基于结构信息的图像质量评价模型

Structural Information-Based Image Quality Assessment Models

针对客观数字图像质量评价方法结构相似度(SSIM)算法对严重失真类型的评价值与主观评价(MOS)值之间存在着明显差异的问题,对几种将SSIM算法与人眼视觉系统(HVS)特性相结合的算法基于图像边缘的ESSIM算法、基于空间域频域敏感的HSSIM算法,以及基于信息内容权值的IWSSIM算法进行了比较。从实验数据可知,ESSIM,HSSIM算法的准确性有不同程度的提高,而IWSSIM模型在比较主客观值相关的单调性方面更为稳定。实验结果为今后结构相似度算法的改进提供了依据,指明研究方向。

The objective assessment algorithm Structural Similarity（SSIM） of image quality was not consistent with the Mean Opinion Score （MOS） in particular types of distoration, such as seriously blurred type, etc. In this paper, three models Edeg-based SSIM, HVS-based SSIM, and IW-SSIM which had considered the characteristics both of HVS and structural information were compared in evaluating the seriously Gaussian blurred images. The experimental datas showed that, the IW-SSIM＇s monotony perfomance was more stable than other models in the relationship between the subjective and objective values, as well as the growing accuracy of the ESSIM and HSSIM. These experimental results are benefit for improvingt the SSIM approach.