the effect is
actually that flatter areas of the image will receive more bits (once again
due to balancing) and hence when psychovisual enhancements are
enabled, artifacts will be masked in textured areas of the image where
they are least visible.
The
57
Guide
Because both psychovisual enhancement methods mask artifacts in textured
areas where they are least visible, the video quality appears to improve when
psychovisual enhancements are enabled. You can visualize the concept by
imagining two characters having a conversation while standing in front of a tree—
the psychovisual enhancement process might enhance details in the characters
faces at the expense of causing some artifacting among the leaves of the tree
where it would be least perceivable.
Experiencing PVE
The best way to
understand the psychovisual enhancements is to watch them working on an
image. The feedback window provides an excellent method of doing this, and
also allows you to configure the degree to which flat and texture psychovisual
Experiencing PVE
enhancements are performed.
Set up a 1-Pass encoding using VirtualDub

and any source video of your choosing with
Fast psychovisual enhancement mode
enabled. Begin encoding and click the Pause
button on the encoder feedback window
when a suitable image to work with appears.
Set Show picture to encoded frame and

enable the pvLumaFlat and pvLumaTexture
controls in the bottom panel.
With the video paused, slowly drag each slider back and forth from zero (no
psychovisual enhancement) to one (full psychovisual enhancement), watching
the image carefully as you do so. Since the aim of psychovisual enhancement is
to mask artifacts where you are least likely to perceive them, you should see that
both effects appear to improve quality when enabled to fuller extents.
By enabling the Data overlaying controls and

configuring them to show the PV level and color
the Full macroblock based also on the PV level it
is possible to view graphically the changes made
in a more statistically-orientated way.
Where information is removed from the image the number overlaid will be
positive and the macroblock tinted red. Where information is added to the image
the number overlaid will be negative and the macroblock tinted blue.
The
58
Guide
Fast psychovisual enhancement
Fast psychovisual enhancement
QUICK GUIDE
Fast psychovisual enhancement mode attempts
to manipulate the DCT co-efficients selectively
so that the resulting noise in the decoded image
is positioned where it will be least visible, for example
in areas of strong texture.
When encoding a multipass video you can save time by enabling

psychovisual enhancements from the second-last pass onwards only. This
will have a minimal impact on quality but can potentially reduce encoding
time substantially.
Fast psychovisual enhancement mode generally produces a more intense

effect than Slow psychovisual enhancement mode.
When choosing which psychovisual enhancement mode to use do not equate

speed with quality. Each psychovisual enhancement mode produces slightly
different results.
Care should be taken when applying psychovisual enhancement to noisy

sources as undesired results can occur. If your source is a particularly old
video or has been acquired via analog capture, you may need to apply
source preprocessing to reduce the level of noise present before applying
psychovisual enhancement.
Psychovisual enhancement should not be applied when encoding as

interlaced. The psychovisual enhancement system is not yet interlace-aware
and may enhance combing artifacts undesirably.
Psychovisual enhancements:
-psy <mode>
Mode is one of:

0 Disable psychovisual enhancement
1 Slow psychovisual enhancement
2 Fast psychovisual enhancement
The
59
Guide
Slow psychovisual enhancement
Slow psychovisual enhancement
QUICK GUIDE
Slow psychovisual enhancement mode analyzes
each block and the blocks surrounding it in turn
to ensure any enhancement does not introduce
significant blocking and ringing artifacts, thus Slow
mode is less likely to introduce artifacts than Fast
mode.
When encoding a multipass video you can save time by enabling

psychovisual enhancements from the second-last pass onwards only. This
will have a minimal impact on quality but can potentially reduce encoding
time substantially.
Slow psychovisual enhancement mode generally produces a less intense

effect than Fast psychovisual enhancement mode and lowers the probability
that visible artifacts will be introduced as a consequence of the process.
When choosing which psychovisual enhancement mode to use do not equate

speed with quality. Each psychovisual enhancement mode produces slightly
different results.
Care should be taken when applying psychovisual enhancement to noisy

sources as undesired results can occur. If your source is a particularly old
video or has been acquired via analog capture, you may need to apply
source preprocessing to reduce the level of noise present before applying
psychovisual enhancement.
Psychovisual enhancement should not be applied when encoding as

interlaced. The psychovisual enhancement system is not yet interlace-aware
and may enhance combing artifacts undesirably.
Psychovisual enhancements:
-psy <mode>
Mode is one of:

0 Disable psychovisual enhancement
1 Slow psychovisual enhancement
2 Fast psychovisual enhancement
The
60
Guide
Source
preprocessing
The
61
Guide
Source Preprocessing
What is Preprocessing?
What is Preprocessing?
In the computer age, digital video can be acquired from a variety of sources. DV
When to use Preprocessing

camcorders, DVDs, TV capture cards and even web cameras have all become
so accessible at the consumer level that for DivX to consistently produce high-
quality output we must first correct video issues attributable to particular sources.
Preprocessing allows the encoder to clean phenomena such as noise, grain,

flickering and color inconsistency from the source video prior to encoding it. An
immediate benefit of preprocessing is an increase in the viewers perception of
quality. With respect to encoding, a reduction in noise and general inconsistency
in the texture from frame-to-frame yields a lower bit-spend when encoding the
image because the DCT is less complex and the motion search is more
successful.
When to use Preprocessing

Preprocessing is most effective when applied to video taken from analog sources
- typical examples being VHS tape, TV capture, and camcorder footage. All of
these sources are likely to suffer from noise—high-frequency variations around
the true color of each pixel. Noise causes the
image to appear fuzzy or speckled.
Digital sources are also susceptible to noise.

DV camcorders and web cameras often amplify
the signal arriving at their image sensor in low-
light conditions, introducing a lot of noise to the
picture. Web-cameras are also notable for an
increase in noise proportional to the resolution, frame-rate and lack of strong
lighting. Even DVD video re-mastered from film can show the fine grain of the
original film media. This particular effect is a special case as it is sometimes
desirable in the encoded video, adding warmth to the picture. DivX includes a
special decoder feature allowing the artificial application of grain to any video,
even when the original grain was removed during encoding.
The
62
Guide
Spatial filtering
Spatial filtering
Pre-processing consists of two filter types, the first is a spatial filter. The spatial
filter is concerned with each pixel in the frame and the pixels surrounding it.
Noise manifests itself as high frequency changes between the colors of adjacent
Temporal filtering
pixels. You can stabilize the colors by applying a low-pass filter (one that rejects
high frequencies) to an image.
Low-pass spatial filtering
Noisy image Sampled data After low-pass filter Filtered image
Temporal filtering
The second type of filter employed by the pre-processing algorithm is a temporal
filter. While the spatial filter can reduce noise within a single frame it does not
consider the changes in each pixel over time, i.e. from frame-to-frame. A good
example of temporal artifacting is flickering video, where even though there may
be little noise within each individual frame, there is temporal noise affecting the
intensity of pixels over time. Temporal noise is also responsible for producing
grain or discoloration patterns that change throughout a video as it is played.
Just as it is possible to perform low-pass filtering within a frame by examining

adjacent pixels, it is also possible to filter the color of each individual pixel by
comparing its color in several previous frames. By applying an adaptive low-pass
filter to co-located pixels throughout time any temporal noise related to lighting or
color tonality between frames can be reduced.
Temporal filtering
The
63
Guide
QUICK GUIDE
Source pre-processing applies spatial and
temporal filtering prior to encoding in order to
reduce noise in the source video.
Four presets are defined allowing control of the filter

intensity. These are light, normal, strong and extreme.
Both light and normal presets are very carefully balanced

and neither should cause visible degradation to the video.
Sources with heavy noise present may require pre-

processing using the strong or extreme presets. These
presets may introduce a visible smoothing effect in the
video, therefore when using these two presets it may be
advisable to process short sample clips under each to find
the best balance between noise reduction and smoothing.
Source pre-processing is recommended when the source video contains

either visible noise, unstable color tonality or flickering.
Source pre-processing is particularly important when encoding from analogue

sources, old films, or consumer-level digital video equipment—especially
when video was filmed under poor lighting conditions.
Source pre-processing:
-pre <Strength> or -pre <Temporal>,<Spatial>,<SpatialPasses>
Strength is one of:

1 Light
2 Normal
3 Strong
4 Extreme
Temporal is the temporal level from 0.0 to 1.0
Spatial is the spatial level from 0.0 to 1.0
SpatialPasses is the number of passes using the Spatial filter from 1 to 3
The
64
Guide
Crop and
Resize
The
65
Guide
Crop and Resize
Crop and Resize

Cropping and resizing via the Encoder
Cropping and resizing via the Encoder
Cropping and resizing are two
of the most basic filters that
can be applied to video, and Cropping
yet not all video applications
provide this functionality. Both
Why crop or resize?

techniques alter the
dimensions of the video area,
Resizing
cropping by discarding areas of the video and resizing by
scaling the video along its horizontal and/or vertical axis.
The DivX encoder ensures that these two filter operations can
be performed from any application capable of compressing
DivX video regardless of whether the application itself includes
this functionality by providing its own crop and resize controls.
The encoder’s internal filters are optimized in terms of quality and performance
and may offer better performance than those in your video application.
Why crop or resize?

You can improve the quality of the encoded video and decrease encoding time
by appropriately cropping and resizing a source.
Cropping is most useful when processing video sources that feature borders
around the exterior of the picture area, a common attribute of both widescreen
DVD video and analogue capture content.
Although the borders themselves are encoded
with relative ease, where the border meets the
picture a high contrast edge is created that
consumes a large number of bits when
encoded.
The
66
Guide
Crop and Resize
Why crop or resize?
This extra bit spend means throughout the video fewer bits are available to
encode the actual picture, causing video quality to suffer. By cropping borders no
bits are wasted outside the actual picture area and thus overall quality is
improved.
Resizing is used in a variety of situations, the simplest case being where there is
a desire to reduce the video dimensions. At lower bitrates reducing video
dimensions may assist in preserving quality by reducing the number of blocks
required to make up the picture - with fewer blocks each block should receive a
greater number of bits on average.
It should be noted that there is a moving balance to be drawn between

dimensions and quality at any particular bitrate. A small reduction in video
dimension will generally mean fewer blocks and greater bit spend per block,
giving higher quality overall. However, if the video dimensions are reduced too
far then the texture and motion within each
Optimal dimensions
block can become so complex that intra-blocks
require more bits to store, and motion search
becomes less effective, Subsequently, this will
produce a greater proportion of intra-blocks,
intra-frames and lower overall quality. At lower
Video dimensions resolutions fine details and sharp edges may
Perceptual quality also become degraded.
Resizing is frequently used to scale video on its

vertical axis to correct the aspect ratio when
encoding from DVD. The aspect ratio describes the
relationship between the horizontal and vertical
dimensions of the video with respect to their
proportionality. If this ratio is not correct video will
appear to be either compressed or stretched
vertically when viewed, Original aspect ratio
leading to characters that
appear either short and fat or
tall and thin.
Higher aspect ratio Lower aspect ratio
The
67
Guide
Crop
Crop and Resize

Crop
QUICK GUIDE
Crop allows you to specify the number of pixels
from each edge of the video that will be
discarded before encoding.
The cropping area is fixed throughout the video.
Crop top
Crop right
Crop left
Crop bottom
Cropping in some applications, such as VirtualDub, can require colorspace

conversions that reduce encoding performance and minimally degrade the
video. Where crop and resize are the only filtering operations required,
greater encoding performance can be achieved using the encoder’s crop
filter.
You should crop any borders surrounding a video before encoding it to avoid
wasting bits. This will improve the overall quality of your video.
Cropping is performed before resizing by the encoder. The dimensions of the

video after both cropping and resizing have been performed must be divisible
by four. For optimal performance the dimensions should be divisible by
sixteen.
Cropping:
-c <left>,<right>,<top>,<bottom>
Left, right, top and bottom are the number of pixels from each edge of the
image that will be discarded before encoding.
The
68
Guide
Resize
Crop and Resize

Resize
Resizing scales the video dimensions on one or both axis.
QUICK GUIDE
You can select the type of sampling that is performed in

the source image to calculate the color of each pixel in the
resized image. Different sampling options create slightly different
effects in the resized image.
Bilinear sampling is best used when reducing an

image. The color of each pixel in the resized image
depends on where the pixel would fall in the source
image.
35% 65% The color is interpolated from the four surrounding

pixels based upon the theoretical position of the new
35%
pixel.
70%
New color =
Because bilinear sampling considers only 2x2 pixels
70% (65% + 35%) +
30% (65% + 35%) in the source and interpolates linearly it produces a
softened image on resizing and can cause pixelation
in enlarged images.
Bicubic sampling is more complex than bilinear

sampling, calculating the color of each pixel in the Linear Bicubic (soft)
resized image from 4x4 pixels in the source.
Instead of simply interpolating linearly between

pixels bicubic sampling weights the contribution of
each pixel in the source using a spline function
based upon the offset of each pixel from the pixel 0 1 0 1 2
being calculated.
Bicubic (normal) Bicubic (sharp)
Changing the spline parameters allows the
sharpness of the resized image to be controlled.
Three presets - soft, normal and sharp—are
available to choose from.
Bicubic sampling produces better quality over 0 1 2 0 1 2

bilinear sampling when enlarging an image but
because more samples are taken and calculations
are more complex for the bicubic method it also takes longer to compute.
Source Bilinear resample Bicubic resample
The
69
Guide
Crop and Resize
Resize
The encoder resize filters are very highly optimized. Consider using them in
place of external resize filters provided with video editing applications.
Resizing in certain applications, such as VirtualDub, can require colorspace
conversions that reduce encoding performance and minimally degrade the
video. Where crop and resize are the only filtering operations required
greater encoding performance will be seen using the encoder’s resize filter.
Cropping is performed before resizing by the encoder. The dimensions of the

video after both cropping and resizing have been performed must be divisible
by four. For optimal performance the dimensions should be divisible by
sixteen.
Bilinear sampling is faster than bicubic sampling and is suitable for reducing
video. While it is acceptable to use bilinear sampling when enlarging, better
quality can be obtained with bicubic sampling.
A common use for resizing is aspect correction when encoding from DVD.
You can easily calculate the dimensions for the correct aspect ratio given the
original video dimensions and the aspect details from the DVD.
Example:
The video dimensions are 704x360 and the DVD box reports 2.35:1
aspect.
To calculate the correct vertical resolution:
(1 / Aspect) x Horizontal dimension
Or:
(1 / (2.35 / 1)) x 704 = 299.57
Recall that dimensions must be divisible by four, and optimally by

sixteen. 704 is already divisible by sixteen, but it is necessary to round
299.57 up to 304.
Resize:
-r <horizontal>,<vertical>,<mode>
Horizontal is the horizontal dimension of the resized video in pixels.
Vertical is the vertical dimension of the resized video in pixels.
Mode is one of:

1 Bilinear (very soft)
2 Bicubic (soft)
3 Bicubic (normal)
4 Bicubic (sharp)
The
70
Guide
MPEG-4 Tools
The
71
Guide
Bi-directional Encoding
What is Bi-directional Encoding?
As discussed in Forward—General concepts, there are three different frame
types available for the DivX encoder to select from. These are intra-frames,
predicted-frames and bi-directional-frames, or I, P and B-frames respectively.
Recall that in an I-frame all blocks are intra-blocks and are encoded as the image
filling the block. In a P-frame blocks can be either intra-blocks or forward
predicted blocks—those described by a vector referencing a matching area of
image in the previous frame (known as forward prediction).
Bi-directional frames can contain blocks that are intra, forwards predicted,
backwards predicted, or both forwards and backwards predicted. This means
that B-frames reference not only the previous frame but the next frame also.
Frame 1 (I) Frame 2 (B) Frame 3 (P) Frame 4 (B) Frame 5 (P)
In the diagram above, frames 2 and 4 are bi-directional frames and any
macroblock may be predicted from either the previous (forward) or next
(backward) frame. If a block is both forwards and backwards predicted then when
motion compensation takes place the block image will be the result of blending
the appropriate areas of the forwards and backwards frames together.
Notice that frame 4 contains some intra-blocks because in the example these
blocks could not be predicted from either the forwards or backwards frame.
The
72
Guide
As described earlier in the guide, predicting blocks consumes far less bits than
encoding them as intra-blocks. Because B-frames can be both forwards and
backwards predicted, they will generally have a lower proportion of intra-blocks
than any other frame type.
Since B-frames themselves are never referenced by any other frame it is

possible for the encoder to use a higher quantizer than is normal for their coding
since it need not be concerned with impacting upon the quality of other frames
due to compound errors.
These two key features of B-frames lead them to offer the greatest compression
ratio of any of the available frame types.
Adaptive Bi-directional Encoding

Adaptive Bi-directional Encoding
The DivX 5.2 encoder adds new support for both adaptive bi-directionally coded
frames, and multiple consecutive adaptive bi-directionally coded frames. The
term “adaptive” refers to intelligent selection of bi-directionally coded frames,
such that bi-directionally coded frames are only used where it is appropriate to do
so.
In a non-adaptive scheme where B-frames are forced for every alternate frame
(IBPBPBP), brief but significant artifacting can occasionally occur should the B-
frame fall between two dissimilar reference frames.
Frame 1 (B) Frame 2 (P) Frame 3 (B) Frame 4 (P) Frame 5 (B)
Non-adaptive B-frame selection

Consider the firing of a gun where a B-frame is selected as the gun flashes. The
flash may only last one frame, so neither the forward or backwards reference
frames will resemble the B-frame. Although an extreme illustration of non-
adaptive schemes, adaptive selection generally produces higher quality video by
avoiding such problems.
The
73
Guide
Multiple consecutive adaptive B-frames
Multiple consecutive adaptive B-frames
Frame 1 (I) Frame 2 (B) Frame 3 (B) Frame 4 (B) Frame 5 (P)
Multiple consecutive B-frames
Where a typical frame sequence using only single consecutive B-frames might
read IBPBPBP, the DivX 5.2 encoder can produce a frame sequence that reads
IBBPBPBBBP, depending on the content being encoded.
Single B-frames offer a very low bit spend, but as more B-frames are placed in
sequence the bit spend associated with each frame rises because B-frames do
not reference each other Therefore, as the picture changes, each new B-frame
must include all the changes from the forwards and backwards reference frames.
Because the following P-frame can’t refer to changes described by the preceding
B-frames, it must describe the complete change from the previous reference
frame. As the number of consecutive B-frames grows, this change becomes
more significant, and so the bit spend for the P-frame tends towards that of an I-
frame, leading to a false economy in many consecutive B-frames.
The
74
Guide
The ordering of frames
Backward prediction introduces issues related to the sequence in which frames
must be encoded and decoded. Whereas forward prediction, as used by P-
frames, simply references the last frame that was decoded, a backwards
predicted block in a B-frame makes reference to some future frame that has not
yet been decoded. Therefore, in order to decode B-frames it is necessary to
decode future frames first. This is why frames may appear out of order in the
multipass log file (see Bitrate mode—Multipass).
Consider the illustration below again. Here the blue arrows represent forwards
prediction in the P-frames and the red arrows represent either forward (left) or
backward (right) prediction in the B-frames.
It is clear from the diagram that the B-frames can make reference to future
frames because there are red arrows pointing to the right. The encoder must re-
order these frames so that they are in the correct order for decoding.
The
75
Guide
Now consider the re-ordered sequence of frames:
Frame 1 (I) Frame 3 (P) Frame 2 (B) Frame 5 (P) Frame 4 (B)
Although Frame 2 still makes reference to both frame 1 and frame 3, frame 3 is
decoded first.
It is now clear that in the re-ordered sequence all frames can be decoded, as
every frame makes reference only to others that have been previously decoded
(all arrows point left). The decoder is responsible for displaying the decoded
frames in the correct order.
The
76
Guide
QUICK GUIDE
Bi-directional encoding lets the encoder to use
the Bi-directional frame-type, supporting both
forwards and backwards motion prediction.
B-frames offer the best compression ratio of all frame

types and are invaluable to producing high-quality video
at low bitrates.
The bi-directional encoding mode can be selected from

the second step of the Select Profile Wizard dialogue
under MPEG-4 Tools.
Support for adaptive multiple consecutive bi-directional encoding is not

required by any DivX Certified device. If you encode your video using
adaptive multiple consecutive bi-directional encoding you may not be able to
play it on DivX Certified devices.
The Handheld profile does not support Bi-directional encoding.
Enabling bi-directional encoding will always increase encoding time.
Bi-directional encoding is not performed when Performance/Quality mode is

set to Fastest, or when the Source Interlace is set to Preserve Interlacing.
The
77
Guide
Bi-directional encoding can greatly improve video quality at low-medium
bitrates. At extremely high bitrates it may be beneficial to disable it.
Bi-directional encoding:
-b <mode>
Enabled only when CLI parameter is present.
Mode is one of:

-1 Original non-adaptive single consecutive B-frames
1 Adaptive single consecutive B-frames
2 Adaptive multiple consecutive B-frames
The
78
Guide
Quarter Pixel
Quarter Pixel
What is Quarter-pixel?
What is Quarter-pixel?
In a digitized video each frame is composed of a 2-
dimensional array of pixels, the small colored
squares that when viewed collectively from a
distance make up the picture.
Recall from Forward-Macroblocks and motion that

during the motion search DivX attempts to locate a
matching image for every block in the current frame
from a similarly sized area in the reference frame. One way to think of the motion
search is to imagine the encoder taking each square of video in the current frame
and overlaying it in various positions in the reference frame until it finds a good
match, continually maneuvering it and evaluating the difference.
The simplest way to perform this operation would be to compare each block in
the current frame against the surrounding area in the reference frame, moving
the block one pixel in any direction at a time. The relative vector offsets from the
origin of the current block during this search might be (0, 0), (0, 1), (0, 2), (1, 0)
and so on.
This scheme would be called whole-pixel because the search was performed
against the reference frame at 1-pixel resolution.
In fact, DivX actually searches at half-pixel resolution by default, achieved by use

of filtering. Applying a filter to the block allows the encoder to create a virtual
block that represents how the block would appear maneuvered half a pixel in any
direction against reference frame. The motion search can then return a vector
accurate to half-pixel resolution. The relative vector offsets from the origin of the
current block during a half-pixel resolution search might be (0, 0), (0, 0.5), (0, 1),
(0, 1.5), (0, 2), (0.5, 0), (0.5, 0.5), (0.5, 1) and so on.
Quarter-pixel, as its name suggests, performs the motion search accurate to

quarter-pixel resolution. Relative vector offsets from the origin of the current
block during a quarter-pixel resolution search might be (0, 0), (0, 0.25), (0, 0.5),
(0, 0.75), (0, 1) and so on.
The
79
Guide
Why use Quarter-pixel?
Quarter Pixel
Why use Quarter-pixel?
Motion vectors accurate to quarter-pixel resolution allow each motion-
compensated frame to be reconstructed more accurately than those accurate to
half-pixel resolution. The resulting video tends to appear sharper and of slightly
higher quality.
Consider that the eye is most sensitive to motion where objects in a scene are
moving very slowly. It would be difficult, for example, to accurately see how many
pixels a rocket had traveled across the screen after firing, but if instead we were
watching a snail crawling in front of the camera then accuracy would be far more
important.
The effect of perspective (where distant objects appear smaller than near
objects) biases the improvements made by quarter-pixel motion search with
respect to an objects depth in the scene. With respect to the image at the camera
objects will appear to move slowly in the distance
where perspective causes space to appear
compressed, and fast near to the camera where
perspective has the least significant effect.
In this scene the

effect of
perspective is apparent—although the car does
not physically shrink as it becomes more distant it
occupies less area in the image. We could say
that distant objects have lower spatial resolution
in the image. Quarter-pixel counter acts this.
Imagine a character walking across the video behind the car while it is positioned
near to the camera. The motion accuracy would not be critical because the
resolution of near objects is already high. In contrast, imagine the same
character walking across the video behind the while it is distant from the camera.
The character would occupy far fewer pixels and motion accuracy would be more
important for fluid and accurate motion.
Therefore, although quarter-pixel reduces the compensation error throughout the

entire video its most significant impact is on the fluidity in motion of objects far
from the camera.
The
80
Guide
Quarter-pixel
Quarter Pixel
Quarter-pixel
QUICK GUIDE
Quarter-pixel increases the resolution of the
motion search to one quarter of one pixel,
double that of the default half-pixel resolution.
Quarter-pixel resolution motion vectors reduce the error in

the motion-compensated frame but increase the search
time, leading to longer encoding times.
Quarter-pixel can be enabled from the second step of the

Select Profile Wizard dialogue under MPEG-4 Tools only
when Disable profiles has
been selected during the first
step.
Support for Quarter-pixel decoding is not required by any DivX Certified

device. If you encode your video using quarter-pixel you may be unable to
play it using DivX Certified devices.
Quarter-pixel motion search resolution can substantially increase encoding

duration.
Quarter-pixel motion search resolution is not supported by the Rate-Distortion

algorithm as used by the Slow Performance/Quality mode. If you enable
Quarter-pixel in combination with these modes the encoder will default to
Standard Performance/Quality mode automatically.
The
81
Guide
Quarter Pixel
Quarter-pixel
If you are not creating content strictly for your personal use be aware of the
decoding implications of quarter-pixel motion accuracy. Slower computers
may perform very poorly when attempting to decode quarter-pixel video—
some may fail to decode in real-time even when the quality slider is at its
minimum setting. Others may require substantially reduced post-processing,
leading to visible artifacting that can counter act the quality improvements
offered by quarter-pixel.
By increasing the accuracy of the motion vectors quarter-pixel also increases

the bit spend on predicted frames. The result can actually be detrimental to
quality at low bitrates. Although the motion compensation will be more
accurate and the picture will appear sharper, fewer free bits will cause the
rate control to enforce higher quantizers throughout the video.
Quarter-pixel:
-q
Enabled only when CLI parameter is present.
The
82
Guide
Global Motion Compensation

What is Global Motion Compensation?
What is Global Motion Compensation?
Global motion compensation reduces the bit spend for motion estimation by
deriving the motion of some blocks from a global motion present in the frame, as
opposed to explicitly specifying a unique
vector for each.
Video sequences frequently contain

features such as panning and zooming
where many of the blocks will share a
common motion attributable to the
behavior of the camera.
Imagine the resulting effect on the

motion vectors if the camera were to pan right across a scene. It would be
reasonable to assume that many of the blocks would share a motion vector
opposite but relative to the panning motion of the camera, with the exception of
any blocks representing objects moving independently in the scene.
Likewise, if the camera were rotated clockwise all of the blocks representing a
static scene would appear to move anti-clockwise around the center of the
picture. If the camera were to zoom in, all blocks would appear to move outwards
from the center of the picture towards the edges. Its entirely possible that the
camera might zoom, rotate and pan at the same time.
During motion estimation, global motion compensation finds and records the
global motion attributable to translation, rotation and zooming of the camera.
Motion vectors for blocks whose motion is consistent with the global motion can
be calculated during decoding as opposed to being explicitly specified, and
hence fewer bits are spent recording motion vectors when global motion
compensation is enabled.
When global motion compensation is not used the encoder remains efficient in
recording translation. Each motion vector is recorded as the difference from the
vector of its neighboring block, thus when vectors are largely consistent
throughout a frame very few bits can be spent recording them.
The
83
Guide

QUICK GUIDE
Global motion compensation reduces the bit
spend for motion estimation by deriving the
motion vectors of some blocks during decoding
from a common global motion present in the scene, as
opposed to specifying a unique vector for each.
Global motion compensation can be enabled from the

second step of the Select Profile Wizard dialogue under
MPEG4 Tools only when Disable profiles has been
selected during the first step.
Support for Global motion compensation is not required by DivX Certified

devices. Although some hardware devices claim to support GMC, this
functionality is not required by the DivX Certified Program or tested by
DivXNetworks.
If you are not creating content strictly for your personal use be aware of the
decoding implications of global motion compensation. Slower computers may
perform poorly when decoding video with global motion compensation.
Global motion compensation can improve video quality by reducing the bits
spent on motion estimation, leading to the rate control reducing quantizers
and thus improving quality.
Global motion compensation
-g
Enabled only when CLI parameter is present
The
84
Guide
Advanced
The
85
Guide
Advanced
Advanced
Scene-change threshold
QUICK GUIDE
As described in Forward—Predicted-frames and
Intra-frames, each frame in a video can be one
of three types: intra, predicted or bi-directional. It
follows that there must be some logic controlling frame
type selection—the encoder is not human and can’t
actually recognize where one scene changes into the
next.
Recall that when encoding predicted or bi-directional

frames the encoder performs the motion search for each
block composing the frame into one or more reference
frames. For P-frames the reference frame is the previous
frame in the video, for B-frames the reference frames are
both the previous and next frames.
Where a block can’t be matched in a reference frame it

will be encoded as an intra-block and instead of prediction
via a motion vector its image will be stored. One way to select where to place
intra-frames is to set some threshold level specifying the percentage of blocks
that must be intra-coded in order for the frame to be encoded as an intra-frame.
This is the scene-change threshold.
The scene-change threshold defines the percentage of blocks not tracked from
the reference frames by the motion search required to trigger a scene-change
detection. When a scene-change is detected a key-frame (intra-frame) will be
used in place of a predicted or bi-directional frame.
100% untracked 0% untracked 9% untracked 4% untracked 14% untracked
Here if the threshold were 13% frame 5 would become a key-frame (intra-frame).
The
86
Guide
Advanced
Note that what the encoder views as a scene-change does not necessarily
correspond with what humans perceive to be a scene-change—Regardless of
which value you configure the scene-change threshold to be, there is no
guarantee that the encoder will key-frame on an actual scene-change, just as
there is no guarantee that the encoder will not key-frame elsewhere.
The scene-change threshold is respected by the encoder only when Fast or

Standard Performance/Quality modes are selected.
Key-frames (intra-frames) are useful throughout a video for a number of

reasons. Where too few blocks can be predicted from one frame to another it
can be more economical to key-frame rather than use one of the predicted
frame types. More frequent key-frames can reduce the seek recovery time in
media players when the viewer skips to a specific point in the video. Due to
the nature of predicted frames, a media player must decode every frame from
the nearest past key-frame to the seek target in order to resume playback at
the specified frame.
Reducing the scene-change threshold will cause an increase in the

proportion of key-frames to predicted frames throughout the video. Key-
frames are the most expensive of all frame types in terms of bit-spend, and
using too many key-frames will starve the encoder of free bits. This will cause
the rate control to return higher quantizers and in turn reduce the overall
video quality.
Raising the scene-change threshold will cause a decrease in the proportion

of key-frames to predicted frames throughout the video. Fewer key-frames
translate to more free bits causing the rate control to return lower quantizers,
in turn increasing the overall quality of the video. However, too few key-
frames will lead to long seek times when the viewer attempts to skip to a
specific point in the video and increase the recovery time should the video
stream become damaged (e.g. when in transit over a broadcast network).
The default scene-change threshold is 50%. It is recommended the scene-

change threshold not be lower than 40% or higher than 60% for most
content.
Scene-change threshold:
-sc <Threshold>
Threshold is the percentage of intra-blocks required to trigger scene-change

detection.
The
87
Guide
Maximum key-frame interval
Advanced
Maximum key-frame interval
QUICK GUIDE
The maximum key-frame interval defines the
maximum number of consecutive frames that
may be a predicted type.
If maximum key-frame interval frames have been

encoded since the last intra-frame, including the current
frame, the current frame will be forced intra.
A fixed limit on the maximum interval between key-frames

is essential to ensure seek and error recovery is prompt
throughout the encoded video –particularly for low-motion
content that might otherwise have a very low proportion of
key-frames.
More frequent key-frames can reduce the seek recovery time in media
players when the viewer skips to a specific point in the video. Due to the
nature of predicted frames, a media player must decode every frame from the
nearest past key-frame to the seek target in order to resume playback at the
specified frame.
Reducing the maximum key-frame interval will cause an increase in the

proportion of key-frames to predicted frames throughout the video. Key-
frames are the most expensive of all frame-types in terms of bit spend, and
using too many key-frames will starve the encoder of free bits causing the
rate control to return higher quantizers, in turn reducing the overall quality of
the video. Conversely, raising the maximum key-frame interval will cause a
decrease in the proportion of key-frames to predicted frames throughout the
video. Fewer key-frames translate to more free bits causing the rate control
to return lower quantizers, in turn increasing the overall quality of the video.
The maximum key-frame interval is limited to 60 frames when Fastest

performance/quality mode is selected.
Maximum key-frame interval:
-key <frames>
Frames is the maximum interval between key-frames (intra-frames),

evaluated inclusive of the current frame.
The
88
Guide
Quantization type
Advanced
Quantization type
QUICK GUIDE
During encoding picture data is converted from
the spatial domain, i.e. color values per pixel
composing the picture, to the frequency domain.
This conversion is performed via the Discrete Cosine

Transformation (DCT) function, which represents the
picture data as a series of cosine waves at different
frequencies.
Low frequency co-efficients generally represent smooth,

gradual color transitions in the picture. High frequency co-
efficients represent sharp, high contrast color transitions,
such as the edges of objects.
Recall from Forward—General Concepts—Quantizers

that the encoder can control image quality by applying
different levels of quantization to the DCT results. In this
way it is possible to balance bit spend against picture quality.
H.263 The quantization matrix controls how different frequency

co-efficients are quantized.
Quantization
Two quantization matrices are available in DivX 5.2:

H.263 and MPEG-2.
Using H.263 quantization, the standard for DivX 5

Frequency
video, all frequency co-efficients are quantized equally.
MPEG2 H.263 quantization is optimized for very low to high
bitrates, and is the recommended quantization matrix.
Quantization
Using MPEG2 quantization high frequency co-efficients

are quantized more than low frequency co-efficients.
MPEG2 quantization improves scalability to extremely
high bitrates, but in general will increase artifacting and
Frequency should be avoided.
Support for MPEG2 quantization is not required by any DivX Certified device.
If you encode your video using MPEG2 quantization you may be unable to
play it using DivX Certified devices.
The
89
Guide
Advanced
Quantization type
Quantization type:
-qm MPEG
MPEG2 quantization will be used when this CLI parameter is present,

otherwise the quantization matrix will be H.263.
The
90
Guide
Interlacing
The
91
Guide
Interlacing
Interlacing
What is interlacing?
What is interlacing?
Interlacing is a method adopted from the old days of analog television that
allowed the frame rate of a video to be doubled by broadcasting only half a frame
at a time. This was achieved by dividing each frame into two fields, each
occupying alternating horizontal lines. We call these the top and bottom fields.
Top field Bottom field
Here you see two fields composing one

frame of video. First the top field is
broadcast and scanned out on odd lines
by the television, and then the bottom
field is broadcast and scanned out on
the even lines, recreating a complete
picture.
An interlaced camera may capture all The complete frame

the even lines in the picture first and
then the odd lines a fraction of a second later, leading to jagged edges that
appear where there is motion in the reconstructed frame.
Consideration of interlacing is important to DivX encoding because some sources

(notably DV and some DVDs) are interlaced and will suffer severe quality
degradation and artifacting if encoded without correctly processing interlacing.
The
92
Guide
Why de-interlace?
Interlacing
Why de-interlace?
During playback of an interlaced video half of one frame is continually being
merged with the previous picture. If the encoder were to treat each frame as
progressive (a complete picture that is not interlaced) then inevitably artifacting
would occur when half of a new frame was merged with the remaining half of the
previous frame, particularly where the new frame differs significantly from the
Maintaining interlacing
last.
Interlacing artifacts therefore appear as

two slightly different pictures being
merged on alternate horizontal lines and
are most visible where there is fast
motion in a scene.
De-interlacing is the process of taking an

interlaced video and recombining the
fields so as to create a progressive, non-
Interlacing artefacts
interlaced output.
Consider that if the encoder were to assume interlaced content was in fact
progressive then when motion occurred in the image interlacing artifacts would
be present. Greater motion causes greater artifacting and inconsistency in the
texture between frames. The motion search must be able to match blocks in the
current frame back to similar areas in the reference frame and an inconsistent
picture will cause this process to fail. Consequently, a higher proportion of blocks
will be intra-coded leading to a higher bit spend per frame, forcing the rate control
to use higher quantizers and lowering the overall quality of the encoded video.
Maintaining interlacing
In certain circumstances, it is desirable to maintain the original interlaced fields
discretely so they can be recreated without artifacting for display on an interlaced
device; playback via a DivX Certified device to an interlaced television, for
example.
To achieve this the encoder can be set to Encode as interlaced. Fields will be
preserved, however encoding each field individually requires a higher bitrate than
is normally required when encoding a progressive source.
The
93
Guide
Source interlace
Interlacing
Source interlace
QUICK GUIDE
Source interlace specifies the interlacing format
of the source video and how interlacing should
be handled by the encoder.
1. Progressive source
The encoder will assume that the source video
is progressive (not interlaced) and no de-
interlacing will be performed prior to encoding.
2. De-interlace source
The encoder will assume that the source video
is interlaced and will de-interlace it, encoding as
progressive.
3. Preserve interlacing
The encoder will assume that the source video is interlaced and will
encode each field in every frame separately.
The Handheld and Portable profiles do not support interlaced video.
DivX 5.2 does not support bi-directional coding when preserving interlacing. If
you choose to preserve interlacing bi-directional coding will be automatically
disabled.
The
94
Guide
Interlacing
Source interlace
If your source video is interlaced you should select to de-interlace all frames
to progressive unless you have a specific reason for preserving interlacing.
Encoding as interlaced requires substantially higher bitrates in order to
achieve equal perceptual quality to progressive content.
All video from DV cameras is interlaced.
The DivX 5.2 decoder does not support real-time de-interlacing during
playback. If you encode as interlaced the interlaced fields will be visible when
played via a progressive display, such as a PC monitor.
If you wish to resize and de-interlace, you must de-interlace before resizing. If
you use the encoder’s resize filter then de-interlacing is always performed
before resizing. If you use an external resize filter you must also use an
external de-interlace filter so that de-interlacing takes place before resizing.
Source interlace:
-d <mode>
Mode is one of:

1 Progressive source (parameter may be omitted)
2 Preserve interlacing
3 De-interlace source
The
95
Guide
Video Buffer
Verifier
The
96
Guide
Video buffer verifier

What is the video buffer verifier?
What is the video buffer verifier?
The encoding profiles available in DivX 5.2 let you create video guaranteed to
play back on DivX® Certified devices. To facilitate this it is necessary to control
aspects of the video stream produced by the encoder in such a way that the
The video buffer verifier explained

capabilities of the target device are not exceeded. The video buffer verifier is part
of this process.
The video buffer verifier forms part of a virtual decoder that is attached to the
encoder, limiting its output in terms of certain device capabilities: The sustainable
rate at which the device and receive the video stream, the buffer associated with
the video stream within the device, and how full that buffer should be before
playback begins.
When playing DivX videos on certain devices (e.g. a fast desktop computer),
these considerations can be overlooked without significant consequences.
However, when encoding for broadcast over an IP network (e.g. the Internet), or
for a hardware device with fixed sustainable throughput, it is critical for
uninterrupted playback that the abilities of the device are not exceeded.
The video buffer verifier explained

Consider a hardware DVD player featuring DivX support. The DVD-ROM drive in
the unit may have a limited sustainable transfer rate and a fixed-size cache
associated with it.
Suppose a sustainable transfer rate of 2000 kbps from the drive and a 4096
kilobit read cache. If the encoder were to encode at 2500 kbps for a long period
this would be 500 kbps in excess of the drive’s ability to read from the disc.
Assuming the 4096 kilobit cache was full to begin with a 500 kbps under-run
would exhaust the buffer and cause playback to pause after around 8 seconds.
As this example illustrates, it is important that the encoder considers the
The
97
Guide

QUICK GUIDE
The video buffer verifier prevents the encoder
from exceeding the ability of a target system to
read and buffer the video stream to prevent
pauses in playback.
Each DivX Certified Profile has specific VBV parameters associated with it
that ensure playback performance on DivX Certified devices. Although it is
possible to alter the VBV parameters in a way that would not negatively
impact playback performance, changing the VBV parameters is not
recommended except by experienced users with specific reasons for doing
so.
The default VBV parameters are:
Profile Bitrate Buffer Initial occupancy

Handheld 712400 262144 196608
Portable 768000 1048576 786432
Home theatre 4000000 3145728 2359296
High-definition 8000000 6291456 4718592
The default initial occupancy is 75% of the buffer.
Using the video buffer verifier it is possible to create MPEG4 video streams
suitable for streaming, even though DivX Pro does not include a streaming
server itself.
By careful configuration of the VBV parameters you may be able to improve

playback on older MPEG-4 compatible DVD players that have not been
certified by DivX Networks.
Although the initial occupancy is always obeyed by the video buffer verifier
and is useful particularly at the beginning of an encoding session, it may not
always be respected by a decoder.
The
98
Guide
Video buffer verifier:
-vbv <bitrate>, <buffer size>, <initial occupancy>
Bitrate is the maximum sustainable read bitrate in bits per second of the
target system.
Buffer size is the size in bits of the buffer associated with the video stream in
the target system.
Initial occupancy is the number of bits from the video stream assumed by the
encoder to have been pre-buffered by the target system before playback
begins.
The
99
Guide
Profiles
The
100
Guide
Profiles
Profiles
What is a profile?
What is a profile?
Why use profiles?

As the DivX® codec grew into a professional-grade standard for low-bitrate
video, hardware manufactures realized its potential in the consumer market and
developed devices claiming to be “DivX compatible”.
In truth these players would typically support some, but not all, versions of DivX,
and then only limited features from these versions. The capabilities of hardware
devices claiming DivX compatibility varied so widely that it became a hit-and-
miss affair attempting to encode DivX video that could be played consistently well
on all of them—a task that often seemed impossible.
Therefore, DivXNetworks designed a system of profiles for DivX® Certified

devices—those tested rigorously test by DivXNetworks’ engineers to ensure they
meet the requirements demanded by the certification program. Associated with
each profile are a set of minimum capabilities a device must possess before it
can attain certification.
Why use profiles?

By selecting one of the four available profile modes from the Select Profile
Wizard, you instruct the encoder to consider the minimum capabilities of devices
certified for that profile. Encoder features unsupported by any given profile will be
automatically disabled and the encoder will produce only video streams suitable
for the certified device.
Creating video streams for DivX Certified devices is as easy as selecting the
profile badge as displayed on the certified device from the Select Profile Wizard
prior to encoding.
The DivX Certified™ Profiles
The
101
Guide
Profiles
Profiles
Profiles
QUICK GUIDE
Profiles enforce that the encoder creates a video
stream compatible with DivX® Certified devices.
Selecting the profile of the DivX Certified device from the

Select Profile Wizard prior to encoding ensures the video
stream will always play correctly on the device.
Any given profile may limit the available encoder options

based upon the minimum requirements defined by the
certification program for that profile. CLI parameters, such
as the video buffer verifier,
will also be configured
automatically when a profile is
selected to reflect the abilities
of the target device.
It is possible to encode video suitable for certified devices even when profiles
are disabled by carefully choosing encoder options.
Certified devices are not required to support Global Motion Compensation or

Quarter-pixel under any profile, and when profiles are disabled you should
take care to set the video buffer verifier CLI parameters appropriately.
Handheld profile does not support Interlacing or Bi-directional encoding.
Portable profile does not support Interlacing.
Quarter-pixel and Global Motion Compensation will always be unavailable

when profiles are enabled. Interlacing will be unavailable when either
handheld or portable profiles are selected. Bi-directional encoding will be
unavailable when Handheld profile is selected.
1-Pass Quality-based mode is unavailable unless profiles are disabled.
When encoding for non-certified players that claim DivX compatibility for
greatest compatibility you should use Home Theatre profile.
For detailed profile information see DivX Certified™ Program—What’s in a

profile?
The
102
Guide
Profiles
Profiles
Profiles:
-profile <number>
Number is one of:

0 Disable profiles
1 Handheld profile
2 Portable profile
3 Home Theatre profile
4 High-definition profile
The
103
Guide
DivX®
Certified
Program
The
104
Guide
DivX® Certified Program

What is the DivX Certified Program?
What is the DivX® Certified Program?
The certification program was developed by DivXNetworks to ensure the ability of
certified devices to play all versions of DivX video without problems.
DivX Certified devices fall into one of four categories, each associated with a
suitable profile that defines its abilities.
►Example Certified Devices
Cell phones, low-end camera phones, PDAs
Multimedia jukeboxes, video cameras, still cameras
DVD players, PVRs (DVD players and recorders), some

types of video cameras
High-definition DVD players, HDTV PVRs
Every certified device is extensively tested by DivXNetworks before being

awarded certification. Only devices that say “DivX Certified” and carry one of the
four official certification logos have been tested by DivXNetworks as capable of
playing all versions of DivX video content.
If a device does not display the logo then do not

assume that it is capable of playing all versions of
DivX video: “If it doesn't say it, it doesn't play it”.
The
105
Guide
What’s in a profile?

What’s in a profile?
Each profile is associated with a set of minimum features a device must support
in order to achieve certification.
The following table lists profile-specific features.
352 x 240 720 x 480

30 fps 30 fps
176 x 144 1280 x 720
Maximum resolution
15 fps 30 fps
352 x 288 720 x 576
25 fps 25 fps
Macroblocks per
1485 9900 40500 108000
second
Maximum average
200 kbps 768 kbps 4000 kbps 8000 kbps
bitrate
Maximum peak
bitrate during any 1 800 kbps 4000 kbps 8000 kbps 32000 kbps
second of video
Minimum VBV buffer

32 KB 138 KB 384 KB 768 KB
size
Bi-directional
No Yes Yes Yes
encoding support
Interlaced video
No No Yes Yes
support
Devices can exceed these minimum requirements but you should never rely
upon this when performing profile encoding.
By selecting the correct profile of the target DivX Certified device from the Select
Profile Wizard prior to encoding you ensure the video stream will always play
correctly on the device.
The
106
Guide
DivX certification requirements

DivX certification requirements
Aside from profile-specific features, there are requirements that all DivX Certified
devices must meet.
The following table lists the generic requirements of certified devices.
Feature Required
All DivX 3.11 movies on 1 CD, anything under 1 mbps average bitrate Yes
All DivX 4 content Yes
DivX 5 content with no GMC and no QPel Yes
DivX video created for Video on Demand Yes
DivX video created on a DivX Certified encoding device Yes
MP3 audio in DivX video both CBR and VBR Yes
DivX 3.11 movies on 2 CDs (high bitrates) No
DivX 5 content with GMC or QPel No
XVID content No
ADPCM audio, PCM audio, Ogg Vorbis audio No
AVI files with bad audio/video interleaving No
Content that is not supported by certified devices can be effortlessly converted

using Dr DivX ( http://www.DivX.com/divx/drdivx ).
The
107
Guide
E.K.G.
The
108
Guide
What is the EKG?
What is the EKG?
We saw in Bitrate mode—Multipass that after each pass of a Multipass encoding
a log file is written containing an analysis of the video encoded during the pass,
used in turn to optimize the rate-control strategy for any successive pass.
Normally the rate-control will use the information from the log file in an unbiased
fashion, lending equal weight to all frames and basing frame quantizers upon the
bitrate and frame complexity. However, the encoder itself can’t conceive of the
actual content of a video as humans can, or accurately determine how the
human eye will perceive the quality of any particular sequence of video.
At low-bitrates it is often desirable to intentionally bias the rate-control to raise or

lower the quality of a particular section of video. For example, when encoding a
movie in its entirety it is likely the viewer will be less concerned with the quality of
the titles than that of the movie itself. It would thus be desirable that the titles are
encoded with a higher quantizer (lower quality), and the bits freed by doing so
used to enhance the quality of the actual movie. Alternatively, where a particular
video sequence demands exceptional quality it is desirable to bias the rate-
control in such a way that a lower quantizer is selected and bits are drawn from
the remainder of the video in order to enhance the sequence.
The EKG makes this possible by

allowing the data in the log file to
be displayed and manipulated
graphically between passes,
displaying the video associated
with the data for ease-of-use.
The EKG offers precision fine

tuning unavailable in other
encoding software.
A complete description of the log

file data was covered in Bitrate
mode—Multipass.
The Electrokompressiongraph™
The
109
Guide
How the EKG works
How the EKG works
As discussed in Forward-General concepts-Quantizers, DivX uses quantizers to
control the picture quality in each frame. Lower quantizers produce higher quality
and bit-spend, while higher quantizers produce lower quality and bit spend.
The aim of the EKG application is to control the modulation value associated with
each frame in the multipass log file. This modulation value actually manipulates
the frame quantizer chosen by the rate-control by acting as a co-efficient.
Frame quantizer = Modulation * Quantizer
Thus if the rate control would normally encode a particular frame at Q=6 but a
modulation of 2.0 was specified the frame would instead be encoded at Q=12. A
modulation of 1.0 would leave the original quantizer unchanged (1.0 x Q = Q),
and a modulation of 0.5 would halve the quantizer so that the frame would be
encoded at Q=3.
Notice that the EKG can’t show the actual quantizer for any particular frame as
this is a decision made on-the-fly as video is encoded by the rate control.
Modulation allows you to manipulate quality in terms of proportionality, not in
terms of fixed quantizers.
Also notice that the relationship between quantizer value and quality is inverse, a
lower quantizer gives rise to higher quality. Therefore, a modulation of 2.0 is the
equivalent of 50% quality, while a modulation of 0.5 is the equivalent of 200%
quality.
The
110
Guide
Using the EKG
Using the EKG
Using the EKG it is possible to display graphically the contents of the multipass
log file after only the 1st pass, however in order to preview the encoded video as
you manipulate the modulation control at least one nth pass must have been
performed. Making best use of the EKG will therefore require at least three
passes—the EKG application being used between passes two and three.
1. Open the EKG application after

encoding two passes of any video.
Select Open from the File menu and
open the multipass log file followed
by the encoded AVI file.
The EKG will display a preview of

the video and graph the data
from the log file. Hover the
mouse over the Legend bar
to see the graph legend.
Click the Swap button to display the

2. contents of the log file numerically.
Hovering the mouse over a frame in
the graph will display its statistical
information. While the drag tool
(hand icon) is selected from the
toolbar at the top of the EKG window
you can safely drag the graph area
using the left mouse button without
making any changes. You can also
use the zoom-in and zoom-out
toolbar buttons to zoom the graph.
The two left-most toolbar buttons add grid-lines to the graph area. You can also
select which statistics are graphed and how they are displayed from the Graph
menu.
The
111
Guide
Using the EKG
Click the Edit button on the
3. toolbar to enter editing mode.
Clicking on the graph will now
operate the modulation
control system.
4. Click on any frame to select it. The

bar will turn purple to indicate that it
has been selected and the
modulation grab-bar will appear.
Drag the bar up or down to adjust

the modulation for the selected
frame. As you do so the modulated quality will appear as a percentage
in the center of the grab-bar.
After you release the grab-bar the new modulation for the
frame will be set and reflected in the orange modulation
line that runs throughout the graph.
You can select multiple frames to modulate

simply by clicking on each before adjusting
the drag-bar.
You can also select an entire sequence of

frames by clicking the first frame, then
holding shift while you click on the last
frame.
5. After making your changes, choose Save Changes from the File menu.
You must now run one more nth pass to incorporate these changes into
your DivX video.
Modulation settings are preserved between passes. You can revise your
modulation settings at a later time if you wish.
The
112
Guide
DivX
Decoder
The
113
Guide
DivX® Decoder
DivX® Decoder
About the decoder
About the decoder
Post-processing
The DivX decoder plays all DivX® 3, 4 and 5 series video, applies special post-
processing effects to enhance video quality, and makes best use of the
advanced features provided by your graphics hardware for improved playback
performance.
The DivX decoder will be used by all third-party media players to play DivX video.
You need only configure it once to impose changes on all of your media players.
It is worth noting before discussing the decoder features that some third-party
MPEG4 decoder filters can commandeer DivX decoding from the genuine DivX
decoder in media players that are based upon Microsoft’s DirectShow platform.
You can check that the genuine DivX decoder is playing your DivX videos simply
by playing any DivX video in a DirectShow based media player (such as
Microsoft’s Windows Media Player) and watching for the DivX logo in the lower
right hand corner of the movie. The logo should appear for several seconds when
playback begins if the genuine DivX decoder filter is in use.
To access the decoder configuration dialogue referred to in the following pages

choose the Config.exe program icon in the DivX programs group on your Start
menu.
Post-processing
Post-processing refers to effects applied to the video after decoding to enhance
the picture in some way.
The DivX decoder contains three post-processing techniques. These are de-
blocking, de-ringing, and the film effect.
The
114
Guide
Deblocking
DivX® Decoder
Deblocking
The most visible artifact occurring in low-
bitrate DivX video is called blocking, and
appears as small 8x8 squares in the video.
Deringing
The artifact is caused by the DCT
algorithm used by DivX to encode the
image, and becomes strongest where
video is encoded at very low bitrates due to
high quantization.
Automatic post-processing
Deblocking blends the transition along the block edges so that individual blocks
become less apparent.
Deringing
Ringing is another artifact caused by the DCT algorithm. Ringing appears as an
outline or shadowing around contrasting edges in the picture.
Ringing is harder to see than blocking

and requires more computation to
correct. By default, the decoder will apply
full deblocking before applying any
deringing, depending on real-time
decoding performance.
Automatic post-processing
The DivX decoder can automatically adjust the level of de-
blocking and de-ringing based upon your computers real-
time playback performance. This prevents the decoder from
dropping frames if the CPU becomes overloaded.
Automatic de-blocking is enabled when Partial auto post-

processing is selected. Automatic deblocking and deringing
are enabled when Full automatic post-processing is
selected.
The
115
Guide
Post-processing levels
DivX® Decoder
Post-processing levels,
The manual de-blocking and de-ringing controls do not control the strength of the
post-processing, but rather where and how it is applied.
Each pixel in a DivX video is composed of three color channels—one luminance

channel and two chrominance channels. The eye is most sensitive to changes in
Film effect
the luminance channel and so de-blocking gives preference to this.
Horizontal Vertical Horizontal Vertical
De-blocking
luminance luminance chrominance chrominance
level
de-blocking de-blocking de-blocking de-blocking
0 (MIN) No No No No
1 Yes No No No
2 Yes Yes No No
3 Yes Yes Yes No
4 (MAX) Yes Yes Yes Yes
De-ringing Horizontal Vertical

level de-ringing de-ringing
0 (MIN) No No
1 Yes No
2 (MAX) Yes Yes
Film effect
The film effect restores warmth to the image and masks the artificial smoothing
effect that can result from low-bitrate encoding by applying a film grain effect to
the picture.
The film effect is particularly useful in counteracting the

effects of strong source
pre-processing applied
during encoding.
The
116
Guide
Quality settings
DivX® Decoder
Quality settings
The quality settings page allows configuration of decoder
options relating to the rendering of DivX video.
►Smooth playback
When a DivX video contains bi-directional
frames smooth playback makes use of a buffer
so the decoder need only decode one frame for
every frame it displays.
If the frame sequence were 1I, 2P, 3B, 4P, 5B

the decoder would decode and buffer 1I, decode
2P and display 1I, decode 3B and display 3B,
decode 4P and display 2P, decode 5B and display 5B, ...
The last frame of the video may be dropped due to the buffer but
because only one frame is decoded at a time smooth playback
enhances performance for slower CPUs.
With smooth playback disabled B-frames are decoded at the same time
as the frames they reference and then displayed at the appropriate
time.
If the frame sequence were 1I, 2P, 3B, 4P, 5B the decoder would
decode and display 1I, decode 2P and 3B and display 3B, wait and
display 2P, decode 4P and 5B and display 5B, wait and display 4P.
No frames are dropped when smooth playback is disabled but greater

demands are placed upon the CPU.
For an explanation of frame orders refer to Bi-directional encoding—

The ordering of frames.
►YUV Extended
When enabled the decoder will output video in YV12 color mode. YV12
is the internal color format used by DivX and thus is the fastest method
of decoding as data can be sent directly to the video card.
When YUV Extended is enabled the Brightness, Contrast and

Saturation controls are unavailable.
The
117
Guide
DivX® Decoder
Quality settings
►Overlay Extended
When enabled the decoder will attempt to use the hardware overlay
instead of the software overlay. The hardware overlay is a feature of
some graphics cards that increases both rendering performance and
quality when video is scaled.
When overlay extended is enabled only one video file may be played
concurrently.
Overlay extended mode lets you modify the pixel aspect

ratio of the display.
►Double Buffering
Double buffering causes the video card to reserve additional memory
for receiving video data. Double buffering improves the smoothness of
playback but may not work with older graphics card that have very little
onboard memory.
►Disable Logo
When enabled, this option will prevent the decoder from overlaying the
DivX Video logo in the lower right-hand corner of the video during the
first few seconds of playback.
►Support generic MPEG4

When enabled the decoder will support playback of various MPEG4-
based video formats, including XVID.
The
118
Guide
Acknowledgements
The
119
Guide
Acknowledgements
Acknowledgments
Thanks and credits
Thanks and credits
Legal
Permission to illustrate this guide with images taken from
Killer Bean 2 was very kindly given by Jeff Lew. See his
work online and read about his DVD, Learning 3D Character
Animation:
http://www.jefflew.com
Thanks to the entire DivX Advanced Research Centre (DARC) team for their
assistance in writing this guide:
 Eugene Kuznetsov
 Andrea Graziani
 John Funnell
 Adam Li
 Adrian Bourke
 Cheng Huang
 Jérôme Rota
 Darrius Thompson
Thanks also to Tom Huntington (Corporate Communications Manager) and

Solange Jacobs-Randolph (Editorial Consultant) for their editorial review and
Scott Green (Art Director) for his cover artwork and other illustrations.
- Alastair Mayo (DigitAl56K)
Legal
Microsoft is a registered trademark of Microsoft Corporation. QuickTime and Macintosh are trademarks of Apple
Computer Inc. VirtualDub is written and maintained by Avery Lee. DivX, “DivX Certified” and the DivX Certified logo are
registered trademarks of DivXNetworks. Killer Bean is the property of Jeff Lew. All other trademarks are the property of
their respective owners.
This guide is Copyright © DigitAl56K/DivXNetworks 2003-2004.
The
120
Guide

Div XUser Guide 521

Uploaded by

Copyright:

Available Formats

Div XUser Guide 521

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Div XUser Guide 521

Uploaded by

Copyright:

Available Formats

Revision 5.2.1.

Video Buffer Verifier

DivX Certified™ Program

The origins of DivX video

Before we begin it is worth noting the outstanding pedigree of DivX video

The origins of DivX ® video

In early 2001, DivXNetworks covertly launched ProjectMayo, a website that

Instead, DivX is a video codec (COmpressor/DECompressor) accessible via

►Microsoft Video For Windows for applications running under Windows

The DivX 5 series encoder is an implementation of the MPEG-4 video standard,

Quick Start Guide

Download VirtualDub before we start. It’s a small download, open-source, and

Quick Start Guide

1. Load VirtualDub by double-clicking on the VirtualDub

In the VirtualDub main window

From left to right the timeline

2. From the File menu choose Open and

4. From the Video menu set Full Processing Mode.

Because we will not be applying filtering here

5. Now we want to configure DivX as the

Select DivX Pro(tm) 5.2

If you made it this far, this is the first time

When you are ready, click Restore

All that remains is to instruct VirtualDub to

Enter a suitable filename and

Here you see the progress bar, how

If the estimated time is quite long and

You will see the feedback

Quick Start Guide

1. Load VirtualDub and configure it as you did in

When you save a video in Multipass, 1st pass

2. Save your AVI file using a suitable file

3. Without closing VirtualDub re-configure the

When you save a video in Multipass, nth

4. Save the AVI file using the same file

After completing your nth pass

Quick Start Guide

►MP3 audio is supported natively by most operating systems

►MP3 audio is supported by DivX Certified hardware players

►CBR MP3 audio is least likely to introduce de-synchronization between

As a rough guide: 128 kbps is generally considered close to CD quality

This represents the worst possible situation with regards to interleaving.

Using VirtualDub, you can access the Interleaving

Ensure “Enable audio/video interleaving” is enabled. The recommended

In direct stream copy mode no processing (other than

In Full processing mode the given audio stream is

Macroblocks and motion

If we were to slow a video down to such an extent that it were possible

Macroblocks and motion

Frame 1 Frame 2 Frame 3

Storing the motion vector of each block from frame-to-frame allows

The process of tracking blocks back to matching areas in the previous

You can see the motion search

While encoding set Show picture to

The white lines overlaid on the

Intra-frames store all macroblocks as images rather than as motion

A predicted frame can contain forward-predicted blocks—blocks that

One or more P-frames may follow an I-frame:

Frame 1 (I) Frame 2 (P) Frame 3 (P)