SlideShare a Scribd company logo
1 of 30
Download to read offline
1/30




    Master Thesis Presentation:
Real-time Rate-distortion Oriented Joint
  Video Denoising and Compression
                 Junsheng Fu
         Signal Processing Department
        Tampere University of Technology




                                           03/12/2011
2/30


Outline



1. A real-time filter

2. Joint video pre-filtering and compression

3. Joint video in-loop filtering and compression




                                                   03/12/2011
3/30




     Part 1:

A real-time filter




                     03/12/2011
4/30


   VBM3D (Video Block-Matching 3D Filtering)




K. Dabov, A. Foi, and K. Egiazarian, “Video denoising by sparse 3D transform-domain collaborative filtering,” in Proc. 15th
European Signal Processing Conference, EUSIPCO, Poznan, Poland, September 2007



                                                                                                                             03/12/2011
5/30


 Filter performance


            Res: 640x480                   Standard
              Sigma:5                       VBM3D

                        Noisy (dB)           34.15
       Vassar         Denoised (dB)          40.81
                       Speed (fps)            1.12
                        Noisy (dB)           34.15
                                                               Computer Platform
     ballroom         Denoised (dB)          41.55             Processor: Intel Core 2 Duo 3GHz
                       Speed (fps)            1.14             Memory: 3.2Gb of RAM




Video sequences vassar and ballroom are from http://www.merl.com/pub/avetro/mvc-testseq/orig-yuv/.




                                                                                                     03/12/2011
6/30


Optimization Task


1. Accelerate filter speed: above 25 fps for video (640 x 480)
   In a computer
    •   Processor: Intel Core 2 Duo 3GHz
    •   Memory: 3.2Gb of RAM


Approaches
1. Simplify VBM3D filter by only using most influential parts for
   noise attenuation.
2. Propose a real-time integer implementation of simplified
   VBM3D.




                                                                    03/12/2011
7/30
                Approach 1:
                What to remove to Simplify VBM3D?

                         Thresholding filtering                        Wiener filtering

                                                                                                    Temporal      Max
  Filters                                                                                           searching   grouping
              Spatial Spatial Temporal Thres- Temporal Spatial Spatial Temporal Wiener Temporal frames           blocks
              search transform search holding transform search transform search filtering transform




 Standard
  VBM3D         +       +         +         +     +        +       +         +        +       +        9           8



*Simplified
  VBM3D         -        -        +         +     +        -       -         -            -   -        5           4


     * Simplified VBM3D means turn off some features of VBM3D filter.


                                                                                                                 03/12/2011
8/30


 Filter performance


            Res: 640x480                   Standard          Simplified
              Sigma:5                       VBM3D             VBM3D

                        Noisy (dB)                     34.15
       Vassar         Denoised (dB)          40.81             38.61
                       Speed (fps)            1.12              7.62
                        Noisy (dB)                     34.15
     ballroom         Denoised (dB)          41.55             37.58
                       Speed (fps)            1.14              7.61




Video sequences vassar and ballroom are from http://www.merl.com/pub/avetro/mvc-testseq/orig-yuv/.




                                                                                                     03/12/2011
9/30
  Approach 2:
  Proposed real-time implementation

  Filter version     Proposed implementation   VBM3D standard version

    Data type                Integer                    float


     Memory            buffer only 4 frames      buffer whole video

 Block matching      Modified diamond search         Full search

Temporal searching
                                4                        5
     frames




                                                                        03/12/2011
10/30


 Practical Results


            Res: 640x480                   Standard         Simplified          Proposed
              Sigma: 5                      VBM3D            VBM3D           Implementation

                        Noisy (dB)                                34.15
       Vassar         Denoised (dB)          40.81             38.61                38.49
                       Speed (fps)           1.12               7.62                30.49
                        Noisy (dB)           34.15             34.15                34.15
     ballroom         Denoised (dB)          41.55             37.58                37.43
                       Speed (fps)           1.14               7.61                29.59




Video sequences vassar and ballroom are from http://www.merl.com/pub/avetro/mvc-testseq/orig-yuv/.




                                                                                                     03/12/2011
11/30


Part 1 summary: A real-time filter


1.   30 time faster than standard VBM3D
2.   Ready to be used for real-time application.
3.   Good video denoising performance in a low noise level.




                                                              03/12/2011
12/30




                 Part 2:

Joint video pre-filtering and compression




                                            03/12/2011
13/30
Typical scheme of Pre-filtering and
compression


                                                          R max
            Filter control               Rate control

              F*                                 Q*
  x(t)       Pre-filtering     y(t)     Video Encoder       z(t)


Usually we have two separate processes for pre-filtering and compression, and we
choose parameters separately.




                                                                              03/12/2011
14/30


Optimization Task

                                        R max
                         Joint Rate
                         control
          F*                                    Q*

  x(t)   Pre-filtering       y(t)     Video Encoder   z(t)




                                                        (1)




                                                              03/12/2011
15/30


Practical Results


Pre-filtering: VBM3D
Video Encoder: H.264/AVC (JM codec)

Experiments modes:
   • Constant quantization mode
   • Constant bitrates mode




                                      03/12/2011
16/30


Practical Results


Constant Quantization Mode
   In compression part, quantization parameters {Q} include:
        • QPI ={21,22,…45}, respectively QPP = QPI + 5
        • Other fixed codec setting
   In filtering part, filtering parameters {F} include:
        • Sigma ={0,0.5,1,…5},
        • Other fixed filter setting
   Full search:
        • For each Q, find the best output under rate-distortion
             framework.




                                                                   03/12/2011
17/30
Practical Results
                       hall.yuv, 25 points, QPI=21:1:45, QPP=26:1:50

             38



             36



             34
    Y-PSNR




             32



             30



             28



             26
                              H.264/AVC compression
                              joint VBM3D prefiltering and H.264/AVC compression
              0   50    100      150   200    250     300   350   400   450
                                        Bitrate (kbit/s)
    PSNR gains up to 0.5 dB.
    Bitrates savings up to 13.4%.
                                                                                   03/12/2011
18/30


Practical Results


Constant Bitrates Mode
   In compression part:
        • Enable the constant bitrates control.
   In filtering part, filtering parameters {F} include:
        • Sigma ={0,0.5,1,…5},
        • Other fixed filter setting
   Full search:
        • For each Q, find the best output under rate-distortion
             framework.




                                                                   03/12/2011
19/30
Practical Results
                               hall.yuv, Constant bitrate=215 kbit/s
          38.5
                           H.264/AVC compression
                           joint VBM3D prefiltering and H.264/AVC compression
           38



          37.5



           37
 Y-PSNR




          36.5



           36



          35.5



           35
             0        50         100            150           200      250   300
                                             frames
           PSNR gains up to 1.2 dB .
                                                                                   03/12/2011
20/30

H.264/AVC, frame 23   VBM3D+H.264/AVC, frame 23




                                                  03/12/2011
21/30

Part 2 summary:
Joint video pre-filtering and compression

 1. Output video frames have less compression artifacts.
 2. Output videos have consistent PSNR gains.
    •   can be up to 1.2 dB under constant bitrates mode.
    •   can be up to 0.5 dB under constant quantization mode.
 3. Can save the bitrates up to 13.4% in comparison with only
    compression.




                                                                03/12/2011
22/30




            Part 3:

Joint video in-loop filtering and
         compression




                                    03/12/2011
23/30
Typical scheme of in-loop filtering and
compression




                                          03/12/2011
24/30


Optimization Task




                    03/12/2011
25/30


Practical Results


Enhanced in-loop filtering: real-time filter from Part 1.
Video Encoder: H.264/AVC (JM codec)

Experiments modes:
   • Inter mode
       • QPI ∈ {21, 22 . . . 45} for I frame, and respective QPP =
         QPI + 5 for P frames.
   • Intra mode
       • QPI ∈ {21, 22 . . . 45} for I frame.




                                                                     03/12/2011
26/30
Practical Results
                         hall.yuv, 24 points, QPI=21:1:45, QPP=26:1:50
              42

              40

              38

              36
     Y-PSNR




              34

              32

              30

              28                              H.264/AVC(intra)
                                              H.264/AVC(intra) + Enhancing filter

                   500      1000      1500      2000       2500    3000      3500
                                        Bitrate (kbit/s)
    PSNR gains up to 0.87 dB.
    Bitrates savings up to 10.5%.
                                                                                    03/12/2011
27/30
Practical Results
                         hall.yuv, 24 points, QPI=21:1:45, QPP=26:1:50


              38


              36


              34
     Y-PSNR




              32


              30


              28


              26                                H.264/AVC(inter)
                                                H.264/AVC(inter) + Enhancing filter
                   200     250      300       350      400     450      500     550
                                          Bitrate (kbit/s)
    PSNR gains up to 0.35 dB.
    Bitrates savings up to 6.3%.
                                                                                      03/12/2011
28/30

Part 3 summary :
Joint video in-loop filtering and compression

  1. This joint approach consistently improves the compression
     performance of H.264/AVC under intra mode, but it gains little
     under inter mode.
  2. Under intra mode
     •   PSNR gains up to 0.87 dB.
     •   Bitrates saving up to 10.5% in comparison with only compression.
  3. Under inter mode
     •   PSNR gains up to 0.35 dB.
     •   Bitrates saving up to 6.3% in comparison with only compression.




                                                                            03/12/2011
29/30


Conclusion


1. We propose
   •   a real-time video filter which has good de-noising performance.
   •   a joint pre-filtering and compression approach
   •   a joint in-loop de-noising and compression approach .
2. Results show that these two joint approaches enhance the
   performance of the H.264/AVC standard:
   •   less compression artifact
   •   increased PSNR
   •   saved the bitrates.




                                                                         03/12/2011
30/30




Thank you!




             03/12/2011

More Related Content

What's hot

Master Thesis Presentation
Master Thesis PresentationMaster Thesis Presentation
Master Thesis Presentation
Wishofnight13
 
Master Thesis, Preliminary Defense
Master Thesis, Preliminary DefenseMaster Thesis, Preliminary Defense
Master Thesis, Preliminary Defense
Jenkins Macedo
 
Thesis Power Point Presentation
Thesis Power Point PresentationThesis Power Point Presentation
Thesis Power Point Presentation
riddhikapandya1985
 
Dissertation Defense Presentation
Dissertation Defense PresentationDissertation Defense Presentation
Dissertation Defense Presentation
Avril El-Amin
 
Thesis Powerpoint
Thesis PowerpointThesis Powerpoint
Thesis Powerpoint
neha47
 
Tactical Data Link (TDL) Training Crash Course
Tactical Data Link (TDL) Training Crash CourseTactical Data Link (TDL) Training Crash Course
Tactical Data Link (TDL) Training Crash Course
Tonex
 

What's hot (20)

Dissertation Proposal Defense
Dissertation Proposal DefenseDissertation Proposal Defense
Dissertation Proposal Defense
 
Dissertation proposal defense slideshow; phenomenology, qualitative
Dissertation proposal defense slideshow; phenomenology, qualitativeDissertation proposal defense slideshow; phenomenology, qualitative
Dissertation proposal defense slideshow; phenomenology, qualitative
 
Dissertation defense ppt
Dissertation defense ppt Dissertation defense ppt
Dissertation defense ppt
 
PhD. Thesis defence Slides
PhD. Thesis defence SlidesPhD. Thesis defence Slides
PhD. Thesis defence Slides
 
Master Thesis Presentation
Master Thesis PresentationMaster Thesis Presentation
Master Thesis Presentation
 
Master Thesis, Preliminary Defense
Master Thesis, Preliminary DefenseMaster Thesis, Preliminary Defense
Master Thesis, Preliminary Defense
 
M.S. Thesis Defense
M.S. Thesis DefenseM.S. Thesis Defense
M.S. Thesis Defense
 
Thesis Power Point Presentation
Thesis Power Point PresentationThesis Power Point Presentation
Thesis Power Point Presentation
 
Oral proposaldefense
Oral proposaldefenseOral proposaldefense
Oral proposaldefense
 
Gary Broils, D.B.A. - Dissertation Defense: Virtual Teaming and Collaboration...
Gary Broils, D.B.A. - Dissertation Defense: Virtual Teaming and Collaboration...Gary Broils, D.B.A. - Dissertation Defense: Virtual Teaming and Collaboration...
Gary Broils, D.B.A. - Dissertation Defense: Virtual Teaming and Collaboration...
 
My Dissertation Proposal Defense
My Dissertation Proposal DefenseMy Dissertation Proposal Defense
My Dissertation Proposal Defense
 
Dissertation Defense Presentation
Dissertation Defense PresentationDissertation Defense Presentation
Dissertation Defense Presentation
 
Ph.D. thesis defense
Ph.D. thesis defensePh.D. thesis defense
Ph.D. thesis defense
 
Thesis Powerpoint
Thesis PowerpointThesis Powerpoint
Thesis Powerpoint
 
Oral defense power point
Oral defense power pointOral defense power point
Oral defense power point
 
Tactical Data Link (TDL) Training Crash Course
Tactical Data Link (TDL) Training Crash CourseTactical Data Link (TDL) Training Crash Course
Tactical Data Link (TDL) Training Crash Course
 
Thesis Oral Defense
Thesis Oral DefenseThesis Oral Defense
Thesis Oral Defense
 
Msc Thesis - Presentation
Msc Thesis - PresentationMsc Thesis - Presentation
Msc Thesis - Presentation
 
PhD Viva presentation /PhD thesis defense/Environmental Science thesis presen...
PhD Viva presentation /PhD thesis defense/Environmental Science thesis presen...PhD Viva presentation /PhD thesis defense/Environmental Science thesis presen...
PhD Viva presentation /PhD thesis defense/Environmental Science thesis presen...
 
Thesis Defense Presentation
Thesis Defense PresentationThesis Defense Presentation
Thesis Defense Presentation
 

Viewers also liked

Ashbaugh dissertation defense presentation
Ashbaugh dissertation defense presentationAshbaugh dissertation defense presentation
Ashbaugh dissertation defense presentation
DRMLAID
 
Powerpoint presentation M.A. Thesis Defence
Powerpoint presentation M.A. Thesis DefencePowerpoint presentation M.A. Thesis Defence
Powerpoint presentation M.A. Thesis Defence
Catie Chase
 
Dissertation oral defense presentation
Dissertation   oral defense presentationDissertation   oral defense presentation
Dissertation oral defense presentation
Dr. Naomi Mangatu
 

Viewers also liked (8)

Thesis tips
Thesis tipsThesis tips
Thesis tips
 
Resource Allocation in Heterogeneous Networks
Resource Allocation in Heterogeneous NetworksResource Allocation in Heterogeneous Networks
Resource Allocation in Heterogeneous Networks
 
Ashbaugh dissertation defense presentation
Ashbaugh dissertation defense presentationAshbaugh dissertation defense presentation
Ashbaugh dissertation defense presentation
 
Finaldefense April2010v10 Web
Finaldefense April2010v10 WebFinaldefense April2010v10 Web
Finaldefense April2010v10 Web
 
Thesis Defense Presentation
Thesis Defense PresentationThesis Defense Presentation
Thesis Defense Presentation
 
My Thesis Defense Presentation
My Thesis Defense PresentationMy Thesis Defense Presentation
My Thesis Defense Presentation
 
Powerpoint presentation M.A. Thesis Defence
Powerpoint presentation M.A. Thesis DefencePowerpoint presentation M.A. Thesis Defence
Powerpoint presentation M.A. Thesis Defence
 
Dissertation oral defense presentation
Dissertation   oral defense presentationDissertation   oral defense presentation
Dissertation oral defense presentation
 

Similar to Master Thesis Presentation

Wavelet video processing tecnology
Wavelet video processing tecnologyWavelet video processing tecnology
Wavelet video processing tecnology
Prashant Madnavat
 
Battle of the Codecs
Battle of the CodecsBattle of the Codecs
Battle of the Codecs
James Uren
 
damaro.ppt
damaro.pptdamaro.ppt
damaro.ppt
Videoguy
 
Avlm 2009 Compression Erik Luyten
Avlm 2009  Compression   Erik LuytenAvlm 2009  Compression   Erik Luyten
Avlm 2009 Compression Erik Luyten
avlm2009avnet
 

Similar to Master Thesis Presentation (20)

Wavelet video processing tecnology
Wavelet video processing tecnologyWavelet video processing tecnology
Wavelet video processing tecnology
 
Optimisation and Compression Intro
Optimisation and Compression IntroOptimisation and Compression Intro
Optimisation and Compression Intro
 
Battle of the Codecs
Battle of the CodecsBattle of the Codecs
Battle of the Codecs
 
Design of Radio Frequency Integrated Circuits for UWB Communications
Design of Radio Frequency Integrated Circuits for UWB CommunicationsDesign of Radio Frequency Integrated Circuits for UWB Communications
Design of Radio Frequency Integrated Circuits for UWB Communications
 
Video Compression Basics - MPEG2
Video Compression Basics - MPEG2Video Compression Basics - MPEG2
Video Compression Basics - MPEG2
 
damaro.ppt
damaro.pptdamaro.ppt
damaro.ppt
 
JASLA_presentation.pdf
JASLA_presentation.pdfJASLA_presentation.pdf
JASLA_presentation.pdf
 
Barcelona keynote web
Barcelona keynote webBarcelona keynote web
Barcelona keynote web
 
Jpeg2000
Jpeg2000Jpeg2000
Jpeg2000
 
Drobo Benchmarks to accompany TMUP 159
Drobo Benchmarks to accompany TMUP 159Drobo Benchmarks to accompany TMUP 159
Drobo Benchmarks to accompany TMUP 159
 
Specsheet sncdh120t
Specsheet sncdh120tSpecsheet sncdh120t
Specsheet sncdh120t
 
Specsheet sncdh120t
Specsheet sncdh120tSpecsheet sncdh120t
Specsheet sncdh120t
 
Avlm 2009 Compression Erik Luyten
Avlm 2009  Compression   Erik LuytenAvlm 2009  Compression   Erik Luyten
Avlm 2009 Compression Erik Luyten
 
Human Behavior Understanding: From Human-Oriented Analysis to Action Recognit...
Human Behavior Understanding: From Human-Oriented Analysis to Action Recognit...Human Behavior Understanding: From Human-Oriented Analysis to Action Recognit...
Human Behavior Understanding: From Human-Oriented Analysis to Action Recognit...
 
Specsheet sncdh120
Specsheet sncdh120Specsheet sncdh120
Specsheet sncdh120
 
Specsheet sncdh120
Specsheet sncdh120Specsheet sncdh120
Specsheet sncdh120
 
Okotech Catwww3
Okotech Catwww3Okotech Catwww3
Okotech Catwww3
 
MPEG video compression standard
MPEG video compression standardMPEG video compression standard
MPEG video compression standard
 
Learning spatiotemporal features with 3 d convolutional networks
Learning spatiotemporal features with 3 d convolutional networksLearning spatiotemporal features with 3 d convolutional networks
Learning spatiotemporal features with 3 d convolutional networks
 
Choosing the Right IP Cores for Low-Latency Video Streaming
Choosing the Right IP Cores for Low-Latency Video StreamingChoosing the Right IP Cores for Low-Latency Video Streaming
Choosing the Right IP Cores for Low-Latency Video Streaming
 

Master Thesis Presentation

  • 1. 1/30 Master Thesis Presentation: Real-time Rate-distortion Oriented Joint Video Denoising and Compression Junsheng Fu Signal Processing Department Tampere University of Technology 03/12/2011
  • 2. 2/30 Outline 1. A real-time filter 2. Joint video pre-filtering and compression 3. Joint video in-loop filtering and compression 03/12/2011
  • 3. 3/30 Part 1: A real-time filter 03/12/2011
  • 4. 4/30 VBM3D (Video Block-Matching 3D Filtering) K. Dabov, A. Foi, and K. Egiazarian, “Video denoising by sparse 3D transform-domain collaborative filtering,” in Proc. 15th European Signal Processing Conference, EUSIPCO, Poznan, Poland, September 2007 03/12/2011
  • 5. 5/30 Filter performance Res: 640x480 Standard Sigma:5 VBM3D Noisy (dB) 34.15 Vassar Denoised (dB) 40.81 Speed (fps) 1.12 Noisy (dB) 34.15 Computer Platform ballroom Denoised (dB) 41.55 Processor: Intel Core 2 Duo 3GHz Speed (fps) 1.14 Memory: 3.2Gb of RAM Video sequences vassar and ballroom are from http://www.merl.com/pub/avetro/mvc-testseq/orig-yuv/. 03/12/2011
  • 6. 6/30 Optimization Task 1. Accelerate filter speed: above 25 fps for video (640 x 480) In a computer • Processor: Intel Core 2 Duo 3GHz • Memory: 3.2Gb of RAM Approaches 1. Simplify VBM3D filter by only using most influential parts for noise attenuation. 2. Propose a real-time integer implementation of simplified VBM3D. 03/12/2011
  • 7. 7/30 Approach 1: What to remove to Simplify VBM3D? Thresholding filtering Wiener filtering Temporal Max Filters searching grouping Spatial Spatial Temporal Thres- Temporal Spatial Spatial Temporal Wiener Temporal frames blocks search transform search holding transform search transform search filtering transform Standard VBM3D + + + + + + + + + + 9 8 *Simplified VBM3D - - + + + - - - - - 5 4 * Simplified VBM3D means turn off some features of VBM3D filter. 03/12/2011
  • 8. 8/30 Filter performance Res: 640x480 Standard Simplified Sigma:5 VBM3D VBM3D Noisy (dB) 34.15 Vassar Denoised (dB) 40.81 38.61 Speed (fps) 1.12 7.62 Noisy (dB) 34.15 ballroom Denoised (dB) 41.55 37.58 Speed (fps) 1.14 7.61 Video sequences vassar and ballroom are from http://www.merl.com/pub/avetro/mvc-testseq/orig-yuv/. 03/12/2011
  • 9. 9/30 Approach 2: Proposed real-time implementation Filter version Proposed implementation VBM3D standard version Data type Integer float Memory buffer only 4 frames buffer whole video Block matching Modified diamond search Full search Temporal searching 4 5 frames 03/12/2011
  • 10. 10/30 Practical Results Res: 640x480 Standard Simplified Proposed Sigma: 5 VBM3D VBM3D Implementation Noisy (dB) 34.15 Vassar Denoised (dB) 40.81 38.61 38.49 Speed (fps) 1.12 7.62 30.49 Noisy (dB) 34.15 34.15 34.15 ballroom Denoised (dB) 41.55 37.58 37.43 Speed (fps) 1.14 7.61 29.59 Video sequences vassar and ballroom are from http://www.merl.com/pub/avetro/mvc-testseq/orig-yuv/. 03/12/2011
  • 11. 11/30 Part 1 summary: A real-time filter 1. 30 time faster than standard VBM3D 2. Ready to be used for real-time application. 3. Good video denoising performance in a low noise level. 03/12/2011
  • 12. 12/30 Part 2: Joint video pre-filtering and compression 03/12/2011
  • 13. 13/30 Typical scheme of Pre-filtering and compression R max Filter control Rate control F* Q* x(t) Pre-filtering y(t) Video Encoder z(t) Usually we have two separate processes for pre-filtering and compression, and we choose parameters separately. 03/12/2011
  • 14. 14/30 Optimization Task R max Joint Rate control F* Q* x(t) Pre-filtering y(t) Video Encoder z(t) (1) 03/12/2011
  • 15. 15/30 Practical Results Pre-filtering: VBM3D Video Encoder: H.264/AVC (JM codec) Experiments modes: • Constant quantization mode • Constant bitrates mode 03/12/2011
  • 16. 16/30 Practical Results Constant Quantization Mode In compression part, quantization parameters {Q} include: • QPI ={21,22,…45}, respectively QPP = QPI + 5 • Other fixed codec setting In filtering part, filtering parameters {F} include: • Sigma ={0,0.5,1,…5}, • Other fixed filter setting Full search: • For each Q, find the best output under rate-distortion framework. 03/12/2011
  • 17. 17/30 Practical Results hall.yuv, 25 points, QPI=21:1:45, QPP=26:1:50 38 36 34 Y-PSNR 32 30 28 26 H.264/AVC compression joint VBM3D prefiltering and H.264/AVC compression 0 50 100 150 200 250 300 350 400 450 Bitrate (kbit/s) PSNR gains up to 0.5 dB. Bitrates savings up to 13.4%. 03/12/2011
  • 18. 18/30 Practical Results Constant Bitrates Mode In compression part: • Enable the constant bitrates control. In filtering part, filtering parameters {F} include: • Sigma ={0,0.5,1,…5}, • Other fixed filter setting Full search: • For each Q, find the best output under rate-distortion framework. 03/12/2011
  • 19. 19/30 Practical Results hall.yuv, Constant bitrate=215 kbit/s 38.5 H.264/AVC compression joint VBM3D prefiltering and H.264/AVC compression 38 37.5 37 Y-PSNR 36.5 36 35.5 35 0 50 100 150 200 250 300 frames PSNR gains up to 1.2 dB . 03/12/2011
  • 20. 20/30 H.264/AVC, frame 23 VBM3D+H.264/AVC, frame 23 03/12/2011
  • 21. 21/30 Part 2 summary: Joint video pre-filtering and compression 1. Output video frames have less compression artifacts. 2. Output videos have consistent PSNR gains. • can be up to 1.2 dB under constant bitrates mode. • can be up to 0.5 dB under constant quantization mode. 3. Can save the bitrates up to 13.4% in comparison with only compression. 03/12/2011
  • 22. 22/30 Part 3: Joint video in-loop filtering and compression 03/12/2011
  • 23. 23/30 Typical scheme of in-loop filtering and compression 03/12/2011
  • 25. 25/30 Practical Results Enhanced in-loop filtering: real-time filter from Part 1. Video Encoder: H.264/AVC (JM codec) Experiments modes: • Inter mode • QPI ∈ {21, 22 . . . 45} for I frame, and respective QPP = QPI + 5 for P frames. • Intra mode • QPI ∈ {21, 22 . . . 45} for I frame. 03/12/2011
  • 26. 26/30 Practical Results hall.yuv, 24 points, QPI=21:1:45, QPP=26:1:50 42 40 38 36 Y-PSNR 34 32 30 28 H.264/AVC(intra) H.264/AVC(intra) + Enhancing filter 500 1000 1500 2000 2500 3000 3500 Bitrate (kbit/s) PSNR gains up to 0.87 dB. Bitrates savings up to 10.5%. 03/12/2011
  • 27. 27/30 Practical Results hall.yuv, 24 points, QPI=21:1:45, QPP=26:1:50 38 36 34 Y-PSNR 32 30 28 26 H.264/AVC(inter) H.264/AVC(inter) + Enhancing filter 200 250 300 350 400 450 500 550 Bitrate (kbit/s) PSNR gains up to 0.35 dB. Bitrates savings up to 6.3%. 03/12/2011
  • 28. 28/30 Part 3 summary : Joint video in-loop filtering and compression 1. This joint approach consistently improves the compression performance of H.264/AVC under intra mode, but it gains little under inter mode. 2. Under intra mode • PSNR gains up to 0.87 dB. • Bitrates saving up to 10.5% in comparison with only compression. 3. Under inter mode • PSNR gains up to 0.35 dB. • Bitrates saving up to 6.3% in comparison with only compression. 03/12/2011
  • 29. 29/30 Conclusion 1. We propose • a real-time video filter which has good de-noising performance. • a joint pre-filtering and compression approach • a joint in-loop de-noising and compression approach . 2. Results show that these two joint approaches enhance the performance of the H.264/AVC standard: • less compression artifact • increased PSNR • saved the bitrates. 03/12/2011
  • 30. 30/30 Thank you! 03/12/2011