Error Concealment for Shape in MPEG4 ObjectBased Coding

1
Error Concealment for Shape in MPEG-4
Object-Based Coding

• Student Jieh-Yuean Hsu
• Advisor Prof. David W. Lin

2
Reference

• 1Meng-Yuan Liu, Real-Time Implementation of
  MPEG-4 Video Encoder Using SIMD-Enhanced Intel
  Process , M.S. thesis, Department of Electrical
  Engineering, Nation Chiao Tung University,
  Hsinchu, Taiwan, July 2004
• 2Cheng Huang and Paul Salama, Error
  Concealment for Shape in MPEG-4 Object-Based
  Video Coding, IEEE Transactions on Image
  Processing, Vol. 14, No. 4, April 2005.

3
Reference

• 3Guido M. Schuster, Aggelos K. Katsaggelos,
  Motion Compensated Shape Error Concealment, IEEE
  Transactions on Image Processing, Vol. 15, NO. 2,
  February 2006

4
Outline

• Introduction of MPEG-4 Object-Based Coding
• Error Concealment
• Boundary Extraction
• Boundary patching
• Boundary Filling
• Experimental Result
• Conclusion

5
Outline

• Introduction of MPEG-4 Object-Based Coding
• Error Concealment
• Boundary Extraction
• Boundary patching
• Boundary Filling
• Experimental Result
• Conclusion

6
VO and VOP

• VO Video Objects
• VOP Video Object Planes
• A sequence of VOPs is referred to as a VO
• I-VOP Intra-coded
• P-VOP Predictive-coded
• B-VOP Bidirectionally predictive-coded

7
VO and VOP
8
VOP information

• Spatial (Intra)
• Shape (Object-Based Only)
• Texture
• Temporal (Inter)
• Motion

9
Structure of VO Encoder
10
Shape Information

• Alpha plane
• Binary alpha plane
• Gray scale alpha plane

11
Binary Mask
16
BAB
Transparent
16
Opaque
Boundary
12
Outline

• Introduction of MPEG-4 Object-Based Coding
• Error Concealment
• Boundary Extraction
• Boundary patching
• Boundary Filling
• Experimental Result
• Conclusion

13
Basic Assumption

• The shape of the objects does not drastically
  change between the current VOP and the reference
  VOP
• It is assumed that the reference VOP is undamaged

14
Boundary Extraction

• Use the 8-neighborhood
• If any pixel in the 8-neighborhood of a current
  pixel doesnt belong the object boundary
• The boundary will be 4-connected

15
Boundary Extraction
16
4-connected Boundary
17
Boundary Patching

• Find the number of endpoints
• A boundary pixel having only one 4-connected
  neighbor
• The number of endpoints must be even
• Partition the endpoints in the current VOP

18
Boundary Patching
Endpoints
A
a
b
c
B
C
19
Boundary Patching

• Patching by Global Motion Compensation
• Some parameter
• Scale
• Centroid

20
Boundary Patching

• orientation angle
• Rotation angle

21
Boundary Patching

• Current VOP Reference VOP

22
Boundary Patching

• Reference VOP Current VOP

23
Boundary Patching
A
A
a
a
b
b
c
c
Mapping
B
B
C
C
Current VOP
Reference VOP
24
Boundary Patching

• Traverse the boundary of the ref. VOP, and record
  the order of endpoints
• ex
• Map each pixel in curve whose endpoints belong to
  different pair
• ex , ,

25
Boundary Patching
A
Recovered pixel
a
b
c
B
C
Reference VOP
Current VOP
26
Boundary Filling

• Filling in the closed boundary of the VOP with
  opaque pixels
• After filling, we can get the recovered VOP

Recovered VOP
27
Outline

• Introduction of MPEG-4 Object-Based Coding
• Error Concealment
• Boundary Extraction
• Boundary patching
• Boundary Filling
• Experimental Result
• Conclusion

28
Experimental Result

• brea_qcif, 30 fps

29
Experimental Result
30
Experimental Result

• Akiyo_cif, 10fps

31
Experimental Result
32
Experimental Result

• brea_qcif, 10 fps

33
Outline

• Introduction of MPEG-4 Object-Based Coding
• Error Concealment
• Boundary Extraction
• Boundary patching
• Boundary Filling
• Experimental Result
• Conclusion

34
Conclusion

• The method for shape concealment based on global
  motion compensation
• It works well for video sequence with slow motion
  and high frame rate
• It can achieve good result, even the shape is
  severely damaged
• Extra information added resulted in 5 increase
  for stream size, and it is acceptable