1 / 21

Improved input-state linearization in video bitrate controllers

Improved input-state linearization in video bitrate controllers. Noam Korem. Presentation outline. Video encoding and rate control Classic models Suggested improvement Simulation results Summary and conclusions. Video encoding and rate control Classic R-Q models

major
Download Presentation

Improved input-state linearization in video bitrate controllers

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Improved input-state linearization in video bitrate controllers Noam Korem

  2. Presentation outline • Video encoding and rate control • Classic models • Suggested improvement • Simulation results • Summary and conclusions • Video encoding and • rate control • Classic R-Q models • Suggested improvement • Simulation results • Summary / Conclusions

  3. Intra Predicted + + + - - - + + + 0 Discrete Cosine Transform + + + IDCT IDCT IDCT qi+1 qi-2 qi qi-1 Iqi-2 Iqi-1 Iqi DCT DCT DCT DCT VLC ..01001.. VLC ..01001.. VLC ..01001.. VLC ..01001.. Generic video encoder* i-2 i-1 i i+1 I P P P • Video encoding and • rate control • Classic R-Q models • Suggested improvement • Simulation results • Summary / Conclusions

  4. Discrete Cosine Transform (DCT) • Resembles Discrete Fourier Transform • Purely real (real transform to real) • Allows representation in the frequency domain, usually more compact Spatial domain Frequency domain

  5. Quantization Quantization scale = 1 Quantization scale = 31

  6. Non linear bitrate controller (Input-state linearization) Non-linear Bit rate controller Ti Linear controller qi=Q(Ti,..) Generic video encoder with rate control mechanism qi Frame encoder bits target bit rate Si

  7. Energy of difference frame i is dependent on the q(i-1) + + - - + + Coded information of frame i is dependent on q(i), q(i-1) IDCT IDCT qi qi+1 Iqi-1 Iqi DCT DCT VLC ..01001.. VLC ..01001.. The problem i-1 i P P + + • Video encoding and • rate control √ • Classic R-Q models • Suggested improvement • Simulation results • Summary / Conclusions

  8. The problem, cont. Classic Rate-Quantization models do not depend on reference frame quantization scale

  9. Ri Δ i-1 i Qi-1 ΔQi-1 i-1(q) Qi Qi-1 Improved R-Q model • Video encoding and • rate control √ • Classic R-Q models √ • Suggested improvement • Simulation results • Summary / Conclusions

  10. Previous frame quantizerdependency Ri(Qi) Ri(Qi-1), Qi=8

  11. Improved R-Q model, cont.

  12. TM5 model: Simulations • Video encoding and • rate control √ • Classic R-Q models √ • Suggested improvement √ • Simulation results • Summary / Conclusions • Encode real videos, compare accuracy of traditional model (TM5) against improved model • ffmpeg open-source video encoder with enhanced quantization and bit rate control used as encoding model (MPEG4). • All test videos are YUV 4:2:0, QVGA (320x240), encoded at 370kbps (74CR)

  13. Simulations • Absolute prediction error is calculated for each frame, for both models • Delta of absolute prediction errors is used as performance measurement

  14. Simulation results Mean Abs Err* TM5: 17.07% TM5i: 15.03% (12% improve) Mean Abs Err* TM5: 7.18% TM5i: 4.8% (33% improve)

  15. Simulation results, cont. Mean Abs Err* TM5: 22.3% TM5i: 19% (14.7% improve) Mean Abs Err* TM5: 16.7% TM5i: 14.1% (15.6% improve)

  16. Simulation results, cont. Mean Abs Err* TM5: 34.2% TM5i: 32.2% (6% improve)

  17. Simulation results, degradation Mean Abs Err* TM5: 13.1% TM5i: 13.2% (1% degrade)

  18. Summary & Conclusions • Video encoding with rate controller scheme presented (camcorders, live streaming) • Video rate and distortion depend on quantization of reference frame, classic models ignore. • Improved R-Q model allows more accurate input-state linearization, demands more calculations • Improved R-Q model does not deliver when scene complexity changes rapidly • Video encoding and • rate control √ • Classic R-Q models √ • Suggested improvement √ • Simulation results √ • Summary / Conclusions

  19. Future work • Prediction quality degrades due to input noise – changes in scene complexity • More accurate complexity estimation (more accurate prediction for Xi) based on: • frame content • motion estimation would improve R-Q model accuracy • Try the model improvement on other R-Q models • A complete model will include past quantization parameters Ri=Ri(Qi,Qi-1,Qi-2,…)

  20. Related work • Video Group, "Test Model 5" JTC1/SC29/WG11 Coding of Moving Pictures and associated Audio MPEG 1994, section 10 • Tihao Chiang and Ya-Qin Zhang "A New Rate Control Scheme Using Quadratic Rate Distortion Model", IEEE International Conference on Image Processing, 1996.   10.1109/ICIP.1996.560604 • Liang-Jin Lin; Ortega, A. “Bit-rate control using piecewise approximated rate-distortion characteristics” Circuits and Systems for Video Technology, IEEE Transactions on, Volume 8, Issue 4, Aug 1998 Page(s):446 - 459 10.1109/76.709411 • Saw, Y.-S.; Grant, P.M.; Hannah, J.M., "Rate-distortion analysis of nonlinear quantisers for MPEG videocoders: sigmoidal and unimodal quantiser control functions" Vision, Image and Signal Processing, IEE Proceedings- Volume 145, Issue 4, Aug 1998 Page(s):249 – 256 • Ma, S.; Wen Gao; Yan Lu; "Rate-distortion analysis for H.264/AVC video coding and its application to rate control“ Circuits and Systems for Video Technology, IEEE Transactions on Volume 15,  Issue 12,  Dec. 2005 Page(s):1533 - 1544 10.1109/TCSVT.2005.857300

  21. Many thanks: Shai Mazor Orly Wigderson Kobi Kohai For the guidance, patience and assistance on all aspects of the project

More Related