A new subsampling-based predictive vector

clangedbivalveAI and Robotics

Oct 19, 2013 (3 years and 9 months ago)

55 views

1

A new subsampling
-
based predictive vector
quantization for image coding

Source: Signal Processing: Image Communication


17 (2002) 477

484

Authors: Ce Zhu

Date: 2002/12/26

Speaker: Chien
-
Fa Li

2

Outline


VQ


SMVQ


PPM


SB
-
PVQ


Experiments


Conclusions

3

VQ

query vector :

X=(x
1
,x
2
,…,x
k
)





Codebook

Original image

d(X,y
1
)

d(X,y
2
)

d(X,y
N
)

Codebook size=
N
*
k

N

k



index
i

(log
2
N

bits)

4

SMVQ

X1

X2

X3

X4

X5

X9

X13

L4

L8

L12

L16

U13

U14

U15

U16

U

L

X

Advantage: Save storage

Problem: Derailment

5

PPM



Advantage: less time, low bit
-
rate



Problem: non
-
smoothness, not better match,


worse flag bit



Z:2x2 codebook, TH
Z



Y:4x4 codebook, TH
Y



flag=0,use 2x2 codebook;


flag=1,use 4x4 codebook




save format :

flag

index

6

Subsampling
-
Base Predictive Vector Quantization

i1

i2

i3

i20

X

1

X

2

X

13


Z:4x2 codebook


Y:4x4 codebook

Use Z codebook


Use Y codebook


save format :

Index z

Index y

save format :

Index z

7

SB
-
PVQ

8

Experiments

9

Experiments

10

Conclusions


the better prediction


the less coding time