An Introduction To Image Compression
An Introduction To Image Compression
An Introduction to
Image Compression
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 3
General Image Storage System
Camera
C Transform to
R-G-B Downsample
Y-Cb-Cr Encoder
coordinate Chrominance
coordinate
Performance
RMSE HDD
Object
PSNR
Monitor
C
Transform to
R-G-B Upsample
R-G-B Decoder
coordinate Chrominance
coordinate
V
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 4
Color Specification
Luminance
Received brightness of the light, which is proportional to the total energy
in the visible band.
Chrominance
Describe the perceived color tone of a light, which depends on the
wavelength composition of light
Chrominance is in turn characterized by two attributes
Hue
Specify the color tone, which depends on the peak wavelength of the light
Saturation
Describe how pure the color is, which depends on the spread or bandwidth of the
light spectrum
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 5
YUV Color Space
In many applications, it is desirable to describe a color in terms
of its luminance and chrominance content separately, to enable
more efficient processing and transmission of color signals
One such coordinate is the YUV color space
Y is the components of luminance
Cb and Cr are the components of chrominance
The values in the YUV coordinate are related to the values in the RGB
coordinate by
Y 0.299 0.587 0.114 R 0
Cb
0.169 0.334 0.500 G
128
Cr 0.500 0.419 0.081 B 128
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 6
Spatial Sampling of Color Component
The three different chrominance downsampling format
(a) 4 : 4 : 4 (b) 4 : 2 : 2 (c) 4 : 2 : 0
W W W
H Y H Y H Y
W W/2 W/2
H/2 Cb
H Cb H Cb
W W/2 W/2
H/2 Cr
H Cr H Cr
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 7
The Flow of Image Compression (1/2)
What is the so-called image compression coding?
To store the image into bit-stream as compact as possible and to display
the decoded image in the monitor as exact as possible
Flow of compression
The image file is converted into a series of binary data, which is called
the bit-stream
The decoder receives the encoded bit-stream and decodes it to reconstruct
the image
The total data quantity of the bit-stream is less than the total data quantity
of the original image
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 8
The Flow of Image Compression (2/2)
Measure to evaluate the performance of image compression
W 1 H 1
f ( x, y) f '( x, y)
2
x 0 y 0
RMSE
Root Mean square error: WH
255
Peak signal to noise ratio: PSNR 20 log 10
n1 MSE
Compression Ratio: Cr
n2
Where n1 is the data rate of original image and n2 is that of the encoded
bit-stream
The flow of encoding
Reduce the correlation between pixels
Quantization
Source Coding
Reduce the
Original correlation Source
Quantization Bitstream
Image between Coding
pixels
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 9
Outline
Image Compression Fundamentals
Reduce correlation between pixels
Quantization and Source Coding
Overview of Image Compression Algorithms
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 10
Reduce the Correlation between Pixels
Orthogonal Transform Coding
KLT (Karhunen-Loeve Transform)
Maximal Decorrelation Process
DCT (Discrete Cosine Transform)
JPEG is a DCT-based image compression standard, which is a lossy coding
method and may result in some loss of details and unrecoverable distortion.
Subband Coding
DWT (Discrete Wavelet Transform)
To divide the spectrum of an image into the lowpass and the highpass
components, DWT is a famous example.
JPEG 2000 is a 2-dimension DWT based image compression standard.
Predictive Coding
DPCM
To remove mutual redundancy between seccessive pixels and encode only the
new information
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 11
Covariance
The covariance between two random variables X and Y, with expected value
E[X]= X and E[Y]= Y is defined as
x [ 1 2 n]T [ E ( x 1) E ( x 2) E ( xn)]T
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 12
The Orthogonal Transform
The Linear Transform The Orthogonal Transform
The forward transform y Ax The forward transform y V x
T
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 13
Transform Coding (1/2)
Original sequence X=(x0,x1) Transformed Sequence Y=(y0,y1)
Height Weight Height Weight
65 170 181.97068384838812 3.416170333863853
75 188 202.40605916474945 0.887250469682385
60 150 161.55397378997284 0.559957738379552
70 170
Y AX 183.8437168154677 -1.219748938970085
56 130 141.51187032497342 -3.223438711671541
80 203 206.13345984096256 -2.999455616319722
68 160 173.822665082968 -3.111447163995635
50 110 120.72055367314222 -5.152467452589058
40 80 89.15897210197947 -7.11882670939854
Transform
200
180
x1
100
cos sin
80
Rotation Matrix : A
60
sin cos
40
20
68
0
0 10 20 30 40 50 60 70 80 90
x0
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 14
Transform Coding (2/2)
Throw the value of weight y1 Inverse Transform
Height Weight Height Weight
Height Weight
^
^ sin cos 0
75 188
60 150
x1 70 170
56 130
Because the other element of the pair
80 203
contained very little information, we could 68 160
discard it without a significant effect on the 50 110
fidelity of the reconstructed sequence 40 80
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 15
Karhunen-Loeve Transform
KLT is the optimum transform coder that is defined as the one that minimizes
the mean square distortion of the reproduced data for a given number of total
bits
The KLT
X: The input vector with size N-by-1
A: The transform matrix with size N-by-N
Y: The transformed vector with size N-by-1, and each components v(k) are mutually uncorrelated
Cxixj: The covariance matrix of xi and xj
Cyiyj: The covariance matrix of yi and yj
The transform matrix A is composed of the eigenvectors of the autocorrelation matrix Cxixj, which
makes the output autocorrelation matrix Cyiyj be composed of the eigenvalues 0, 1,... N 1 in
the diagonal direction. That is
Inverse DCT
2 N 1 N 1 (2 x 1)u (2 y 1)v
f ( x, y ) C (u )C (v) F (u, v)cos cos
N u 0 v 0 2N 2N
for x 0,..., N 1 and y 0,..., N 1 where N 8
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 17
Discrete Cosine Transform (2/2)
The 8-by-8 DCT basis
u
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 18
Discrete Wavelet Transform (1/2)
Subband Coding
The spectrum of the input data is decomposed into a set of bandlimitted
components, which is called subbands
Ideally, the subbands can be assembled back to reconstruct the original spectrum
without any error
The input signal will be filtered into lowpass and highpass components
through analysis filters
The human perception system has different sensitivity to different frequency
band
The human eyes are less sensitive to high frequency-band color components
The human ears is less sensitive to the low-frequency band less than 0.01 Hz and
high-frequency band larger than 20 KHz
y0(n)
H0 () H1 ()
-----------------------
-----------------------
h0(n) ↓ 2 ↑ 2 g0(n)
x(n) ^
x(n) Lowband Highband
Analysis Synthesis +
---------
w
h1(n) ↓ 2 ↑ 2 g1(n)
0 Pi/2 Pi
y1(n)
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 19
Discrete Wavelet Transform (2/2)
1D DWT applied alternatively to vertical and horizontal direction line by line
The LL band is recursively decomposed, first vertically, and then horizontally
1D scaling function ( x), ( y ) 2D scaling function ( x, y ) ( x) ( y )
1D wavelet function ( x), ( y ) 2D wavelet function ( x, y) ( x) ( y)
D
Rows
H ( x, y) ( x) ( y)
Columns h (m) ↓ 2 WD ( j , m, n)
h (n) ↓ 2
V ( x, y) ( y) ( x)
h (m) ↓ 2 WV ( j , m, n)
W ( j 1, m, n)
h (m) ↓ 2 WH ( j , m, n)
h (n) ↓ 2
h (m) ↓ 2 WH ( j , m, n)
LL LH
L H
HL HH
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 20
DPCM (1/3)
DPCM CODEC
u[ n ] e[n] e [n] Communication e [n] u [ n]
Quantizer
Channel
uˆ[ n ]
uˆ[ n ] u [ n]
Predictor Predictor
With Delay With Delay
The variance of the error signal is less than the variance of the
original signal.
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 23
Differential Coding - JPEG (1/2)
Transform Coefficients
DC coefficient
AC coefficients
Because there is usually strong correlation between the DC coefficients of adjacent
8×8 blocks, the quantized DC coefficient is encoded as the difference from the DC
term of the previous block
The other 63 entries are the AC components. They are treated separately from the
DC coefficients in the entropy coding process
0 1 5 6 14 15 27 28
2 4 7 13 16 26 29 42
3 8 12 17 25 30 41 43
9 11 18 24 31 40 44 53
10 19 23 32 39 45 52 54
20 22 33 38 46 51 55 60
21 34 37 47 50 56 59 61
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 24
Differential Coding - JPEG (2/2)
We set DC0 = 0.
DC of the current block DCi will be equal to DCi-1 + Diffi .
Therefore, in the JPEG file, the first coefficient is actually the difference
of DCs. Then the difference is encoded with Huffman coding algorithm
together with the encoding of AC coefficients
… Blocki-1 Blocki …
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 25
Outline
Image Compression Fundamentals
Reduce correlation between pixels
Quantization and Source Coding
Overview of Image Compression Algorithms
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 26
Quantization and Source Coding
Quantization
The objective of quantization is to reduce the precision and to achieve
higher compression ratio
Lossy operation, which will result in loss of precision and unrecoverable
distortion
Source Coding
To achieve less average length of bits per pixel of the image.
Assigns short descriptions to the more frequent outcomes and long
descriptions to the less frequent outcomes
Entropy Coding Methods
Huffman Coding
Arithmetic Coding
Run Length Coding
Dictionary Codes
Lempel-Ziv77
Lempel-Ziv 78
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 27
Source Coding
Sequence of source
symbols ui Sequence of code symbols ai
Source Encoder
i 1 i 1 i 1
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 28
Huffman Coding (1/2)
The code construction process has a complexity of O(Nlog2N)
Huffman codes satisfy the prefix-condition
Uniquely decodable: no codeword is a prefix of another codeword
Codeword
Codeword X Probability
length
01 00 1 0
2 01 1 0.25 0.3 0.45 0.55 1
10 01 00 1
2 10 2 0.25 0.25 0.3 0.45
11 10 01
3 11 3 0.2 0.25 0.25
000 11
3 000 4 0.15 0.2
001
3 001 5 0.15
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 30
Arithmetic Coding (1/4)
Shannon-Fano-Elias Coding
We take X={1,2,…,m}, p(x)>0 for all x.
1
Modified cumulative distribution function F P (a )
a x 2
P( x)
1
The codeword of symbol x has l(x) log 1 bits
p ( x)
Codeword is the binary value of F ( x ) with l(x) bits
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 32
Arithmetic Coding (3/4)
Symbol Probability Sub-interval Input String : l l u u r e ?
k 0.05 [0.00,0.05) l l u u r e ?
l 0.2 [0.05,0.25)
u 0.1 [0.20,0.35) 0.0713348389
w 0.05 [0.35,0.40)
=2-4+2-7+2-10+2-15+2-16
e 0.3 [0.40,0.70)
r 0.2 [0.70,0.90) Codeword : 0001001001000011
? 0.2 [0.90,1.00)
e e e e e e e e
0.40
w w w w w w w w
0.35
u u u u u u u u
0.25
l l l l l l l l
0.05
k k k k k k k k
0
0 0.05 0.06 0.070 0.0710 0.07128 0.07132 0.0713336
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 33
Arithmetic Coding (4/4)
Symbol Probability Huffman codeword Input String : l l u u r e ?
k 0.05 10101
l 0.2 01
Huffman Coding 18 bits
u 0.1 100 Codeword : 01,01,100,100,00,11,1101
w 0.05 10100
e 0.3 11 Arithmetic Coding 16 bits
r 0.2 00
? 0.2 1011 Codeword : 0001001001000011
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 34
Zero-Run-Length Coding-JPEG (1/2)
The notation (L,F)
L zeros in front of the nonzero value F
EOB (End of Block)
A special coded value means that the rest elements are all zeros
If the last element of the vector is not zero, then the EOB marker will not be added
An Example:
1. 57, 45, 0, 0, 0, 0, 23, 0, -30, -16, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, ..., 0
2. (0,57) ; (0,45) ; (4,23) ; (1,-30) ; (0,-16) ; (2,1) ; EOB
3. (0,57) ; (0,45) ; (4,23) ; (1,-30) ; (0,-16) ; (2,1) ; (0,0)
4. (0,6,111001);(0,6,101101);(4,5,10111);(1,5,00001);(0,4,0111);(2,1,1);(0,0)
5. 1111000 1111001 , 111000 101101 , 1111111110011000 10111 ,
11111110110 00001 , 1011 0111 , 11100 1 , 1010
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 35
Zero-Run-Length Coding-JPEG (2/2)
Huffman table of Luminance AC coefficients
run/category code length code word
0/0 (EOB) 4 1010
15/0 (ZRL) 11 11111111001
0/1 2 00
... … …
0/6 7 1111000
... … …
0/10 16 1111111110000011
1/1 4 1100
1/2 5 11011
... … …
1/10 16 1111111110001000
2/1 5 11100
... … …
4/5 16 1111111110011000
... … …
15/10 16 1111111111111110
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 36
Dictionary Codes
Dictionary based data compression algorithms are based on the
idea of substituting a repeated pattern with a shorter token
Dictionary codes are compression codes that dynamically
construct their own coding and decoding tables “on the fly” by
looking at the data stream itself
It is not necessary for us to know the symbol probabilities
beforehand. These codes take advantage of the fact that, quite
often, certain strings of symbols are “frequently repeated” and
these strings can be assigned code words that represent the
“entire string of symbols”
Two series
Lempel-Ziv 77: LZ77, LZSS, LZBW
Lempel-Ziv 78: LZ78, LZW, LZMW
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 37
Lempel Ziv 77 Algorithm (1/4)
Search Buffer: It contains a portion of LZ77 Compression Algorithm
the recently encoded sequence. searches the search buffer for the longest
Look-Ahead Buffer: It contains the match
next portion of the sequence to be
encoded. If (longest match is found and all the
Once the longest match has been found, characters are compared)
the encoder encodes it with a triple Output <Cp, Cl, Cs>
<Cp, Cl, Cs> Shift window Cl characters
Cp :the offset or position of the longest
match from the lookahead buffer ELSE
Cl :the length of the longest matching Output <0, 0, Cs>
string
Shift window 1 character
Cs :the codeword corresponding to the
symbol in the look-ahead buffer that END
follows the match
The size of sliding window : N
Search Buffer Look-Ahead Buffer
…… b a b a a c a a c a b ……
Coded Text Text to be read
(Cp, Cl, Cs)
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 39
Lempel Ziv 77 Algorithm (3/4)
Codeword = (15,4,I) for “LZ77I”
Sliding Window of N characters
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8
L Z 7 7 T y p e I s O l d e s t L Z 7 7 I s v x
y p e I s O l d e s t L Z 7 7 I ss vv x O I d e s
e I s O l d e s t L Z 7 7 I s v xx O l d e s t Z
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 41
Lempel Ziv 78 Algorithm (1/3)
The LZ78 algorithm parsed a Lempel Ziv 78 Algorithm
string into phrases, where each Step 1: In the parsing context, search
phrase is the shortest phrase the longest previously parsed phrase P
not seen so far matching the next encoded substring.
The multi-character patterns Step 2: Identify this phrase P by its
are of the form: C0C1 . . . Cn- index L in a list of phrases, and place
1Cn. The prefix of a pattern the index on the code string. Go to the
consists of all the pattern innovative context.
characters except the last: Step 3: In the innovative context,
C0C1 . . . Cn-1 concatenate next character C to the code
This algorithm can be viewed string, and form a new parsed phrase
as building a dictionary in the P‧C.
form of a tree, where the nodes Step 4: Add phrase P‧C to the end of the
corresponding to phrases seen list of parsed phrases as (L,C)
so far Return to the Step 1.
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 42
Lempel Ziv 78 Algorithm (2/3)
Advantages
Asymptotically, the average length of the codeword per source symbol is
not greater than the entropy rate of the information source
The encoder does not know the probabilities of the source symbol
beforehand
Disadvantage
If the size of the input goes to infinity, most texts are considerably shorter
than the entropy of the source. However, due to the limitation of memory
in modern computer, the resource of memory would be exhausted before
compression become optimal. This is the bottleneck of LZ78 needs to be
overcame
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 43
Lempel Ziv 78 Algorithm (3/3)
Input String: ABBABBABBBAABABAA
Parsed String: A, B, BA, BB, AB, BBA, ABA, BAA
Output Codes: (0,A), (0,B), (2,A), (2,B), (1,B), (4,A), (5,A), (3,A)
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 44
Outline
Image Compression Fundamentals
Reduce correlation between pixels
Quantization and Source Coding
Overview of Image Compression Algorithms
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 45
JPEG
B
The JPEG Encoder
G
R Chrominance
YVU color 8X8
Downsampling
Image coordinate FDCT
(4:2:2 or 4:2:0)
Huffman
zigzag
Encoding
Quantizer Bit-stream
Differential Huffman
Quantization Coding Encoding
Table
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 46
Quantization in JPEG
Quantization is the step where we actually throw away data.
Luminance and Chrominance Quantization Table
lower numbers in the upper left direction
large numbers in the lower right direction
The performance is close to the optimal condition
F (u, v)
Quantization F (u, v)Quantization round
Q ( u , v )
Dequantization F (u, v) deQ F (u, v)Quantization Q(u, v)
16 11 10 16 24 40 51 61 17 18 24 47 99 99 99 99
12 12 14 19 26 58 60 55 18 21 26 66 99 99 99 99
14 13 16 24 40 57 69 56 24 26 56 99 99 99 99 99
14 17 22 29 51 87 80 62
QY
47 66 99 99 99 99 99 99
18
QC
22 37 56 68 109 103 77 99 99 99 99 99 99 99 99
24 35 55 64 81 104 113 92 99 99 99 99 99 99 99 99
49 64 78 87 103 121 120 101 99 99 99 99 99 99 99 99
72 92 95 98 112 100 103 99 99 99 99 99 99 99 99 99
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 47
JPEG 2000
B
The JPEG 2000 Encoder
G
R JPEG 2000
Forward Bit-stream
Component 2D DWT Quantization EBCOT
Image
Transform
Rate-
Context Arithmetic
Distortion
Modeling Coding
Control
Tier-1
Tier-2
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 48
Quantization in JPEG 2000
| ab(u, v) |
Quantization coefficients qb (u , v ) sign au (u , v ) floor b
ab(u,v) : the wavelet coefficients of subband b
Rb b b
Quantization step size b 2 1 11
2
Rb: the nominal dynamic range of subband b
εb: number of bits alloted to the exponent of the subband’s coefficients
μb: number of bits allotted to the mantissa of the subband’s coefficients
Reversible wavelets
Uniform deadzone scalar quantization with a step size of Δb =1 must be
used
Irreversible wavelets
The step size is specified in terms of an exponentεb, 0≦εb<25 , and a
mantissaμb , 0≦μb<211
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 49
Bitplane Scanning
The decimal DWT coefficients can be converted into signed binary format, so
the DWT coefficients are decomposed into many 1-bit planes.
In one 1-bit-plane
Significant
A bit is called significant after the first bit ‘1’ is met from MSB to LSB
Insignificant
The bits ‘0’ before the first bit ‘1’ are insignificant
n
Sign 0 MSB
n insignificant
MSB 0
significant
coding First 1 appear 1
order
0
LSB 1
1 LSB
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 50
Scanning Sequence
The scanning order of the bit-plane
Sample
Each element of the bit-plane is called a sample
Column stripe
Four vertical samples can form one column stripe
Full stripe
The 16 column stripes in the horizontal direction can form a full stripe
Code block 64-bits wide
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61
Stripe height of 4
2 6 10 14 18 22 26 30 34 38 42 46 50 54 58 62
3 7 11 15 19 23 27 31 35 39 43 47 51 55 59 63
4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64
65 ...
66 ...
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 51
The Context Window
Current sample
The “curr” is the sample which is to be coded
The other 8 samples are its neighbor samples
The diagonal sample
The samples in the diagonal direction
The vertical samples
The samples in the vertical direction
The horizontal samples
The samples in the horizontal direction
Stripe i-1
1d 2v 3d
4h curr 5h
Stripe i ……
6d 7v 8d
Stripe i+1
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 52
Arithmetic Coder in JPEG 2000
The decision and context data generated from context formation is coded in
the arithmetic encoder
The arithmetic encoder used by JPEG 2000 standard is a binary coder
More Possible Symbol (MPS): If the value of input is 1
Less Possible Symbol (LPS): If the value of input is 0
MPS : More Possible Interval
LPS : Less Possible Interval
MPS
A : The probability distribution of present interval
A C : The bottom of present interval
LPS
C
New MPS
LPS Code MPS LPS new A
New LPS
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 53
Rate Distortion Optimization
For meeting a target bit-rate or transmission time, the packaging
process imposes a particular organization of coding pass data in
the output code-stream
The rate-control assures that the desired number of bytes used by
the code-stream while assuring the highest image quality
possible
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 54
Shape-Adaptive Image Compression
Both the JPEG and JPEG 2000 image compression standard can achieve great
compression ratio, however, both of them do not take advantage of the local
characteristics of the given image effectively
Instead of taking the whole image as an object and utilizing transform coding,
quantization, and entropy coding to encode this object, the SAIC algorithm segments
the whole image into several objects, and each object has its own local characteristic
and color
Because of the high correlation of the color values in each image segment, the SAIC
can achieve better compression ratio and quality than conventional image compression
algorithm Boundary Descriptor
Boundary Quantization
Boundary
Transform And
Coding Entropy Coding
Image
Bit-stream
Segmentation
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 56
Thank You
Q&A
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 57
Question 1
In Slide DPCM(3), what’s the orthogonal principle?
Proof:
Given E[( S 0 Sˆ 0) Si ] 0
E[( S 0 Sˆ 0) Sˆ 0]
E[( S 0 Sˆ 0)(a1S 1 a 2 S 2 ... anSn)]
a1E[( S 0 Sˆ 0) S 1] a 2 E[( S 0 Sˆ 0) S 2] anE[( S 0 Sˆ 0) Sn]
00 0
0
NTU, GICE, MD531, DISP Lab An Introduction to Image Compression Wei-Yi Wei 58