Research On 3D-SSIM Based Subjective Quality Optimization For HEVC

To meet people’s higher and higher demand of video quality, ITU-T and ISO jointly developed a new generation of video coding standard:High Efficiency Video Coding (HEVC).Compared with the previous H.264, HEVC can achieve higher compression performance and subjective quality. In terms of the existing video coding framework, SSE and SAD are applied to measure the distortion in Rate Distortion Optimization (RDO). However, results from the researches show that Sum of Squared Error (SSE) and Sum of Absolute Difference (SAD) fail to represent human eyes’ subject feelings to video quality effectively.But many researches show that SSE and SAD can’t reflect people’s subject feelings to video quality well. Therefore, the improvement of subjective quality in video coding is of great significance in practical applications. The main purpose of this thesis is to study a coding strategy that may optimize the subjective quality in HEVC framework, moreover, such a new method is proposed.This thesis mainly focuses on the following aspects:1) After 3D Structural Similarity(3D-SSIM) with low computational complexity and well correlation with DMOS was analyzed, two 3D-SSIM based new methods(3 DSSIM pzc and 3DSSIMpdc) are proposed to take temporal correlation and the content of the picture blocks in 3D block into consideration. Experimental results show that the proposed method 3 DSSIMp and 3 DSSIM pzc are both more relevant to subjective scores than 3D-SSIM. When compare the quality of the video in different distorton of the same sence,3DSSIMpzc outperforms 3D-SSIM.2) After analyzing the shortage of 3DSSIMpdc based RDO, a 1/(3D-SSIM) based lamda coefficient self-adaptive adjustment method for RDO of each block is proposed. After analyze the relationship between 3D-SSIM and SSE, a adjustment factor is figure out to turn the 1/(3D-SSIM) based RDO problem into a SSE based problem in which the adjustment factor related to content of picture block is used. The experiences on HM16.2 shows coding performance of the proposed method improved 1%(SCENIC, Subjective Comparison of ENcoders based on fltted Curves)on average for intra mode and improved 8.73%(SCENIC) on average for inter mode.