Chinese Journal of Computers   Full Text
Title	Improved Inter Prediction Based on Structural Similarity for H.264
Authors	YANG Chun-Ling1) WANG Hua-Xing2) LIANG Rong-Kun1)
Address	1)(School of Electronic and Information Engineering, South China University of Technology, Guangzhou 510640) 2)(Ericsson(China) Communication Company Ltd., Beijing 100016)
Year	2009
Issue	No.8(1603—1610)
Abstract &   Background	Abstract H.264/AVC achieves higher compression efficiency by employing multiple modes inter prediction, rate-distortion(RD)optimal mechanism and other new techniques. Distortion metric plays an important role in video compression performance. Structural similarity(SSIM) is a new image quality assessment method, which is more consistent with Human Vision Systems(HVS). This paper proposes to adopt SSIM as the distortion metric in the inter prediction cost functions, named “improved inter prediction method based on SSIM”(IPBSS). It is an improved method of the authors’ previous work on MEBSS(Motion Estimation Method Based on Structural Similarity). Simulation results show that the proposed IPBSS can averagely save bit rate more than 13% while maintaining almost the same video quality with QP=10, 20 and 30. That is a better result than the authors’ previous work MEBSS. Keywords H.264； inter prediction； distortion metric； structural similarity(SSIM)； rate-distortion optimization(RDO) Background Rate-distortion optimization(RDO) plays a vital role in video compression applications, and it is widely researched in resent years. In literature, RDO for video compression can be classified into two categories. The first category computes the theoretical RD function based on a given statistic model for video data. And the second category uses an operational RD function, which is computed based on the data to be compressed. Such as ρ domain based RDO (proposed by Chen L et al), context based RDO (proposed by Zhang J et al) and Laplacian distribute based RDO (proposed by Li X et al). Both of the two categories above are based on the PSNR-Rate framework. However the challenge for designing a method under this framework is that PSNR do not correlate well with HVS. So, in this paper a new RDO method-IPBSS, which is based on structure similarity, is proposed. Simulation results indicate that the IPBSS can average save bit rate more than 13%, while the video quality almost remain the same. This work is supported in part by research project from National Natural Science Foundation of China (No.60402015), and research project from Guangdong Natural Science Foundation of China (No.06025642). These projects aim at construction methods of image quality assessment and its application in video coding and image compression. The research interests of the group include image quality assessment, image/video coding. The group has proposed several kind of SSIM applications, such as SSIM based JPEG2000 encoding, and SSIM based H.264 inter frame encoding is presented in this paper.