18
Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004, pp. 9-14 Speaker: Chia-Chun Wu ( 吳吳吳 ) Date : 2004/09/08

Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

Embed Size (px)

Citation preview

Page 1: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

Improvements to the JPEG-LS prediction scheme

Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos

Source : Image and Vision Computing. Vol. 22, No. 1, 2004, pp. 9-14

Speaker: Chia-Chun Wu ( 吳佳駿 )

Date : 2004/09/08

Page 2: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

2

Outline

• Introduction to JPEG-LS

• JPEG-LS prediction scheme

• Improvements JPEG-LS prediction scheme

• Experimental results

• Conclusions

• Comments

Page 3: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

3

Introduction to JPEG-LS

c b d

a

Pixel value x in original imagePixel value x in original image

Compute difference = x - Compute difference = x - ̂x

c b d

a x

Encode using entropy coding Encode using entropy coding

Determine predictive value

for x with prediction scheme

Determine predictive value

for x with prediction scheme

Page 4: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

4

JPEG-LS prediction scheme

ˆIn smooth, , when a = b = c = dx a

c b d

a x

In nonsmoth,

min , , if max ,

ˆ max , , if min ,

, if min a,b max ,

a b c a b

x a b c a b

a b c c a b

ˆ: the predictive valuex50 50 50

50 50

31 20 40

30 20

JPEG-LS predictive template

Page 5: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

5

JPEG-LS prediction scheme

c a

b x

c a

b a

c a

b b

max( , )c a b

a b

b a

100 99

50 52

100 50

99 50

100 99

50 50

max( , )c a b a b

b a

100 50

99 52

max( , )c a b

Horizontal edge

Vertical edge

Page 6: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

6

JPEG-LS prediction scheme

c a

b x

c a

b b

c a

b a

min( , )c a b

a b

b a

40 100

41 60

40 41

100 100

40 100

41 100

min( , )c a b a b

b a

40 41

100 60

min( , )c a b

Vertical edge

Horizontal edge

Page 7: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

7

min( , ), max( , )

ˆ max( , ), min( , )

,

a b if c a b

x a b if c a b

a b c otherwise

JPEG-LS prediction scheme

c a

b x

c a

b a+b-c

ˆ 99 + 101 - 100 = 100 x

100 99

101 50

100 99

101 100

min , max ,a b c a b

100 49

101 51

100 49

101 50

ˆ 49 + 101 - 100 = 50 x

Page 8: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

8

Improvements JPEG-LS

ˆ3

a b dx

1 1

2

max , OR min a,b

AND

c a b T c T

abs a b T

1 2, Ideally T T

c b d

a x

200 50 53

50 52

200 50 53

50 51

200 50 5

50 30

200 50 8

50 36

ˆ (50+50+53)/3=153/3=51 x ˆ (50+50+8)/3=108/3=36 x

Diagonal edge Diagonal edgeT1=20, T2=0 T1=20, T2=0

Page 9: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

9

Experimental results

JPEG_LS Proposed

JPEG_LS

% Improvement_of_PMSE

PMSE PMSE*100

PMSE

N

2

ii 1

ˆxPMSE

ix

N

78 90 74

64 145 142

100 63 97

70 87 76

60 140 135

100 59 99

2 2 2 2 2 2 2 2 2PMSE=(8 3 2 4 5 7 0 4 2 ) / 9

(64 9 4 16 25 49 0 16 4) /

9

187 0 / 9 2 .7

Predictive mean squared error

N = 9

Page 10: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

10

Experimental resultsTable 1 PMSE values and lossless compression ratios of all test images, when compressed using JPEG-LS

Page 11: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

11

Experimental results2 2Clown and Camera, 15 and 5T T

Fig. 1. Performance at T1 = 11

Page 12: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

12

Experimental results

Fig. 2. Performance at T1 = 19

2 2Clown and Camera, 15 and 5T T

Page 13: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

13

Experimental results

Fig. 3. Optimum performance graphs

Page 14: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

14

Experimental resultsTable 2 Optimum performance and settings for proposed algorithm

Page 15: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

15

Experimental results

Fig. 4. Predictive error images (a) using JPEG-LS, (b) using proposed scheme

Page 16: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

16

Conclusions

• It maintains the low complexity requirement of JPEG-LS standard.

• The proposed a modifications reduce the PMSE up to 15%.

Page 17: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

17

Comments

• 如何根據影像不同的特性,動態的去調整門檻值 T1 及 T2 ?

• 是否可以利用統計分析的方法,去找出適用於一般影像的門檻值 T1 及 T2 ?

Page 18: Improvements to the JPEG-LS prediction scheme Authors: S. Bedi, E. A. Edirisinghe, and G. Grecos Source : Image and Vision Computing. Vol. 22, No. 1, 2004,

18

Sample images

Lena Baboon