IJMLC 2018 Vol.8(5): 442-446 ISSN: 2010-3700
DOI: 10.18178/ijmlc.2018.8.5.726

Fast CU Determination Algorithm Based on Convolutional Neural Network for HEVC

Takafumi Katayama, Tian song, Wen Shi, Xiantao Jiang, and Takashi Shimamoto

Abstract—High efficiency video coding (HEVC) is the current video coding standard. HEVC achieved very high coding efficiency compared with previous video coding standards. However, the increasing of the computational complexity and the hardware implementation difficulty are the critical problems for HEVC. In this paper, we propose a fast coding unit (CU) size decision algorithm for HEVC based on convolutional neural network. The proposed fast algorithm contribute to decrease no less than two CU partition modes in each coding tree unit for full rate-distortion optimization processing, thereby reducing the encoder hardware complexity. Moreover, our algorithm only use the texture information and it does not depend on the correlations among CU depths or spatially nearby CUs. It is friendly to the parallel processing of RDO. The proposed algorithm is evaluated by the reference software of HEVC (HM16.7). The simulation results show that the proposed algorithm can achieve over 66.7% computation complexity reduction comparing to the original HEVC algorithm.

Index Terms—High efficiency video coding (HEVC), intra coding, convolutional neural network (CNN).

Takafumi Katayama, Wen Shi, Tian Song, and Takashi Shimamoto are with the Department of Electrical and Electronics Engineering of Tokushima University, Tokushima city, Tokushima, 770-8506 Japan (e-mail: katayama@ee.tokushima-u.ac.jp, tiansong@ee.tokushima-u.ac.jp).
Xiantao Jiang is with the Department of Information Engineering Shanghai Maritime University, Pudong New Dist., Shanghai 201306, P. R. China (e-mail: xtjiang@shmtu.edu.cn).

[PDF]

Cite: Takafumi Katayama, Tian song, Wen Shi, Xiantao Jiang, and Takashi Shimamoto, "Fast CU Determination Algorithm Based on Convolutional Neural Network for HEVC," International Journal of Machine Learning and Computing vol. 8, no. 5, pp. 442-446, 2018.

General Information

  • ISSN: 2010-3700 (Online)
  • Abbreviated Title: Int. J. Mach. Learn. Comput.
  • Frequency: Bimonthly
  • DOI: 10.18178/IJMLC
  • Editor-in-Chief: Dr. Lin Huang
  • Executive Editor:  Ms. Cherry L. Chen
  • Abstracing/Indexing: Scopus (since 2017), EI (INSPEC, IET), Google Scholar, Crossref, ProQuest, Electronic Journals Library.
  • E-mail: ijmlc@ejournal.net