Adaptiv Huffman Coding

Adaptive Huffman Coding CSEP 590 Data Compression
Autumn 2007 One pass During the pass calculate the frequencies Update the Huffman tree accordingly
Coder new Huffman tree computed after transmitting the symbol Decoder new Huffman tree computed after receiving the symbol
Adaptive Huffman Coding
Symbol set and their initial codes must be known ahead of time. Need NYT (not yet transmitted symbol) to indicate a new leaf is needed in the tree.
CSEP 590 - Lecture 2 - Autumn 2007 2
Optimal Tree Numbering

a : 5, b: 2, c : 1, d : 3
Weight the Nodes

a : 5, b: 2, c : 1, d : 3
11 5 6 3 3
a d
a d
1 2
CSEP 590 - Lecture 2 - Autumn 2007
Number the Nodes

a : 5, b: 2, c : 1, d : 3
11 5 5 a 3 1
Adaptive Huffman Principle

In an optimal tree for n symbols there is a numbering of the nodes y1<y2<... <y2n-1 such that their corresponding weights x1,x2, ... , x2n-1 satisfy:
7
6
6
3
3
2
d c 1 b 2
x1 < x2 < ... < x2n-1 siblings are numbered consecutively
And vice versa

That is, if there is such a numbering then the tree is optimal. We call this the node number invariant.
Number the nodes as they are removed from the priority queue.
Initialization
Symbols a1, a2, ... ,am have a basic prefix code, used when symbols are first encountered. Example: a, b ,c, d, e, f, g, h, i, j
0 0 0 1 0 1 1 0 1 0 1 0 1 0 1 1 1
Initialization
The tree will encode up to m + 1 symbols including NYT. We reserve numbers 1 to 2m + 1 for node numbering. The initial Huffman tree consists of a single node weight
0 NYT
2m + 1
b c
f 0 g
node number
h i
j
7 CSEP 590 - Lecture 2 - Autumn 2007 8
Coding Algorithm
1. If a new symbol is encountered then output the code for NYT followed by the fixed code for the symbol. Add the new symbol to the tree. 2. If an old symbol is encountered then output its code. 3. Update the tree to preserve the node number invariant.
Decoding Algorithm
1. Decode the symbol using the current tree. 2. If NYT is encountered then use the fixed code to decode the symbol. Add the new symbol to the tree. 3. Update the tree to preserve the node number invariant.
10
Updating the Tree

1. Let y be leaf (symbol) with current weight x.* 2. If y the root update x by 1, otherwise, 3. Exchange y with the largest numbered node with the same weight (unless it is the parent).** 4. Update x by 1 5. Let y be the parent with its weight x and go to 2. *We never update the weight of NYT ** This exchange will preserve the node number invariant output = 000
fixed code
Example
aabcdad in alphabet {a,b,..., j}
0 NYT
21
fixed code for a
12
Example
aabcdad
1
0
Example
aabcdad
1
0
21
1
21
1
0 19 NYT
1 a
20
0 19 NYT
1 a
20
output = 000
output = 0001
Example
aabcdad
1
0
Example
aabcdad
2
0
21
1
21
1
0 19 NYT
2 a
20
0 19 NYT
2 a
20
NYT
fixed code for b
output = 0001
output = 00010001
Example
aabcdad
2
0
Example
aabcdad
2
0
21
1
21
1
0
0
19
1
2 a
20
0
0 0 17 NYT
19
1
2 a
20
0 17 NYT
0 a b
18
1 a b
18
output = 00010001
output = 00010001
Example
aabcdad
2
0
Example
aabcdad
3
0
21
1
21
1
1
0
19
1
2 a
20
0
1 0 17 NYT
19
1
2 a
20
0 17 NYT
1 a b
18
1 a b
18 fixed code for c
NYT output = 00010001

output = 0001000100010
Example
aabcdad
0
Example
aabcdad
0
3 1
0
21
1
3 1
0
21
1
19
1
2 a
20
0
0
19
1
2 a
20
0
0
17
1
a 1 b
18
0 NYT
17
1
a 1 b
18
0 NYT
a 16 15 0 c
a 16 15 1 c
output = 0001000100010
output = 0001000100010
Example
aabcdad
0
Example
aabcdad
0
3 1
0
21
1
3 2
0
21
1
19
1
2 a
20
1
0
19
1
2 a
20
1
0
17
1
a 1 b
18
0 NYT
17
1
a 1 b
18
0 NYT
1 16 15 a c
1 16 15 a c
output = 0001000100010
output = 0001000100010
Example
aabcdad
0
Example
aabcdad
2 4
0
21
1
4 2
0
21
1
19
1
19
1
2 a
20
0
2 a
20
1 0
0
17
1
a 1 b
18
1
0
17
1
a 1 b
18
0 NYT
1 16 15 a c
1 16 15 a
1
fixed code for d NYT
0 NYT
13
0 a d
14
output = 0001000100010000011
output = 0001000100010000011
Example
aabcdad
2
0
Example
aabcdad
4
0
4
0
21
1
21
1
19
1
2 a
20
0
2 1
0
19
1
2 a
20 exchange!
1
0
17
1
a 1 b
18
1
0
17
1
a 1 b
18
0
0
1 16 15 a
1
1 16 15 a
1
0 NYT
13
1 a d
14
0 NYT
13
1 a d
14
output = 0001000100010000011
output = 0001000100010000011
Example
aabcdad
2
0
Example
aabcdad
4
0
4
0
21
1
21
1
19
1
2 a 1
20
0
2 1 b
0
19
1
2 a 2
20 exchange!
1 b
17
0
18
1
17
0
18
1
1
0
a 16 15 1
1
1 0 NYT
a 16 15 1
1
0 NYT
13
1 a d
14
13
1 a d
14
output = 0001000100010000011
output = 0001000100010000011
Example
aabcdad
4
0
Example
aabcdad
4
0
21
1
21
1
2 a
19
0
20
1
2 a
19
0
20
1
1 b
17
0
2 1
18
1
1 b
17
0
2 1
18
1
15
1
1 a c
16
0
1 16 15 a
1
0 NYT
13
1 a d
14
0 NYT
13
1 a d
14
output = 0001000100010000011
output = 0001000100010000011
Example
aabcdad
5
0
Example
Note: the first a is coded as 000, the second as 1, and the third as 0
21
1
aabcdad
5
0
21
1
2 a
19
0
20
1
3 a
19
0
20
1
1 b
17
0
2 1
18
1
1 b
17
0
2 1
18
1
15
1
1 a c
16
0
1 16 15 a
1
0 NYT
13
1 a d
14
0 NYT
13
1 a d
14
output = 00010001000100000110
output = 00010001000100000110
Example
aabcdad
6
0
Example
aabcdad
6
0
21
1
21
1
3 a
19
0
20
1
exchange! 18
1
3 a
19
0
20
1
1 b
17
0
2 1
1 d
17
0
2 1
18
1
a 16 15 1
1
a 16 15 1
1
0 NYT
13
1 a d
14
0 NYT
13
1 a b
14
output = 000100010001000001101101
output = 000100010001000001101101
Example
aabcdad
6
0
Example
aabcdad
6
0
21
1
21
1
3 a
19
0
20
1
3 a
19
0
20
1
2 d
17
0
2 1
18
1
2 d
17
0
2 1
18
1
15
1
1 a c
16
0
1 16 15 a
1
0 NYT
13
1 a b
14
0 NYT
13
1 a b
14
output = 000100010001000001101101
output = 000100010001000001101101
Example
aabcdad
7
0
Data Structure for Adaptive Huffman

a b c d . . . j NYT
39
21
1 0
7
1
3 a
19
0
20
1
3 a
4
0 1
2 d
17
0
2 1
18
1
2 d
0 0
2
1
1. Fixed code table 2. Binary tree with parent pointers 3. Table of pointers nodes into tree 4. Doubly linked list to rank the nodes 1 a c
15
1
1 a c
16
1
1
0 NYT
13
1 a b
14
0 NYT
a 1 b
output = 000100010001000001101101
CSEP 590 - Lecture 2 - Autumn 2007 CSEP 590 - Lecture 2 - Autumn 2007 40
In Class Exercise
Decode using adaptive Huffman coding assuming the following fixed code
0 0 0 1 0 1 1 0 1 0 1 0 1 1
Huffman Summary

41
b c
00110000
Statistical compression algorithm Prefix code Fixed-to-variable length code Optimization to create a best code Symbol merging Context Adaptive coding Decoder and encoder behave almost the same Need for data structures and algorithms

Adaptiv Huffman Coding

Uploaded by

Copyright:

Available Formats

Adaptiv Huffman Coding

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Adaptiv Huffman Coding

Uploaded by

Copyright:

Available Formats

Adaptive Huffman Coding CSEP 590 Data Compression

Adaptive Huffman Coding

Optimal Tree Numbering

Weight the Nodes

CSEP 590 - Lecture 2 - Autumn 2007

CSEP 590 - Lecture 2 - Autumn 2007

Number the Nodes

Adaptive Huffman Principle

x1 < x2 < ... < x2n-1 siblings are numbered consecutively

And vice versa

CSEP 590 - Lecture 2 - Autumn 2007

CSEP 590 - Lecture 2 - Autumn 2007

CSEP 590 - Lecture 2 - Autumn 2007

CSEP 590 - Lecture 2 - Autumn 2007

Updating the Tree

fixed code for a

CSEP 590 - Lecture 2 - Autumn 2007

fixed code for b

18 fixed code for c

NYT output = 00010001

fixed code for d NYT

Data Structure for Adaptive Huffman

You might also like