A study on fast intermode decision using statistical hypothesis testing for scalable video coding

Abstract

Scalable Video Coding, SVC is a new video compression technology, called H.264 Scalable Extension which is being standardized by Joint Video Team (JVT), a joint group of MPEG (Moving Picture Experts Group) and VCEG (Video Coding Expert Group). SVC has layered structures to support temporal, spatial and quality scalabilities. It adopts the hierarchical B-picture structure to support temporal scalability. Since the base layer of SVC is H.264|AVC compatible, it uses 7 different block sizes of 16$\times$16, 16$\times$8, 8$\times$16, 8$\times$8, 8$\times$4, 4$\times$8 and 4$\times$4 for motion estimation and mode decision. In SVC, the hierarchical B-picture utilizes only B-picture except key pictures such as I and P pictures within a group of picture. So, computational complexity is higher compared with H.264|AVC because B-picture encoding has to consider two directional motion vectors in LIST0 and LIST1 and multiple reference frames in both directions. In this thesis, a fast inter mode decision scheme for scalable video coding which is based on statistical hypothesis testing is introduced. The hypothesis testing is performed on 16x16 and 8x8 blocks to decide early termination from the further processing for RD-optimized mode decision by comparing the pixel values of the current blocks and those of motion-compensated blocks. The proposed scheme exhibits an effective early termination behavior in inter mode decision, thus leading to significant reduction in computational complexity with small increment in bit amounts an content-independent decision based on a statistical framework. It also allows for with small amount of reduction in visual quality