Investigations of Machine Learning Algorithms for High Efficiency Video Coding (HEVC)

Abstract

The growing demand of high-resolution video on portable devices, the applications require higher coding efficiency, high throughput and low power for handling heterogenous types of video signals. This paper presents a survey on possibility of applying Machine Learning (ML) models in H.265/ HEVC video encoder unit. Higher computational complexity with respect to motion estimation, coding, and parallel processing architectures are required for HEVC. The existing HEVC algorithms are based on spatial temporal relationship which requires dynamic video sequences handling for fast changes in scenes. This paper focuses on the possible realization of machine learning algorithms for Rate Control (RC) in video sequences, Coding Unit (CU) depth decision, Neural network-based Motion Estimation and Compensation, adaptive de-blocking filter for reducing blocking artifacts and task driven semantic coding for real time video applications. The algorithms are surveyed with respect to the learning process used in various units of HEVC encoders and summarized in terms of parameters achieved and datasets used in the existing literature.