Chapter 9

Image Compression
Standards
9.1 The JPEG Standard
9.2 The JPEG2000 Standard
9.3 The JPEG-LS Standard
9.4 Bi-level Image Compression
Standards
9.5 Further Exploration
Ze-Nian Li & Mark Drew, Fundamentals 1of Multimedia, Prentice Hall, 2003Li & Drew

Fundamentals of Multimedia, Chapter 9

9.1 The JPEG Standard

JPEG is an image compression standard that was developed by the
Joint Photographic Experts Group. JPEG was formally accepted as
an international standard in 1992.
JPEG is a lossy image compression method. It employs a transform
coding method using the DCT (Discrete Cosine Transform).
An image is a function of i and j (or conventionally x and y) in the
spatial domain. The 2D DCT is used as one step in JPEG in order to
yield a frequency response which is a function F(u, v) in the spatial
frequency domain, indexed by two integers u and v.

Li & Drew

Fundamentals of Multimedia, Chapter 9

Observations for JPEG

Image Compression

The effectiveness of the DCT transform coding method

in JPEG relies on 3 major observations:
Observation 1: Useful image contents change relatively
slowly across the image, i.e., it is unusual for intensity
values to vary widely several times in a small area, for
example, within an 88 image block.
much of the information in an image is repeated, hence
spatial redundancy.
3

Li & Drew

Fundamentals of Multimedia, Chapter 9

Observations for JPEG Image

Compression
(contd)
Observation 2: Psychophysical
experiments suggest that humans are
much less likely to notice the loss of very high spatial frequency
components than the loss of lower frequency components.
the spatial redundancy can be reduced by largely reducing
the high spatial frequency contents.

Observation 3: Visual acuity (accuracy in distinguishing closely

spaced lines) is much greater for gray (black and white) than for
color.
chroma subsampling (4:2:0) is used in JPEG.
4

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.1: Block diagram for JPEG encoder.

5

Li & Drew

Fundamentals of Multimedia, Chapter 9

9.1.1 Main Steps in JPEG Image Compression

Transform RGB to YIQ or YUV and subsample color.
DCT on image blocks.
Quantization.
Zig-zag ordering and run-length encoding.
Entropy coding.

Li & Drew

Fundamentals of Multimedia, Chapter 9

DCT on image blocks

Each image is divided into 8 8 blocks. The 2D
DCT is applied to each block image f(i, j), with
output being the DCT coefficients F(u, v) for each
block.
Using blocks, however, has the effect of isolating
each block from its neighboring context. This is
why JPEG images look choppy (blocky) when a
high compression ratio is specified by the user.
7

Li & Drew

Fundamentals of Multimedia, Chapter 9

Quantization
F (u , v)
F (u, v) round

Q(u , v)

(9.1)

F(u, v) represents a DCT coefficient, Q(u, v) is a quantization matrix

Fentry,
(u , v) and

represents the quantized DCT

coefficients which JPEG will use in the succeeding entropy coding.

The quantization
compression.

step

is

the

main

source

for

loss

in

JPEG

The entries of Q(u, v) tend to have larger values towards the lower right corner.
This aims to introduce more loss at the higher spatial frequencies a
practice supported by Observations 1 and 2.
Table 9.1 and 9.2 show the default Q(u, v) values obtained from psychophysical
studies with the goal of maximizing the compression ratio while minimizing
perceptual losses in JPEG images.
8

Li & Drew

Fundamentals of Multimedia, Chapter 9

Table 9.1 The Luminance Quantization Table

16
12
14
14
18
24
49
72

11
12
13
17
22
35
64
92

10
14
16
22
37
55
78
95

16 24 40
19 26 58
24 40 57
29 51 87
56 68 109
64 81 104
87 103 121
98 112 100

51
61
60 55
69
56
80
62
103
77
113
92
120 101
103
99

Table 9.2 The Chrominance Quantization Table

17 18 24 47
99 99
18 21 26 66
99 99
24 26 56 99
99 99
47 66 99 99
99 99
99 99 99 99
99 999
99 99 99 99

99 99
99 99
99 99
99 99
99 99
Li & Drew

99 99

Fundamentals of Multimedia, Chapter 9

An 8 8 block from the Y image of Lena

200
175
200
178
203
185
200
187
200
187
200
187
205

202
175
203
175
200
175
200
187
205
175
200
175
200

189 188 189 175

515
-16
-12
-8
0
0
3
-2

198 188 189 182

200 195 200 187
200 200 197 187
200 200 195 188

65
3
6
3
-2
-3
-2
5

-12 4 1 2
2 0 0 -11
11 -1 3 0
-4 2 -2 -3
7 -5 4 0
-1 0 4 1
-3 3 3 -1
-2 4 -2 2
F(u, v)

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

Fig. 9.2:
JPEG
compression for a smooth image block.
200 200
200
190

199 200 191 187

10

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.2 (contd): JPEG compression for a smooth image block.

11

Li & Drew

Fundamentals of Multimedia, Chapter 9

Another 8 8 block from the Y image of Lena

70 70 100 70 87 87 150 187

-80 -40 89 -73 44 32 53 -3
85 100 96 79 87 154 87 113
-135 -59 -26 6 14 -3 -13 -28
100 85 116 79 70 87 86 196
47 -76 66 -3 -108 -78 33 59
136 69 87 200 79 71 117 96
-2 10 -18 0 33 11 -21 1
161 70 87 200 103 71 96 113
-1 -9 -22 8 32 65 -36 -1
161 123 147 133 113 113 85
5 -20 28 -46 3 24 -30 24
161
6 -20 37 -28 12 -35 33 17
146 147 175 100 103 103 163
-5 -23 33 -30 17 -5 -4 20
187
F(u, v)
156 146 189 70 113 161 163
Fig. 9.2: JPEG compression for a smooth image block.
197
f(i, j)
12

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.3 (contd): JPEG compression for a textured image block.

13

Li & Drew

Fundamentals of Multimedia, Chapter 9

Run-length Coding (RLC) on AC

coefficients
RLC aims to turn the
F (u , v)

values into sets {#-zeros-to-skip, next non-zero value}.

To make it most likely to hit a long run of zeros: a zig-zag scan is used to turn the 88

Fmatrix
(u , v )

into a 64-vector.

Fig. 9.4: Zig-Zag Scan in JPEG.

14

Li & Drew

Fundamentals of Multimedia, Chapter 9

DPCM on DC coefficients
The DC coefficients are coded separately from
the

AC

ones.

Differential

Pulse

Code

modulation (DPCM) is the coding method.

If the DC coefficients for the first 5 image blocks
are 150, 155, 149, 152, 144, then the DPCM
would produce 150, 5, -6, 3, -8, assuming di =
DCi+1 DCi, and d0 = DC0.
15

Li & Drew

Fundamentals of Multimedia, Chapter 9

Entropy Coding
The DC and AC coefficients finally undergo an entropy coding step to gain a
possible further compression.
Use DC as an example: each DPCM coded DC coefficient is represented by (SIZE,
AMPLITUDE), where SIZE indicates how many bits are needed for representing
the coefficient, and AMPLITUDE contains the actual bits.
In the example were using, codes 150, 5, 6, 3, 8 will be turned into
(8, 10010110), (3, 101), (3, 001), (2, 11), (4, 0111) .
SIZE is Huffman coded since smaller SIZEs occur much more often.

AMPLITUDE is

not Huffman coded, its value can change widely so Huffman coding has no
appreciable benefit.
16

Li & Drew

Fundamentals of Multimedia, Chapter 9

Table 9.3 Baseline entropy coding details

SIZ
E

sizeAMPLITUDE
category.

-1, 1

-3, -2, 2, 3

-7..-4, 4..7

-15..-8, 8..15

10

-1023..-512, 512..1023

17

Li & Drew

Fundamentals of Multimedia, Chapter 9

9.1.2 Four Commonly Used

JPEG Modes

Sequential Mode the default JPEG mode,

implicitly assumed in the discussions so far. Each
graylevel image or color image component is
encoded in a single left-to-right, top-to-bottom
scan.
Progressive Mode.
Hierarchical Mode.
Lossless Mode discussed in Chapter 7, to be
replaced by JPEG-LS (Section 9.3).
18

Li & Drew

Fundamentals of Multimedia, Chapter 9

Progressive Mode
Progressive JPEG delivers low quality versions of the image
quickly, followed by higher quality passes.
1.Spectral selection: Takes advantage of the spectral
(spatial frequency spectrum) characteristics of the DCT
coefficients: higher AC components provide detail information.
Scan 1: Encode DC and first few AC components, e.g., AC1, AC2.
Scan 2: Encode a few more AC components, e.g., AC3, AC4, AC5.
...
Scan k: Encode the last few ACs, e.g., AC61, AC62, AC63.

19

Li & Drew

Fundamentals of Multimedia, Chapter 9

2.

Progressive Mode
(Contd)

Successive

approximation:

Instead

of

gradually encoding spectral bands, all DCT

coefficients are encoded simultaneously but
with their most significant bits (MSBs) first.
Scan 1: Encode the first few MSBs, e.g., Bits 7, 6, 5,
4.
Scan 2: Encode a few more less significant bits, e.g.,
Bit 3.
...
Scan m: Encode the least significant bit (LSB), Bit 0.
20

Li & Drew

Fundamentals of Multimedia, Chapter 9

Hierarchical Mode
The encoded image at the lowest resolution is
basically a compressed low-pass filtered image,
whereas the images at successively higher
resolutions

provide

additional

details

(differences from the lower resolution images).

Similar to Progressive JPEG, the Hierarchical JPEG
images can be transmitted in multiple passes
progressively improving quality.
21

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.5: Block diagram for Hierarchical JPEG.

22

Li & Drew

Fundamentals of Multimedia, Chapter 9

Encoder for a Three-level

Hierarchical JPEG

1. Reduction of image resolution:

Reduce resolution of the input image f (e.g., 512512) by a factor of 2 in each dimension to obtain f2 (e.g., 256
256). Repeat this to obtain f4 (e.g., 128 128).

2. Compress low-resolution image f4:

Encode f4 using any other JPEG method (e.g., Sequential, Progressive) to obtain F4.

3. Compress difference image d2:

(a) Decode F4 to obtain
E(

).

(b) Encode difference

D 2.

. Use any interpolation method to expand

f
4
d 2 f 2 E ( f 4 )

f
4

to be of the same resolution as f2 and call it

f
4

using any other JPEG method (e.g., Sequential, Progressive) to generate

4. Compress difference image d 1:

(a) Decode D2 to obtain

d2 ; add it to E(f 4

) to get

f E ( f ) d
2
4
2

which is a version of f2 after compression

and decompression.
(b) Encode difference
D 1.

d1 f E ( f 2 )

using any other JPEG method (e.g., Sequential, Progressive) to generate

23

Li & Drew

Fundamentals of Multimedia, Chapter 9

Decoder for a Three-level

Hierarchical JPEG
1. Decompress the encoded low-resolution image F4:
Decode F4 using the same JPEG method as in the

encoder tof 4
f
2. Restore image
2

UseE ( f 4 ) d2
f
3. Restore image

Use E ( f 2 ) d1

obtain

at the intermediate resolution:

f2
to obtain

at the original resolution:

f%
to obtain
24

.
Li & Drew

Fundamentals of Multimedia, Chapter 9

9.1.3 A Glance at the JPEG

Bitstream

Fig. 9.6: JPEG bitstream.

25

Li & Drew

Fundamentals of Multimedia, Chapter 9

9.2 The JPEG2000

Standard
Design Goals:
To provide a better rate-distortion tradeoff and improved
subjective image quality.
To provide additional functionalities lacking in the current
JPEG standard.
The JPEG2000 standard addresses the following problems:
Lossless and Lossy Compression: There is currently no
standard that can provide superior lossless compression
and lossy compression in a single bitstream.
26

Li & Drew

Fundamentals of Multimedia, Chapter 9

Low Bit-rate Compression: The current JPEG

standard
offers
excellent
rate-distortion
performance in mid and high bit-rates. However,
at bit-rates below 0.25 bpp, subjective distortion
becomes unacceptable. This is important if we
hope to receive images on our web- enabled
ubiquitous
devices,
such
as
web-aware
wristwatches and so on.
Large Images: The new standard will allow
image resolutions greater than 64K by 64K
without tiling. It can handle image size up to 2 32
1.
Single Decompression Architecture: The
current JPEG standard has 44 modes, many of
which are application specific and not used by
the majority of JPEG decoders.
27

Li & Drew

Fundamentals of Multimedia, Chapter 9

Transmission in Noisy Environments: The

new standard will provide improved error
resilience
for
transmission
in
noisy
environments such as wireless networks and
the Internet.
Progressive Transmission: The new standard
provides seamless quality and resolution
scalability from low to high bit-rate. The target
bit-rate and reconstruction resolution need not
be known at the time of compression.
Region of Interest Coding: The new
standard allows the specification of Regions of
Interest (ROI) which can be coded with superior
quality than the rest of the image. One might
like to code the face of a speaker with more
quality than the surrounding furniture.

28

Li & Drew

Fundamentals of Multimedia, Chapter 9

Computer Generated Imagery: The current JPEG

standard is optimized for natural imagery and does
not perform well on computer generated imagery.
Compound Documents: The new standard offers
metadata mechanisms for incorporating additional
non-image data as part of the file. This might be
useful for including text along with imagery, as one
important example.

In addition, JPEG2000 is able to handle up to

256

channels

of

information

whereas

the

current JPEG standard is only able to handle

three color channels.
29

Li & Drew

Fundamentals of Multimedia, Chapter 9

Properties of JPEG2000 Image

Compression

Uses Embedded Block Coding with Optimized Truncation (EBCOT)

algorithm which partitions each subband LL, LH, HL, HH produced
by the wavelet transform into small blocks called code blocks.
A separate scalable bitstream is generated for each code block
improved error resilience.

Fig. 9.7: Code block structure of EBCOT.

30

Li & Drew

Fundamentals of Multimedia, Chapter 9

Main Steps of JPEG2000 Image Compression

Embedded

Block

coding

and

bitstream

generation.

Post

compression

rate

distortion

(PCRD)

optimization.

Layer formation and representation.

31

Li & Drew

Fundamentals of Multimedia, Chapter 9

Embedded Block Coding and

Bitstream Generation
1.Bitplane coding.

2. Fractional bitplane coding.

32

Li & Drew

Fundamentals of Multimedia, Chapter 9

1. Bitplane Coding
Uniform dead zone quantizers are used with successively smaller
interval sizes. Equivalent to coding each block one bitplane at a time.

Fig. 9.8: Dead zone quantizer. The length of the dead zone is 2. Values
inside the dead zone are quantized to 0.

33

Li & Drew

Fundamentals of Multimedia, Chapter 9

Blocks are further divided into a sequence of 16 16 subblocks.

The significance of sub-blocks are encoded in a significance
map P where p(Bi[j]) denote the significance of sub-block Bi[j]
at bitplane P.
A quad-tree structure is used to identify the significance of
sub-blocks one level at a time.

Bi0 [ j] Bi [ j]
The
by identifying the sub-blocks
B t [ jtree
] structure
B t 1[2is
j constructed
z]
i

z{0,1}2

with leaf nodes, i.e.,

using recursion:

The higher levels are built

, 0 t T.

34

Li & Drew

Fundamentals of Multimedia, Chapter 9

Bitplane Coding
Primitives
Four different primitive coding methods that employ context based
arithmetic coding are used:
Zero Coding: Used to code coefficients on each bitplane that are
not yet significant.

hi [k ]

Horizontal:

Vertical:

vi [k ]
di [k ]

Diagonal:

i [k1 z, k2 ], with 0 hi [k ] 2.

i [k1 , k2 z ], with 0 vi [k ] 2.

z{1, 1}

z{1, 1}

z1 , z2 {1, 1}

i [k1 z1 , k2 z2 ], with 0 d i [k ] 4.

35

Li & Drew

Fundamentals of Multimedia, Chapter 9

Table 9.4 Context assignment for the zero coding

primitive.
LL, LH and HL
subbands

HH subband

Label

hi[k]

vi[k]

di[k]

di[k]

hi[k]+vi[k]

>1

>1

>1

>0

>0

>0

>2

36

Li & Drew

Fundamentals of Multimedia, Chapter 9

Run-length coding: Code runs of 1-bit

significance values.

Four conditions must

be met:
Four consecutive samples must be insignificant.
The samples must have insignificant neighbors.
The samples must be within the same subblock.
The horizontal index k1 of the first sample must
be even.
37

Li & Drew

Fundamentals of Multimedia, Chapter 9

Sign coding: Invoked at most once when a coefficients

goes from being insignificant to significant.
The sign bits i[k] from adjacent samples contains
substantial dependencies.
The conditional distribution of i[k] is assumed to be the
same as i[k].
hi [k ]
is 0 if both horizontal neighbors are insignificant, 1
if at least one horizontal neighbor is positive, or 1 if at
least one horizontal neighbor is negative.
v [k ]
i

is defined similarly for vertical neighbors.

If i [k ]
is the sign prediction, the binary symbol coded
using the relevant
.
[context
k ] [k is
]
i

38

Li & Drew

Fundamentals of Multimedia, Chapter 9

Table 9.5 Context assignment for the sign coding

primitive
Labe
l

i [k ]

vi [k ]

hi [k ]

-1

-1

-1

-1

-1

-1

-1

-1

-1

-1

39

Li & Drew

Fundamentals of Multimedia, Chapter 9

Magnitude refinement: Code the value

ip [kof
]
given that i[k] 2p+1.
%[k ]
i

changes from 0 to 1 after the magnitude

refinement primitive is first applied to si[k].

p [k ] is coded with context 0%
if
[k ] hi [k ] vi [k ] 0
i
,
with context
1 if
and
%
i [k ] 0
hi [k ] vi [k ] 0
, and
with
context 2 if
.
%
[
k
]

1
i
40

Li & Drew

Fundamentals of Multimedia, Chapter 9

2. Fractional Bitplane
Coding
Divides code block samples into smaller subsets having different
statistics.
Codes one subset at a time starting with the subset expecting to
have the largest reduction in distortion.
Ensures each code block has a finely embedded bitstream.
(P1 p )
Four different passes are used: forward significance propagation
p

(P2 )
pass
; reverse significance propagation
pass
(P3 p )
(P4 p ) normalization pass
refinement
pass
; and
41

; magnitude
.
Li & Drew

Fundamentals of Multimedia, Chapter 9

Forward Significance
Propagation Pass
Sub-block samples are visited in scan-line order and
insignificant samples and samples that do not satisfy the
neighborhood requirement are skipped.
For the LH, HL, and LL subbands, the neighborhood
requirement is that at least one of the horizontal
neighbors has to be significant.
For the HH subband, the neighborhood requirement is that
at least one of the four diagonal neighbors is significant.
42

Li & Drew

Fundamentals of Multimedia, Chapter 9

Reverse Significance Propagation Pass

This pass is identical top

P1

except that it proceeds in the reverse order. The

neighborhood requirement is relaxed to include samples that have at least one

significant neighbor in any direction.

Magnitude Refinement Pass

This pass encodes samples that are already significant but have not been coded in the
previous two passes. Such samples are processed with the magnitude refinement
primitive.

Magnitude Refinement Pass

The value

[k ]
p

of all samples not considered in the previous three coding passes

are coded i using the sign coding and run-length coding primitives as appropriate. If a
sample is found to be significant, its sign is immediately coded using the sign coding
primitive.
43

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.9: Appearance of coding passes and quad tree codes in each blocks
embedded bitstream.

44

Li & Drew

Post Compression Rate

Distortion
Goal: (PCRD) Optimization

Fundamentals of Multimedia, Chapter 9

Produce a truncation of the independent bitstream of each

code block in an optimal way such that distortion is
minimized, subject to the bit-rate constraint.

For each truncated embedded bitstream of code block Bi

having
rate
R ni
i

truncation

Dini
point ni,

with distortion
the

overall

distortion

and
of

the

reconstructed image is (assuming distortion is additive)

D Dini
i

45

(9.3)
Li & Drew

Fundamentals of Multimedia, Chapter 9

The optimal selection of truncation points ni can be

formulated into a minimization problem subject to the
following constraint:

R R R
ni
i

max

(9.5)

{ni }

For some , any set of truncation point

minimizes

D ( ) R ( ) D

ni
i

ni
i

that

(9.6)

is optimal in the rate-distortion sense.

46

Li & Drew

Fundamentals of Multimedia, Chapter 9

The distortion-rate slopes given by the ratios

Si jk

Di jk

jk
Ri

(9.5)

is strictly decreasing.
This allows the optimization problem be solved by a simple
selection through an enumeration j1 < j2 < ... of the set of
feasible truncation points.

ni max jk N i | Si jk

47

(9.6)

Li & Drew

Fundamentals of Multimedia, Chapter 9

Layer Formation and

Representation

JPEG2000 offers both resolution and quality scalability

through the use of a layered bitstream organization and a
two-tiered coding strategy.
The first tier produces the embedded block bit-streams while
the second tier compresses block summary information.
n1i
i

R
The quality layer Q1 contains the initial
q

bytes of each code

q 1

block
Lq RBni i and
R ni the
0 other layers Qq contain the incremental
i

contribution

from code block Bi.

48

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.10: Three quality layers with eight blocks each.

49

Li & Drew

Fundamentals of Multimedia, Chapter 9

Region of Interest Coding

in JPEG2000

Goal:

Particular regions of the image may contain important information,

thus should be coded with better quality than others.

Usually implemented using the MAXSHIFT method which scales

up the coefficients within the ROI so that they are placed into
higher bit-planes.
During the embedded coding process, the resulting bits are
placed in front of the non-ROI part of the image. Therefore,
given a reduced bit-rate, the ROI will be decoded and refined
before the rest of the image.
50

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.11: Region of interest (ROI) coding of an image using a circularly

shaped ROI. (a) 0.4 bpp, (b) 0.5 bpp, (c) 0.6bpp, and (d) 0.7 bpp.
51

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.12: Performance comparison for JPEG and JPEG2000 on

different image types. (a): Natural images.
52

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.12: Performance comparison for JPEG and JPEG2000 on

different image types. (b): Computer generated images.
53

Li & Drew

Fundamentals of Multimedia, Chapter 9

Fig. 9.12: Performance comparison for JPEG and JPEG2000 on

different image types. (c): Medical images.
54

Li & Drew

Fundamentals of Multimedia, Chapter 9

(a)
Fig. 9.13: Comparison of JPEG and JPEG2000. (a) Original image.
55

Li & Drew

Fundamentals of Multimedia, Chapter 9

(b)

(c)
Fig. 9.13 (Contd): Comparison of JPEG and JPEG2000. (b) JPEG (left) and JPEG2000 (right) images
compressed at 0.75 bpp. (c) JPEG (left) and JPEG2000 (right) images compressed at 0.25 bpp.
56

Li & Drew

Fundamentals of Multimedia, Chapter 9

9.3 The JPEG-LS

Standard

JPEG-LS is in the current ISO/ITU standard for lossless or near lossless

compression of continuous tone images.
It is part of a larger ISO effort aimed at better compression of medical
images.
Uses the LOCO-I (LOw COmplexity LOssless Compression for Images)
algorithm proposed by Hewlett-Packard.
Motivated by the observation that complexity reduction is often more
important than small increases in compression offered by more complex
algorithms.
Main Advantage: Low complexity!
57

Li & Drew

Fundamentals of Multimedia, Chapter 9

The LOCO-I algorithm makes uses of context

modelling.
The idea of context modelling is to take
advantage of the structure within the input
source the conditional probabilities.

Fig. 9.14: JPEG-LS Context Model.

58

Li & Drew

Fundamentals of Multimedia, Chapter 9

9.4 JBIG and JBIG-2: Bi-level

Image
Compression
Standards
Main Goal:
Enables the handing of documents
in electronic form.
Primarily used to code scanned images of printed or hand-written
text, computer generated text, and facsimile transmissions.
JBIG is a lossless compression standard. It also offers progressive
encoding/decoding capability, the resulting bitstream contains a
set of progressively higher resolution images.
JBIG-2 introduces model-based coding similar to context-based
coding. It supports lossy compressions well.
59

Li & Drew