基于视觉阈值的结构相似度图像质量评价算法  

Image quality assessment based on human visibility threshold theory and structural similarity

作　　者：胡媛媛[1,2] 牛夏牧[1] 

机构地区：[1]哈尔滨工业大学深圳研究生院,深圳518055 [2]深圳大学信息工程学院,深圳518060

出　　处：《深圳大学学报（理工版）》2010年第2期185-191,共7页Journal of Shenzhen University(Science and Engineering)

基　　金：国家自然科学基金资助项目(60832010);广东省自然科学基金资助项目(9151806001000025)

摘　　要：针对空间域结构相似度测量(image measure of structural similarity,SSIM)没有考虑人类视觉系统亮度掩蔽和对比度掩蔽对视觉感知的影响,提出基于人类视觉阈值理论的结构相似度图像质量评价算法(visibility threshold SSIM,VTSSIM).通过计算局部标准方差,将图像分为平坦区域、细节区域和一般区域;基于图像视觉阈值的临界可视误差分布,确定平坦区域中的亮度掩蔽极端区域和细节区域中的对比度掩蔽极端区域;分别对这两个极端区域及其他区域进行视觉阈值加权处理,得到整幅图像的质量评价.结果表明,VTSSIM比SSIM更符合人眼视觉特性,尤其适合评价失真较小的图像.In the experiment on images with just noticeable distortion （JND）, it was found that the influences of the luminance adaptation and contrast masking on visual perception were not taken into consideration in the image measure of structural similarity （SSIM）. To overcome such limitation, the visibility threshold SSIM （VTSSIM） algorithm was proposed in the paper. First, the prototype image was divided into smooth, detailed and common image blocks according to its local standard deviations. Then, two extreme regions, i. e. , the luminance adaptation extreme regions for smooth blocks and contrast masking extreme regions for detailed blocks, were determined by the JND profile. Finally, the SSIM maps of the extreme regions together with those of the remaining blocks were weighted to get the ultimate quality assessment of VTSSIM. Experimental results show that VTSSIM is consistent with the judgments of human observers than SSIM, especially for the images with less distortion.

关 键 词：图像处理 图像质量评价 结构相似度测量 人类视觉系统 亮度掩蔽 对比度掩蔽 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP37[自动化与计算机技术—计算机科学与技术]