基于梯度结构相似度的AVS帧间模式选择算法  

AVS INTER MODE DECISION ALGORITHM BASED ON GSIM

作　　者：白伟[1] 

机构地区：[1]太原广播电视大学,山西太原030006

出　　处：《计算机应用与软件》2015年第12期105-107,113,共4页Computer Applications and Software

基　　金：山西省青年科技研究基金项目(2010021019)

摘　　要：AVS帧间模式选择率失真优化使用绝对误差和(SAD)作为失真度度量,方法简单,但不能很好地符合人眼视觉系统(HVS);最近提出的结构相似度(SSIM)图像质量评价方法更符合HVS的特性,但不能很好地评价严重模糊的降质图像,基于梯度幅度值的结构相似度图像质量评价方法(GSIM)可解决此问题。然而率失真优化使用GSIM作为失真度度量,计算复杂,不利于实时编码。针对SAD不能很好符合HVS和GSIM计算复杂的问题,采用SAD和GSIM的结合作为失真度度量,并利用帧间预测模式间的相关性,提出新的帧间预测模式选择算法。此算法取SAD的最优值和次优值,计算其差值,根据差值和阈值的比较判断是否需要GSIM计算,其中阈值的选择根据预测模式间相关性自适应确定。需要计算GSIM值时,根据GSIM的最优值和次优值的情况判断是否需要对帧间模式进行修正。实验结果表明,该算法较传统算法GSIM值增加0.0007,PSNR降低0.1 db,编码时间增加1.24%。与传统算法相比,该算法有较好的主观质量,客观质量几乎不变,编码时间增加很少。The rate distortion optimisation for AVS inter mode decision uses the sum absolute difference（ SAD） as the distortion metric,it is simple,but is not consistent with human vision system（ HVS） quite well. The structural similarity（ SSIM） image quality evaluation method proposed recently accords with the HVS more,but yet has some deficiencies in assessing badly blurred degraded images. The image quality evaluation method of the gradient-magnitude-based structure similarity（ GSIM） can solve the problem. The rate distortion optimisation using GSIM can be consistent with HVS quite well,but its complexity does not conducive to real-time coding. The paper uses the combination of SAD and GSIM as the distortion metric in light of the above problems of them,and makes use of the correlation between interframe prediction modes to present new interframe prediction mode decision algorithm. This algorithm calculates the difference of optimal value and suboptimal value of SAD,and judges whether the calculation of GSIM is needed according to the comparison of the difference value and threshold value,in it the threshold is adaptively determined based on the correlation between prediction modes. When to calculate GSIM value is needed,according to the optimal values and suboptimal values of GSIM it will determine whether or not to correct the inter modes. Experimental results show that compared with traditional algorithm the GSIM value can get up to 0. 0007 increase,the PSNR drops 0. 01 d B,and the encoding time increases 1. 24%. The algorithm has better subjective quality than the traditional algorithm,while the objective quality has rare reduction and the encoding time is just slightly increased.

关 键 词：AVS 梯度结构相似度 率失真优化 帧间模式选择 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]