MULTIMEDIA COMMUNICATION (18EC743)

MULTIMEDIA COMMUNICATION - 18EC743

MODULE – 3:
Text and Image Compression: Introduction, Compression principles, Text compression, Image
compression 8Hours

TEXT BOOK:

1. Multimedia Communications: Applications, Networks, Protocols and Standards, Fred Halsall,

Pearson Education, Asia, Second Indian reprint 2002.

REFERENCE BOOKS:

1. “Multimedia Fundamentals: Vol 1 - Media Coding, and Content Processing”, Ralf Steinmetz, Klara
Narstedt, Pearson Education, 2004.
 MULTIMEDIA COMMUNICATION (18EC743)

TEXT AND IMAGE COMPRESSION

INTRODUCTION:
In many Multimedia applications a technique known as compression is applied to the source
information prior to itstransmission.
Compression done to achieve 2 things:
1. Reduce the volume of information to be transmitted for text, fax, and images.
2. Reduce the bandwidth required for the transmission of speech, audio, and video.
COMPRESSION PRINCIPLES:
Compression algorithms based on the following compression principles:
1. Source encoders and destination decoders
2. Lossless and lossy compression
3. Entropy encoding
4. Source encoding
Source encoders and destination decoders:
❖ Prior to transmitting the source information relating to a particular multimedia application, a compression
algorithm is applied to it.
❖ In order for the destination to reproduce the original source information or, in some instances, a nearly
exact copy of it a matching decompression algorithm must be applied to it.
❖ The application of the compression algorithm is the main function carried out by the source encoder.
❖ The application of the decompression algorithm is the main function carried out by the destination decoder.
❖ In applications which involve two computers communicating with each other, the time required to perform
the compression and decompression algorithms is not always critical both algorithms are normally
implemented in software within the two computers as shown in figure below.

❖ An application which uses this approach is the compression of text and/or image files.

Lossless and lossy compression:

❖ Compression algorithms can be classified as being either lossless or lossy.
 MULTIMEDIA COMMUNICATION (18EC743)

Lossless compression algorithm:

❖ The aim is to reduce the amount of source information to be transmitted in such a way that, when the
compressed information is decompressed, there is no loss of information.
❖ Ex.: transfer of a text file over a network, since no, part of the source information is lost during either the
compression or decompression operations.
Lossless compression algorithm:
❖ The aim is normally not to reproduce an exact copy of the source information after decompression but
rather a version of its which is perceived by the recipient as a true copy.
❖ With such algorithms the higher level of compression being applied to the source information the more
approximate the received version becomes.
❖ Ex.: transfer of digitized images, audio and video streams. In such cases, the sensitivity of the human eye
or ear is such that any fine details that may be missing from the original source signal after decompression
are not detectable.

Entropy encoding:
It is lossless and independent of the type of information that is being compressed. Concerned
solely with how the information is represented.
Ex.: Run-length encoding and Statistical encoding.

Run-length encoding:

Typical applications: When source information comprises long substrings of the same character or
binarydigit.
Here, instead of transmitting source string in the form of independent codewords or bits source string is
transmitted in the form of different set of codewords whichindicates:
Particular character or bit being transmitted and
Number of characters/bits in thesubstring
Providing the destination to know the set of codewords being used destination simply, interprets each
codeword received and outputs the appropriate number of characters or bits.
Ex.: An application involves transmission of long strings of binary bits that comprise limited number of
substrings.
Each substring can be assigned a separate codeword, total bit string is then transmitted in the form of
the string of codewords selected from the codeword set.
Ex.: Transmission of the binary strings produced by the scanner in a facsimile machine.
When scanning typed documents in many instances scanner produces long substrings of either binary
0s or 1s.
Instead of transmitting directly, they are sent in the form of a string of codewords each indicating both
the bit 0 or 1 and the number of strings in the substring.
 MULTIMEDIA COMMUNICATION (18EC743)

Ex.: if the output of the scanner was 000000011111111110000011… then it can be represented as
(0,7) (1,10) (0,5)(1,2) …
Since, only the two binary digits 0 and 1 are involved and if, first substring always comprises binary 0s
then, the string could be represented as 7, 10, 5, 2 …
To send this in the digital form, individual decimal digits would be sent in their binary form.
Assuming the fixed number of bits per codeword number of bits per codeword would be determined by
the largest possiblesubstring.

Statistical encoding:
Applications use a set of codewords to transmit the sourceinformation.
Ex.: Set of ASCII codewords are often used for the transmission of strings of characters.
Normally, all the codewords in the set comprise a fixed number of binary bits.
Ex.: 7 bits in the case of ASCII.
In many applications: Symbols (characters) and hence codewords that are present in the source
information do not occur with the same frequency of occurrence, that is with equal probability.
Ex.: Consider a string of text in which character A may occur more frequently than character P which again,
will occur more frequently than character Z, and so on.
Statistical encoding exploits this property by using the set of variable length codewords, with the shortest
codewords used to represent the most frequently occurring symbols.
Use of variable-length codewords is not quite as straight-forward. Destination must know the set of
codewords being used by the source as in run-length encoding.
When variable-length codewords used, in order for the decoding operation to be carried out correctly. It
is necessary to ensure that a shorter codeword in the set does not form the start/beginning of a longer
codeword otherwise; the decoder will interpret the string on the wrong codeword boundaries.
Prefix property codeword set are used, which avoids the above happening.
Ex.: Huffman encoding algorithm.
Source entropy: Theoretical minimum average number of bits that are required to transmit a particular
source stream and is given by Entropy,

Efficiency of a encoding scheme:

ratio of the entropy of the source to the average number of bits per codeword that are required with
the scheme
 MULTIMEDIA COMMUNICATION (18EC743)
Source encoding:
Uses a particular property of the source information to produce an alternative form of representation
which are:
1. Compressed version of the original form
2. Is more amenable to the application of compression.

Differential encoding:
Used extensively in applications where, the amplitude of a value or signal covers a large range but
Difference in amplitude between successive values/signals is relatively small.
To exploit this property of the source information: A set of smaller codewords can be used each of which
indicates only difference in amplitude between the current value/signal being encoded and the
immediately preceding value/signal instead of using a relatively large codewords to represent the
amplitude of each value/signal.
Ex.: If the digitization of analog signal requires say, 12 bits to obtain the required dynamic range but, the
maximum difference in amplitude between successive samples of thesignal requires only 3 bits then, by
using only the difference values a saving of 75% on transmission bandwidth can be saved.
Differential encoding can be:
lossless or lossy.
Selection of the mode (lossy or lossless) depends on the number of bits used to encode the difference
values.
Can be lossless if the number of bits used is sufficient to cater for the maximum difference value.

Can be lossy if the number of bits used is not sufficient to cater for the maximum difference value.

Transform encoding:
Involves transforming the source information from one form into another form ,which lending it to more
readily to the application of compression.
In general, there is no loss of information associated with the transformation operation.
Applications: image and video applications.
Ex.: Digitization of a continuous tone monochromatic image produces a 2-D matrix of pixel values each of
which represents the level of gray in a particular position of the image.
Scenario: Moving from one position in the matrix to the next magnitude of each pixel value can be varying.
So, if a scan if performed across the pixel locations rate of change in magnitude will vary from zero, if all the
pixel values remain the same to a low rate of change if, say one half is different from the next half through
to a high rate of change if each pixel magnitude changes from one location to the next as shown in the figure
below.
Spatial frequency: It is the rate of change in magnitude as one traverses the matrix.

For any particular image there will be a mix of different spatial frequencies whose amplitudes are
determined by the related changes in magnitude of the pixels.
 MULTIMEDIA COMMUNICATION (18EC743)

In practice Human eye is less sensitive to the higher spatial frequency components associated with an
image than, the lower frequency components. If the amplitude of the higher frequency components falls
below a certain amplitude threshold they will not be detected by the eye.

Thus, in terms of compression: If we can transform the original spatial form of representation
into an equivalent representation involving spatial frequency components then, more readily
identifying and eliminating those higher frequency components which the eye cannot detect can
be done. It reduces the volume of information to be transmitted without degrading the perceived
quality of the original image.
DCT (Discrete Cosine Transform):
Transformation of 2-D matrix of pixel values into an equivalent matrix of spatial
frequency components can be carried out from this mathematical technique.
Transformation using DCT is lossless apart, from small rounding errors in the mathematics.
Once the equivalent matrix of spatial frequency components known as coefficients has
been derived then any frequency components in the matrix whose amplitude is less than
a defined threshold can be dropped it is at this point DCT operation, will become lossy. Fig.
shows – the basic principle behind the DCT.
 MULTIMEDIA COMMUNICATION (18EC743)

Text Compression:
3 different types of text unformatted, formatted and hypertext are all represented as strings of
characters selected from a defined set.
The strings comprise alphanumeric characters which are interspersed with additional control
character.
Different types of text use and interpret the latter in different ways.
Compression algorithm associated with text must be lossless.
Due to the above reason we use entropy encoding in particular, statistical encoding methods.
Two types of Statistical encoding methods are used with text:
➢ Use single character as the basis of deriving an optimum set of codewords.
▪ Ex.: Huffman and arithmetic coding algorithm.
Use variable length strings of characters.
▪ Ex.: Lempel-Ziv (LZ)algorithm.
Two types of coding are used with Text:

Static coding:
➢ Intended for applications in which the text to be compressed has known characteristics in
terms of the characters used and their relative frequencies of occurrence.
➢ Using above an optimum set of variable length codewords (instead of fixed length) is
derived with the shortest codewords used to have more frequently occurring characters.
➢ Resulting set of codewords are then used for all subsequent transfers involving this
 MULTIMEDIA COMMUNICATION (18EC743)

type of text.

Dynamic coding(adaptive coding):

➢ Intended for more general applications type of text transferring may vary from one
transfer to another.
➢ In this case, optimum set of codewords also likely to vary from one transfer to another.
➢ Codeword set used to transfer a particular text string is derived as the transfer takes place
this can be done by building up knowledge of both the characters that are present in the
text and their relative frequency of occurrence dynamically as the characters are being
transmitted.
➢ Codewords used change as a transfer takes place in such a way that receiver is able to
dynamically compute the same set of codewords that are being used at each point during
a transfer.

Static Huffman coding:

Character string to be transmitted is first analyzed for character types and their relative frequency
determined.
Coding operation involves creating an unbalanced tree with some branches shorter than others.
Degree of imbalance is a function of relative frequency of occurrence of the characters larger
the spread, more unbalanced the tree.
Resulting tree obtained from above process is known as Huffman code tree.

Huffman code tree:

Basics:
It is a binary tree with branches assigned a value 0 of1.
Base of the tree is normally, the geometric top in practice known as root node.
Point at which the branch divides is known branch node.
Leaf node is a termination point of a branch for which the symbols being encoded are assigned.
Figure Below is an Example of Huffman code tree corresponding to string of characters –
AAAABBCD
 MULTIMEDIA COMMUNICATION (18EC743)

As each branch divides binary value 0 & 1 are assigned to each new branch a 0 for left and 1 for
right.
Codewords used for each character (shown in leaf nodes) are got by tracing the path from root
node to each leaf node and forming a string of binary values associated with each branch traced.
Set of codes associated with this tree is 4*1+2*2+1*3+1*3=14bits is required to transmit the
complete string – AAAABBCD.
Figure below shows Huffman tree with information frequency of occurrence of each characters
here characters listed in a column in decreasing weight order.

Tree construction:
Huffman tree can be constructed from the set of codewords, the actual codewords, however
start at the root and not the leaf node.
Optimality check for tree (hence for Codewords):
1. Listing the resulting weights of all the leaf and branch nodes in the tree starting with the
smallest weight and proceeding from left to right and from bottom to top.
 MULTIMEDIA COMMUNICATION (18EC743)

2. Tree is optimal if the resulting list increments in weight order.

Each character in its encoded form has a variable number of bits; received bitstream must be
interpreted (decoded) in a bit-oriented way rather than on fixed 7/8 bit boundaries.
Shorter codeword will never form the start of a longer codeword which is a prefix property.
Prefix property: received bitstream can be decoded by carrying out a recursive search bit by bit
until each valid codeword is found.

Decoding Algorithm:
Figure below shows the flowchart ofa decoding algorithm.
Assumption: Table of codewords are available at the receive and also had corresponding ASCII
codeword.
Received bit stream is held in the variable BIT-STREAM.
Variable CODEWORD - used to hold the bits in each codeword - while it is being constructed.
From the flow chart once the codeword is identified corresponding ASCII codeword is

written into the variable RECEIVE_BUFFER.

Procedure - repeats until all the bits in the received string have been processed.
Figure below shows example of the decoded string.
 MULTIMEDIA COMMUNICATION (18EC743)

Huffman code tree and hence, codewords varies for different sets of characters being transmitted
for the receiver to perform the decoding operation, it must know the codewords relating to the
data being transmitted can be done in 2 ways:
1. Codewords relating to the next set of data are sent before the data is transmitted.
▪ Advantage: adaptive compression since, Codewords can be changed to suit the
type of data being transmitted.
▪ Disadvantage: overhead of having to send the new set of codewords (and
characters corresponding) whenever a new type of data is to be sent.
2. Receiver knows in advance what codewords are being used.
▪ Here the receiver to have one or more different sets of codewords and Sender to
indicate to the receiver through an agreed message which codeword set to use for
the next set of data.
▪ Ex.: word processor and bibliography in text books frequency of occurrence of
characters is used to construct the Huffman code tree for alphabet. If this data is
being sent transmitter and receiver automatically use this set of codewords

The Huffman code tree corresponding to the derived set of code-words is given in Figure
 MULTIMEDIA COMMUNICATION (18EC743)

below and, as we can see, this is the optimum tree since all leaf and branch nodes
increment in numerical order.

(a) Average number of bits per codeword using Huffman coding is:
2(2×0.25) + 2(3×0.14) +4(4×0.055) = 2.72 bits per codeword which is 99.8% of the Shannon
value.
Using fixed-length binary codewords:
There are 8 characters –A through H – and hence 3 bits per codeword is sufficient which is 90.7%
of the Huffman value.
Using 7-bit ASCII codewords:
7 bits per codeword which is 38.86% of the Huffman value.

Dynamic/ Adaptive Huffman Coding:

Huffman coding requires prior statistical knowledge about the information source i.e, both
transmitter and receiver should know the table of codewords relating to the data being
transmitted.
Such information is often not available and it is particularly true in multimedia application, where
future data is unknown before its arrival which is a disadvantage associated with dynamic
Huffman coding.
Ex: Live audio and video streaming.
The above can be solved by using dynamic Huffman coding, where the transmitter and receiver
build the Huffman tree and hence codeword table dynamically as the characters are being
transmitted/received.
With the method, if the character to be transmitted is currently present in the tree its codeword
is determined and send in the normal way.
 MULTIMEDIA COMMUNICATION (18EC743)
If the character is not present, if it is first occurrence it is transmitted in its uncompressed form.

The encoder updates its Huffman tree either by incrementing the frequency of occurrence of
the transmitted character or by introducing the new character into the tree.
Procedure:

ENCODER: DECODER:
Initial_code (); initial_code();
While not EOF while not EOF
{ {
Get(c); decode(c);
Encode(c); output(c);
Update_tree(c); update_tree(c);
} }

Initial_code assign symbols with some initially agreed upon codes, without any prior knowledge
of the frequency counter for them.
Update_tree is a procedure for constructing an adaptive Huffman tree. It does two things: it
increments the frequency counts for the symbols and updates the configuration of the tree.
The Huffman tree must always maintain its “Sibling Property”- that is, all nodes are arranged in
the order from Left to Right, Bottom to Top.
❖ If the Sibling Property is about to be violated to update the tree by rearranging the
nodes.
➢ When a swap is necessary, the farthest node with count N is swapped with the node
whose count has just been increased to N+1. Note that if the node with count
N is not a leaf node- it is the root of a subtree the entire subtree will go with it during the
swap.
The encoder and decoder must use exactly the same Initial_code and update_tree routines.
Ex: AADCCDD
 MULTIMEDIA COMMUNICATION (18EC743)

To improve the implementation of the algorithm, we adopt an additional rule: if any

character/symbol is to be sent the first time, it must be preceded by a special symbol NEW. The
initial code for NEW is 0. The count for NEW is always kept as 0.

ADVANTAGES:
1) Saving in transmission bandwidth start only when characters begin to repeat themselves.
Savings with text files can be significant.
2) Used in number of common applications that involves the transmission of text.
MULTIMEDIA COMMUNICATION (18EC743)
MULTIMEDIA COMMUNICATION (18EC743)
Arithmetic Coding:
Huffman coding achieves the Shannon value only if the character/symbol probabilities are all
integer powers of 1/2.
In contrast, the codeword produced using arithmetic coding always achieve the Shannon value.
It yields a Single Codeword for each encoded string of characters however Huffman coding
yields a single codeword for each encoded characters in a string.
Arithmetic coding, however, is more complicated than Huffman coding.
Algorithm:
BEGIN:
 Low = 0.0; high 1.0; range = 1.0; While (Symbol != terminator)
{
get (Symbol);
Low = low + range x range_low (Symbol); High = low + range x range_high (Symbol); range
= high – low;
}
Output a code so that low ≤ Code < high; END

A message is represented by a half-open interval [a,b) here a & b are real numbers between 0 &
1.
Initially the interval is [0,1). When the message becomes longer, the length of the interval
shortens and the number of bits needed to represent the interval increases.
At the end of each character string making up a message, a known character is sent which, in this
example, is a period (.) . When this is decoded at the receiving side, the decoder interprets this
as the end of the string/message.
To illustrate how the coding operation takes place, consider the transmission of a message
comprising a string of characters with probabilities

e = 0.3, n = 0.3, t = 0.2, w = 0.1, . = 0.1

The word needed to be transmitted is went.

Calculations:
High=low+(range × range_high(symbol))

Encoded version of the character string went. Is a single codeword in the range
Problems on Arithmetic Coding:

1. Illustrate the operation of arithmetic coding, consider the transmission of a message comprising a
string of characters with probabilities as given below:

e= 0.3, n= 0.3, t= 0.2, w=0.1, .= 0.1

The word needed to be transmitted is ‘etnw.’ Solution:
Formula: High= low + [range * range_high(symbol)]

1) To transmit ‘e’, range= 0.3 - 0.0= 0.3

e = 0 + (0.3 * 0.3) = 0.09

n = 0.09 + (0.3 * 0.3) = 0.18
t = 0.18 + (0.3 * 0.2) = 0.24
w = 0.24 + (0.3 * 0.1) = 0.27
. = 0.27 + (0.3 * 0.1) = 0.3

2) To transmit ‘t’, range= 0.24 - 0.18= 0.06

e = 0.18 + (0.06 * 0.3) = 0.198

n = 0.198 + (0.06 * 0.3) = 0.216
t = 0.216 + (0.06 * 0.2) = 0.228
w = 0.228 + (0.06 * 0.1) = 0.234
. = 0.234 + (0.06 * 0.1) = 0.24
 3) To transmit ‘n’, range= 0.216 - 0.198= 0.018

e = 0.198 + (0.018* 0.3) = 0.2034

n = 0.2034 + (0.018 * 0.3) = 0.2088
t = 0.2088 + (0.018* 0.2) = 0.2124
w = 0.2124 + (0.018* 0.1) = 0.2142
. = 0.2142 + (0.018* 0.1) = 0.216

4) To transmit ‘w’, range= 0.2142 - 0.2124= 0.0018

e = 0.2124 + (0.0018* 0.3) = 0.21294

n = 0.21294 + (0.0018* 0.3) = 0.21348
t = 0.21348 + (0.0018* 0.2) = 0.21384
w = 0.21384 + (0.0018* 0.1) = 0.21402
. = 0.21402 + (0.0018* 0.1) = 0.2142

• Since we encounter period (.), the transmission is stopped.

• Therefore, the codeword for the string ‘etnw.’ should lie in the range given below: 0.2124 ≤
codeword ≥ 0.2142
• Any value sent between the above range will be interpreted as the string ‘etnw.’
Decoding Process:
• The decoder knows the set of characters that are present in the encoded messages it receives as
well as the segment to which each characters has been assigned and its related range (i.e.,
probability distribution).
• Taking this string as a starting point, the decoder can follow the same procedure as that followed
by the character string relating to each received codeword.
• Ex: say the received codeword = 0.81611.
• Seeing the above output decoder interprets 1st character as ‘w’ since it is in the range of 0.8 to
0.9.
• It then expands the above interval further and determines that the 2 nd character is ‘e’ as 0.8161
as it is in the range 0.8 to 0.833.
• The decoder repeats the process until it decodes the terminator character then it stops.

Precision Issue:
• In this coding, number of decimal digits in the final codeword increases linearly with the
number of character in the string to be encoded.
• Maximum number of characters in a string is determined by the precision with which floating
numbers are represented in the source and destination computers.
• Thus, a complete message may be first fragmented into multiple smaller strings.
• Each string is then encoded separately and the resulting set of codewords is sent as a block of
floating-point number each in a known format.

Lempel-Ziv coding:
• Advantage:
➢ The Lempel-Ziv (LZ) compression algorithm, instead of using single characters as the basis
of the coding operation, uses strings of characters.
▪ Ex., for the compression of text, a table containing all the possible character strings
of words that occur in the text to be transferred is held by both the encoder and
decoder.
• Operation:
➢ As each word occurs in the text, instead of sending the word as a set of individual ASCII
codewords, the encoder sends only the index of where the word is stored in the table and,
on receipt of each index, the decoder uses this to access the corresponding word/string
of characters from the table and proceeds to reconstruct the text into its original form.
➢ Thus the table is used as a dictionary and the LZ algorithm is known as dictionary-based
compression algorithm.
• Scenario:
➢ Most word processing packages have a dictionary associated with them which is
 used for both spell checking and for the compression of text. Typically, they contain 25000
words and hence 15bits which has 32768 combinations are required to encode the index.
➢ To send the word “multimedia” with such a dictionary would require just 15 bits
instead of 70 bits with 7-bit ASCII codewords. This results in a compression ratio of 4.7:1.
➢ From above we can analyze that shorter words will have a lower compression ratio and
longer words have a higher ratio.
• Requirement:
1. A copy of the dictionary is held by both the encoder and the decoder.
2. Above is acceptable for the transmission of text which has been created using a standard
word-processing package.
• Limitation:
➢ It can be relatively inefficient if the text to be transmitted comprises only a small subset
of the words stored in the dictionary.
• Solution for limitation:
➢ Hence a variation of the LZ algorithm has been developed which allows the dictionary to
be built up dynamically by the encoder and decoder as the compressed text is being
transferred. In this way, the size of the dictionary is often a better match to the number
of different words in the text being transmitted than if a standard dictionary was used.

Lempel-Ziv-Welsh coding:
• Principle of operation:
➢ Here encoder and decoder will build the contents of the dictionary dynamically as the
text is being transferred.
➢ Initially, the dictionary held by both the encoder and decoder contains only the
character set example ASCII that has been used to create the text.
➢ The remaining entries in the dictionary are then built up dynamically by both the
encoder and decoder and contain the words that occur in the text.
• Scenario:
➢ Let the character set comprises 128 characters and the dictionary is limited to, say, 4096
entries, then the first 128 entries would contain the single characters that make up the
character set and the remaining 3968 entries would each contain strings of two or more
characters that make up the words in the text being transferred.
➢ The more frequently the words stored in the dictionary that occur in the text, the higher
the level of compression.
➢ In order to describe how the dictionary is built up, let us assume that the text to be
compressed starts with the string:
 This is simple as it is…..
➢ Since the idea is for the dictionary to contain only words, then only strings of characters
that consist of alphanumeric characters are stored in the dictionary and all the other
characters in the set are interpreted as word delimiters.
➢ Initially, the dictionary held by both the encoder and decoder contains only the individual
characters from the character set being used; for example, the 128 characters in the ASCII
character set.
➢ The first word in the example text is sent by the encoder using the index of each of the
four characters T,h,i and s.
➢ When the encoder reads the next character from the string the first space (SP) character
it determines that this is not an alphanumeric character. It therefore transmits the
character using its index as before but, in addition, interprets it as terminating the first
word and hence stores the preceding four characters in the next available (free) location
in the dictionary.
➢ In the same way decoder, on receipt of the first five indices/codewords, reads the
character stored at each index and commences to reconstruct the text.
➢ When it determines that the fifth character is a space character, it interprets this as a word
delimiter and proceeds to store the word This in its dictionary.
➢ The same procedure is followed by both the encoder and decoder for transferring the
other words except the encoder, prior to sending each word in the form of single
characters, first checks to determine if the word is currently stored in its dictionary and, if
it is, it sends only the index for the word.
➢ Similarly the decoder, since it also has the word stored in its dictionary, uses the index to
access the string of characters that make up the word.
➢ Thus with the example text string, after the space character following the second
occurrence of the word is, the contents of the dictionary held by both the encoder and
the decoder will be as shown in Figure below. As we can see, since this is the
second occurrence of the word is, it is transferred using only the index of where it is stored
in the dictionary (129).
• Accuracy Assessment:
➢ As we can deduce from this example, a key issue in determining the level of compression
that is achieved, is the number of entries in the dictionary since this, in turn, determines
the number of bits that are required for the index.
➢ With a static dictionary, the number of entries is fixed and, for the example we identified
earlier, a dictionary containing 25000 words requires 15 bits to encode the index.
➢ When building the dictionary dynamically, however, the question arises as to how many
entries should be provided for the dictionary.
➢ Clearly, if too few entries are provided then the dictionary will contain only a subset of
the words that occur in the text while if too many are provided, then it will contain empty
spaces which, in turn, makes the index unnecessarily long.
➢ In order to optimize the number of bits used for the index, at the commencement of each
transfer the number of entries is set to a relatively low value but, should the available space
become full, then the number of entries is allowed to increase incrementally.
➢ E.x., in an application that uses 128 characters in the basic character set, then both the
encoder and decoder would start with, say, 256 entries in the dictionary.
➢ This requires an index/codeword length of 8 bits and the dictionary would provide
 space for the 128 characters in the character set and a further 128 locations for words
that occur in the text.
➢ When this number of locations becomes insufficient, on detecting this, both the encoder
and decoder would double the size of their dictionary to 512 locations.
➢ Clearly, this necessitates an index length of 9 bits and so from this point, the encoder uses
9-bit codewords.
➢ However, since the decoder had also doubled the size of its own directory, it expects 9-
bit codewords from this point.
➢ In this way, the number of entries in the dictionary more accurately reflects the number
of different words in the text being transferred and hence optimizes the number of bits
used for each index/codeword.
➢ The procedure is shown in diagrammatic form in figure 3.8(b).
➢ In this example it is assumed that the last entry in the existing table at location 255 is the
word fish and the next word in the text that is not currently in the dictionary is pond.

Algorithm (Compression):
BEGIN
s = next input character; while not EOF
{ c = next input character;
if s + c exists in the dictionary s = s + c;
else
{ output the code for s;
add string s + c to the dictionary with a new code; s = c;
}
}
output the code for s; END
Problem:
1. Compress the following string “ABABBABCABABBA” using LZW Algorithm.
Solution:
Let’s start with a very simple dictionary (also referred to as a “string table”), initially containing
only 3 characters, with codes as follows:

Code String

1 A
 2 B

3 C

Now if the input string is “ABABBABCABABBA”, the LZW compression algorithm works as follows:

S C Output Code string

1 A

2 B

3 C

---------------------------------------------------------------------------------------------------------------------

A B 1 4 AB

B A 2 5 BA

A B

AB B 4 6 ABB

B A

BA B 5 7 BAB

B C 2 8 BC

C A 3 9 CA

A B

AB A 4 10 ABA

A B

AB B

ABB A 6 11 ABBA

A EOF 1

The output codes are: 1 2 4 5 2 3 4 6 1. Instead of sending 14 characters, only 9 codes need to be sent
(compression ratio = 14/9 = 1.56).
ALGORITHM (Decompression):
BEGIN s = NIL;
while not EOF
{
k = next input code;
entry = dictionary entry for k; output entry;
if (s != NIL)
add string s + entry[0] to dictionary with a new code; s = entry;
} END
Problem:
2. Input codes to the decoder are 1 2 4 5 2 3 4 6 1. Decode the string using LZW algorithm. The initial
string table is identical to what is used by the encoder.
Solution:

S k Output Code string

1 A

2 B

3 C

NIL 1 A

A 2 B 4 AB

B 4 AB 5 BA

AB 5 BA 6 ABB

BA 2 B 7 BAB

B 3 C 8 BC

C 4 AB 9 CA

AB 6 ABB 10 ABA
 ABB 1 A 11 ABBA

A EOF

• Apparently, the output string is “ABABBABCABABBA”, a truly lossless result!

3. Compress the following string “ABCBCABCEAB” using LZW Algorithm.
Let’s start with a very simple dictionary (also referred to as a “string table”), initially containing
only 3 characters, with codes as follows:

Code String

1 A

2 B

3 C

4 E

Solution: Now if the input string is “ABCBCABCEAB”, the LZW compression algorithm works as
follows:

S C Output Code String

1 A

2 B

3 C

4 E

--------------------------------------------------------------------------------------------------------------------

A B 1 5 AB

B C 2 6 BC

C B 3 7 CB

B C
 BC A 6 8 BCA

A B

AB C 5 9 ABC

C E 3 10 CE

E A 4 11 EA

A B

AB EOF 5

• The output codes are: 1 2 3 6 5 3 4 5. Instead of sending 11 characters, only 8 codes need to be
sent (compression ratio = 11/8 = 1.375).
4. Compress the following string “MYMEMYMO” using LZW Algorithm
Solution: Let’s start with a very simple dictionary (also referred to as a “string table”), initially
containing only 4 characters, with codes as follows:

Code String

1 M

2 Y

3 E

4 O

Now if the input string is “MYMEMYMO”, the LZW compression algorithm works as follows:

S C Output Code String

1 M

2 Y

3 E

4 O
M Y 1 5 MY

Y M 2 6 YM

M E 1 7 ME

E M 3 8 EM

M Y

MY M 5 9 MYM

M O 1 10 MO

O EOF 4

The output codes are: 1 2 1 3 5 1 4. Instead of sending 8 characters, only 6 codes need to be sent
(compression ratio = 8/6 = 1.33).

Decoding:

• Input codes to the decoder are 1 2 1 3 5 1 4 . Decode the string using LZW algorithm.

S k Output Code String

1 M

2 Y

3 E

4 O

NIL 1 M

M 2 Y 5 MY

Y 1 M 6 YM
M 3 E 7 ME

E 5 MY 8 EM

MY 1 M 9 MM

M 4 O 10 MO

O EOF

The output string is “ MYMEMYMO ”.

Image Compression:
• Images can be of two basic types:
1. Computer-generated (also known as graphical) images and
2. Digitized images (of both documents and pictures).
• Both types are displayed (and printed) in the form of a two-dimensional matrix of individual
picture elements, but a graphical image is represented differently in the computer file system.
• As, this is in the form of a program this type of representation requires considerably less memory
(and hence transmission bandwidth) than the corresponding matrix of picture elements,
whenever possible, graphics are transferred across a network in this form.
• Digitized documents and pictures, however, once digitized, the only form of representation is as
a two-dimensional matrix of picture elements.
• With respect to compression:
1. When transferring graphical images which are represented in their program form, a
lossless compression algorithm must be used.
2. When the created image/graphic is to be transferred across the network in its bit- map
form, then this is normally compressed prior to its transfer.
• To transfer digitized images two types of compression algorithm normally are employed:
1. A combination of run-length and statistical encoding. It is lossless and is used for the
transfer of the digitized documents generated by scanners such as those used in facsimile
machines.
2. A combination of transform, differential, and run-length encoding and has been
developed for the compression of both bitonal and color digitized pictures.
Graphics interchange format (GIF):
• The graphics interchange format (CIF) is used extensively with the Internet for the representation
and compression of graphical images.
• Color images comprising 24-bit pixels are supported 8 bits each for R. G, and B — GIF reduces the
number of possible colors that are present by choosing the 256 colors from the original set of 224
colors that match most closely those used in the original image.
• The resulting table of colors therefore consists of 256 entries, each of which contains a 24- bit
color value.
• Thus instead of sending each pixel as a 24-bit value, only the 8-bit index to the table entry that
contains the closest match color to the original is sent. This results in a compression ratio-of 3:1.
• The table of colors can relate either to:
1. Whole image — in which case it is referred to as the global color table
2. To a portion of the image, when it is referred to as a local color table.
• The contents of the table are sent across the network together with the compressed image data
and other information such as the screen size and aspect ratio in a standardized format.
• The principles of the scheme are shown in Figure below

• From the figure above we see that, the LZW coding algorithm can be used to obtain further levels
of compression.
• This works by extending the basic color table dynamically as the compressed image data is being
encoded and decoded. As long strings of the same color — are detected, these are entered into
the color table after the 256 selected colors.
• In this application, since each entry in the color table comprises 24 bits, in order to save memory,
to represent each string of pixel values just the corresponding string of 8-bit indices to the basic
color table are used.
• Normally, since the basic table contains 256 entries, an initial table size of 512 entries is selected
which allows for up to 256 common strings to be stored.
• As with text compression, however, should more strings be found, then the number of
 entries in the table is allowed to increase incrementally by extending the length of the index by
1 bit as shown below.

• GIF also allows an image to be stored and subsequently transferred over the network in an
interlaced Mode.
• This can be useful when transferring images over either low bit rate channels or the Internet
which provides a variable transmission rate.
• With this mode, the compressed image data is organized so that the decompressed image is built
up in a progressive way as the data arrives.
• To achieve this, the compressed data is divided into four groups as shown in Figure below and, as
we can see, the first contains 1/8 of the total compressed image data, the second a further 1/8,
the third a further 1/4, and the last the remaining 1/2.
Tagged image file format (TIFF):
• It supports pixel resolutions of up to 48 bits —16 bits each for R, G, and B.
• Is intended for the transfer of both images and digitized documents.
• The image data can be stored and hence transferred over the network in a number of different
formats.
• The particular format being used is indicated by a code number and these range from the
uncompressed format (code number 1) through to LZW-compressed which is code number 0Code
numbers 2, 3, and 4 are intended for use with digitized documents.
• Use the same compression algorithms that are used in facsimile machines.
• The LZW compression algorithm that is. used is the same as that used with GIF. It starts with a
basic color table containing 256 colors and the table can be extended to contain up to 4096
entries containing common strings of pixels in the image being transferred.
• A standard format is used for the transfer of both the color table and the compressed image data.

JPEG:
• JPEG is an acronomy for Joint Photographic Expert Group.
• It is the widely adapted standards relating to the compression of digitized pictures.
• There are 5 main stages they are:
1. Image/block preparation
2. Forward DCT
3. Quantization
4. Entropy Encoding
5. Frame Building
• Developed by ISO and ITU-T for United Nation
• Figure below shows Encoder schematic
Image/block preparation:
• The source image is made of one or more 2-D matrices of values.
• For continuous tone monochrome image, a single 2-D matrix is required to store the set of 8-bit
grey level values that represent the images.
• For color image, if a CLUT is used just a single matrix of values is required.
• If the image is represented in an R, G, B format three matrices are required, one each for the R,
G, and B quantized values.
• For color images the alternative form of representation known as Y, Cb, Cr can optionally be used.
• The above representation in turn allows the two matrices that contain the digitized chrominance
components to be smaller in size than the Y matrix so producing a reduced form of representation
over that of R, G, B form.

• The four alternative forms of representation are as shown in figure above.

• Once the source image format has been selected and prepared, the set of values in each matrix
are compressed separately using the DCT.
• But before performing the DCT on each matrix, another step known as block preparation is
carried out.
• It would be too time consuming to compute the DCT of the total matrix in a single step so each
matrix is first divided into a set of smaller 8 x 8 sub matrices.
 • Each is known as block and is fed sequentially to the DCT which transforms each block
separately as shown below.

Forward DCT:
Transforms the pixel information from spatial domain to frequency domain with the DCT.

• Each pixel value is quantized using 8 bits which produces a value in the range 0 to 255 for the
intensity values — R, G, B or Y.
• The value in the range —128 to +127 for the two chrominance values— Cb and Cr.

• To compute DCT, however, all the values are centered around zero by subtracting 128 from
each intensity values.
• Then, if the input 2-D matrix is represented by : P[x,y] and the transform matrix is F[i,j],then
DCT of each 8 x 8 block f values is computed by:

F[i, j]= C(i)C(j) Cos Cos

for i , j=0
Here C(i) and C(j) =
 =1 for all other values of i and x, y, i , and j all vary from 0 through 7.

• Consider a typical image say of 640 x480 pixels.

• If block size is 8 x 8 pixels, then image will have 80 x 60 or 4800 block each of which, for a screen
width of say, 16 inches (400 mm), will occupy a square of only 0.2x0.2 inches (5x 5 mm).
• Those regions of picture that contains a single color will generate a set of transformed blocks all
of which will have firstly, the same DC coefficient and secondly AC coefficient with them.
• Also only those areas of picture which contain color transition that will generate a set of
transformed blocks with different DC coefficient and a large number of AC coefficients with them.
• The above features are exploited in the quantization and entropy encoding phases of the
compression algorithm.

Quantization:
• The values from the DCT are initially floating point; they are change to integers by quantization.
• During the DCT phase there is very little loss of information the main source of information loss
occurs during the quantization and entropy encoding stages where the compression takes place.
• Human eye responds primarily to the DC coefficient and the lower spatial frequency coefficients
i.e. having higher magnitude volume.
• If the magnitude of a higher frequency coefficient is below a certain threshold, the eye will not
detect it.
• The above property is exploited in the quantization phase by dropping those spatial frequency
coefficients in the transformed matrix whose amplitudes are less than a defined threshold value.
• Quantization process aims to reduce size of the DC and AC coefficients so that less bandwidth is
required for their transmission.
• For the above point a division operation is performed using the defined threshold value as the
divisor.
• If the resulting (rounded) quotient is zero, the coefficient is less than the threshold value. If it is
non-zero, this indicates the number of times the coefficient is greater than the threshold rather
than its absolute value.
• The sensitivity of the eye varies with spatial frequency, which implies that the amplitude
threshold below which the eye will detect a particular spatial frequency also varies.
• Threshold values used vary for each of the 64 DCT coefficients which are held in a two-
dimensional matrix known as the quantization table with the threshold value to be used with a
particular DCT coefficient in the corresponding position in the matrix.
• The choice of threshold values is important and, in practice, is a compromise between the level
of compression that is required and the resulting amount of information loss that is acceptable.
• JPEG standard includes two default quantization table values:
1. One for use with the luminance coefficients and
2. The other for use with the two sets of chrominance coefficients.
• It also allows for customized tables, to be used and sent with the compressed image.
• An example set of threshold values is given in the quantization table shown in Figure below
together with a set of DCT coefficients and their corresponding quantizedvalues.
 • Number of points can be concluded from the values shown in the tables:
1. The computation of the quantized coefficients involves rounding the quotients to the
nearest integer value.
2. The threshold values used, in general, increase in magnitude with increasing spatial
frequency.
3. The DC coefficient in the transformed matrix is largest.
4. Many of the higher-frequency coefficients are zero.
• It is the last two points that are exploited during the encoding stage.

Entropy encoding:
➢ It has 4 stages:
1. Vectoring
2. Differential encoding
3. Run-length encoding
4. Huffman encoding.
Vectoring:
• Since the output of the quantization stage is a 2-D matrix of values, in order to apply any entropy
encoding to the set of values in the matrix, we must first represent the values in the form of a 1-
D vector and this operation is known as vectoring.
• In vectoring operation in order to exploit the presence of the large number of zeroes in the
quantization matrix, a zig-zag scan of the matrix is used as shown in the figure below.
 • With this type of scan, the DC coefficient and lower frequency AC coefficients both horizontal
and vertical are scanned first.
• All the high frequency coefficients are in sequential order so making this form of representation
more suitable for compression.
• Two differential encoding schemas are applied in parallel to the values in the vector.
1. Differential encoding to DC-coefficient only
2. Run-length encoding to AC-coefficient

Differential encoding:
• In this stage only the difference in magnitude of the DC coefficient in a quantized block
relatively to the value in the preceding block is encoded.
• In this way number of bits required to encode DC coefficient is reduced significantly.
• Ex: if the sequence of DC coefficients in consecutive quantized blocks-one per block- were
12,13,11,11,10,…..
• Corresponding difference amount will be, 12, 1,-2, 0,-1…
• The difference values are then encoded in the form (SSS,value)
• Where,
SSS = number of bits needed to encode the value. Value = actual bits that represent the value.
 Value SSS Encoded value

12 4 1100[ binary
form]

1 1 1

-2 2 01[contemporary
form]

0 0

1 1 0

Run-length encoding:
• The remaining 63 values in the vector are the AC coefficients and, because of the zig-zag scan,
the vector contains long strings of zeros within it.
• To exploit this feature, the AC coefficients are encoded in the form of a string of pairs of values.
Each pair is made up of (skip, value)
where
skip = number of zeros in the run and value = next non-zero coefficient.
• Hence the 63 values in the vector would be encoded as: (0,6) (0,7) (0,3) (0,3) (0,3) (0,2) (0,2) (0,2)
(0,2) (0,0)
• Note that the final pair (0,0) indicates the end of the string for this block and that all the remaining
coefficients in the block are zero. Also, that the value field is encoded in the form SSS/ value.

Huffman encoding :
• Huffman coding is used to encode the output of both the differential and run-length encoders.
• For the differential-encoded DC coefficients in the block, the bits in the SSS field are not sent in
their unsigned binary form but are sent in a Huffman-encoded form. This is done so that the bits
in the SSS field have the prefix property and this enables the decoder to determine
unambiguously the first SSS field from the received encoded bitstream.
• For each of the run-length encoded AC coefficients in the block. The bits that make up the skip
and SSS fields are treated as a single (composite) symbol and this is then encoded using either the
default table of Huffman codewords or a table of codewords that is sent with the encoded
bitstream.
• This is done so that the strings of encoded composite symbols all have the prefix property
 so that the decoder can interpret the received bitstream on the correct coefficient boundaries.
• To enable the decoder to discriminate between the skip and SSS fields, each combination of the
two fields is encoded separately and the composite symbol is then replaced by the equivalent
Huffman codeword.
• Because of the use of variable-length codewords in the various parts of the entropy encoding
stage, this is also known as the variable-length coding (VLC) stage.

Frame building:
• In order for the decoder in the remote computer to be able to interpret all the different fields
and tables that make up the bitstream, it is necessary to delimit each field and set of table values
in a defined way. The JPEG standard, therefore, includes a definition of the structure of the total
bitstream relating to a particular image/picture. This is known as a frame and its outline structure
is shown in Figure below.

Set of Huffman CW’s for the block

• The role of the frame builder is to encapsulate all the information relating to an encoded
image/picture in this format and, the structure of a frame is hierarchical.
• At the top level, the complete frame-plus-header is encapsulated between a start-of-frame and
an end-of-frame delimiter which allows the receiver to determine the start and end of all the
information relating to a complete image/picture.
• The frame header contains a number of fields that include:
1. The overall width and height of the image in pixels
2. The number and type of components that are used to represent the image (CLUT,
 R/G/B, Y/Cb/Cr )
3. The digitization format used (4:2:2, 4:2:0 etc.).
• At the second level, a frame consists of a number of components each of which is known as a
scan. These are also preceded by a header which contains fields that include:
1. The identity of the components (R/G/B etc.)
2. The number of bits used to digitize each component
3. The quantization table of values that have been used to encode each component.
• Typically, each scan/component comprises one or more segments each of which can contain a
group of (8 x 8) blocks preceded by a header. This contains the Huffman table of values that have
been used to encode each block in the segment should the default tables not be used. In this
way, each segment can be decoded independently of the others which overcome the possibility
of bit errors propagating and affecting other segments.
• Hence each complete frame contains all the information necessary to enable the JPEG decoder
to identify each field in a received frame and then perform the corresponding decoding
operation.

JPEG decoding:
• JPEG decoder is made up of a number of stages which are simply the corresponding decoder
sections of those used in the encoder.
• The time to carry out the decoding function is similar to that used to perform the encoding.
• On receipt of the encoded bitstream the frame decoder first identifies the control information
and tables within the various headers.
• It then loads the contents of each table into the related table and passes the control information
to the image builder.
• It then starts to pass compressed bitstream to the Huffman decoder which carries out the
corresponding decompression operation using either the default or the preloaded table of
codewords.
• The two decompressed streams containing the DC and AC coefficients of each block are then
passed to the differential and run-length decoders respectively. The resulting matrix of values is
then dequantized using either the default or the preloaded values in the quantization table.
• Each resulting block of 8 x 8 spatial frequency coefficients is passed in turn to the
inverse DCT which transforms them back into their spatial form using the expression:

P[x,y]= ∑ ∑ C(i) C(j) F[i,j] cos cos

 Where C (i) and C (j) = 1/ for i, j= 0
= 1 for all other values of i and j.
• The image builder then reconstructs the original image from these blocks using the control
information passed to it by the frame decoder.
Compression ratio:
1. JPEG standard is relatively complicated owing to the number of encoding/decoding stages
associated with it, compression ratios in excess of 20:1 can be obtained while still retaining a
good quality output image/picture. This level of compression, however, applies to pictures
whose content is relatively simple - that is, have relatively few color transitions and
2. For more complicated pictures, compression ratios nearer to 10:1 are more common.

• Finally, as with the GIF, it is also possible to encode and rebuild the image the image in a
progressive way by first sending an outline of the image and progressively adding more detail to
it. This can be achieved in the following ways:
1. Progressive mode:
In this mode, first the DC and low-frequency coefficients of each block are sent and
then the higher-frequency coefficients;
2. Hierarchical mode:
In this mode, the total image is first sent using a low resolution - for example 320
x 240 - then at a higher resolution such as 640 x 480.