We propose a machine learning based fast Coding Unit (CU) depth decision method for High Efficiency Video Coding (HEVC), which optimizes the complexity allocation at CU level with given Rate-Distortion (RD) cost constraints. Firstly, a flexible CU depth decision structure is presented. Then, a three-output joint classifier consists of multiple binary classifiers with different parameters is designed to control the risk of false prediction. Finally, a sophisticated RD-complexity model is derived to determine the optimal parameters for the joint classifier, which is capable of minimizing the complexity in each CU depth at given RD degradation constraints. Comparative experiments over various sequences show that the proposed algorithm can reduce the computational complexity from 28.82% to 70.93%, and 51.45% on average when compared with the original HM model. The Bjøntegaard delta peak signal-to-noise ratio and Bjøntegaard delta bit rate are -0.061 dB and 1.98% on average, which is negligible. The overall performance of the proposed algorithm outperforms those of state-of-the-art schemes.

Large volumes of video data requires highly effective video coding;

High complexity of video encoder requires more computing resources, power consumption, and high production cost.

Compression efficiency is improved at the cost of intensive computational complexity (dozens times)

To minimize the prediction error, the optimal mode shall be chosen among multiple candidates;

Try all mode candidates and select the best one via cost comparison, brutal-forces

Proposed a flexible CU depth decision structure, three-output joint classifier and sophisticated RD-complexity model to minimize the complexity at given RD degradation constraints

Decision problem Modelling

Basic idea: model the mode decision problem in HEVC as classification problems and then solved by learning algorithms

Optimal Parameter Determination

Basic idea: minimize the coding complexity subject to negligible RD degradation while determining the optimal hyperplanes.

Verified on 21 Video Sequences, the proposed CU decision algorithm achieves 51.94% complexity reduction on average, the BDBR increase is 1.77% and the BDPSNR is -0.059dB,which is negligible. It outperforms the state-of-the-art schemes.

Y. Zhang, S. Kwong, X. Wang, Z. Pan, H. Yuan, and L. Xu, Machine Learning Based Coding Unit Depth Decisions for Flexible Complexity Allocation in High Efficiency Video Coding, IEEE Transactions on Image Processing, vol.24, no.7,pp.2225-2238, July. 2015. (SCI IF 3.625)

Y. Zhang, Z. Pan, N. Li, X. Wang, G. Jiang, and S. Kwong, Effective Data Driven Coding Unit Size Decision Approaches for HEVC INTRA Coding, IEEE Transactions on Circuits and Systems for Video Technology, Nov., 2016, (submitted.)