Kap 5

CHAPTER 5
Lossless Compression
Computers can handle many different kinds of information like text, equations,
games, sound, photos, and film. Some of these information sources require a
huge amount of data and may quickly fill up your hard disk or take a long time
to transfer across a network. For this reason it is interesting to see if we can
somehow rewrite the information in such a way that it takes up less space. This
may seem like magic, but does in fact work well for many types of information.
There are two general classes of methods, those that do not change the informa-
tion, so that the original file can be reconstructed exactly, and those that allow
small changes in the data. Compression methods in the first class are called loss-
less compression methods while those in the second class are called lossy com-
pression methods. Lossy methods may sound risky since they will change the
information, but for data like sound and images small alterations do not usually
matter. On the other hand, for certain kinds of information like for example text,
we cannot tolerate any change so we have to use lossless compression methods.
In this chapter we are going to study lossless methods; lossy methods will be
considered in a later chapter. To motivate our study of compression techniques,
we will first consider some examples of technology that generate large amounts
of information. We will then study two lossless compression methods in detail,
namely Huffman coding and arithmetic coding. Huffman coding is quite simple
and gives good compression, while arithmetic coding is more complicated, but
gives excellent compression.
In section 5.3.2 we introduce the information entropy of a sequence of sym-
bols which essentially tells us how much information there is in the sequence.
This is useful for comparing the performance of different compression strate-
gies.
73
74 CHAPTER 5. LOSSLESS COMPRESSION
5.1 Introduction
The potential for compression increases with the size of a file. A book typically
has about 300 words per page and an average word length of four characters. A
book with 500 pages would then have about 600 000 characters. If we write in
English, we may use a character encoding like ISO Latin 1 which only requires
one byte per character. The file would then be about 700 KB (kilobytes)1 , includ-
ing 100 KB of formatting information. If we instead use UTF-16 encoding, which
requires two bytes per character, we end up with a total file size of about 1300 KB
or 1.3 MB. Both files would represent the same book so this illustrates straight
away the potential for compression, at least for UTF-16 encoded documents.
On the other hand, the capacity of present day hard disks and communication
channels are such that a saving of 0.5 MB is usually negligible.
For sound files the situation is different. A music file in CD-quality requires
44 100 two-byte integers to be stored every second for each of the two stereo
channels, a total of about 176 KB per second, or about 10 MB per minute of
music. A four-minute song therefore corresponds to a file size of 40 MB and a
CD with one hour of music contains about 600 MB. If you just have a few CDs
this is not a problem when the average size of hard disks is approaching 1 TB
(1 000 000 MB or 1 000 GB). But if you have many CDs and want to store the
music in a small portable player, it is essential to be able to compress this in-
formation. Audio-formats like Mp3 and Aac manage to reduce the files down to
about 10 % of the original size without sacrificing much of the quality.
Not surprisingly, video contains even more information than audio so the
potential for compression is considerably greater. Reasonable quality video re-
quires at least 25 images per second. The images used in traditional European
television contain 576 × 720 small coloured dots, each of which are represented
with 24 bits2 . One image therefore requires about 1.2 MB and one second of
video requires about 31MB. This corresponds to 1.9 GB per minute and 112 GB
per hour of video. In addition we also need to store the sound. If you have more
than a handful of films in such an uncompressed format, you are quickly going
to exhaust the capacity of even quite large hard drives.
These examples should convince you that there is a lot to be gained if we
can compress information, especially for video and music, and virtually all video
formats, even the high-quality ones, use some kind of compression. With com-
pression we can fit more information onto our hard drive and we can transmit
information across a network more quickly.
1 Here we use the SI prefixes, see Table 4.1.

2 This is a digital description of the analog PAL system.
5.2. HUFFMAN CODING 75
00000000000000000000000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000111000000000000000000000
00000000000000000000000000000000000000000
Figure 5.1. A representation of a black and white image, for example part of a scanned text document.
5.1.1 Run-length coding

We have already seen that English text stored in UTF-16 encoding can be com-
pressed to at least half the number of bits by encoding in UTF-8 instead. Another
simple strategy that sometimes work well is run-length encoding. One example
is text documents scanned in black and white. This will produce an image of
each page represented by a two-dimensional array of a large number of inten-
sity values. At a point which is covered by text the intensity value will be 1, at
all other points the value will be 0. Since most pages of text contain much more
white than black, there will be long sequences of 0s in the intensity array, see
figure 5.1. A simple way to compress such a file is to replace a sequence of n
consecutive 0s by a code like n. Programs that read this file must then of course
know that this code is to be interpreted as n consecutive 0s. As long as the se-
quences of 0s are long enough, this can lead to significant compression of the
image file.
5.2 Huffman coding

The discussion in section 5.1 illustrates both the potential and the possibility of
compression techniques. In this section we are going to approach the subject in
more detail and describe a much used technique.
Before we continue, let us agree on some notation and vocabulary.
Definition 5.1 (Jargon used in compression). A sequence of symbols is called

a text and is denoted x = {x 1 , x 2 , . . . , x m }. The symbols are assumed to be taken
from an alphabet that is denoted A = {α1 , α2 , . . . , αn }, and the number of times
that the symbol αi occurs in x is called its frequency and is denoted by f (αi ).
For compression each symbol αi is assigned a binary code c(αi ), and the text
x is stored as the bit-sequence z obtained by replacing each character in x by
its binary code. The set of all binary codes is called a dictionary or code book.
If we are working with English text, the sequence x will just be a string of
letters and other characters like x = {h,e, l, l, o, , a, g, a, i, n, .} (the character after
’o’ is space, and the last character a period). The alphabet A is then the ordinary
Latin alphabet augmented with the space character, punctuation characters and
digits, essentially characters 32–127 of the ASCII table, see Table 4.2. In fact, the
ASCII codes define a dictionary since it assigns a binary code to each character.
However, if we want to represent a text with few bits, this is not a good dictionary
because the codes of very frequent characters are no shorter than the codes of
the characters that are hardly ever used.
In other contexts, we may consider the information to be a sequence of bits
and the alphabet to be {0, 1}, or we may consider sequences of bytes in which
case the alphabet would be the 256 different bit combinations in a byte.
Let us now suppose that we have a text x = {x 1 , x 2 , . . . , x m } with symbols taken
from an alphabet A . A simple way to represent the text in a computer is to
assign an integer code c(αi ) to each symbol and store the sequence of codes
{c(x 1 ), c(x 2 ), . . . , c(x m )}. The question is just how the codes should be assigned.
Small integers require fewer digits than large ones so a good strategy is to
let the symbols that occur most frequently in x have short codes and use long
codes for the rare symbols. This leaves us with the problem of knowing the
boundary between the codes. Huffman coding uses a clever set of binary codes
which makes it impossible to confuse the codes even though they have different
lengths.
Fact 5.2 (Huffman coding). In Huffman coding the most frequent symbols in
a text x get the shortest codes, and the codes have the prefix property which
means that the bit sequence that represents a code is never a prefix of any
other code. Once the codes are known the symbols in x are replaced by their
codes and the resulting sequence of bits z is the compressed version of x.
Example 5.3. This may sound a bit vague, so let us consider an example. Sup-
pose we have the four-symbol text x = DBACDBD of length 7. We note that
the different symbols occur with frequencies f (A) = 1, f (B ) = 2, f (C ) = 1 and

f (D) = 3. We use the codes
c(D) = 1, c(B ) = 01, c(C ) = 001, c(A) = 000. (5.1)
We can then store the text as
z = 1010000011011, (5.2)
altogether 13 bits, while a standard encoding with one byte per character would
require 56 bits. Note also that we can easily decipher the code since the codes
have the prefix property. The first bit is 1 which must correspond to a ’D’ since
this is the only character with a code that starts with a 1. The next bit is 0 and
since this is the start of several codes we read one more bit. The only character
with a code that start with 01 is ’B’ so this must be the next character. The next
bit is 0 which does not uniquely identify a character so we read one more bit. The
code 00 does not identify a character either, but with one more bit we obtain the
code 000 which corresponds to the character ’A’. We can obviously continue in
this way and decipher the complete compressed text.
Compression is not quite as simple as it was presented in example 5.3. A
program that reads the compressed code must clearly know the codes (5.1) in
order to decipher the code. Therefore we must store the codes as well as the
compressed text z. This means that the text must have a certain length before it
is worth compressing it.
5.2.1 Binary trees

The description of Huffman coding in fact 5.2 is not at all precise since it does
not state how the codes are determined. The actual algorithm is quite simple,
but requires a new concept.
Definition 5.4 (Binary tree). A binary tree T is a finite collection of nodes

where one of the nodes is designated as the root of the tree, and the remain-
ing nodes are partitioned into two disjoint groups T0 and T1 that are also trees.
The two trees T0 and T1 are called the subtrees or children of T . Nodes which
are not roots of subtrees are called leaf nodes. A connection from one node to
another is called an edge of the tree.
An example of a binary tree is shown in figure 5.2. The root node which is
shown at the top has two subtrees. The subtree to the right also has two subtrees,
both of which only contain leaf nodes. The subtree to the left of the root only has
one subtree which consists of a single leaf node.
Figure 5.2. An example of a binary tree.
5.2.2 Huffman trees

It turns out that Huffman coding can conveniently be described in terms of a
binary tree with some extra information added. These trees are usually referred
to as Huffman trees.
Definition 5.5. A Huffman tree is a binary tree that can be associated with an
alphabet consisting of symbols {αi }ni=1 with frequencies f (αi ) as follows:
1. Each leaf node is associated with exactly one symbol αi in the alphabet,
and all symbols are associated with a leaf node.
2. Each node has an associated integer weight:
(a) The weight of a leaf node is the frequency of the symbol.

(b) The weight of a node is the sum of the weights of the roots of the
node’s subtrees.
3. All nodes that are not leaf nodes have exactly two children.
4. The Huffman code of a symbol is obtained by following edges from the

root to the leaf node associated with the symbol. Each edge adds a bit
to the code: a 0 if the edge points to the left and a 1 if it points to the
right.
Example 5.6. In figure 5.3 the tree in figure 5.2 has been turned into a Huffman
tree. The tree has been constructed from the text CCDACBDC with the alphabet
{A,B,C,D} and frequencies f (A) = 1, f (B ) = 1, f (C ) = 4 and f (D) = 2. It is easy
to see that the weights have the properties required for a Huffman tree, and by
8
0 1
4 4
C 0 1
2 2
D
0 1
1 1
A B
Figure 5.3. A Huffman tree.
following the edges we see that the Huffman codes are given by c(C ) = 0, c(D) =
10, c(A) = 110 and c(B ) = 111. Note in particular that the root of the tree has
weight equal to the length of the text.
We will usually omit the labels on the edges since they are easy to remember:
An edge that points to the left corresponds to a 0, while an edge that points to
the right yields a 1.
5.2.3 The Huffman algorithm

In example 5.6 the Huffman tree was just given to us; the essential question is
how the tree can be constructed from a given text. There is a simple algorithm
that accomplishes this.
Algorithm 5.7 (Huffman algorithm). Let the text x with symbols {αi }ni=1 be
given, and let the frequency of αi be f (αi ). The Huffman tree is constructed
by performing the following steps:
1. Construct a one-node Huffman tree from each of the n symbols αi and

its corresponding weight; this leads to a collection of n one-node trees.
2. Repeat until the collection consists of only one tree:
(a) Choose two trees T0 and T1 with minimal weights and replace
them with a new tree which has T0 as its left subtree and T1 as
its right subtree.
3. The tree remaining after the previous step is a Huffman tree for the
given text x.
Most of the work in algorithm 5.7 is in step 2, but note that the number of
trees is reduced by one each time, so the loop will run at most n times.
The easiest way to get to grips with the algorithm is to try it on a simple
example.
Example 5.8. Let us try out algorithm 5.7 on the text ’then the hen began to eat’.
This text consists of 32 characters, including the five spaces. We first determine
the frequencies of the different characters by counting. We find the collection of
one-node trees
4 3 5 3 1 1 2 1 5
t h e n b g a o t
where the last character denotes the space character. Since ’b’ and ’g’ are two
characters with the lowest frequency, we combine them into a tree,
4 3 5 3 2 2 1 5
t h e n a o t
1 1
b g
The two trees with the lowest weights are now the character ’o’ and the tree we
formed in the last step. If we combine these we obtain
4 3 5 3 3 2 5
t h e n a t
2 1
o
1 1
b g
Now we have several choices. We choose to combine ’a’ and ’h’,
4 5 5 3 3 5
t e n t
3 2 2 1
h a o
1 1
b g
At the next step we combine the two trees with weight 3,
4 5 5 6 5
t e t
3 2 3 3
h a n
2 1
o
1 1
b g
Next we combine the ’t’ and the ’e’,
9 5 6 5
t
4 5 3 2 3 3
t e h a n
2 1
o
1 1
b g
We now have two trees with weight 5 that must be combined
9 10 6
4 5 5 5 3 3
t e t n
3 2 2 1
h a o
1 1
b g
25
10 15
5 5 6 9
t
3 2 3 3 4 5
h a n t e
2 1
o
1 1
b g
Figure 5.4. The Huffman tree for the text ’then the hen began to eat’.
Again we combine the two trees with the smallest weights,
10 15
5 5 6 9
t
3 2 3 3 4 5
h a n t e
2 1
o
1 1
b g
By combining these two trees we obtain the final Huffman tree in figure 5.4.
From this we can read off the Huffman codes as
c(h) = 000, c(b) = 10000, c(n) = 101,

c(a) = 001, c(g ) = 10001, c(t ) = 110,
c(t) = 01, c(o) = 1001, c(e) = 111.
so we see that the Huffman coding of the text ’then the hen began to eat’ is
110 000 111 101 01 110 000 111 01 000 111 101 01 10000
111 10001 001 101 01 110 1001 01 111 001 110
The spaces and the new line have been added to make the code easier to read;
on a computer these will not be present.
The original text consists of 25 characters including the spaces. Encoding
this with standard eight-bit encodings like ISO Latin or UTF-8 would require
400 bits. Since there are only nine symbols we could use a shorter fixed width
encoding for this particular text. This would require five bits per symbol and
would reduce the total length to 125 bits. In contrast the Huffman encoding
only requires 75 bits.
5.2.4 Properties of Huffman trees

Huffman trees have a number of useful properties, and the first one is the prefix
property, see fact 5.2. This is easy to deduce from simple properties of Huffman
trees.
Proposition 5.9 (Prefix property). Huffman coding has the prefix property:
No code is a prefix of any other code.
Proof. Suppose that Huffman coding does not have the prefix property, we will
show that this leads to a contradiction. Let the code c 1 be the prefix of another
code c 2 , and let n i be the node associated with the symbol with code c i . Then
the node n 1 must be somewhere on the path from the root down to n 2 . But then
n 2 must be located further from the root than n 1 , so n 1 cannot be a leaf node,
which contradicts the definition of a Huffman tree (remember that symbols are
only associated with leaf nodes).
We emphasise that it is the prefix property that makes it possible to use vari-
able lengths for the codes; without this property we would not be able to decode
an encoded text. Just consider the simple case where c(A) = 01, c(B ) = 010 and
c(C ) = 1; which text would the code 0101 correspond to?
In the Huffman algorithm, we start by building trees from the symbols with
lowest frequency. These symbols will therefore end up the furthest from the root
and end up with the longest codes, as is evident from example 5.8. Likewise, the
symbols with the highest frequencies will end up near the root of the tree and
therefore receive short codes. This property of Huffman coding can be quanti-
fied, but to do this we must introduce a new concept.
Note that any binary tree with the symbols at the leaf nodes gives rise to a
coding with the prefix property. A natural question is then which tree gives the
coding with the fewest bits?
Theorem 5.10 (Optimality of Huffman coding). Let x be a given text, let T

be any binary tree with the symbols of x as leaf nodes, and let `(T ) denote
the number of bits in the encoding of x in terms of the codes from T . If T ∗
denotes the Huffman tree corresponding to the text x then
`(T ∗ ) ≤ `(T ).
Theorem 5.10 says that Huffman coding is optimal, at least among coding
schemes based on binary trees. Together with its simplicity, this accounts for
the popularity of this compression method.
5.3 Probabilities and information entropy

Huffman coding is the best possible among all coding schemes based on binary
trees, but could there be completely different schemes, which do not depend
on binary trees, that are better? And if this is the case, what would be the best
possible scheme? To answer questions like these, it would be nice to have a way
to tell how much information there is in a text.
5.3.1 Probabilities rather than frequencies

Let us first consider more carefully how we should measure the quality of Huff-
man coding. For a fixed text x, our main concern is how many bits we need to
encode the text, see the end of example 5.8. If the symbol αi occurs f (αi ) times
and requires `(αi ) bits and we have n symbols, the total number of bits is
n
X
B= f (αi )`(αi ). (5.3)
i =1
However, we note that if we multiply all the frequencies by the same constant,
the Huffman tree remains the same. It therefore only depends on the relative
frequencies of the different symbols, and not the length of the text. In other
words, if we consider a new text which is twice as long as the one we used in
example 5.8, with each letter occurring twice as many times, the Huffman tree
would be the same. This indicates that we should get a good measure of the
quality of an encoding if we divide the total number of bits used by the length of
5.3. PROBABILITIES AND INFORMATION ENTROPY 85
the text. If the length of the text is m this leads to the quantity
n f (α )
i
`(αi ).
X
B̄ = (5.4)
i =1 m
If we consider longer and longer texts of the same type, it is reasonable to be-
lieve that the relative frequencies of the symbols would converge to a limit p(αi )
which is usually referred to as the probability of the symbol αi . As always for
probabilities we have ni=1 p(αi ) = 1.
P
Instead of referring to the frequencies of the different symbols in an alpha-

bet we will from now on refer to the probabilities of the symbols. We can then
translate the bits per symbol measure in equation 5.4 to a setting with probabil-
ities.
Observation 5.11 (Bits per symbol). Let A = {α1 , . . . , αn } be an alphabet

where the symbol αi has probability p(αi ) and is encoded with `(αi ) bits.
Then the average number of bits per symbol in a text encoded with this al-
phabet is
n
X
b̄ = p(αi )`(αi ). (5.5)
i =1
Note that the Huffman algorithm will work just as well if we use the prob-
abilities as weights rather than the frequencies, as this is just a relative scaling.
In fact, the most obvious way to obtain the probabilities is to just divide the fre-
quencies with the number of symbols for a given text. However, it is also pos-
sible to use a probability distribution that has been determined by some other
means. For example, the probabilities of the different characters in English have
been determined for typical texts. Using these probabilities and the correspond-
ing codes will save you the trouble of processing your text and computing the
probabilities for a particular text. Remember however that such pre-computed
probabilities are not likely to be completely correct for a specific text, particu-
larly if the text is short. And this of course means that your compressed text will
not be as short as it would be had you computed the correct probabilities.
In practice, it is quite likely that the probabilities of the different symbols
change as we work our way through a file. If the file is long, it probably contains
different kinds of information, as in a document with both text and images. It
would therefore be useful to update the probabilities at regular intervals. In the
case of Huffman coding this would of course also require that we update the
Huffman tree and therefore the codes assigned to the different symbols. This
may sound complicated, but is in fact quite straightforward. The key is that the
decoding algorithm must compute probabilities in exactly the same way as the
compression algorithm and update the Huffman tree at exactly the same posi-
tion in the text. As long as this requirement is met, there will be no confusion as
the compression end decoding algorithms will always use the same codes.
5.3.2 Information entropy

The quantity b̄ in observation 5.11 measures the number of bits used per symbol
for a given coding. An interesting question is how small we can make this num-
ber by choosing a better coding strategy. This is answered by a famous theorem.
Theorem 5.12 (Shannon’s theorem). Let A = {α1 , . . . , αn } be an alphabet

where the symbol αi has probability p(αi ). Then the minimal number of bits
per symbol in an encoding using this alphabet is given by
n
X
H = H (p 1 , . . . , p n ) = − p(αi ) log2 p(αi ).
i =1
where log2 denotes the logarithm to base 2. The quantity H is called the in-
formation entropy of the alphabet with the given probabilities.
Example 5.13. Let us return to example 5.8 and compute the entropy in this
particular case. From the frequencies we obtain the probabilities
c(t ) = 4/25, c(n) = 3/25, c(a) = 2/25,

c(h) = 3/25, c(b) = 1/25, c(o) = 1/25,
c(e) = 1/5, c(g ) = 1/25, c(t) = 1/5.
We can then compute the entropy to be H ≈ 2.93. If we had a compression al-

gorithm that could compress the text down to this number of bits per symbol,
we could represent our 25-symbol text with 74 bits. This is only one bit less than
what we obtained in example 5.8, so Huffman coding is very close to the best we
can do for this particular text.
Note that the entropy can be written as
n
X ¡ ¢
H= p(αi ) log2 1/p(αi ) .
i =1
If we compare this expression with equation (5.5) we see that a compression

strategy would reach the compression rate promised by the entropy if the length
5.4. ARITHMETIC CODING 87
of the code for the symbol αi was log2 1/p(αi ) . But we know that this is just the
¡ ¢
number of bits in the number 1/p(αi ). This therefore indicates that an optimal
compression scheme would represent αi by the number 1/p(αi ). Huffman cod-
ing necessarily uses an integer number of bits for each code, and therefore only
has a chance of reaching entropy performance when 1/p(αi ) is a power of 2 for
all the symbols. In fact Huffman coding does reach entropy performance in this
situation, see exercise 3.
5.4 Arithmetic coding

When the probabilities of the symbols are far from being fractions with pow-
ers of 2 in their denominators, the performance of Huffman coding does not
come close to entropy performance. This typically happens in situations with
few symbols as is illustrated by the following example.
Example 5.14. Suppose that we have a two-symbol alphabet A = {0, 1} with the
probabilities p(0) = 0.9 and p(1) = 0.1. Huffman coding will then just use the
obvious codes c(0) = 0 and c(1) = 1, so the average number of bits per symbol is
1, i.e., there will be no compression at all. If we compute the entropy we obtain
H = −0.9 log2 0.9 − 0.1 log2 0.1 ≈ 0.47.
So while Huffman coding gives no compression, there may be coding methods

that will reduce the file size to less than half the original size.
5.4.1 Arithmetic coding basics

Arithmetic coding is a coding strategy that is capable of compressing files to a
size close to the entropy limit. It uses a different strategy than Huffman coding
and does not need an integer number of bits per symbol and therefore performs
well in situations where Huffman coding struggles. The basic idea of arithmetic
coding is quite simple.
Idea 5.15 (Basic idea of arithmetic coding). Arithmetic coding associates se-
quences of symbols with different subintervals of [0, 1). The width of a subin-
terval is proportional to the probability of the corresponding sequence of
symbols, and the arithmetic code of a sequence of symbols is a floating-point
number in the corresponding interval.
To illustrate some of the details of arithmetic coding, it is easiest to consider

an example.
0 1
0 0.8 1
00 01 10 11
0.64 0.96
110 111
000 001 010 011 100 101
0.512 0.768 0.928 0.996
Figure 5.5. The basic principle of arithmetic coding applied to the text in example 5.16.
Example 5.16 (Determining an arithmetic code). We consider the two-symbol

text ’00100’. As for Huffman coding we first need to determine the probabilities
of the two symbols which we find to be p(0) = 0.8 and p(1) = 0.2. The idea is
to allocate different parts of the interval [0, 1) to the different symbols, and let
the length of the subinterval be proportional to the probability of the symbol. In
our case we allocate the interval [0, 0.8) to ’0’ and the interval [0.8, 1) to ’1’. Since
our text starts with ’0’, we know that the floating-point number which is going to
represent our text must lie in the interval [0, 0.8), see the first line in figure 5.5.
We then split the two subintervals according to the two probabilities again.
If the final floating point number ends up in the interval [0, 0.64), the text starts
with ’00’, if it lies in [0.64, 0.8), the text starts with ’01’, if it lies in [0.8, 0.96), the
text starts with ’10’, and if the number ends up in [0.96, 1) the text starts with ’11’.
This is illustrated in the second line of figure 5.5. Our text starts with ’00’, so the
arithmetic code we are seeking must lie in the interval [0, 0.64).
At the next level we split each of the four sub-intervals in two again, as shown
in the third line in figure 5.5. Since the third symbol in our text is ’1’, the arith-
metic code must lie in the interval [0.512, 0.64). We next split this interval in the
two subintervals [0.512, 0.6144) and [0.6144, 0.64). Since the fourth symbol is ’0’,
we select the first interval. This interval is then split into [0.512, 0.59392) and
[0.59392, 0.6144). The final symbol of our text is ’0’, so the arithmetic code must
lie in the interval [0.512, 0.59392).
We know that the arithmetic code of our text must lie in the half-open inter-
val [0.512, 0.59392), but it does not matter which of the numbers in the interval
we use. The code is going to be handled by a computer so it must be repre-
sented in the binary numeral system, with a finite number of bits. We know that
any number of this kind must be on the form i /2k where k is a positive inte-
ger and i is an integer in the range 0 ≤ i < 2k . Such numbers are called dyadic
numbers. We obviously want the code to be as short as possible, so we are look-

ing for the dyadic number with the smallest denominator that lies in the inter-
val [0.512, 0.59392). In our simple example it is easy to see that this number is
9/16 = 0.5625. In binary this number is 0.10012 , so the arithmetic code for the
text ’00100’ is 1001.
Example 5.16 shows how an arithmetic code is computed. We have done all
the computations in decimal arithmetic, but in a program one would usually use
binary arithmetic.
It is not sufficient to be able to encode a text; we must be able to decode
as well. This is in fact quite simple. We split the interval [0, 1] into the smaller
pieces, just like we did during the encoding. By checking which interval contains
our code, we can extract the correct symbol at each stage.
5.4.2 An algorithm for arithmetic coding

Let us now see how the description in example 5.16 can be generalised to a sys-
tematic algorithm in a situation with n different symbols. An important tool in
the algorithm is a function that maps the interval [0, 1] to a general interval [a, b].
Observation 5.17. Let [a, b] be a given interval with a < b. The function
g (z) = a + z(b − a)
will map any number z in [0, 1] to a number in the interval [a, b]. In particular
the endpoints are mapped to the endpoints and the midpoint to the midpoint,
a +b
g (0) = a, g (1/2) = , g (1) = b.
2
We are now ready to study the details of the arithmetic coding algorithm.
As before we have a text x = {x 1 , . . . , x m } with symbols taken from an alphabet
A = {α1 , . . . , αn }, with p(αi ) being the probability of encountering αi at any
given position in x. It is much easier to formulate arithmetic coding if we in-
troduce one more concept.
Definition 5.18 (Cumulative probability distribution). Let A = {α1 , . . . , αn } be

an alphabet where the probability of αi is p(αi ). The cumulative probability
distribution F is defined as
j
X
F (α j ) = p(αi ), for j = 1, 2, . . . , n.
i =1
The related function L is defined by L(α1 ) = 0 and
L(α j ) = F (α j ) − p(α j ) = F (α j −1 ), for j = 2, 3, . . . , n.
It is important to remember that the functions F , L and p are defined for the
symbols in the alphabet A . This means that F (x) only makes sense if x = αi for
some i in the range 1 ≤ i ≤ n.
The basic idea of arithmetic coding is to split the interval [0, 1) into the n
subintervals
£ ¢ £ ¢ £ ¢ £ ¢
0, F (α1 ) , F (α1 ), F (α2 ) , . . . , F (αn−2 ), F (αn−1 ) , F (αn−1 ), 1 (5.6)
£ ¢
so that the width of the subinterval F (αi −1 ), F (αi ) is F (αi ) − F (αi −1 ) = p(αi ).
If the first symbol is x 1 = αi , the arithmetic code must lie in the interval [a 1 , b 1 )
where
a 1 = p(α1 ) + p(α2 ) + · · · + p(αi −1 ) = F (αi −1 ) = L(αi ) = L(x 1 ),

b 1 = a 1 + p(αi ) = F (αi ) = F (x 1 ).
The next symbol in the text is x 2 . If this were the first symbol of the text,
£ ¢
the desired subinterval would be L(x 2 ), F (x 2 ) . Since it is the second symbol we
must map the whole interval [0, 1) to the interval [a 1 , b 1 ] and pick out the part
£ ¢
that corresponds to L(x 2 ), F (x 2 ) . The mapping from [0, 1) to [a 1 , b 1 ) is given by
g 2 (z) = a 1 + z(b 1 − a 1 ) = a 1 + zp(x 1 ), see observation 5.17, so our new interval is
h ¡ ¢ ¡ ¢´ £ ¢
[a 2 , b 2 ) = g 2 L(x 2 ) , g 2 F (x 2 ) = a 1 + L(x 2 )p(x 1 ), a 1 + F (x 2 )p(x 1 ) .
£ ¢
The third symbol x 3 would be associated with the interval L(x 3 ), F (x 3 ) if it
were the first symbol. To find the correct subinterval, we map [0, 1) to [a 2 , b 2 )
with the mapping g 3 (z) = a 2 + z(b 2 − a 2 ) and pick out the correct subinterval as
h ¡ ¢ ¡ ¢´
[a 3 , b 3 ) = g 3 L(x 3 ) , g 3 F (x 3 ) .
This process is then continued until all the symbols in the text have been pro-
cessed.
With this background we can formulate a precise algorithm for arithmetic
coding of a text of length m with n distinct symbols.
Algorithm 5.19 (Arithmetic coding). Let the text x = {x 1 , . . . , x m } be given,

with the symbols being taken from an alphabet A = {α1 , . . . , αn }, with prob-
abilities p(αi ) for i = 1, . . . , n. Generate a sequence of m subintervals of [0, 1):
£
1. Set [a 0 , b 0 ] = 0, 1).
2. For k = 1, . . . , m
(a) Define the linear function g k (z) = a k−1 + z(b k−1 − a k−1 ).
h ¡ ¢ ¡ ¢´
(b) Set [a k , b k ] = g k L(x k ) , g k F (x k ) .
The arithmetic code of the text x is the midpoint C (x) of the interval [a m , b m ),
i.e., the number
am + bm
,
2
truncated to l ¡ ¢m
− log2 p(x 1 )p(x 2 ) · · · p(x m ) + 1
binary digits. Here dwe denotes the smallest integer that is larger than or equal
to w.
A program for arithmetic coding needs to output a bit more information

than just the arithmetic code itself. For the decoding we also need to know ex-
actly which probabilities were used and the ordering of the symbols (this influ-
ences the cumulative probability function). In addition we need to know when
to stop decoding. A common way to provide this information is to store the
length of the text. Alternatively, there must be a unique symbol that terminates
the text so when we encounter this symbol during decoding we know that we
are finished.
Let us consider another example of arithmetic coding in a situation with a
three-symbol alphabet.
Example 5.20. Suppose we have the text x = {AC B BC A AB A A} and we want to
encode it with arithmetic coding. We first note that the probabilities are given
by
p(A) = 0.5, p(B ) = 0.3, p(C ) = 0.2,
so the cumulative probabilities are F (A) = 0.5, F (B ) = 0.8 and F (C ) = 1.0. This
means that the interval [0, 1) is split into the three subintervals
[0, 0.5), [0.5, 0.8), [0.8, 1).

The first symbol is A, so the first subinterval is [a 1 , b 1 ) = [0, 0.5). The second sym-
bol is C so we must find the part of [a 1 , b 1 ) that corresponds to C . The mapping
from [0, 1) to [0, 0.5) is given by g 2 (z) = 0.5z so [0.8, 1] is mapped to
£ ¢
[a 2 , b 2 ) = g 2 (0.8), g 2 (1) = [0.4, 0.5).
The third symbol is B which corresponds to the interval [0.5, 0.8). We map [0, 1)
to the interval [a 2 , b 2 ) with the function g 3 (z) = a 2 + z(b 2 − a 2 ) = 0.4 + 0.1z so
[0.5, 0.8) is mapped to
£ ¢
[a 3 , b 3 ) = g 3 (0.5), g 3 (0.8) = [0.45, 0.48).
Let us now write down the rest of the computations more schematically in a
table,
£ ¢
g 4 (z) = 0.45 + 0.03z, x 4 = B,[a 4 , b 4 ) = g 4 (0.5), g 4 (0.8) = [0.465, 0.474),
£ ¢
g 5 (z) = 0.465 + 0.009z, x 5 = C , [a 5 , b 5 ) = g 5 (0.8), g 5 (1) = [0.4722, 0.474),
£ ¢
g 6 (z) = 0.4722 + 0.0018z, x 6 = A, [a 6 , b 6 ) = g 6 (0), g 6 (0.5) = [0.4722, 0.4731),
£ ¢
g 7 (z) = 0.4722 + 0.0009z, x 7 = A, [a 7 , b 7 ) = g 7 (0), g 7 (0.5) = [0.4722, 0.47265),
£ ¢
g 8 (z) = 0.4722 + 0.00045z, x 8 = B, [a 8 , b 8 ) = g 8 (0.5), g 8 (0.8) = [0.472425, 0.47256),
g 9 (z) = 0.472425 + 0.000135z, x 9 = A,
£ ¢
[a 9 , b 9 ) = g 9 (0), g 9 (0.5) = [0.472425, 0.4724925),
g 10 (z) = 0.472425 + 0.0000675z, x 10 = A,
£ ¢
[a 10 , b 10 ) = g 10 (0), g 10 (0.5) = [0.472425, 0.47245875).
The midpoint M of this final interval is
M = 0.472441875 = 0.011110001111000111112 ,
and the arithmetic code is M rounded to

l ¢m
− log2 p(A)5 p(B )3 p(C )2 + 1 = 16
¡
bits. The arithmetic code is therefore the number
C (x) = 0.01111000111100012 = 0.472427,
but we just store the 16 bits 0111100011110001. In this example the arithmetic
code therefore uses 1.6 bits per symbol. In comparison the entropy is 1.49 bits
per symbol.
5.4.3 Properties of arithmetic coding

In example 5.16 we chose the arithmetic code to be the dyadic number with the
smallest denominator within the interval [a m , b m ). In algorithm 5.19 we have
chosen a number that is a bit easier to determine, but still we need to prove that
the truncated number lies in the interval [a m , b m ). This is necessary because
when we throw away some of the digits in the representation of the midpoint,
the result may end up outside the interval [a m , b m ]. We combine this with an
important observation on the length of the interval.
Theorem 5.21. The width of the interval [a m , b m ) is
b m − a m = p(x 1 )p(x 2 ) · · · p(x m ) (5.7)
and the arithmetic code C (x) lies inside this interval.
Proof. The proof of equation(5.7) is by induction on m. For m = 1, the length

is simply b 1 − a 1 = F (x 1 ) − L(x 1 ) = p(x 1 ), which is clear from the last equation in
Definition 5.18. Suppose next that
b k−1 − a k−1 = p(x 1 ) · · · p(x k−1 );
we need to show that b k − a k = p(x 1 ) · · · p(x k ). This follows from step 2 of algo-
rithm 5.19,
¡ ¢ ¡ ¢
b k − a k = g k F (x k ) − g k L(x k )
¡ ¢
= F (x k ) − L(x k ) (b k−1 − a k−1 )
= p(x k )p(x 1 ) · · · p(x k−1 ).
In particular we have b m − a m = p(x 1 ) · · · p(x m ).

Our next task is to show that the arithmetic code C (x) lies in [a m , b m ). Define
the number µ by the relation
1
= b m − a m = p(x 1 ) · · · p(x m ) or µ = − log2 p(x 1 ) · · · p(x m ) .
¡ ¢
2 µ
In general µ will not be an integer, so we introduce a new number λ which is the

smallest integer that is greater than or equal to µ,
l ¢m
λ = dµe = − log2 p(x 1 ) · · · p(x m ) .
¡
This means that 1/2λ is smaller than or equal to b m − a m since λ ≥ µ. Consider

the collection of dyadic numbers D λ on the form j /2λ where j is an integer in
am C (x) M bm
× ◦ × ×
k −1 2k − 1 k
2λ 2λ+1 2λ
am C (x) M bm
× × ◦ ×
k 2k + 1 k +1
2µ 2µ+1 2µ
Figure 5.6. The two situations that can occur when determining the number of bits in the arithmetic code.
the range 0 ≤ j < 2λ . At least one of them, say k/2λ , must lie in the interval
[a m , b m ) since the distance between neighbouring numbers in D λ is 1/2λ which
is at most equal to b m − a m . Denote the midpoint of [a m , b m ) by M . There are
two situations to consider which are illustrated in figure 5.6.
In the first situation shown in the top part of the figure, the number k/2λ
is larger than M and there is no number in D λ in the interval [a m , M ]. If we
form the approximation M̃ to M by only keeping the first λ binary digits, we
obtain the number (k − 1)/2λ in D λ that is immediately to the left of k/2λ . This
number may be smaller than a m , as shown in the figure. To make sure that the
arithmetic code ends up in [a m , b m ) we therefore use one more binary digit and
set C (x) = (2k −1)/2λ+1 , which corresponds to keeping the first λ+1 binary digits
in M .
In the second situation there is a number from D λ in [a m , M ] (this was the
case in example 5.16). If we now keep the first λ digits in M we would get C (x) =
k/2λ . In this case algorithm 5.19 therefore gives an arithmetic code with one
more bit than necessary. In practice the arithmetic code will usually be at least
thousands of bits long, so an extra bit does not matter much.
Now that we know how to compute the arithmetic code, it is interesting to

see how the number of bits per symbol compares with the entropy. The number
of bits is given by
l ¡ ¢m
− log2 p(x 1 )p(x 2 ) · · · p(x m ) + 1.
Recall that each x i is one of the n symbols αi from the alphabet so by properties
of logarithms we have
¡ ¢ Xn
log2 p(x 1 )p(x 2 ) · · · p(x m ) = f (αi ) log2 p(αi )
i =1
where f (αi ) is the number of times that αi occurs in x. As m becomes large we

know that f (αi )/m approaches p(αi ). For large m we therefore have that the
number of bits per symbol approaches
1l ¡ ¢m 1 1 ¡ ¢ 2
− log2 p(x 1 )p(x 2 ) · · · p(x m ) + ≤ − log2 p(x 1 )p(x 2 ) · · · p(x m ) +
m m m m
1 X n 2
=− f (αi ) log2 p(αi ) +
m i =1 m
n
X
≈− p(αi ) log2 p(αi )
i =1
= H (p 1 , . . . , p n ).
In other words, arithmetic coding gives compression rates close to the best pos-
sible for long texts.
Corollary 5.22. For long texts the number of bits per symbol required by the
arithmetic coding algorithm approaches the minimum given by the entropy,
provided the probability distribution of the symbols is correct.
5.4.4 A decoding algorithm

We commented briefly on decoding at the end of section 5.4.1. In this section
we will give a detailed decoding algorithm similar to algorithm 5.19.
We will need the linear function that maps an interval [a, b] to the interval
[0, 1], i.e., the inverse of the function in observation 5.17.
Observation 5.23. Let [a, b] be a given interval with a < b. The function
y −a
h(y) =
b−a
will map any number y in [a, b] to the interval [0, 1]. In particular the end-
points are mapped to the endpoints and the midpoint to the midpoint,
¡ ¢
h(a) = 0, h (a + b)/2 = 1/2, h(b) = 1.
Linear functions like h in observation 5.23 play a similar role in decoding

as the g k s in algorithm 5.19; they help us avoid working with very small inter-
vals. The decoding algorithm assumes that the number of symbols in the text is
known and decodes the arithmetic code symbol by symbol. It stops when the
correct number of symbols have been found.
Algorithm 5.24. Let C (x) be a given arithmetic code of an unknown text x

of length m, based on an alphabet A = {α1 , . . . , αn } with known probabilities
p(αi ) for i = 1, . . . , n. The following algorithm determines the symbols of the
text x = {x 1 , . . . , x m } from the arithmetic code C (x):
1. Set z 1 = C (x).
2. For k = 1, . . . , m
(a) Find the integer i such that L(αi ) ≤ z k < F (αi ) and set
£ ¢
[a k , b k ) = L(αi ), F (αi ) .
(b) Output x k = αi .
(c) Determine the linear function h k (y) = (y − a k )/(b k − a k ).
(d) Set z k+1 = h k (z k ).
The algorithm starts by determining which of the n intervals

£ ¢ £ ¢ £ ¢ ¢ £ ¢
0, F (α1 ) , F (α1 ), F (α2 ) , . . . , F (αn−2 ) , F (αn−1 ) , F (αn−1 ), 1
it is that contains the arithmetic code z 1 = C (x). This requires a search among
the cumulative probabilities. When the index i of the interval is known, we
know that x 1 = αi . The next step is to decide which subinterval of [a 1 , b 1 ) =
£ ¢
L(αi ), F (αi ) that contains the arithmetic code. If we stretch this interval out to
[0, 1) with the function h k , we can identify the next symbol just as we did with
the first one. Let us see how this works by decoding the arithmetic code that we
computed in example 5.16.
Example 5.25 (Decoding of an arithmetic code). Suppose we are given the arith-
metic code 1001 from example 5.16 together with the probabilities p(0) = 0.8
and p(1) = 0.2. We also assume that the length of the code is known, the proba-
bilities, and how the probabilities were mapped into the interval [0, 1]; this is the
typical output of a program for arithmetic coding. Since we are going to do this
manually, we start by converting the number to decimal; if we were to program

arithmetic coding we would do everything in binary arithmetic.
The arithmetic code 1001 corresponds to the binary number 0.10012 which
is the decimal number z 1 = 0.5625. Since this number lies in the interval [0, 0.8)
we know that the first symbol is x 1 = 0. We now map the interval [0, 0.8) and the
code back to the interval [0, 1) with the function
h 1 (y) = y/0.8.
We find that the code becomes
z 2 = h 1 (z 1 ) = z 1 /0.8 = 0.703125
relative to the new interval. This number lies in the interval [0, 0.8) so the second
symbol is x 2 = 0. Once again we map the current interval and arithmetic code
back to [0, 1) with the function h 2 and obtain
z 3 = h 2 (z 2 ) = z 2 /0.8 = 0.87890625.
This number lies in the interval [0.8, 1), so our third symbol must be a x 3 = 1. At
the next step we must map the interval [0.8, 1) to [0, 1). From observation 5.23
we see that this is done by the function h 3 (y) = (y −0.8)/0.2. This means that the
code is mapped to
z 4 = h 3 (z 3 ) = (z 3 − 0.8)/0.2 = 0.39453125.
This brings us back to the interval [0, 0.8), so the fourth symbol is x 4 = 0. This
time we map back to [0, 1) with the function h 4 (y) = y/0.8 and obtain
z 5 = h 4 (z 4 ) = 0.39453125/0.8 = 0.493164.
Since we remain in the interval [0, 0.8) the fifth and last symbol is x 5 = 0, so the
original text was ’00100’.
5.4.5 Arithmetic coding in practice

Algorithms 5.19 and 5.24 are quite simple and appear to be easy to program.
However, there is one challenge that we have not addressed. The typical symbol
sequences that we may want to compress are very long, with perhaps millions
or even billions of symbols. In the coding process the intervals that contain the
arithmetic code become smaller for each symbol that is processed which means
that the ends of the intervals must be represented with extremely high precision.
A program for arithmetic coding must therefore be able to handle arbitrary pre-
cision arithmetic in an efficient way. For a time this prevented the method from
being used, but there are now good algorithms for handling this. The basic idea
is to organise the computations of the endpoints of the intervals in such a way
that early digits are not influenced by later ones. It is then sufficient to only work
with a limited number of digits at a time (for example 32 or 64 binary digits). The
details of how this is done is rather technical though.
Since the compression rate of arithmetic coding is close to the optimal rate
predicted by the entropy, one would think that it is often used in practice. How-
ever, arithmetic coding is protected by many patents which means that you have
to be careful with the legal details if you use the method in commercial software.
For this reason, many prefer to use other compression algorithms without such
restrictions, even though these methods may not perform quite so well.
In long texts the frequency of the symbols may vary within the text. To com-
pensate for this it is common to let the probabilities vary. This does not cause
problems as long as the coding and decoding algorithms compute and adjust
the probabilities in exactly the same way.
5.5 Lempel-Ziv-Welch algorithm

The Lempel-Ziv-Welch algorithm is named after the three inventors and is usu-
ally referred to as the LZW algorithm. The original idea is due to Lempel and Ziv
and is used in the LZ77 and LZ78 algorithms.
LZ78 constructs a code book during compression, with entries for combina-
tions of several symbols as well as for individual symbols. If, say, the ten next
symbols already have an entry in the code book as individual symbols, a new
entry is added to represent the combination consisting of these next ten sym-
bols. If this same combination of ten symbols appears later in the text, it can be
represented by its code.
The LZW algorithm is based on the same idea as LZ78, with small changes
to improve compression further.
LZ77 does not store a list of codes for previously encountered symbol com-
binations. Instead it searches previous symbols for matches with the sequence
of symbols that are presently being encoded. If the next ten symbols match a
sequence 90 symbols earlier in the symbol sequence, a code for the pair of num-
bers (90, 10) will be used to represent these ten symbols. This can be thought of
as a type of run-length coding.
5.6 Lossless compression programs

Lossless compression has become an important ingredient in many different
contexts, often coupled with a lossy compression strategy. We will discuss this
5.6. LOSSLESS COMPRESSION PROGRAMS 99
in more detail in the context of digital sound and images in later chapters, but
want to mention two general-purpose programs for lossless compression here.
5.6.1 Compress
The program compress is a much used compression program on UNIX plat-
forms which first appeared in 1984. It uses the LZW-algorithm. After the pro-
gram was published it turned out that part of the algorithm was covered by a
patent.
5.6.2 gzip
To avoid the patents on compress, the alternative program gzip appeared in
1992. This program is based on the LZ77 algorithm, but uses Huffman coding
to encode the pairs of numbers. Although gzip was originally developed for
the Unix platform, it has now been ported to most operating systems, see www.
gzip.org.
Exercises
5.1 In this exercise we are going to use Huffman coding to encode the text
’There are many people in the world’, including the spaces.
a) Compute the frequencies of the different symbols used in the text.

b) Use algorithm 5.7 to determine the Huffman tree for the symbols.
c) Determine the Huffman coding of the complete text. How does the
result compare with the entropy?
5.2 We can generalise Huffman coding to numeral systems other than the bi-
nary system.
a) Suppose we have a computer that works in the ternary (base-3) nu-

meral system; describe a variant of Huffman coding for such ma-
chines.
b) Generalise the Huffman algorithm so that it produces codes in the
base-n numeral system.
5.3 In this exercise we are going to do Huffman coding for the text given by
x = {AB AC ABC A}.
a) Compute the frequencies of the symbols, perform the Huffman algo-

rithm and determine the Huffman coding. Compare the result with
the entropy.
b) Change the frequencies to f (A) = 1, f (B ) = 1, f (C ) = 2 and compare

the Huffman tree with the one from (a).
5.4 Recall from section 4.3.1 in chapter 4 that ASCII encodes the 128 most
common symbols used in English with seven-bit codes. If we denote the
alphabet by A = {αi }128
i =1
, the codes are
c(α1 ) = 0000000, c(α2 ) = 0000001, c(α3 ) = 0000010, ...

c(α127 ) = 1111110, c(α128 ) = 1111111.
Explain how these codes can be associated with a certain Huffman tree.
What are the frequencies used in the construction of the Huffman tree?
5.5 In this exercise we use the two-symbol alphabet A = {A, B }.
a) Compute the frequencies f (A) and f (B ) in the text
x = {A A A A A A AB A A}
and the probabilities p(A) and p(B ).

b) We want to use arithmetic coding to compress the sequence in (a);
how many bits do we need in the arithmetic code?
c) Compute the arithmetic code of the sequence in (a).
5.6 The four-symbol alphabet A = {A, B,C , D} is used throughout this exer-
cise. The probabilities are given by p(A) = p(B ) = p(C ) = p(D) = 0.25.
a) Compute the information entropy for this alphabet with the given
probabilities.
b) Construct the Huffman tree for the alphabet. How many bits per
symbol is required if you use Huffman coding with this alphabet?
c) Suppose now that we have a text x = {x 1 , . . . , x m } consisting of m sym-
bols taken from the alphabet A . We assume that the frequencies of
the symbols correspond with the probabilities of the symbols in the
alphabet.
How many bits does arithmetic coding require for this sequence and
how many bits per symbol does this correspond to?
d) The Huffman tree you obtained in (b) is not unique. Here we will fix
a tree so that the Huffman codes are
c(A) = 00, c(B ) = 01, c(C ) = 10, c(D) = 11.
Compute the Huffman coding of the sequence ’ACDBAC’.

5.6. LOSSLESS COMPRESSION PROGRAMS 101
e) Compute the arithmetic code of the sequence in (d). What is the

similarity with the result obtained with Huffman coding in (d)?
5.7 The three-symbol alphabet A = {A, B,C } with probabilities p(A) = 0.1,
p(B ) = 0.6 and p(C ) = 0.3 is given. A text x of length 10 has been encoded
by arithmetic coding and the code is 1001101. What is the text x?
5.8 We have the two-symbol alphabet A = {A, B } with p(A) = 0.99 and p(B ) =
0.01. Find the arithmetic code of the text
99 times
z }| {
A A A · · · A A A B.
5.9 The two linear functions in observations 5.17 and 5.23 are special cases of
a more general construction. Suppose we have two nonempty intervals
[a, b] and [c, d ], find the linear function which maps [a, b] to [c, d ].
Check that your solution is correct by comparing with observations 5.17
and 5.23.

Kap 5

Uploaded by

Copyright:

Available Formats

Kap 5

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

Kap 5

Uploaded by

Copyright:

Available Formats

CHAPTER 5

1 Here we use the SI prefixes, see Table 4.1.

5.1.1 Run-length coding

5.2 Huffman coding

Definition 5.1 (Jargon used in compression). A sequence of symbols is called

the different symbols occur with frequencies f (A) = 1, f (B ) = 2, f (C ) = 1 and

c(D) = 1, c(B ) = 01, c(C ) = 001, c(A) = 000. (5.1)

We can then store the text as

5.2.1 Binary trees

Definition 5.4 (Binary tree). A binary tree T is a finite collection of nodes

Figure 5.2. An example of a binary tree.

5.2.2 Huffman trees

2. Each node has an associated integer weight:

(a) The weight of a leaf node is the frequency of the symbol.

4. The Huffman code of a symbol is obtained by following edges from the

Figure 5.3. A Huffman tree.

5.2.3 The Huffman algorithm

1. Construct a one-node Huffman tree from each of the n symbols αi and

2. Repeat until the collection consists of only one tree:

Now we have several choices. We choose to combine ’a’ and ’h’,

At the next step we combine the two trees with weight 3,

Next we combine the ’t’ and the ’e’,

We now have two trees with weight 5 that must be combined

Again we combine the two trees with the smallest weights,

c(h) = 000, c(b) = 10000, c(n) = 101,

5.2.4 Properties of Huffman trees

Theorem 5.10 (Optimality of Huffman coding). Let x be a given text, let T

5.3 Probabilities and information entropy

5.3.1 Probabilities rather than frequencies

Instead of referring to the frequencies of the different symbols in an alpha-

Observation 5.11 (Bits per symbol). Let A = {α1 , . . . , αn } be an alphabet

5.3.2 Information entropy

Theorem 5.12 (Shannon’s theorem). Let A = {α1 , . . . , αn } be an alphabet

c(t ) = 4/25, c(n) = 3/25, c(a) = 2/25,

We can then compute the entropy to be H ≈ 2.93. If we had a compression al-

If we compare this expression with equation (5.5) we see that a compression

5.4 Arithmetic coding

H = −0.9 log2 0.9 − 0.1 log2 0.1 ≈ 0.47.

So while Huffman coding gives no compression, there may be coding methods

5.4.1 Arithmetic coding basics

To illustrate some of the details of arithmetic coding, it is easiest to consider

0.512 0.768 0.928 0.996

Example 5.16 (Determining an arithmetic code). We consider the two-symbol

numbers. We obviously want the code to be as short as possible, so we are look-

5.4.2 An algorithm for arithmetic coding

Definition 5.18 (Cumulative probability distribution). Let A = {α1 , . . . , αn } be

The related function L is defined by L(α1 ) = 0 and

L(α j ) = F (α j ) − p(α j ) = F (α j −1 ), for j = 2, 3, . . . , n.

a 1 = p(α1 ) + p(α2 ) + · · · + p(αi −1 ) = F (αi −1 ) = L(αi ) = L(x 1 ),

Algorithm 5.19 (Arithmetic coding). Let the text x = {x 1 , . . . , x m } be given,

A program for arithmetic coding needs to output a bit more information

[0, 0.5), [0.5, 0.8), [0.8, 1).

The midpoint M of this final interval is

and the arithmetic code is M rounded to

bits. The arithmetic code is therefore the number

C (x) = 0.01111000111100012 = 0.472427,

5.4.3 Properties of arithmetic coding

Theorem 5.21. The width of the interval [a m , b m ) is

b m − a m = p(x 1 )p(x 2 ) · · · p(x m ) (5.7)