Chapter Six

Image compression
 Image Compression
Compression Fundamentals
•Data compression refers to the process of reducing the amount of data required to
represent given quantity of information.
•Data refers to the means by which the information is conveyed.
•Various amounts of data can represent the same amount of information.
•Sometimes the given data contains some data which has no relevant information, or
restates/repeats the known information.
•It is thus said to contain data redundancy.
•Data redundancy is the central concept in image compression and can be
mathematically defined.
Image Compression
Compression Fundamentals
• Information versus Data

REDUNDANT
DATA
IMAGE

INFORMATION

IMAGE = INFORMATION + REDUNDANT DATA

 Image Compression
Compression Fundamentals
•Given n1 and n2 denoting the information-carrying units in two data sets that represent the same
information/image.
•The Relative data redundancy RD of the first data set, n1 ,is defined by:
1
R  1C
D
R

•CR refers to the compression ratio and is defined C R  n1

by: n2

•If n1 = n2 , then CR=1 and RD=0, indicating that the first representation of the information contains no

redundant data.

• A typical compression ratio around 10 or(10:1) indicates that 90% (RD=0.9) of the data in the first data set is

redundant.
 Image Compression
Data Redundancy: There are three main data redundancies used in image compression.
•Coding redundancy
•Interpixel redundancy
•Psychovisual redundancy
•Coding Redundancy : A code is a system of symbols (i.e. bytes, bits) that represents information.
•Each piece of information is represented by a set of code symbols.
•The gray level histogram of an image can be used in construction of codes to reduce the data used to
represent it. Given the normalized histogram of a gray level image where,
n
pr (rk )  nk k  0,1, 2,...., L
1
•rk is the pixel values defined in the interval [0,1] and pr(k) is the probability of occurrence of rk.

• L is the number of gray levels. nk is the number of times that kth gray level appears and n is the total number of
Image Compression
•Coding Redundancy
•Average number of bits required to represent each pixel is given by:

L1
  l(rk ) pr
Lavg k
k 1 (r )

•Where, l(rk) is the number of bits used to represent each value of rk.
•An 8 gray level image has the following gray level distribution.

8 gray levels Fixed 3-bit code Variable length code

Image Compression
•Coding Redundancy
•The average number of bit used for fixed 3-bit code:

•The average number of bits used for variable-length code in this particular example:

•The compression ratio:

1
•The relative Data Redundancy: RD   0.099 ~
1 1.11 %10
 Image Compression
•Coding Redundancy
•In this example the suggested variable-length coding gets rid of the ~10% redundant data of the fixed 3-bit code.

•The following graph shows the relationship between the histogram of an image, pr(rk) and l2(rk) which are inversely

proportional.
•The shortest code words are assigned to the most frequent (high probability) gray levels
•The longest code words are assigned to the least frequent (low probability) gray levels
•Data compression is achieved by assigning fewer bits to more probable gray levels than the less probable gray levels.
 Image Compression
•Interpixel Redundancy
•This type of redundancy is related with the interpixel correlations within an image.
•Much of the visual contribution of a single pixel is redundant and can be guessed from the values of its

neighbors.

•Given a 1024x343 binary image

•Consider a line crossing the image at line 100

•The respective line of 1024 bits can be represented by the Run-length

code given at the bottom.

•Note that in this line there is 8 regions that are 1 or 0 with the specified

run-length.
•Total of 11 bits (1 bit for thresholded value and 10 bit for the run length)

Line 100: (1,63)(0,87)(1,37)(0,5)(1,4)(0,556)(1,62)(0,210) can be used to represent each of these 8 neighborhoods.

 Image Compression
•Interpixel Redundancy
•The resulting compression ratio and respective relative redundancy is given by:
1-bit for each pixel

C R  (1024)(343)(1)  2.63
(12166)
(11)
Determined by thresholding each line and counting the run-length regions

1
RD  
1 2.63 0.62
•The relative redundancy is %62 obtained only by using correlation among the pixels (interpixel

dependencies) in the given image.

•This method can be extended to gray level images.
 Image Compression
•Psychovisual Redundancy

•Certain information has relatively less importance for the quality of image perception.

•This information is said to be psychovisually redundant.

•Unlike coding and interpixel redundancies, the psychovisual redundancy is related with the real/quantifiable

visual information.

• Its elimination results a loss of quantitative information. However psychovisually the loss is negligible.

•Removing this type of redundancy is a lossy process and the lost information cannot be recovered.

•The method used to remove this type of redundancy is called quantization which means the mapping of a

broad range of input values to a limited number of output values.

 Image Compression
•Psychovisual Redundancy
•The following example shows how an 8-bit image can be reduced to 4-bit image.

8-bit image 4-bit image Uniform quantization 4-bit image IGS quantization No
Undesired contouring effect contouring effect
 Image Compression
•Psychovisual Redundancy
•The improved gray-scale quantization (IGS) is one of the possible quantization procedures
and summarized in the following table.

Add

Add

•The IGS Quantization Procedure:

•Add the Least Significant Nibble (4 bits) of the previous sum to the current 8-bit pixel value.

•If the MSN of a given 8-bit pixel is 11112 than add zero instead.
•Declare the Most Significant Nibble of the sum to be the 4-bit IGS code.
Image Compression
•Quality Measure of a Compressed Image (Fidelity Criteria):
•The removal of psychovisual redundancy removes real/quantitative information and provides lossy image

compression.
•The quality of such images can be evaluated by objective and subjective methods.
•The objective quality measures:
•The mean-square-error between 2 images (original versus the compressed)
2

 ˆf (x, y)  f (x,

M 1N
1 1
e MSE 
MN x0 y0

•The root mean-square-error:

y) 
 ˆf (x, y)  f (x, 
M 1N
1 1
eRMSE  2
MN
y)
x0 y0
Image Compression
•Quality Measure of a Compressed Image:
•The objective quality measures:
•Peak Signal to Noise Ratio (PSNR) – in decibel (dB):

(2 B 1)2
PSNR  10 B is the number of bits used for
log1 0 eMSE each pixel. (i.e.8 bits)

•The mean-square signal-to-noise-ratio:

M 1 N
1

SNRms  M 1 N
 fˆ(x, y)2


1
ˆf (x, y)  f (x, 
x0 y0

x0 y0
y) 2
Image Compression
• Subjective evaluations used for Image quality rating.
• Human observers can be used to provide subjective evaluations.
 Image Compression Models
•The Source Encoder and Decoder:

•The Source Encoder reduces/eliminates any coding, interpixel or psychovisual redundancies. It contains 3 processes:
• Mapper: Transforms the image into array of coefficients reducing interpixel redundancies. This is a reversible

process which is not lossy.

• Quantizer: This process reduces the accuracy and hence psychovisual redundancies of a given image. This process is

irreversible and therefore lossy.

• Symbol Encoder: This is the source encoding process where fixed or variable-length code is used to represent mapped

and quantized data sets.

• This is a reversible process (not lossy). Removes coding redundancy by assigning shortest codes for the most frequently

occurring output values.

 Image Compression Models
•The Source Encoder and Decoder:

•The Source Decoder contains two components.

•Symbol Decoder: This is the inverse of the symbol encoder and reverse of the variable-length coding is applied.
•Inverse Mapper : Inverse of the removal of the interpixel redundancy.
•The only lossy element is the Quantizer which removes the psychovisual redundancies causing irreversible

loss. Every Lossy Compression methods contains the quantizer module.

•If error-free compression is desired the quantizer module is removed.
 Image Compression
Information Theory-Entropy
•Measuring Information: The information in an image can be modeled as a probabilistic process, where we first develop a

statistical model of the image generation process. The information content (entropy) can be estimated based on this model.
•The information per source (symbol or pixel), which is also referred as entropy is calculated by:
J

E   P(a j ) log
P(aprobabilities.
•Where P(aj) refers to the source symbol/pixel j) J refers to the number of symbols or different
pixel values. j 1
•For example, given the following Image segment:

21 21 21 95 169 243 243 243

21 21 21 95 169 243 243 243

21 21 21 95 169 243 243 243
21 21 21 95 169 243 243 243
 Image Compression
Information Theory-Entropy
•Measuring Information: The entropy of the given 8-bit image segment can be calculated by:

Gray Level Count Probability

21 12 3/8
95 4 1/8
169 4 1/8
243 12 3/8

•The entropy of this image is calculated by:

J

E   P(a j ) log P(a j )  [(3 / 8) log( 3 / 8)  (1/ 8)

log(1/ 8)
j 1

 (1/ 8) log(1/ 8)  (3 / 8) log( 3

/ 8)]
 1.81 bits / pixel
 Image Compression
Error-Free Compression

•Error-free compression is generally composed of two relatively independent operations: (1) reduce the interpixel

redundancies and (2) introduce a coding method to reduce the coding redundancies.

•The coding redundancy can be minimized by using a variable-length coding method where the shortest codes are assigned

to most probable gray levels.

•The most popular variable-length coding method is the Huffman Coding.

•Huffman Coding: The Huffman coding involves the following 2 steps.

• 1) Create a series of source reductions by ordering the probabilities of the symbols and combining the lowest probability

symbols into a single symbol and replace in the next source reduction.

• 2) Code each reduced source starting with the smallest source and working back to the original source. Use 0 and 1 to

code the simplest 2 symbol source.

 Image Compression
Error-Free Compression
•Huffman Coding:
1) Huffman source reductions:
ai’s corresponds to the available gray levels in
a given image.

2) Huffman code assignments:

The first code assignment is done for a2
with the highest probability and the last
assignments are done for a3 and
a5 with the lowest probabilities.
First code

Last code
 Image Compression
Error-Free Compression
• Huffman Coding: Note that the shortest codeword (1) is given for the symbol/pixel with the highest
probability (a2). The longest codeword (01011) is given for the symbol/pixel with the lowest probability (a5).
•The average length of the code is given by:

Lavg  (0.4)(1)  (0.3)(2)  (0.1)(3)  (0.1)(4)  (0.06)(5)  (0.04)(5)

 2.2 bits / symbol
•The entropy of the source is given by:
J

E   P(a j ) log P(a j )  2.14 bits /

symbol.
j 1
•The resulting Huffman coding efficiency is %97.3 (2.14/2.2). Note that Huffman Coding is not optimal and
many more efficient versions of it as well as other variable-length coding methods can be used.
 Image Compression
Lossy Compression
•Unlike the error-free compression, lossy encoding is based on the concept of compromising the accuracy of
the reconstructed image in exchange for increased compression.
•The lossy compression method produces distortion which is irreversible. On the other hand, very high

compression ratios ranging between 10:1 to 50:1 can be achieved with visually indistinguishable from the
original. The error-free methods rarely give results more than 3:1.
•Transform Coding: Transform coding is the most popular lossy image compression method which operates

directly on the pixels of an image.

•The method uses a reversible transform (i.e. Fourier, Cosine transform) to map the image into a set of
transform coefficients which are then quantized and coded.
•The goal of the transformation is to decorrelate the pixels of a given image block such the most of the

information is packed into smallest number of transform coefficients.

 Image Compression
Lossy Compression
•Transform Coding:

A Transform Coding System: encoder and

decoder.
•Transform Selection: The system is based on discrete 2D transforms. The choice of a transform in a given application

depends on the amount of the reconstruction error that can be tolerated and computational resources available.
 Image Compression
DCT-based JPEG (Joint Photographic Expert Group) Standard

Original Compressed
image image
8X8 Coefficients-to-Symbols Entropy
Quantizer
DCT Map Coder

Encoder
Image Compression
DCT-based JPEG Standard

Consider the following 8-bit image with size 512 x 512 pixels.
I m a ge C ompression
I ma ge osl -e

DCT-based JPEG Standard

8x8 Block to be processed

 I m a ge C ompression
I ma ge osl -e

DCT-based JPEG Standard Lowest frequency

Hihgest frequency

DCT

69 71 74 76 89 106 111 122 717.6 0.2 0.4 -19.8 -2.1 -6.2 -5.7 -7.6
59 70 61 61 68 76 88 94 -99.0 -35.8 27.4 19.4 -2.6 -3.8 9.0 2.7
82 70 77 67 65 63 57 70 51.8 -60.8 3.9 -11.8 1.9 4.1 1.0 6.4
97 99 87 83 72 72 68 63 30.0 -25.1 -6.7 6.2 -4.4 -10.7 -4.2 -8.0
91 105 90 95 85 84 79 75 22.6 2.7 4.9 3.4 -3.6 8.7 -2.7 0.9
92 110 101 106 100 94 87 93 15.6 4.9 -7.0 1.1 2.3 -2.2 6.6 -1.7
89 113 115 124 113 105 100 110 0.0 5.9 2.3 0.5 5.8 3.1 8.0 4.8
104 110 124 125 107 95 117 116 -0.7 -2.3 -5.2 -1.0 3.6 -0.5 5.1 -0.1

Step 1:Discrete Cosine Transform (DCT)

 I m a ge C
I ma ge osl -e

ompression
DCT-based JPEG Standard
Step 2:Quantization Procedure
Quantization Matrix
717.6 0.2 0.4 -19.8 -2.1 -6.2 -5.7 -7.6 16 11 10 16 24 40 51 61
12 12 14 19 26 58 60 55
-99.0 -35.8 27.4 19.4 -2.6 -3.8 9.0 2.7
14 13 16 24 40 57 69 56


51.8 -60.8 3.9 -11.8 1.9 4.1 1.0 6.4


14 17 22 29 51 87 80 62
30.0 -25.1 -6.7 6.2 -4.4 -10.7 -4.2 -8.0
18 22 37 56 68 109 103 77
22.6 2.7 4.9 3.4 -3.6 8.7 -2.7 0.9
24 35 55 64 81 104 113 92
15.6 4.9 -7.0 1.1 2.3 -2.2 6.6 -1.7
49 64 78 87 103 121 120 101
0.0 5.9 2.3 0.5 5.8 3.1 8.0 4.8
72 92 95 98 112 100 103 99
-0.7 -2.3 -5.2 -1.0 3.6 -0.5 5.1 -0.1

45 0 0 -1 0 0 0 0
-8 -3 2 1 0 0 0 0
4 -5 0 0 0 0 0 0
2 -1 0 0 0 0 0 0
Quantized

1 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 8x8 block
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
I m a ge C ompression
I ma ge osl -e

DCT-based JPEG Standard

Step3: Coefficient-to-Symbol Mapping

45 0 0 -1 0 0 0 0
-8 -3 2 1 0 0 0 0
4 -5 0 0 0 0 0 0
2 -1 0 0 0 0 0 0
1 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0

Input Zigzag scan procedure

Result = 45,0,-8,4,-3,0,-1,2,-5,2,1,-1,0,1,0,0,0,0,0,0,1,EOB

EOB symbol denotes the end-of-block condition

 I m a ge C ompression
I ma ge osl -e

DCT-based JPEG Standard

Given the symbols below,

Result = 45,0,-8,4,-3,0,-1,2,-5,2,1,-1,0,1,0,0,0,0,0,0,1,EOB

Step 4: Entropy Coding

• Symbols are encoded using mostly Huffman coding.
• Huffman coding is a method of variable length coding in
• which shorter codewords are assigned to the more frequently occurring symbols.

1110101101 010 1011000 ……………..…………1010

 I m a ge C ompression
I ma ge osl -e

DCT-based JPEG Standard

•Once the encoded file is received the decoding is the inverse process given below.

Compressed
image
Inverse Merging Decompressed
Symbol decoder 8x8 subimages
DCT image

Decoder
I m a ge C
I ma ge osl -e

ompression
DCT-based JPEG Standard
•Compressed image examples for changing compression ratios.

Size: 263224Bytes Size: 5728 Bytes CR = 46 Size: 11956 Bytes CR = 22

Size: 15159 Bytes CR = 17 Size: 18032 Bytes CR = 15 Size: 20922 Bytes CR = 13