ECE/CS 5578 Multimedia Communication

Lec 05: Arithmetic Coding I

Zhu Li
Dept of CSEE, UMKC
http://l.web.umkc.edu/lizhu
Office: FH560E, Email: [email protected], Ph: x 2346.
Course web:http://sites.google.com/view/ece5578

slides created with WPS Office Linux and EqualX LaTex equation editor

Z. Li, ECE/CS 5578 Multimedia Communciation p.1

 Outline
 Lecture 04 ReCap
 Arithmetic Coding
 About Homework-1 and Lab

Z. Li, ECE/CS 5578 Multimedia Communciation p.2

 JPEG Coding
 Block (8x8 pel) based coding
 DCT transform to find sparse *
representation
 Quantization reflects human visual
system
 Zig-Zag scan to convert 2D to 1D
string
=
 Run-Level pairs to have even more
compact representation
 Hoffman Coding on Level Category Quant Table:

 Fixed on the Level with in the

category

Z. Li, ECE/CS 5578 Multimedia Communciation p.3

 Coding of AC Coefficients
Zigzag scanning:
 Example
8 24 -2 0 0 0 0 0
-31 -4 6 -1 0 0 0 0
0 -12 -1 2 0 0 0 0
0 0 -2 -1 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0

 Example: zigzag scanning result

24 -31 0 -4 -2 0 6 -12 0 0 0 -1 -1 0 0 0 2 -2 0 0 0 0 0 -1 EOB
 (Run, level) representation:
 (0, 24), (0, -31), (1, -4), (0, -2), (1, 6), (0, -12), (3, -1), (0, -1),
(3, 2), (0, -2), (5, -1), EOB

Z. Li, ECE/CS 5578 Multimedia Communciation p.4

 Coding of AC Coefficients

Run / Base Run / Base … Run / Base codeword

Catg. codeword Catg. Codeword Cat.
EOB 1010 - - … ZRL 1111 1111 001
0/1 00 1/1 1100 … 15/1 1111 1111 1111 0101
0/2 01 1/2 11011 … 15/2 1111 1111 1111 0110
0/3 100 1/3 1111001 … 15/3 1111 1111 1111 0111
0/4 1011 1/4 111110110 … 15/4 1111 1111 1111 1000
0/5 11010 1/5 11111110110 … 15/5 1111 1111 1111 1001
… … … … … … …
 ZRL: represent 16 zeros when number of zeros exceeds 15.
 Example: 20 zeros followed by -1: (ZRL), (4, -1).
 (Run, Level) sequence: (0, 24), (0, -31), (1, -4), mapped to Run/Cat
 Run/Cat. Sequence: 0/5, 0/5, 1/3, …
24 is the 24-th entry in Category 5  (0, 24): 11010 11000
-4 is the 3-th entry in Category 3  (1, -4): 1111001 011
Z. Li, ECE/CS 5578 Multimedia Communciation p.5
 Outline
 Lecture 04 ReCap
 Arithmetic Coding
 Basic Encoding and Decoding
 Uniqueness and Efficiency
 Scaling and Incremental Coding
 Integer Implementation
 About Homework-1 and Lab

Z. Li, ECE/CS 5578 Multimedia Communciation p.6

 Arithmetic Coding – The SciFi Story
 When I was in my 5th grade….

Aliens visit earth….

how to write home on what we see ?

Z. Li, ECE/CS 5578 Multimedia Communciation p.7

 Huffman Coding
Replacing an input symbol with a codeword
Need a probability distribution
Hard to adapt to changing statistics
Need to store the codeword table
Minimum codeword length is 1 bit

Arithmetic Coding
 Replace the entire input seq with a single floating-
point number
 Does not need the probability distribution
 Adaptive coding is very easy
 No need to keep and send codeword table
 Fractional codeword length per symbol
Z. Li, ECE/CS 5578 Multimedia Communciation p.8
 Introduction

Recall table look-up decoding of Huffman code

 N: alphabet size
 L: Max code word length 1

 Divide [0, 2^L] into N intervals 00

 Granularity: 1/2^L
 One interval for one symbol 010 011
 Interval size is roughly
proportional to symbol prob. 000 010 011 100

 Arithmetic coding applies this idea recursively

 Normalizes the range [0, 2L] to [0, 1], in continous values.
 Map an input sequence (multiple symbols) to a unique tag in
[0, 1).
abcd…..
dcba….. 0 1

Z. Li, ECE/CS 5578 Multimedia Communciation p.9

 Arithmetic Coding
Symbol set and prob: a (0.8), b(0.02), c(0.18)
Disjoint and complete partition of the range [0, 1)
0 1
[0, 0.8), [0.8, 0.82), [0.82, 1)
a b c
Each interval corresponds to one symbol
Interval size is proportional to symbol probability

 The first symbol restricts the tag

0 1
position to be in one of the intervals
 The reduced interval is partitioned
0 1
recursively as more symbols are
processed.
0 1
 Observation: once the tag falls into an interval, it
never gets out of it
Z. Li, ECE/CS 5578 Multimedia Communciation p.10
 Some Questions to think about:
Why compression is achieved this way?
How to implement it efficiently?
How to decode the sequence?
Why is it better than Huffman code?

Z. Li, ECE/CS 5578 Multimedia Communciation p.11

 Example:

Symbol Prob. 1 2 3
1 0.8
2 0.02 0 0.8 0.82 1.0

3 0.18  Map to real line range [0, 1)

 Order does not matter
 Decoder needs to use the same order
• Disjoint but complete partition:
• 1: [0, 0.8): 0, 0.799999…9
• 2: [0.8, 0.82): 0.8, 0.819999…9
• 3: [0.82, 1): 0.82, 0.999999…9

Z. Li, ECE/CS 5578 Multimedia Communciation p.12

 Concept of the encoding (not practical)
 Input sequence: “1321”
1 2 3
Range 1
0 0.8 0.82 1.0
1 2 3
Range 0.8
0 0.64 0.656 0.8

1 2 3
Range 0.144
0.656 0.7712 0.77408 0.8

1 2 3
Range 0.00288
0.7712 0.773504 0.7735616 0.77408
Termination: Encode the lower end or midpoint to signal the end.
Difficulties: 1. Shrinking of interval requires very high precision for long sequence.
2. No output is generated until the entire sequence has been processed.
Z. Li, ECE/CS 5578 Multimedia Communciation p.13
 Encoder Pseudo Code
 Cumulative Density Function (CDF) Probability Mass Function
 For continuous distribution: 0.4
x
0.2 0.2 0.2
FX ( x) = P( X £ x) = ò p( x)dx
-¥

 For discrete distribution: 1 2 3 4 X

i
FX (i ) = P ( X £ i ) = å P( X = k )
k = -¥ CDF 0.8
1.0

0.4
P( X = i ) = FX (i ) - FX (i - 1). 0.2
 Properties:
X
 Non-decreasing 1 2 3 4
 Piece-wise constant
 Each segment is closed at the lower end.
Z. Li, ECE/CS 5578 Multimedia Communciation p.14
 Encoder Pseudo Code
 Keep track of LOW=0.0, HIGH=1.0;
while (not EOF) {
 LOW, HIGH, RANGE n = ReadSymbol();
 Any two are RANGE = HIGH - LOW;
sufficient, e.g., HIGH = LOW + RANGE * CDF(n);
LOW = LOW + RANGE * CDF(n-1);
LOW and RANGE. }
output LOW;

Input HIGH LOW RANGE

Initial 1.0 0.0 1.0
1 0.0+1.0*0.8=0.8 0.0+1.0*0 = 0.0 0.8
3 0.0 + 0.8*1=0.8 0.0 + 0.8*0.82=0.656 0.144
2 0.656+0.144*0.82=0.77408 0.656+0.144*0.8=0.7712 0.00288
1 0.7712+0.00288*0=0.7712 0.7712+0.00288*0.8=0.773504 0.002304

Z. Li, ECE/CS 5578 Multimedia Communciation p.15

 Concept of the Decoding
 Arithmetic Decoding (conceptual)
Suppose encoder encodes the lower end: 0.7712 Zoom
1 2 3 In by * CDF
Decode 1
0 0.8 0.82 1.0
1 2 3
Decode 3 * CDF
0 0.64 0.656 0.8

1 2 3
Decode 2 * CDF
0.656 0.7712 0.77408 0.8

1 2 3
Decode 1
0.7712 0.773504 0.7735616 0.77408

Drawback: need to recalculate all thresholds each time.

Z. Li, ECE/CS 5578 Multimedia Communciation p.16
 Simplified Decoding (Floating Pt Ops)
x - low
 Normalize RANGE to [0, 1) each time x¬
 No need to recalculate the thresholds.
range
Receive 0.7712 1 2 3
Decode 1

x =(0.7712-0) / 0.8 0 0.8 0.82 1.0

= 0.964
1 2 3
Decode 3

0 0.8 0.82 1.0

x =(0.964-0.82) / 0.18
= 0.8 1 2 3
Decode 2
x =(0.8-0.8) / 0.02 0 0.8 0.82 1.0
=0
Decode 1.
1 2 3
Stop.

0 0.8 0.82 1.0

Z. Li, ECE/CS 5578 Multimedia Communciation p.17
 Decoder Pseudo Code
Low = 0; high = 1;
x = GetEncodedNumber();
While (x ≠ low) {
n = DecodeOneSymbol(x);
output symbol n;
x = (x - CDF(n-1)) / (CDF(n) - CDF(n-1));
};
1
00

010 011

000 010 011 100

But this method still needs high-precision floating point operations.
It also needs to read all input at the beginning of the decoding.

Z. Li, ECE/CS 5578 Multimedia Communciation p.18

 Arithmetic Coding in Python and C
Python has a torchac package for this:
 https://github.com/fab-jul/torchac
 backend in C/C++, had a python wrapper around
 will do a tutorial on this next week
Paras’s LSTM driven AC for sequence encoding
 https://sce.umkc.edu/faculty-
sites/lizhu/teaching/2021.spring.video/hw/lstm_ac.zip
 This can be used to code DNA/RNA seq data, an option for course
project.
C/C++ implementation of Arithmetic Coding
 SFU Prof. Jie Liang’s code:

Z. Li, ECE/CS 5578 Multimedia Communciation p.19

 Outline
 Lecture 04 ReCap
 Arithmetic Coding
 Basic Encoding and Decoding
 Uniqueness and Efficiency
 Scaling and Incremental Coding
About Homework-1 and Lab

Z. Li, ECE/CS 5578 Multimedia Communciation p.20

 Uniqueness
1 2 3
Range 0.00288
0.7712 0.773504 0.7735616 0.77408
Termination: Encode the midpoint (or lower end) to signal the end.
How to represent the final tag uniquely and efficiently?
Answer: Take the binary value of the tag T(X) and truncate to
(X is the sequence {x1, …, xm}, not individual symbol)
é 1 ù 1 bit longer than
l ( X ) = êlog ú + 1 bits. Shannon code
ê p( X ) ú
1
Proof: Assuming midpoint tag: T(X) = F(X - 1) + p ( X ), p(X) > 0.
2
First show the truncated code is unique, that is, code is within [F(X-1), F(X)).

1). ëT(X)û l ( X ) £ T ( X ) < F ( X )

So ëT(X)û l ( X ) is below the high end of the interval.
Z. Li, ECE/CS 5578 Multimedia Communciation p.21
 Uniqueness and Efficiency
F(X)
Prove the code is uniquely decodable (prefix free):
Any code with prefix
ëT(X) û l ( X ) is in ëT(X)û l ( X )
1 ö 1
é +
êë ëT(X)û l ( X ) , ëT(X)û l ( X ) + 2l ( X ) ÷ø 2l ( X )

Need to show that this is in [F(X-1), F(X) ):

ëT(X)û l ( X )
We already show ëT(X)û l ( X ) ³ F ( X - 1).
F(X-1)
1
Only need to show F ( X ) - ëT(X)û l ( X ) >
2l ( X )
p( X ) 1
F ( X ) - ëT(X)û l ( X ) > F(X ) -T(X ) = > l( X )
2 2
é 1 ù
ëT(X)û l ( X ) is prefix free if l ( X ) = êlog ú + 1 bits.
ê p( X ) ú

Z. Li, ECE/CS 5578 Multimedia Communciation p.22

 Uniqueness and Efficiency

æé 1 ù ö æ 1 ö
2).
-çç ê log
p ( X )
ú +1 ÷÷ -çç log +1 ÷÷
p( X ) ø 1
2 -l ( X ) = 2 èê ú ø
£2 è
= p( X )
2
By def,
1
T(X) = F(X - 1) + p ( X ), p(X) > 0.
2 F(X)
1
T(X) - F(X - 1) = p( X ) ³ 2-l(X)
2
Together with T(X)- êëT(X) úû l ( X ) £ 2 - l ( X ) T(X)

£ 2-l ( X )
ëT(X)û l ( X ) ³ F ( X - 1). ³ 2-l ( X )
ëT(X)û l ( X )
F(X-1)
Thus F ( X - 1) £ ëT(X)û l ( X ) < F ( X )

So the truncated code is still in the interval. This proves the uniqueness.
Z. Li, ECE/CS 5578 Multimedia Communciation p.23
 Uniqueness and Efficiency
Efficiency of arithmetic code:

m é 1 ù X 1m : { x1 ,..., xm }
l ( X ) = êlog
1 m ú
+ 1 bits.
ê p( X 1 ) ú

æé 1 ù ö
{ 1
m
ç
1
m
}
L = E p( X )l ( X ) = å P( X )ç êlog 1
m
m ú
p( X 1 ) ú
+ 1 ÷÷
èê ø
æ 1 ö
£ å P( X )çç log1
m
m
+ 1 + 1 ÷
÷ = H ( X m
1 )+2
è p( X 1 ) ø

l(X) is the bits to code a sequence {x1,x2, …, xm}.

Assume iid sequence, H ( X 1m ) = mH ( X )
L 2
H (X ) £ £ H (X ) + L/m  H(X) for large m, stronger than prev
m m results.
Z. Li, ECE/CS 5578 Multimedia Communciation p.24
 Uniqueness and Efficiency
Comparison with Huffman code:
 Expected length of Huffman code:

*
H ( X ) £ L £ H ( X ) +1

 Huffman code can reduce the overhead by

jointly encoding more symbols, but needs
much larger alphabet size - not practical.
 Arithmetic coding is more efficient for longer
sequences.

Z. Li, ECE/CS 5578 Multimedia Communciation p.25

 Binary Arithmetic Coding

Arithmetic coding is slow in general:

To decode a symbol, we need a series of decisions and
multiplications:
While (Tag > LOW + RANGE * Sum(n) / N - 1) {
n++;
}

 The complexity is greatly reduced if we have only two

symbols: 0 and 1.
symbol 0 symbol 1

0 x 1

 Only two intervals: [0, x), [x, 1)

Z. Li, ECE/CS 5578 Multimedia Communciation p.26

 Encoding of Binary Arithmetic Coding
HIGH ¬ LOW + RANGE ´ CDF (n)
LOW ¬ LOW + RANGE ´ CDF (n - 1)
LOW = 0, HIGH = 1 Prob(0)=0.6. Sequence: 0110

LOW = 0, HIGH = 0.6

0 0.6 1

LOW = 0.36, HIGH = 0.6

0 0.36 0.6

LOW = 0.504, HIGH = 0.6

0.36 0.504 0.6

LOW = 0.504, HIGH = 0.5616

0.504 0.5616 0.6
Only need to update LOW or HIGH for each symbol.
Z. Li, ECE/CS 5578 Multimedia Communciation p.27
 Decoding of Binary Arithmetic Coding

Tag

0 0.6 1

General case (integer implementation):

While (Tag > LOW + RANGE * Sum(n) / N - 1)) {
n++;
}
Binary case: only one condition to check

if (Tag > LOW + RANGE * Sum(Symbol0) / N - 1) {

n = 1;
} else {
n = 0;
}
Z. Li, ECE/CS 5578 Multimedia Communciation p.28
 Applications of Binary Arithmetic Coding
Increasingly popular:
 JBIG, JBIG2, JPEG2000, H.264
Convert non-binary signals into binary:
 Golomb-Rice Code: used in H.264.
Bit-plane coding: used in JPEG2000.
B = [B0, B1, ……, BK-1]: binary representation of B.
Chain rule:
H(B) = H(B0) +H(B1 | B0) + …… + H(BK-1 | B0, … BK-2).
To code B0, needs P0(0): Prob(B0=0).
To code B1, needs P1(0 | i): Prob(B1=0 | B0 = i), i = 0, 1.
……
 More details:
 AVC Binary Adaptive Arithmetic Coding: CABAC
 HEVC Arithmetic Coding:

Z. Li, ECE/CS 5578 Multimedia Communciation p.29

 Outline
 Lecture 04 ReCap
 Arithmetic Coding
 Basic Encoding and Decoding
 Uniqueness and Efficiency
 Scaling and Incremental Coding
About Homework-1 and Lab

Z. Li, ECE/CS 5578 Multimedia Communciation p.30

 Scaling and Incremental Coding
Problems of Previous examples:
 Need high precision
 No output is generated until the entire sequence is encoded.
 Decoder needs to read all input before decoding.

 Key Observation:
 As the RANGE reduces, many MSB’s of LOW and HIGH become
identical:
 Example: Binary form of 0.7712 and 0.773504:
0.1100010.., 0.1100011..,
 We can output identical MSB’s and re-scale the rest:
  Incremental encoding
 Can achieve infinite precision with finite-precision integers.
 Three scaling scenarios: E1, E2, E3

 Important Rules: Apply as many scalings as possible before

further encoding and decoding.
Z. Li, ECE/CS 5578 Multimedia Communciation p.31
 E1(lower half) and E2(higher half) Scaling

E1: [LOW HIGH) in [0, 0.5) 0 0.5 1.0

 LOW: 0.0xxxxxxx (binary),
 HIGH: 0.0xxxxxxx.

0 0.5 1.0
 Output 0, then shift left by 1 bit
 [0, 0.5) [0, 1): E1(x) = 2 x

 E2: [LOW HIGH) in [0.5, 1) 0 0.5 1.0

 LOW: 0.1xxxxxxx,
 HIGH: 0.1xxxxxxx.
0 0.5 1.0
 Output 1, subtract 0.5,
shift left by 1 bit
 [0.5, 1) [0, 1): E2(x) = 2(x - 0.5)
0 0.5 1.0
 The 3rd scaling, E3(mid), will be studied later:
LOW < 0.5, HIGH > 0.5, but range < 0.5.

Z. Li, ECE/CS 5578 Multimedia Communciation p.32

 Encoding with E1 and E2 Symbol Prob.
1 0.8
Input 1
2 0.02
0 0.8 1.0
3 0.18
Input 3
0 0.656 0.8 E2: Output 1
Input 2 2(x – 0.5)
0.312 0.5424 0.54816 0.6 E2: Output 1

0.0848 0.09632
E1: 2x, Output 0
0.1696 0.19264 E1: Output 0

0.3392 0.38528 E1: Output 0

0.6784 0.77056 E2: Output 1

Input 1
Encode any value
0.3568 0.54112 in the tag, e.g., 0.5
Output 1
0.3568 0.504256 All: 1100011 (0.7734)
Z. Li, ECE/CS 5578 Multimedia Communciation 33
 To verify
LOW = 0.5424 (0.10001010... in binary),
HIGH = 0.54816 (0.10001100... in binary).
So we can send out 10001 (0.53125)
 Equivalent to E2E1E1E1E2
After left shift by 5 bits:
 LOW = (0.5424 – 0.53125) x 32 = 0.3568
 HIGH = (0.54816 – 0.53125) x 32 = 0.54112
 Same as the result in the last page.

(In this example, suppose 7 bits are enough for the

decoding)

Z. Li, ECE/CS 5578 Multimedia Communciation p.34

 Comparison with Huffman
 Rule: Complete all possible scaling before
Symbol Prob.
encoding the next symbol
1 0.8
2 0.02
Input Symbol 1 does not cause any output 3 0.18
Input Symbol 3 generates 1 bit
Input Symbol 2 generates 5 bits
 Symbols with larger probabilities generates less
number of bits.
 Sometimes no bit is generated at all
 Advantage over Huffman coding
 Large probabilities are desired in arithmetic coding
 Can use context-adaptive method to create larger probability
and to improve compression ratio.
Z. Li, ECE/CS 5578 Multimedia Communciation p.35
 Incremental Decoding
If input is 1100011, the 1st symbol can be decoded without ambiguity after 5 bits
are read. When 6 bits are read, the status is:
Read 5 bits: Decode 1.
After reading 6 bits:
0 0.8 1.0 Tag: 110001, 0.765625
No scaling.
0 0.656 0.8 Decode 3, E2 scaling:
Shift out 1 bit, read 1 bit
Tag: 100011 (0.546875)
0.312 0.5424 0.54816 0.6
Decode 2, E2 scaling
Tag: 000110 (0.09375)
0.0848 0.09632 E1: Tag: 001100 (0.1875)

0.1696 0.19264 E1: Tag: 011000 (0.375)

0.3392 0.38528 E1: Tag: 110000 (0.75)

0.6784 0.77056 E2: Tag: 100000 (0.5)

0.3568 0.54112 Decode 1

Z. Li, ECE/CS 5578 Multimedia Communciation p.36
 Encoding Pseudo Code with E1, E2
 Rule: Complete all possible scalings Symbol Prob. CDF
before further decoding. Adjust LOW, 1 0.8 0.8
HIGH and Tag together. 2 0.02 0.82

(For floating-point implementation) 3 0.18 1

EncodeSymbol(n) {
//Update variables
RANGE = HIGH - LOW;
HIGH = LOW + RANGE * CDF(n);
LOW = LOW + RANGE * CDF(n-1);

//Apply all possible scalings before encoding the //next symbol

while LOW, HIGH in [0, 0.5) or [0.5, 1) {
Output 0 for E1 and 1 for E2
scale LOW, HIGH by E1 or E2 rule
}
}

Z. Li, ECE/CS 5578 Multimedia Communciation p.37

 Decoding Pseudo Code with E1, E2

(For floating-point implementation)

DecodeSymbol(Tag) {
RANGE = HIGH - LOW;
n = 1;
While ( (tag - LOW) / RANGE >= CDF(n) ) {
n++;
}
Symbol Prob. CDF
HIGH = LOW + RANGE * CDF(n); 1 0.8 0.8
LOW = LOW + RANGE * CDF(n-1);
2 0.02 0.82
//keep scaling before decoding next symbol 3 0.18 1
while LOW, HIGH in [0, 0.5) or [0.5, 1) {
scale LOW, HIGH by E1 or E2 rule
read one more bit, update Tag
}
return n;
}

Z. Li, ECE/CS 5578 Multimedia Communciation p.38

 E3 Scaling: [0.25, 0.75)[0, 1)
 If RANGE straddles 1/2, E1 and E2 cannot be applied,
but the range can be quite small
 Example: LOW=0.4999, HIGH=0.5001
Binary: LOW=0.01111…., HIGH=0.1000…
 We may not have enough bits to represent the interval.

0.25 0.5 0.75

0 1
 E3 Scaling:
[0.25, 0.75) [0, 1):
0 0.5 1
E3(x) = 2(x - 0.25)

Z. Li, ECE/CS 5578 Multimedia Communciation p.39

 Example
 Previous example without E3:
Input 1
0 0.8 1.0
Input 3
0 0.656 0.8 (E2: Output 1)
Input 2
0.312 0.5424 0.54816 0.6 E2: Output 1

0.0848 0.09632
E1: Output 0
0.1696 0.19264
 With E3:
0.312 0.6

0.124 0.5848 0.59632 0.7 E3: (x-0.25)x2

Input 2
E2: Output 1
0.1696 0.19264
 state after E2°E3 = state after E1°E2, but outputs are different…
Z. Li, ECE/CS 5578 Multimedia Communciation 40
 Encoding Operation with E3

 Without E3:
Input 2
0.312 0.5424 0.54816 0.6 E2: Output 1

0.0848 0.09632 E1: Output 0

0.1696 0.19264
 With E3:
0.312 0.6

0.124 0.5848 0.59632 0.7 E3 (no output)

Input 2
E2: Output 1
0.1696 0.19264 Output 0 here!

 Don’t send anything when E3 is used, but send a 0 after E2:

 Same output, same final state  Equivalent operations

Z. Li, ECE/CS 5578 Multimedia Communciation 41

 Decoding for E3 Input 1100011
Read 6 bits:
Tag: 110001 (0.765625)
0 Without E3: 0.8 1.0 Decode 1

0 0.656 0.8 Decode 3, E2 scaling

Tag: 100011 (0.546875)
0.312 0.5424 0.54816 0.6 Decode 2, E2 scaling
Tag: 000110 (0.09375)
0.0848 0.09632 E1:
Tag: 001100 (0.1875)
0.1696 0.19264
 With E3:
0.312 0.6 Tag: 100011 (0.546875)

0.124 0.5848 0.59632 0.7 E3:

Tag: 100110 (0.59375)
Decode 2, E2 scaling
Tag: 001100 (0.1875)
0.1696 0.19264
 Apply E3 whenever it is possible, everything else is same.
Z. Li, ECE/CS 5578 Multimedia Communciation p.42
 Summary of Different Scalings

0 0.25 0.5 0.75 1.0

Need E1 scaling

Need E2 scaling

Need E3 scaling

No scaling is required.
Continue to
encode/decode the next
symbol.

Z. Li, ECE/CS 5578 Multimedia Communciation p.43

 Summary
 VLC is the real world image coding solution
 Elements of Hoffman and Golomb coding schemes are
incorporated
 JPEG: introduced DC prediction , AC zigzag scan, run-level
VLC
 H264: introduced reverse order coding.

Z. Li, ECE/CS 5578 Multimedia Communciation p.44