Fast H.264 Encoding Based on Statistical Learning

Abstract

In this paper, we propose an efficient video coding system that applies statistical learning methods to reduce the computational cost in H.264 encoder. The proposed method can be applied to many coding components in H.264, like intermode decision, multi-reference motion estimation, intra-mode prediction. First, representative features are extracted from video to build the learning models. Then, an off-line pre-classification approach is used to determine the best results from the extracted features, thus a significant amount of computation is reduced based on the classification strategy. The proposed statistical learning based approach is applied to the aforementioned three main components and a novel framework of learning based H.264 encoder is proposed to speed up the computation. Experimental results show that the motion estimation (ME) time of the proposed system is significantly speed up with twelve times faster than the H.264 encoder with a conventional fast ME algorithm, and the total encoding time of the proposed encoder is greatly reduced with about four times faster than the fast encoder EPZS in the H.264 reference code with negligible video quality degradation.