1
Ferdous A. Sohel Ferdous A. Sohel , Prof. Laurence S. Dooley and Dr. Gour C. Karmakar , Prof. Laurence S. Dooley and Dr. Gour C. Karmakar Gippsland School of Computing and Information System Gippsland School of Computing and Information System Monash University, Churchill. Victoria 3842 Monash University, Churchill. Victoria 3842 Contact address Contact address E-mail: [email protected] u Reference: [1]. A.K. Katsaggelos, L.P. Kondi, F.W. Meier, J. Ostermann, and G.M. Schuster, MPEG-4 and Rate-Distortion-Based Shape-Coding Techniques,” Proceedings of IEEE, vol. 86, pp. 1126-1154, June 1998. Shape Codec Which one is BETTER ? QUALITY A D B G H K BK BA WHAT IS THE MINIMUM DISTANCE OF POINT B FROM LINE AD ?? So The SHORTEST ABSOLUTE DISTANCE alone cannot do ALL 6 8 10 12 14 16 18 2 3 4 5 6 7 8 What is the distortion at the corners ?? 1 pel 3.1 pel !! ! Region 3 Region 2 Region 1 F E A D B C P Q s r Step 1 If the shape point is in region 1 (opposite side of the perpend shortest absolute distance as the distortion metric, d (A, Else if it is in region 2 same side of the perpendicular lines use the direct Euclidean distortion measure d (A, Shape_poin Else use the direct Euclidean distortion measure d (D, Shape_ (1) (2) 2 , 1 , 2 , 1 , , 1 , , 1 , 1 , , 1 1 ) ( ) ( | ) )( ( ) )( ( | ) , , ( y k y k x k x k x k x k y k y y k y k x k x k k s s s s s s s q s s s q q s s d 2 2 1 2 2 1 2 1 , y y x x p p d Step 2 Object Shape (class one) (pel) ORD-ADMSC (distortion) ORD-original (distortion) Class one (pel) Class two (pel 2 ) Class one (pel) Class two (pel 2 ) Miss America (Neck) 1 1 3.62 1.414 4.43 Miss America (Lip) 1 1 3.76 1.414 4.43 Fish 2 2 15.66 2.2 17.73 Fish 3 3 23.2 4 17.73 Butterfly 1 1 9.67 10.77 90.37 Butterfly 2 2 28.1 13.93 102.3 Synthetic 1 1 1.48 3.16 6.38 30 35 40 45 15 20 25 30 30 35 40 45 15 20 25 30 0 20 40 60 0 10 20 30 40 50 60 70 80 90 100 0 20 40 60 0 10 20 30 40 50 60 70 80 90 100 80 90 100 110 120 110 115 120 125 130 135 140 80 90 100 110 120 115 120 125 130 135 140 80 85 90 95 100 105 110 80 85 90 95 100 80 85 90 95 100 105 110 80 85 90 95 100 Given maximum distortion = 1 pel [1]- D_max=1.4pel DMSC- D_max=1 pel [1]- D_max=10 pel DMSC-D_mx=1 pe 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 Given D_max = 3 pel [1]- D_max >3pel DMSC- D_max=3 pel 6 8 10 12 14 16 18 2 3 4 5 6 7 8 6 8 10 12 14 16 18 2 3 4 5 6 7 8 [1]- D_max=4.5 pel DMSC- D_max=2 pel Given D_max = 2 pel Results Abstract Efficient encoding of object boundaries has become increasingly prominent in areas such as content-based storage and retrieval, studio and television post- production facilities, mobile communications and other real-time multimedia applications. The way distortion between the actual and approximated shapes is measured however, has a major impact upon the quality of the shape coding algorithms. In existing shape coding methods, the distortion measure do not generate an actual distortion value, so this paper proposes a new distortion measure, called a modified distortion measure for shape coding (DMSC) which incorporates an actual perceptual distance. The performance of the Operational Rate Distortion optimal algorithm [1] incorporating DMSC has been empirically evaluated upon a number of different natural and synthetic arbitrary shapes. Both qualitative and quantitative results confirm the superior results in comparison with the ORD algorithm for all test shapes, without any increase in computational complexity. Challenges: Measure the actual distance of a particular point from an edge. Hence, measure the accurate distortion at that particular shape point while approximating the shape using polygons. Motivation: Operational-Rate-Distortion based Shape coding is a challenging task. A correct distortion metric is important to ensure the quality of the encoding system. The existing ORD shape coding algorithms (e.g., [1]) use the shortest absolute distance as the distortion metric. It cannot always calculate the distortion correctly specially, for the shapes having sharp corners. Conclusions: Accurately calculate the distortion. Can be seamlessly integrated into the ORD algorithms [1]. Makes guarantee on the distortion measurements. Has the same computational complexity order of the Shortest absolute distance metric. How DMSC Works: Step 1: Find the Position of the shape point with respect to the approximating polygon. - Draw perpendicular lines through the end-points of the approximating line. - Determine the region where the shape point belongs to. Step 2: Measure the distance in accordance with the region (position of the shape). Step 3: Use this distortion hence forth in the ORD algorithms [1].

Ferdous A. Sohel, Prof. Laurence S. Dooley and Dr. Gour C. Karmakar Gippsland School of Computing and Information System Monash University, Churchill

Embed Size (px)

Citation preview

Page 1: Ferdous A. Sohel, Prof. Laurence S. Dooley and Dr. Gour C. Karmakar Gippsland School of Computing and Information System Monash University, Churchill

Ferdous A. SohelFerdous A. Sohel, Prof. Laurence S. Dooley and Dr. Gour C. Karmakar, Prof. Laurence S. Dooley and Dr. Gour C. KarmakarGippsland School of Computing and Information SystemGippsland School of Computing and Information System

Monash University, Churchill. Victoria 3842Monash University, Churchill. Victoria 3842

Contact addressContact address

E-mail: [email protected]

Reference: [1]. A.K. Katsaggelos, L.P. Kondi, F.W. Meier, J. Ostermann, and G.M. Schuster, “MPEG-4 and Rate-Distortion-Based Shape-Coding Techniques,” Proceedings of IEEE, vol. 86, pp. 1126-1154, June 1998.

ShapeCodec

Which one is

BETTER ?

QUALITY

A D

BG

HK

BKBA

WHAT IS THE MINIMUM DISTANCE OF POINT B FROM LINE AD ??

So The SHORTEST ABSOLUTE DISTANCE alone cannot do ALL

6 8 10 12 14 16 182

3

4

5

6

7

8

What is the distortion at the corners ??

1 pel

3.1 pel !!!

Region 3Region 2 Region 1F E

A D

B CP Q

s r

Step 1

If the shape point is in region 1 (opposite side of the perpendicular lines) use the shortest absolute distance as the distortion metric, d (A, D, Shape_point). Else if it is in region 2 same side of the perpendicular lines and close to End point A use the direct Euclidean distortion measure d (A, Shape_point). EQ: (2). Else use the direct Euclidean distortion measure d (D, Shape_point). EQ: (2).

(1)

(2)

2,1,

2,1,

,1,,1,1,,11

)()(

|))(())((|),,(

ykykxkxk

xkxkykyykykxkxkk

ssss

sssqsssqqssd

221

22121, yyxxppd

Step 2

Object Shape (class one) (pel)

ORD-ADMSC (distortion) ORD-original (distortion)

Class one (pel) Class two (pel2) Class one (pel) Class two (pel2)

Miss America (Neck)

1 1 3.62 1.414 4.43

Miss America (Lip) 1 1 3.76 1.414 4.43

Fish 2 2 15.66 2.2 17.73

Fish 3 3 23.2 4 17.73

Butterfly 1 1 9.67 10.77 90.37

Butterfly 2 2 28.1 13.93 102.3

Synthetic 1 1 1.48 3.16 6.38

Synthetic 2 2 4 4.47 14.95

30 35 40 45

15

20

25

30

30 35 40 45

15

20

25

30

0 20 40 60

0

10

20

30

40

50

60

70

80

90

1000 20 40 60

0

10

20

30

40

50

60

70

80

90

100

80 90 100 110 120

110

115

120

125

130

135

14080 90 100 110 120

115

120

125

130

135

140

80 85 90 95 100 105 110

80

85

90

95

10080 85 90 95 100 105 110

80

85

90

95

100

Given maximum distortion = 1 pel

[1]- D_max=1.4pel DMSC- D_max=1 pel

[1]- D_max=10 pel DMSC-D_mx=1 pel

0 10 20 30 40 50 60 70

0

10

20

30

40

50

60

70

0 10 20 30 40 50 60 70

0

10

20

30

40

50

60

70

Given D_max = 3 pel [1]- D_max >3pel DMSC- D_max=3 pel

6 8 10 12 14 16 182

3

4

5

6

7

8

6 8 10 12 14 16 182

3

4

5

6

7

8

[1]- D_max=4.5 pel DMSC- D_max=2 pel

Given D_max = 2 pel

Results

Abstract

Efficient encoding of object boundaries has become increasingly prominent in areas such as content-based storage and retrieval, studio and television post-production facilities, mobile communications and other real-time multimedia applications. The way distortion between the actual and approximated shapes is measured however, has a major impact upon the quality of the shape coding algorithms. In existing shape coding methods, the distortion measure do not generate an actual distortion value, so this paper proposes a new distortion measure, called a modified distortion measure for shape coding (DMSC) which incorporates an actual perceptual distance. The performance of the Operational Rate Distortion optimal algorithm [1] incorporating DMSC has been empirically evaluated upon a number of different natural and synthetic arbitrary shapes. Both qualitative and quantitative results confirm the superior results in comparison with the ORD algorithm for all test shapes, without any increase in computational complexity.

Challenges: Measure the actual distance of a particular point from an edge. Hence, measure the accurate distortion at that particular shape point while approximating the shape using polygons.

Motivation: Operational-Rate-Distortion based Shape coding is a challenging task. A correct distortion metric is important to ensure the quality of the encoding system. The existing ORD shape coding algorithms (e.g., [1]) use the shortest absolute distance as the distortion metric. It cannot always calculate the distortion correctly specially, for the shapes having sharp corners.

Conclusions: Accurately calculate the distortion. Can be seamlessly integrated into the ORD algorithms [1]. Makes guarantee on the distortion measurements. Has the same computational complexity order of the Shortest absolute distance metric.

How DMSC Works:Step 1: Find the Position of the shape point with respect to the approximating polygon. - Draw perpendicular lines through the end-points of the approximating line. - Determine the region where the shape point belongs to.Step 2: Measure the distance in accordance with the region (position of the shape).Step 3: Use this distortion hence forth in the ORD algorithms [1].