摘要： 结构相似度(SSIM)是一种衡量2幅图像相似度的重要指标,但是SSIM对于严重模糊的图像评价结果准确度不高。在基于SSIM的图像质量评价方法基础上,考虑到纹理信息是图像的重要组成部分以及人眼对图像的纹理信息部分十分敏感,提出基于离散余弦变换(DCT)域 的纹理结构相似度质量评价(TSSIM)。该评价方法在DCT域中提取图像的纹理信息,并通过改进SSIM中的结构信息来确定纹理区域的失真,计算TSSIM值。实验结果表明,该方法得到的Pearson相关系数为0.893 42,Spearman等级相关系数为0.925 07,比SSIM的评价结果更准确,符合人眼的主观感受。

关键词: 结构相似度, 人类视觉系统, 离散余弦变换域, 纹理结构相似度, Pearson相关系数, Spearman等级相关系数

Abstract: Structure Similarity(SSIM) is an important indicator of measuring the similarity of two images,but there still remain some deficiencies when assessing badly blurred images.In this paper,a new method using image texture combines with SSIM,which takes the importance of the texture information to images and human eye perception structure into account.The algorithm combines texture features in Discrete Cosine Transform(DCT) domain with SSIM to acquire the Texture Structure Similarity(TSSIM).Experimental results for Pearson correlation coefficient is 0.893 42,and the Spearman rank correlation coefficient is 0.925 07,which show that the proposed method works better in the aspect of subjective visual impression than SSIM.

Key words: Structure Similarity(SSIM), Human Visual System(HVS, Spearman rank correlation coefficient

中图分类号: 

• TP391.4