Predictive Grayscale Image Coding Scheme Using VQ and BTC

Predictive Grayscale Image Coding Scheme Using VQ and BTC

Author: Yu-Chen HuSource: Fundamenta Informaticae 78 (2007) 239–255

Speaker: Meng-Jing TsaiDate: 2011.12.29

2

Outline• Introduction• The proposed scheme– Block prediction technique– Vector quantization (VQ)– Block truncation coding (BTC)

• Conclusion

3

Introduction• A predictive image compression scheme that

combines the advantages of vector quantization and moment preserving block truncation coding to provide good image quality at a low bit rate.

4

• VQ is widely used because it has a simple decoding structure and requires a low bit rate. However, a lot of computational costs are needed in the codebook design procedure and the image coding procedure.

• BTC has the advantages of requiring very low computational cost and providing good

• reconstructed image quality. However, a high bitrate is needed in BTC.

5

• To reduce the bit rate

6

Vector quantization (VQ)image vector :x=(x1,x2,…,xk)

…

…

Codebook

Original image

d(x,y1)

d(x,y2)

d(x,yN)

NiiyY 1

),...,,( 111 11 kyyy

),...,,( 222 11 kyyy

),...,,(11 kNNN yyy

),(min,...,2,1

min iNi

yxdd

index of ymin

7

VQ encoding procedure• Find Closest codeword

Codebook012

255254253

IndexImage vector(21,44,45,78) 2

(24,45,47,76)

(33,32,97,98)(78,22,77,20)

(20,55,53,26)

(20,25,17,11)(12,14,13,16)

8

VQ encoding procedure

ImageIndex table

w

Index table

h

9

VQ decoding procedure

ImageIndex table

wh

Index table

10

Block truncation coding (BTC)• 區塊臨界值 =區塊平均值

11

qqmxb

qmqxa

mx

m

iix

個數大於等於臨界值的像素 : q

sizeblock :m

11

≤

12

Block prediction technique

13

≤

14

≤≤

15

BTC

251 7 5 43 169 12 0 16 255 247 200 37 47 56 114 121

98.75

block 44 aConsider

th

x

xx

x

204 17 17 17 204 17 17 17

204 204 204 17 17 17 204 204

: isblock tedReconstruc

1 0 0 01 0 0 01 1 1 00 0 1 1

: is map-Bit

2042.204177.16

795.92

baq

16

≤

17

18

19

• Experimental results of image quality of related BTC scheme and VQ with different codebook sizes

20

• Results of block prediction technique for BTC and VQ with different search range sr

21

22

23

Conclusion• This scheme exploits the block similarities

among neighboring image blocks to improve the bit rate of traditional hybrid methods based on VQ and BTC.

• By choosing adequate controlling thresholds, the proposed scheme provides good image quality at a low bit rate.