SVM based approach for complexity control of HEVC intra coding

• Published in: Signal processing. Image communication 2021-04, Vol.93, p.116177, Article 116177
• Main Authors: Pakdaman, Farhad, Yu, Li, Hashemi, Mahmoud Reza, Ghanbari, Mohammad, Gabbouj, Moncef
• Format: Article
• Language: English
• Subjects: Adaptive algorithms; Coding; Complexity; Complexity control; Control methods; Decision making; Efficiency; HEVC; Image classification; Image processing; Intra coding; Machine learning; Partitioning; Support vector machines; SVM; Video compression; Video transmission
• ISSN: 0923-5965; 1879-2677
• DOI: 10.1016/j.image.2021.116177

The High Efficiency Video Coding (HEVC) is adopted by various video applications in recent years. Because of its high computational demand, controlling the complexity of HEVC is of paramount importance to appeal to the varying requirements in many applications, including power-constrained video coding, video streaming, and cloud gaming. Most of the existing complexity control methods are only capable of considering a subset of the decision space, which leads to low coding efficiency. While the efficiency of machine learning methods such as Support Vector Machines (SVM) can be employed for higher precision decision making, the current SVM-based techniques for HEVC provide a fixed decision boundary which results in different coding complexities for different video content. Although this might be suitable for complexity reduction, it is not acceptable for complexity control. This paper proposes an adjustable classification approach for Coding Unit (CU) partitioning, which addresses the mentioned problems of complexity control. Firstly, a novel set of features for fast CU partitioning is designed using image processing techniques. Then, a flexible classification method based on SVM is proposed to model the CU partitioning problem. This approach allows adjusting the performance-complexity trade-off, even after the training phase. Using this model, and a novel adaptive thresholding technique, an algorithm is presented to deliver video encoding within the target coding complexity, while maximizing the coding efficiency. Experimental results justify the superiority of this method over the state-of-the-art methods, with target complexities ranging from 20% to 100%. •Accurately controlling complexity of video coding using machine learning.•Flexible classification allows adjusting the performance after training.•Hand-crafted features better represent block partitioning decision.•Offline learning with online probability modeling improves compression performance.