The document provides an overview of different file formats and techniques for databending or distorting digital images in reversible and irreversible ways. It explores compressing, decompressing, and corrupting raw image files, bitmaps, GIFs, PNGs and more. Specific techniques discussed include opening raw files with incorrect header data, copying/pasting image data, using text editors to reformat image data, and manipulating interlacing and color values. The goal is to "deconstruct and create new, brutalist (audio)visual works" through exploiting file formats.
2. Type to enter text
Tango between a corrupt format and its user. ! ! ! !
Contents ! the
!
There is not sufficient data. Please enter data art is a practice that studies and researches
Glitch
! Static!
vernacular of file formats in exploitative manners to ! !
You have too Noise Art > Filter art > when Cool becomes Hot >>>>>>>
! ! Uncompressed!!
deconstruct and create new, brutalist (audio)visual works.
m u c h
However, glitch artists often the ! ! Lossless! ! !
c h a r a c t e r ! Glitches are hot; proof can be found on MTV, Flickr, in go beyond this formal
approach; coffee table that the glitch Lossy! not exists
A p p l y club and in the bookstore. While thethey realize book Glitch:
! ! does ! !
! Moving image! !
g a u s s i a n Designing Imperfection without human perception and therefore have a more
(2009) has introduced the glitch
inclusive approach to digital material. ! ! MOV ! ! !
blur. design aesthetic to the world of latte drinking designers, and
! ! DV ! ! !
The materiality of glitch art is constantly mutating; it exists
Kanye West used glitches to sing about his imperfect love
! ! WMV! hand the
!
life, the awkward, shy and physically ugly celebrate underrelies on the one
as an unstable assemblage that the
! ! AVI! !
You still have header "Glitched: Nerdcore for life". operation and content of the apparatus (the
construction,
! ! ! the writer/artist,
! !
quirks! Clone Glitch has transformed from cool and on Its no more then a work,
medium) to hot. the other hand the
s t a m p f o r brightly colored bubblegum wrapper that doesn't by the any
and the interpretation ask for reader and/or user (the
meaning).
s m o o t h involvement, or offers any stimulus. Inside I find gum that I
surface keep chewing, hoping for some new explosion of good taste.
But the more I chew, the less tasty / rubbery it gets. Glitch
design fulfills an average, imperfect stereotype, a filter or
Your file has commodity that echoes a "medium is the message" standard.
i n v a l i d
m a r k e r s . Luckily (or naturally?), the "No Content - Just Imperfection"
E n t e r n e w slogan of this kind of hot glitch design is complimented by
markers cool glitches. In The Laws ofthe materiality of the glitch art is not (just) the digital
Thus, Cool (2004), Alan Liu asks
himself What is "Cool"? He describes follows the vernacular of file formats, nor the
material that that cool is the ellipsis
of knowing whats cool ...and withholding that idea. Those that constantly changing
machine it appears upon, but a
Y o u r insist on asking, are definitely uncool. depends on the interactions between text,
construct that
dimensions social, esthetical, political and economic dynamics and the
do n o t Unruly and defiant as it point is, I viewgive my which on cool
thus of will from take the different actors make
meaning.
correspond. glitches. Cool glitches are the glitches that do not just focus
C h a n g e on a static end product, Digital artists a exploit theirpersonal materials thus also
but (also) on process, a digital
dimensions exploration or a narrative element (that or reflects (showing
metaphorically oftencritically critically the medium in a
on a medium). critical state or criticizing the medium and its inherent
norms) and not just formally.
ERROR Cool is in a constant state of and moreis the genre starting to use the term "glitch
more flux, as people are of "cool
glitch art", which finally exists as think it has become apparent that the question of
genre", I an assemblage that relies
on first of all the construction,constitutes and content of how a genre should be
what operation a genre, and the
G o t o d a t a apparatus (the medium)studied needs to bework, the in Glitch Studies. Also, in the
and secondly the included writer/
therapy and artist, and the interpretation by a "glitch genre", I think there is a need to research
case of the reader and/or user (the
repair your meaning). In the end, there processone stylization of cool” -- the point where the
the is “no of definition of glitch
registry glitch art. formal creation of glitches are not unknown, new
utterances but are becoming stabilized, new commodities
and even filters.
In an effort to make what was once cool now hot, and visa
Y o u r versa, and to take what kind of study involves more then just a vernacular of
This happened in the Designing
file formats or a research into technology but also includes
k e y f r a m e s Imperfection book a step further, I made this Vernacular of
are missing. File Formats, in which I study ways to exploit and deconstruct the history of the
culture, individuals, politics and
technology.
Your codecs the organizations of file formats into new, brutalist designs.
are not
supported. But I still wonder if there is really anything consistent within
…I am waiting for the first "Glitchs not dead" hoodie in H&M.
And because fans are as bad as the ignorant, If so, then I think it is the critical use
the glitch art "genre". for the sake of
being bad, I will definitelyof error, perceivedHoera! perceived, real or designed. And
wear the hoodie. or non
Please I think when I watch a glitched video, or any other glitch
respect the software. Now it is too little too late - youwhat not try most interesting to look for: what
work, this is did I find hard
enough - you are just not good enough. [system shut downplay a role in the work - does the work
critical elements indefinitely]
Dear mr compression I write a 1000 poems to you Is this whator does it show the technology in a
criticize something, they call
critical state?
progress? Warmly yours, the noxious angel of history 2
3. <-- Photoshop .RAW
UNCOMPRESSED
3channel, 2020x1138. (with
scan lines from the original
video)
Raw image files contain
minimally processed data
RAW
(pixels) from the image
sensor of either a digital
camera, image scanner, or
motion picture film scanner.
The file header of a RAW
image typically contains
information concerning the
byte-ordering of the file, the
camera sensor information
and other image metadata like
exposure setting, camera/
scanner/lens model, date (and,
optionally, place) of shoot/
scan, format, size, number of
colors, and other information
needed to display the image.
It is possible to save a
RAW image file without a
header (choose header=0).
When the interleaved RAW
image is saved without a
header the computer
doesnʼt know the
dimensions or any other
crucial information that is
needed to reconstruct the
image out of the image
data.
<-- Photoshop .RAW, (h=0)
I opened a 3 channel
interleaved RAW document
as a 1 channel (1 color)
interleaved document.
(reversible databend)
When you open the image,
softwares like for instance
photoshop will ask you for this
data, giving you the
opportunity to
“bend” (reversible) the image.
At this moment, you will be
able to choose the dimensions,
the amount of channels and if
the image will be displayed
interleaved or non-interleaved.
By entering another value then
the original, the image will be
displayed in a distorted way.
<-- Photoshop .RAW, (h=0)
interleaved, I opened
a 3 channel interleaved
document
but entered a 3
slightly smaller
value for the width.
(reversible databend)
4. In the case of a RAW image
file, interleaving and non-
interleaving refer to the order in
which the RGB color values of
every pixel are stored.
In an interleaved Raw
image, the data is stored in
a RGBRGBRGB sequence.
<-- Photoshop .RAW, (h=0)
I opened the 3 channel
interleaved RAW document
as a 3 channel non-
interleaved 8 bit document.
(reversible databend)
<-- Photoshop .RAW, (h=0)
I opened a 3 channel
interleaved RAW image in
Microsoft Word (Convert to
Text Only) and saved it. This
technique reformats the
image data into a Microsoft
Word document, changing
some of the values and
adding / deleting some extra
data. In the image this shows
by abrupt discolorations, and
image general image shifts.
(Stallioʼs Wordpad effect)
(irreversible databend)
<-- Photoshop .RAW, (h=0)
I opened a 3 channel non-
interleaved raw document
with smaller hight
(databend, reversible)
When the image is saved in
non-interleaved array, the RGB
values are not ordered
sequentially but have their own
ʻlayersʼ with
4
5. <-- Photoshop .RAW, (h=0)
I opened a 3 channel non-
interleaved document but
entered a slightly smaller
value for the width (1 pixel)
(reversible databend)
<-- Photoshop .RAW, (h=0)
I opened a 3 channel non-
interleaved RAW document
with a much smaller width
(reversible databend)
<-- Photoshop .RAW, (h=0)
I opened a 3 channel non-
interleaved RAW image in
Microsoft Word (Convert to
Text Only) and saved it
(wordpad, textpad and
other text editors could also
work). This technique
reformats the image data
into a Microsoft Word
document, changing some
of the values and adding /
deleting some extra data.
In the image this shows by
abrupt discolorations, and
image general image shifts.
(Stallioʼs Wordpad effect)
(irreversible databend)
5
6. <-- Bitmap (.bmp)
bend by copy pasting a bit of
BMP the image data over and over.
(irreversible databend)
The BMP file format is
uncompressed; every bit that
indexes a bitmap pixel value
is packed within a linear row,
"upside-down" with respect to
normal image raster scan
order, starting in the lower
right corner, advancing row
by row from the bottom to the
top.
<-- Low quality Bitmap (.bmp)
Random data replacement
changed the values of the
indexed colors in the color
palette (irreversible databend)
This is why, when you copy-
paste some of the image data
the lower part of the image
will still be intact, while the
upper part of the image only
shifts (and sometimes
discolors)
In BMP files, and many other
bitmap file formats the color
palette consists of a block of
bytes (a table or palette)
listing the colors available for
use in a particular indexed-
color image.
Each pixel in the image is
described by a number of bits
(1-32 bit color depth) that
index a single color from the
color palette, that is described
right after the header.
The BMP color palette uses
the interleaved RGB color
model. In this model, a color
depends on different
intensities (from 0 to 255) of
the primary RGB colors. A
color is thus defined by the
final intensities of R+G+B.
When you copy-paste the
image data, the intensity data
From B can (for instance)
shift to the R, creating sudden
discoloration.
<-- .BMP Wordpad effect
In comparison to the RAW
image, the image now waves
towards the right. This is
because the BMP
raster format is 6
saved from left to
right, top to bottom.
(irreversible databend)
7. <-- Graphics Interchange
LOSSLESS
Format (.gif), 8 colors
restricted pattern
(with dither)
Graphics Interchange Format
is a bitmap image format that
GIF
supports 8 bits per pixel and
can therefore consist of no
more then 256 colors.
The format supports
animation.
Dither (the grainy blocky
artifacts) is an intentionally
applied form of noise used to
“randomize quantization
error”; the difference between
the actual analog value and
quantized digital value. This
error is caused by truncation
(the discarding of less
significant information).
Dither thus helps to prevent
from large-scale patterns
such as "banding" (stepwise
rendering of smooth
gradations in brightness or
hue). Moreover, the not
available colors are
approximated because the
human eye perceives the
diffusion as a mixture of the
colors. This creates the
illusion of color depth.
<-- Graphics Interchange
Format (.gif), non-interlaced,
8 colors, restricted pattern
(with dither)
(irreversible databend)
<-- Graphics Interchange
Format (.gif), interlaced,
animated 8 colors, restricted
pattern (with dither)
(irreversible databend)
The gif format uses a 4 pass
one dimensional interlacing
strategy. This means that one
half of the image, consisting of
every other row of pixels is
rendered after the other half.
In the image on the left this
shows through a gradual
displacement during weaving
(the putting together of the two
layers), which resulted in a
second “ghost image” (or
combing artifacts with jagged
edges).
7
8. <-- Portable Network
PNG
Graphics (PNG). interlaced
(irreversible databend)
pre-compression: filtering
(prediction)
1-2 of the 7 stages before the
image is reconstructed.
(irreversible databend)
Like GIF PNG is a lossless
compressed raster format,
which means that it
represents an image as a
two-dimensional array of
colored pixels.
PNG is a bitmapped image
format that employs lossless
data compression and offers
a 7-pass 2-dimensional
interlacing scheme—the
Adam7 algorithm.
This is more sophisticated
than GIF's 1-dimensional, 4-
pass scheme, and often
allows for a clearer low-
resolution image to be
visible earlier in the transfer.
This is visible in image 1
which just passed its first
stage of the 7 part
interlacing scheme. In this
stage a part of the image is
rendered almost flawless,
while the further it gets
rendered, the more the
corrupted data becomes
visible.
<-- PNG. interlaced. stage
3/4. (irreversible databend)
8
<-- PNG. interlaced. stage
5/6.(irreversible databend)
9. <-- PNG. interlaced
(irreversible databend)
The 7-pass de-interlacing
of the corrupted image
sometimes stops early (as
early as the first pass).
<-- PNG. interlaced
(irreversible databend)
<-- PNG. non-interlaced
(irreversible databend)
9
10. <-- Truevision Advanced
TARGA
Raster Graphics Adapter;
(.TGA) is an initialism for
Truevision Graphics Adapter.
(can be lossy or lossless)
16 bits Lossless RLE
compressed
(irreversible databend)
Targa recognizes
over half a dozen image file
formats, some of which are
more widely used than
others.
I have databend a Tagra
compressed file by searching
and replacing + adding some
data. Apparent are shifts of
some blocks and some small
color shifts.
To understand how these
shifts went the way they went,
is however to complex and I
am not going to try to
understand or explain this in
this pdf.
<-- Targa (.tga)
24 bits Lossless RLE
compressed
(irreversible databend)
<-- Targa (.tga)
32 bits Lossless RLE
compressed
(irreversible databend)
10
11. <-- Targa (.tga)
32 bits Lossless RLE
compressed
(irreversible databend)
<-- Tagged Image File
Format (.TIF) or in short
TIFF TIFF, with white
background layer
(irreversible databend)
Just like Targa, TIFF is a
very complex compression.
I have had some really
interesting diverse
experiences with this file
format, but I find it very had
to get grips on the reason
why they come to the
surface the way they do.
Thats why these are real
glitch bends to me.
<-- TIFF
With white background
layer
(irreversible databend)
11
12. <-- TIFF
interleaved macintosh Save
Image Pyramid
(irreversible databend)
<-- TIFF
interleaved macintosh Save
Image Pyramid
(irreversible databend)
<-- TIFF
interleaved macintosh Save
Image Pyramid
(irreversible databend)
12
14. <-- Joint Photographic
Experts Group (.JPG)
LOSSY JPEG (lossy)
severely downsampled so
that the 8x8 macroblocks
(and quantization error)
are apparent.
(irreversible databend)
A JPG compression
consists of 6 subsequent
steps:
1. Color space
transformation
2. Downsampling
3. Block splitting
4. Discrete cosine
transform
5. Quantization
6. Entropy coding
1. Initially, images have to
be transformed from the
RGB color space to
another color space (called
Y′CbCr), that consists of
three components that are
handled separately; the Y
(luma or brightness) and
the Cb and Cr values
(chroma or color values,
which are divided into hue
and saturation).
2. Because the human eye
doesnʼt perceives small
differences within the Cb
and Cr space very well,
these elements are
downsampled.
3. After the color space
transformation, the image
is split into tiles or
macroblocks. Rectangular
regions of the image that
are transformed and
encoded separately.
<-- 8 × 8 DCT basis
patterns of a JPG.
4. Next, a Discrete Cosine
Transform (which works
similar to the Fourier
Transform function,
exploited in datamoshing
and macroblock studies) is
used to create a frequency
spectrum, to transform the
8×8 blocks to a
combination of the 64 two-
dimensional DCT basis
functions or patterns (as
differentiated by the red
lines).
14
15. 5. During the Quantization
step, the highest brightness-
frequency variations become
a base line (or 0-value), while
small positive and negative
frequency differentiations get
a value, which take many
fewer bits to represent.
<-- high frequency mapping
from which basic values are
derived.
6. finally, entropy coding is
applied. Entropy coding is a
special form of lossless data
compression that involves
arranging the image
components in a "zigzag"
order. This allows the
quantized coefficient table to
be rewritten in a zigzag order
to a sequence of frequencies.
A run-length encoding (RLE)
algorithm groups similar
frequencies together and after
that, via "Huffman coding"
organizes what is left.
Because the RGB color
values are described in such
a complex algorithms, some
random data replacement
often results into dramatic
discoloration and other
effects.
<-- low res JPG, Baseline
standard.
(irreversible databend)
<-- low res JPG, Progressive
(irreversible databend)
The very high compression
ratio of this jpg effects the
quality of the image and the
size of the artifacts.
When using quantization with
block-based coding, as in
these JPEG-compressed
images, several types of
often unwanted artifacts can
appear, for instance ringing or
ghosting. In the bend image to
the left, the low quality and
corruption have made these
artifacts more apparent.
15
16. <-- low res JPG, Progressive
(irreversible databend)
Ringing is (often) the result of
the loss of high frequency
components, but can also be
used to “enhance” the image,
because this artifact put an
emphasis on the edges (in
moving image these artifacts
can also be referred to as
mosquito noise).
Ghosting is an artifact that
appear when a part of the
image is somehow doubled
(often by refraction of a
television signal within the
atmosphere).
'Blocking' / Staircase artifacts
appear most clearly along the
curving edges, as a result of
the 8×8 jpg blocks.
Blockiness in "busy" regions is
sometimes also referred to as
quilting or checker-boarding.
The JPG to the left shows
some typical color distortion
(color mismatching) after
random data replacement.
It also shows some less
typical JPG artifacts called
jaggies.
"Jaggies" is the informal
name for artifacts in raster
images. They are often the
result from poor aliasing,
which happens when a JPG
signal reconstruction after
downsampling has
produced only high
frequency outcomes.
<-- JPG, Progressive
(irreversible databend)
<-- JPG, Progessive
(irreversible databend)
16
17. <-- JPG, Baseline standard
(irreversible databend)
<-- JPG, Baseline
(irreversible databend)
<-- Joint Photographic
JPEG 2000
Experts Group committee in
2000; low res JPEG 2000.
(irreversible databend)
The JPEG 2000 standard was
mainly developed because of
the many edge and blocking
artifacts of the JPG format.
JPEG 2000 has improved
scalability and edit-ability.
In JPG 2000, after the color
transformation step, the
image is split into so-called
tiles, rectangular regions of
the image that are
transformed and encoded
separately.
Tiles can be any size, and it is
also possible to consider the
whole image as one single
tile. This results into
a collection of 17
sub-bands which
represent several
approximation scales.
18. <-- JPEG 2000.
(irreversible databend)
A sub-band is a set of
coefficients that represent
aspects of the image
associated with a certain
frequency range as well as a
spatial area of the image.
The quantized sub-bands are
split further into precincts,
rectangular regions in the
wavelet domain.
A wavelet is a wave-like
oscillation with an amplitude
that starts out at zero,
increases, and then
decreases back to zero. It
can typically be visualized as
a "brief oscillation" like one
might see recorded by a
seismograph or heart
monitor.
Precincts are split further into
code blocks. Code blocks are
located in a single sub-band
and have equal sizes.
The chrominance
components can be, but do
not necessarily have to be,
down-scaled in resolution;
in fact, since the wavelet
transformation already
separates images into
scales, downsampling is
more effectively handled by
dropping the finest wavelet
scale.
<-- JPEG 2000
(irreversible databend)
<-- Low quality JPEG 2000
(irreversible databend)
18
19. <-- Quicktime movie (.mov)
MOVING IMAGE Compression: none
256 colors,
least quality
The low quality and the
compression to fewer colors
MOV introduced posterization
artifacts.
Posterization of an image
means the conversion of a
continuous gradation of
tones to several regions of
fewer tones, with abrupt
changes from one tone to
the another.
Posterization may be
deliberate or may be an
unintended artifact of color
quantization.
<-- Quicktime movie (.mov)
Compression: Animation
256 greys, least quality
codec
displaced scan lines and
posterization artifacts.
(I donʼt know why the scan
lines displaced themselves)
<-- Quicktime movie
(.mov) Compression:
Animation 1000 colors,
least quality codec
"Banding artifacts," or “false
contours” result from color
quantization within digitally
compressed images.
The lines in the image can
also be referred to as 10-bit
Banding artifacts.
19
20. <-- DV (Digital Video) tape
DV
rewind (shows macroblocks)
The artifact to the left shows
the macroblocks the footage
consists of. This artifact is
also referred to as screen
tearing (a video artifact that
appears when information
from two or more different
frames is shown a single
screen draw).
The lossy compressed video
image is framed
fundamentally different from
analog or RAW video footage.
First of all, the frames no
longer rely upon raw pixels.
Instead, macroblocks have
become one of the
elementary components of the
lossy compressed moving
image (at least under current
standard codecs).
Lossy compressed video
often depends on luminance
(brightness) and chrominance
(coloring) thresholds arranged
within 16x16 pixel (more or
less) macroblocks within the
keyframes (the I-frames) of an
image sequence. The
thresholds (or frequencies) of
chrominance and luminance
depend on an oscillating
cosine function (following
Fourier Transform).
The exploitation/bending of
this FFT and its chrominance
and luminance values shows
in the discolored macroblocks,
that now show the otherwise
mostly obscured macroblock
structures of digital video.
<-- DV, pro50 DV,
macroblock bend
(rearranged macroblock data)
“checkerboarding”
<-- DV pro50 DV,
compression: pro50 DV
bend luminance (brightness)
and chrominance (coloring)
values.
Re-organizations of color
(chrominance) structures.
The matrix of macroblocks is
still completely intact but the
chrominance values are off.
Also referred to as
“mosaicking”, “pixelating” or
“quilting” artifacts.
20
21. <-- Windows Media Video
WMV
(.wmv) compression: two
pass VBR Constrained
bitrate: 5kbps (datamoshed)
The handling of space and
time within video is
significantly different between
the linear analog or RAW
footage and lossy
compressed footage. Besides
the macroblock structures
within digital film material, the
footage is also no longer
based on a linear series of
discrete images (a
sequence); instead the video
consists of different kinds of
frames (I-frames or
reference/key frames, P-
frames or forward-predicted
frames and B-frames or bi-
directional frames), of which
only the keyframe possesses
a complete matrix of
macroblocks.
When a video is encoded,
each frame is stored as an I
frame (the keyframce) or a P/
B frame. An I frame is like a
JPEG image, it holds the still
image in its entirety. P and B
frames are the smart frames
that allow videos to be
compressed. They store only
the differences between the
current frame and the last
frame.
A motion prediction algorithm
is used to calculate the
difference between the P-
and B-frames. These frames
consist of motion vectors that
index only the difference in
position (the offset) of the
macroblocks between the
original and the next frame.
<-- Windows Media Video
(.wmv) compression: two
pass VBR Constrained
bitrate: 5kbps (datamoshed)
<-- Windows Media Video
(.wmv) compression: two
pass VBR Constrained
bitrate: 5kbps (datamoshed)
A recently popularized wave
of video artworks was based
on the deletion of keyframes
and the exploitation of the
vector motion of P-frames.
Which was also referred to
as“datamoshing”, “pixel
bleeding” or simply
“compression art”.
The effect you see in the
datamosh videos is what
happens when you store only
the differences between
frames, ie. when all I-Frame
or keyframe references
have been
deleted. 21
22. <-- Windows Media Video
(.wmv) Compression: two
pass VBR Constrained
bitrate: 5kbps (datamoshed)
In this type of compression,
artifacts tend to remain on
several generations of
decompressed frames, and
move with the optic flow of
the image, leading to a
peculiar effect, part way
between a painting effect r
smear that moves with
objects in the scene.
AVI
The DIVX file compression
follows a similar encoding
as the WMV compression.
And is more often used.
<-- AVI (.avi)
Compression: DIVX
(datamoshed)
<-- AVI (.avi)
Compression: DIVX
(datamoshed)
22
23. <-- AVI (.avi) compression:
none, 16 colors, least quality
(dither)
Color quantization performs a
multi-scale analysis on the
neighborhood of each pixel, to
determine the presence and
scale of banding artifacts, and
probabilistically dithers the color
of the pixel. The overall effect is
to "break down" the false
contours
<-- AVI (.avi) compression:
Cinepak, 256 colors
lowest quality
Color quantization
decreasing color depth
<-- AVI (.avi) compression:
none, black and white
best quality
The image shows a lot of
jaggies.
!
23
24. <-- AVI (.avi) compression:
none, 16 grays
least quality
<-- AVI Cinepak
compression 256 grays,
lowest quality.
Cinepak is my most favorite
compression. The
compression is based on
vector quantization, which
results in blocky artifacting
at low bitrates.
Cinepak divides a movie
into key images and intra-
coded images. Each image
is divided into a number of
horizontal bands which
have individual 256-color
palettes transferred in the
key images. Each band is
subdivided into 4x4 pixel
blocks. The compressor
uses vector quantization to
determine the one or two
band palette colors which
best match each block and
encodes runs of blocks as
either one color byte or two
color bytes plus a 16-bit
vector which determines
which pixel gets which color.
The cinepak compression is
not widely supported
anymore, and this is why,
when opening the image in
a player, the indexed colors
of the image are often
associated with an incorrect
color palette, which results
into very unexpected
colorations of the image.
<-- AVI Cinepak
compression, 256 grays,
lowest quality. Opened in
VLC player that
associated the
video with a 24
purple/black/turquoise
palette.