An Efficient Logo Insertion Method for Video Coding

  • Slides: 16
Download presentation
An Efficient Logo Insertion Method for Video Coding in HEVC Yunchang Li (Speaker), Zhijie

An Efficient Logo Insertion Method for Video Coding in HEVC Yunchang Li (Speaker), Zhijie Huang, Jun Sun Peking University, Beijing 29 th Sept 2019

Outline o Introduction o Related Works o Proposed Method o Results & Conclusion MMSP

Outline o Introduction o Related Works o Proposed Method o Results & Conclusion MMSP 2019 2/15

Introduction o Video without Logo n Raw material n Clean o Video with Logo

Introduction o Video without Logo n Raw material n Clean o Video with Logo n Copyright announcement n Advertising MMSP 2019 3/15

Introduction o Encoding Scheme MMSP 2019 4/15

Introduction o Encoding Scheme MMSP 2019 4/15

Related Works o Previous Coding Standards o H. 264/AVC n Slice partition o H.

Related Works o Previous Coding Standards o H. 264/AVC n Slice partition o H. 265/HEVC n Region partition n Reuse Mode & MV information [1] D. Xu, P. Nasiopoulos. “Logo Insertion Transcoding for H. 264/AVC Compressed Video, ” IEEE International Conference on Image Processing, 2009: 3693– 3696. [2] Q. Jing, P. Xu, J. Sun, Z. Guo. “Efficient Logo Insertion Method for High-Resolution H. 265/HEVC Compressed Video, ” Pacific Rim Conference on Multimedia, 2018: 674 -682. MMSP 2019 5/15

Proposed Method (Intra Frame) o Region Partition 64 64 Logo 64 Re-encoding region Lossless

Proposed Method (Intra Frame) o Region Partition 64 64 Logo 64 Re-encoding region Lossless coding region 64 MMSP 2019 Information reusing region 6/15

Proposed Method (Intra Frame) o Lossless Region Refine n Use loosy coding if the

Proposed Method (Intra Frame) o Lossless Region Refine n Use loosy coding if the reconstructed pixels are p Not much different p Not referenced n Error propagation model p Determine whether CUs satisfy the above conditions MMSP 2019 7/15

Proposed Method (Intra Frame) o Error Propagation Model n Propagation Area 4 4 A

Proposed Method (Intra Frame) o Error Propagation Model n Propagation Area 4 4 A B C D E F G H 4 I E J E K F L F C D E 4 J K L F A B C D E F G H 0 2 4 4 4 2 0 0 Angular mode 31 for a 4× 4 block MMSP 2019 I A B C D I J K L 2 2 2 2 Non-angular mode for a 4× 4 block 8/15

Proposed Method (Intra Frame) o Error Propagation Model n Basic Unit (BU) 1 p

Proposed Method (Intra Frame) o Error Propagation Model n Basic Unit (BU) 1 p Propagation area (S) p Distortion (D) LOGO n Use lossy coding if p Σ S�D < T 64 0 p Threshold (T) 1 1 0 64 MMSP 2019 9/15

Proposed Method (Inter Frame) o Region Partition n Re-encoding region p CUs locate in

Proposed Method (Inter Frame) o Region Partition n Re-encoding region p CUs locate in the logo area p CUs reference the logo area n Information reusing region MMSP 2019 10/15

Proposed Method (Inter Frame) o Quality Maintenance n Record costs information of the original

Proposed Method (Inter Frame) o Quality Maintenance n Record costs information of the original sequence p SATD p RD-COST n Determine quality-loss blocks p Compare with the recorded cost n Re-encode MMSP 2019 11/15

Proposed Method o Information Reuse n Splitting information n Prediction modes n Motion vectors

Proposed Method o Information Reuse n Splitting information n Prediction modes n Motion vectors n Residual coefficients MMSP 2019 12/15

Results & Conclusion o Experiment Setup n HM 16. 20 n Standard testing sequences

Results & Conclusion o Experiment Setup n HM 16. 20 n Standard testing sequences from Class A to E n Logo size: 30× 30 or 60× 60 pixels n Logo location: top-left corner n Processors: Intel Xeon E 5 -1680, RAM 64 GB MMSP 2019 13/15

Results & Conclusion o Performance n All-Intra MMSP 2019 Class Revolution Sequences People. On.

Results & Conclusion o Performance n All-Intra MMSP 2019 Class Revolution Sequences People. On. Street Traffic Basketball. Drive BQTerrace B Cactus 1920× 1080 Kimono Park. Scene Basketball. Drill BQMall C 832× 480 Party. Scene Race. Horses Basketball. Pass Blowing. Bubbles D 416× 240 BQSquare Race. Horses Four. People E Johnny 1280× 720 Kristen. And. Sara Average A 2560× 1600 Our Jing BDRate[%] 6. 87 5. 66 11. 23 7. 29 9. 61 13. 79 5. 67 13. 88 11. 98 5. 91 7. 94 22. 14 16. 28 9. 43 15. 88 13. 84 34. 91 28. 68 13. 39 SR 21. 3 x 22. 0 x 26. 4 x 20. 6 x 20. 9 x 31. 3 x 22. 2 x 19. 2 x 18. 9 x 15. 5 x 19. 2 x 16. 0 x 15. 5 x 14. 1 x 15. 6 x 20. 6 x 25. 7 x 24. 6 x 20. 5 x BDRate[%] 0. 58 1. 29 3. 28 0. 36 0. 13 3. 07 0. 81 2. 12 3. 16 0. 77 1. 17 7. 16 1. 79 2. 43 3. 31 1. 27 1. 47 1. 24 1. 97 SR 26. 8 x 28. 4 x 33. 4 x 26. 0 x 27. 0 x 38. 8 x 28. 4 x 24. 6 x 22. 7 x 20. 4 x 24. 0 x 19. 6 x 18. 8 x 18. 0 x 19. 0 x 25. 0 x 29. 7 x 28. 7 x 25. 5 x 14/15

Results & Conclusion o Performance n Random-Access MMSP 2019 Class Revolution A 2560× 1600

Results & Conclusion o Performance n Random-Access MMSP 2019 Class Revolution A 2560× 1600 Sequences People. On. Street Traffic Basketball. Drive BQTerrace B Cactus 1920× 1080 Kimono Park. Scene Basketball. Drill C BQMall 832× 480 Party. Scene Race. Horses Basketball. Pass Blowing. Bubbles D 416× 240 BQSquare Race. Horses Four. People E Johnny 1280× 720 Kristen. And. Sara Average BDRate[%] 1. 19 0. 75 1. 38 0. 91 0. 69 0. 84 1. 09 0. 95 1. 72 1. 07 1. 81 2. 87 1. 81 1. 74 3. 22 1. 19 1. 39 1. 13 1. 43 SR 76. 5 x 122. 5 x 44. 2 x 60. 2 x 80. 3 x 74. 1 x 47. 4 x 61. 8 x 30. 5 x 7. 7 x 15. 8 x 4. 6 x 14. 8 x 20. 3 x 5. 0 x 98. 2 x 94. 0 x 81. 7 x 52. 2 x 15/15

Thank You! Contact us: liyunchang@pku. edu. cn

Thank You! Contact us: liyunchang@pku. edu. cn