Transcript DC 4 : Source Coding – Compression

Source CodingCompression
Most Topics from Digital CommunicationsSimon Haykin
Chapter 9
9.1~9.4
Fundamental Limits on Performance

Given an information source, and a noisy channel
1) Limit on the minimum number of bits
per symbol
2) Limit on the maximum rate for reliable
communication
 Shannon’s theorems
Information Theory

Let the source alphabet,
S  {s0, s1 , .. , sK -1}
with the prob. of occurrence
P(s  sk )  pk ,

K -1
k  0,1, .. , K -1
and
p
k 0
Assume the discrete memory-less source (DMS)
What is the measure of information?
k
1
Uncertainty, Information, and Entropy
(cont’)
Interrelations between info., uncertainty or surprise
No surprise
no information
1
(  Info. 
)
Pr ob.
If A is a surprise and B is another surprise,
then what is the total info. of simultaneous A and B
Info.( A  B)  Info.( A)  Info.( B)
The amount of info may be related to the inverse of
the prob. of occurrence.
1
 I ( Sk )  log( )
pk
Property of Information
1) I (s )  0 for p  1
k
k
2)
I (sk )  0 for 0  pk  1
3)
4) I (sk )  I (si ) for p k  pi
I (sk si )  I (sk )  I (si ), if sk and si statist. indep.
*
Custom is to use logarithm of base 2
Entropy (DMS)

Def. : measure of average information
contents per source symbol
The mean value of I (sk ) over S,
H ( S )  E[I ( sk )] 
The property of H
K-1
 pk I (sk ) 
k 0
K-1
 pk log 2 (
k 0
1
)
pk
0  H (S )  log2 K , where K is radix ( # of symbols)
1) H(S)=0, iff pk  1 for some k, and all other pi ' s  0
No Uncertainty
2) H(S)= log 2 K , iff pk  1 for all k
K
Maximum Uncertainty
Extension of DMS (Entropy)

Consider blocks of symbols rather them individual symbols
Coding efficiency can increase if higher order DMS are used
H(Sn) means having Kn disinct symbols where K is the # of
distinct symbols in the alphabet
Thus H(Sn) = n H(S)

Second order extension means H(S2)





Consider a source alphabet S having 3 symbols i.e. {s0, s1, s2}
Thus S2 will have 9 symbols i.e. {s0s0, s0s1, s0s2, s1s1, …,s2s2}
Average Length
For a code C with associated probabilities p(c) the average
length is defined as
la (C)   p(c)l (c)
cC
We say that a prefix code C is optimal if for all prefix
codes C’, la(C)  la(C’)
Relationship to Entropy
Theorem (lower bound): For any probability
distribution p(S) with associated uniquely decodable
code C,
H ( S )  la (C)
Theorem (upper bound): For any probability
distribution p(S) with associated optimal prefix code
C,
la (C)  H ( S ) 1
Coding Efficiency

Coding Efficiency



From Shannon’s Theorem



n = Lmin/La
where La is the average code-word length
La >= H(S)
Thus Lmin = H(S)
Thus

n = H(S)/La
Kraft McMillan Inequality
Theorem (Kraft-McMillan): For any uniquely decodable code
C,
 l ( c)
2
1

cC
Also, for any set of lengths L such that
l
2
 1
l L
there is a prefix code C such that
l (ci )  li (i  1,...,| L|)
NOTE: Kraft McMillan Inequality does not tell us
whether the code is prefix-free or not
Uniquely Decodable Codes
A variable length code assigns a bit string (codeword)
of variable length to every message value
e.g. a = 1, b = 01, c = 101, d = 011
What if you get the sequence of bits
1011 ?
Is it aba, ca, or, ad?
A uniquely decodable code is a variable length code in
which bit strings can always be uniquely decomposed
into its codewords.
Prefix Codes
A prefix code is a variable length code in which no
codeword is a prefix of another word
e.g a = 0, b = 110, c = 111, d = 10
Can be viewed as a binary tree with message values at the
leaves and 0 or 1s on the edges.
0
1
0 1
a
0
1
b
c
d
Some Prefix Codes for Integers
n
1
2
3
4
5
6
Binary
..001
..010
..011
..100
..101
..110
Unary
0
10
110
1110
11110
111110
Split
1|
10|0
10|1
110|00
110|01
110|10
Many other fixed prefix codes:
Golomb, phased-binary, subexponential, ...
Data compression implies sending or storing a
smaller number of bits. Although many methods are
used for this purpose, in general these methods can
be divided into two broad categories: lossless and
lossy methods.
Data compression methods
Run Length Coding
Introduction – What is RLE?

Compression technique


Represents data using value and run length
Run length defined as number of consecutive equal values
e.g
1110011111
RLE
130215
Values
Run Lengths
Introduction


Compression effectiveness depends on input
Must have consecutive runs of values in order to maximize
compression

Best case: all values same


Worst case: no repeating values


Can represent any length using two values
Compressed data twice the length of original!!
Should only be used in situations where we know for sure have
repeating values
Run-length encoding example
Run-length encoding for two symbols
Encoder – Results
Input: 4,5,5,2,7,3,6,9,9,10,10,10,10,10,10,0,0
Output: 4,1,5,2,2,1,7,1,3,1,6,1,9,2,10,6,0,2,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1…
Valid Output
Output Ends Here
Best Case:
Input: 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
Output: 0,16,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1…
Worst Case:
Input: 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
Output: 0,1,1,1,2,1,3,1,4,1,5,1,6,1,7,1,8,1,9,1,10,1,11,1,12,1,13,1,14,1,15,1
Huffman Coding
Huffman Codes
Invented by Huffman as a class assignment in 1950.
 Used in many, if not most compression algorithms such
as gzip, bzip, jpeg (as option), fax compression,…
 Properties:





Generates optimal prefix codes
Cheap to generate codes
Cheap to encode and decode
la=H if probabilities are powers of 2
Huffman Codes
Huffman Algorithm
 Start with a forest of trees each consisting of a single
vertex corresponding to a message s and with weight
p(s)

Repeat:


Select two trees with minimum weight roots p1 and p2
Join into single tree by adding root with weight p1 + p2
Example
p(a) = .1, p(b) = .2, p(c ) = .2, p(d) = .5
a(.1)
(.3)
b(.2)
c(.2)
(.5)
d(.5)
(1.0)
1
0
(.5) d(.5)
a(.1) b(.2)
(.3)
c(.2)
1
0
Step 1
(.3)
c(.2)
a(.1) b(.2)
0
1
Step 2
a(.1) b(.2)
Step 3
a=000, b=001, c=01, d=1
Encoding and Decoding
Encoding: Start at leaf of Huffman tree and follow path
to the root. Reverse order of bits and send.
Decoding: Start at root of Huffman tree and take branch
for each bit received. When at leaf can output message
and return to root.
There are even faster methods that
can process 8 or 32 bits at a time
(1.0)
1
0
(.5) d(.5)
1
0
(.3)
c(.2)
0
1
a(.1) b(.2)
Huffman codes Pros & Cons

Pros:


The Huffman algorithm generates an optimal prefix code.
Cons:




If the ensemble changes the frequencies and probabilities change 
the optimal coding changes
e.g. in text compression symbol frequencies vary with context
Re-computing the Huffman code by running through the entire file in
advance?!
Saving/ transmitting the code too?!
Lempel-Ziv (LZ77)
Lempel-Ziv Algorithms
LZ77 (Sliding Window)
 Variants: LZSS (Lempel-Ziv-Storer-Szymanski)
 Applications: gzip, Squeeze, LHA, PKZIP, ZOO
LZ78 (Dictionary Based)
 Variants: LZW (Lempel-Ziv-Welch),
LZC (Lempel-Ziv-Compress)
 Applications:
compress, GIF, CCITT (modems), ARC, PAK

Traditionally LZ77 was better but slower, but the gzip version is
almost as fast as any LZ78.
Lempel Ziv encoding
Lempel Ziv (LZ) encoding is an example of a
category of algorithms called dictionary-based
encoding. The idea is to create a dictionary (a table)
of strings used during the communication session. If
both the sender and the receiver have a copy of the
dictionary, then previously-encountered strings can
be substituted by their index in the dictionary to
reduce the amount of information transmitted.
Compression
In this phase there are two concurrent events:
building an indexed dictionary and compressing a
string of symbols. The algorithm extracts the smallest
substring that cannot be found in the dictionary from
the remaining uncompressed string. It then stores a
copy of this substring in the dictionary as a new entry
and assigns it an index value. Compression occurs
when the substring, except for the last character, is
replaced with the index found in the dictionary. The
process then inserts the index and the last character
of the substring into the compressed string.
An example of Lempel Ziv encoding
Decompression
Decompression is the inverse of the compression
process. The process extracts the substrings from the
compressed string and tries to replace the indexes
with the corresponding entry in the dictionary, which
is empty at first and built up gradually. The idea is
that when an index is received, there is already an
entry in the dictionary corresponding to that index.
An example of Lempel Ziv decoding