An intra prediction skip scheme based On SATD And motion vector magnitude features in H.264|MPEG-4 Part 10 AVC

The computational complexity of AVC encoders has been dramatically increased due to the adoption of various coding tools. Among them, intra prediction is one of the computationally expensive tools for enhancing coding efficiency. The intra 4 x 4 prediction occupies large portions of the total encoding time despite of its small portion of being selected as the final block mode in inter-slice coding. This paper presents a fast intra prediction skip method that decides whether or not to perform the intra 4 x 4 prediction is performed beforehand. Intra prediction is avoided in P-pictures by inspecting both the motion vector magnitudes and the SATD values. Our proposed fast intra prediction skip method effectively checks the two features for the possibility of intra prediction as the final decision block mode. The proposed method significantly enhances the encoding time of AVC by effectively performing a two-class classification with negligible amounts of PSNR drops and bitrate increases. Experimental results are shown for the performance of the proposed method with various video sequences with different motion and texture characteristics.

1. Introduction

Video coding is an essential tool for storage and transmission of video contents which sizes are usually too large. For this reason, various video coding methods have been developed and many video coding standards have been specified such as MPEG 1, 2, 4, H.26x series for last twenty years. Among these standards, H.264│MPEG-4 Part 10 Advanced Video Coding (AVC) [1] [3] shows a good coding efficiency and is being widely adopted for many application areas.