Reinforcement learning for video encoder control in HEVC

Philipp Helle, Heiko Schwarz, Thomas Wiegand, Klaus Muller

Research output: Chapter in Book/Report/Conference proceedingConference contribution

5 Citations (Scopus)

Abstract

In todays video compression systems, the encoder typically follows an optimization procedure to find a compressed representation of the video signal. While primary optimization criteria are bit rate and image distortion, low complexity of this procedure may also be of importance in some applications, making complexity a third objective. We approach this problem by treating the encoding procedure as a decision process in time and make it amenable to reinforcement learning. Our learning algorithm computes a strategy in a compact functional representation, which is then employed in the video encoder to control its search. By including measured execution time into the reinforcement signal with a lagrangian weight, we realize a trade-off between RD-performance and computational complexity controlled by a single parameter. Using the reference software test model (HM) of the HEVC video coding standard, we show that over half the encoding time can be saved at the same RD-performance.

Original languageEnglish
Title of host publicationProceedings - 2017 International Conference on Systems, Signals and Image Processing, IWSSIP 2017
PublisherIEEE Computer Society
ISBN (Electronic)9781509063444
DOIs
Publication statusPublished - 2017 Jun 30
Externally publishedYes
Event24th International Conference on Systems, Signals and Image Processing, IWSSIP 2017 - Poznan, Poland
Duration: 2017 May 222017 May 24

Other

Other24th International Conference on Systems, Signals and Image Processing, IWSSIP 2017
CountryPoland
CityPoznan
Period17/5/2217/5/24

Fingerprint

Reinforcement learning
Image compression
Image coding
Learning algorithms
Computational complexity
Reinforcement

Keywords

  • Computational Complexity
  • Cost Sensitive Classification
  • HEVC
  • Machine Learning
  • Optimization
  • Rate-Distortion Optimization
  • Reinforcement Learning
  • Video Coding

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Networks and Communications
  • Signal Processing
  • Software

Cite this

Helle, P., Schwarz, H., Wiegand, T., & Muller, K. (2017). Reinforcement learning for video encoder control in HEVC. In Proceedings - 2017 International Conference on Systems, Signals and Image Processing, IWSSIP 2017 [7965586] IEEE Computer Society. https://doi.org/10.1109/IWSSIP.2017.7965586

Reinforcement learning for video encoder control in HEVC. / Helle, Philipp; Schwarz, Heiko; Wiegand, Thomas; Muller, Klaus.

Proceedings - 2017 International Conference on Systems, Signals and Image Processing, IWSSIP 2017. IEEE Computer Society, 2017. 7965586.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Helle, P, Schwarz, H, Wiegand, T & Muller, K 2017, Reinforcement learning for video encoder control in HEVC. in Proceedings - 2017 International Conference on Systems, Signals and Image Processing, IWSSIP 2017., 7965586, IEEE Computer Society, 24th International Conference on Systems, Signals and Image Processing, IWSSIP 2017, Poznan, Poland, 17/5/22. https://doi.org/10.1109/IWSSIP.2017.7965586
Helle P, Schwarz H, Wiegand T, Muller K. Reinforcement learning for video encoder control in HEVC. In Proceedings - 2017 International Conference on Systems, Signals and Image Processing, IWSSIP 2017. IEEE Computer Society. 2017. 7965586 https://doi.org/10.1109/IWSSIP.2017.7965586
Helle, Philipp ; Schwarz, Heiko ; Wiegand, Thomas ; Muller, Klaus. / Reinforcement learning for video encoder control in HEVC. Proceedings - 2017 International Conference on Systems, Signals and Image Processing, IWSSIP 2017. IEEE Computer Society, 2017.
@inproceedings{624900375e8640d2817789c3ea3b37eb,
title = "Reinforcement learning for video encoder control in HEVC",
abstract = "In todays video compression systems, the encoder typically follows an optimization procedure to find a compressed representation of the video signal. While primary optimization criteria are bit rate and image distortion, low complexity of this procedure may also be of importance in some applications, making complexity a third objective. We approach this problem by treating the encoding procedure as a decision process in time and make it amenable to reinforcement learning. Our learning algorithm computes a strategy in a compact functional representation, which is then employed in the video encoder to control its search. By including measured execution time into the reinforcement signal with a lagrangian weight, we realize a trade-off between RD-performance and computational complexity controlled by a single parameter. Using the reference software test model (HM) of the HEVC video coding standard, we show that over half the encoding time can be saved at the same RD-performance.",
keywords = "Computational Complexity, Cost Sensitive Classification, HEVC, Machine Learning, Optimization, Rate-Distortion Optimization, Reinforcement Learning, Video Coding",
author = "Philipp Helle and Heiko Schwarz and Thomas Wiegand and Klaus Muller",
year = "2017",
month = "6",
day = "30",
doi = "10.1109/IWSSIP.2017.7965586",
language = "English",
booktitle = "Proceedings - 2017 International Conference on Systems, Signals and Image Processing, IWSSIP 2017",
publisher = "IEEE Computer Society",

}

TY - GEN

T1 - Reinforcement learning for video encoder control in HEVC

AU - Helle, Philipp

AU - Schwarz, Heiko

AU - Wiegand, Thomas

AU - Muller, Klaus

PY - 2017/6/30

Y1 - 2017/6/30

N2 - In todays video compression systems, the encoder typically follows an optimization procedure to find a compressed representation of the video signal. While primary optimization criteria are bit rate and image distortion, low complexity of this procedure may also be of importance in some applications, making complexity a third objective. We approach this problem by treating the encoding procedure as a decision process in time and make it amenable to reinforcement learning. Our learning algorithm computes a strategy in a compact functional representation, which is then employed in the video encoder to control its search. By including measured execution time into the reinforcement signal with a lagrangian weight, we realize a trade-off between RD-performance and computational complexity controlled by a single parameter. Using the reference software test model (HM) of the HEVC video coding standard, we show that over half the encoding time can be saved at the same RD-performance.

AB - In todays video compression systems, the encoder typically follows an optimization procedure to find a compressed representation of the video signal. While primary optimization criteria are bit rate and image distortion, low complexity of this procedure may also be of importance in some applications, making complexity a third objective. We approach this problem by treating the encoding procedure as a decision process in time and make it amenable to reinforcement learning. Our learning algorithm computes a strategy in a compact functional representation, which is then employed in the video encoder to control its search. By including measured execution time into the reinforcement signal with a lagrangian weight, we realize a trade-off between RD-performance and computational complexity controlled by a single parameter. Using the reference software test model (HM) of the HEVC video coding standard, we show that over half the encoding time can be saved at the same RD-performance.

KW - Computational Complexity

KW - Cost Sensitive Classification

KW - HEVC

KW - Machine Learning

KW - Optimization

KW - Rate-Distortion Optimization

KW - Reinforcement Learning

KW - Video Coding

UR - http://www.scopus.com/inward/record.url?scp=85026663364&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85026663364&partnerID=8YFLogxK

U2 - 10.1109/IWSSIP.2017.7965586

DO - 10.1109/IWSSIP.2017.7965586

M3 - Conference contribution

AN - SCOPUS:85026663364

BT - Proceedings - 2017 International Conference on Systems, Signals and Image Processing, IWSSIP 2017

PB - IEEE Computer Society

ER -