快速提升的结构相似度IQA算法  

Fast Lifting IQA Algorithm of the Structure Similarity Index

作　　者：高丽霞[1] 刘国军[1] 张选德[1] Gao Lixia;Liu Guojun;Zhang Xuande(School of Mathematics and Statistics, Ningxia University, Yinchuan 750021)

机构地区：[1]宁夏大学数学统计学院,银川750021

出　　处：《计算机辅助设计与图形学学报》2017年第7期1314-1323,共10页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61461043;61362029);宁夏自然科学基金(NZ1616)

摘　　要：客观图像质量评价(IQA)的目的是设计与主观评价算法尽可能一致的数学模型来度量图像的质量.针对结构相似度(SSIM)和其他一些算法的局限性,考虑到梯度可以反映图像的边缘纹理等结构信息,提出一种快速的全参考型IQA算法,即提升的梯度加权结构相似度(GWL-SSIM)算法.首先定义图像局部块的广义梯度;然后利用其相似性及图像对比度相似性和结构相似性得到局部质量的特征映射图;最后采用广义梯度加权的池化策略得到IQA模型,广义梯度能很好地刻画视觉感知系统的非线性属性,而加权策略模拟了其对图像不同成分感知的差异性.在6个公开数据库中进行数值实验的结果表明,GWL-SSIM算法计算效率高,并取得了与目前流行算法相当的评价效果.Image quality assessment(IQA)aims to design mathematical models for measuring an imagequality well consistent with subjective evaluations.Considering both the limitations of structure similarity(SSIM)index and other methods,and the advantages of the gradient on characterizing the edge and textureinformation of an image,we put forward a kind of fast full-reference IQA algorithm,namely gradientweighted lifting SSIM(GWL-SSIM)method.Firstly,we define the generalized gradient of local image block.Then employ the generalized gradient similarity,contrast similarity and structural similarity to obtain a featuremapping image of local quality.Finally,we adopt the weighted pooling strategy by generalized gradientinformation to obtain an image evaluation model.In our algorithm,the generalized gradient well describesthe nonlinearity of human visual system(HVS)for image perception,while the weighted strategy simulatesthe diversity how HVS perceive the different ingredients of an image.Numerical experiments,performed onsix public databases,demonstrate that GWL-SSIM can achieve high computational efficiency and have aconsiderable evaluation result compared with state-of-the-art IQA metrics.

关 键 词：图像质量评价 广义梯度特征 结构相似度 池化策略 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]