基于流形特征相似度的感知图像质量评价  

Manifold Feature Similarity Based Perceptual Image Quality Assessment

作　　者：王朝云[1] 蒋刚毅[1] 郁梅[1,2] 陈芬[1] 

机构地区：[1]宁波大学信息科学与工程学院,宁波315211 [2]南京大学计算机软件新技术国家重点实验室,南京210093

出　　处：《自动化学报》2016年第7期1113-1124,共12页Acta Automatica Sinica

基　　金：国家高技术研究发展计划(863计划)(2015AA015901);国家自然科学基金(U1301257;61271270;61311140262);浙江省自然科学基金(LY15F010005;LY16F010002)资助~~

摘　　要：图像质量评价(Image quality assessment,IQA)的目标是利用设计的计算模型得到与主观评价一致的结果,而人类视觉感知特性是感知图像质量评价的关键.大量研究发现,认知流形和拓扑连续性是人类感知的基础即人类感知局限在低维流形之上.基于图像低维流形特征分析,本文提出了基于流形特征相似度(Manifold feature similarity,MFS)的全参考图像质量评价方法.首先,利用正交局部保持投影算法来模拟大脑的视觉处理过程获取最佳映射矩阵进而得到图像的低维流形特征,通过流形特征的相似度来表征两幅图像的结构差异,从而反映感知质量上的差异.其次,考虑亮度失真对人眼视觉感知的影响,通过图像块均值计算亮度相似度并用于评价图像的亮度失真;最后,结合两个相似度得到图像的客观质量评价值.在四个公开图像测试库上的实验结果表明,所提出方法与现有代表性的图像质量方法相比总体上具有更好的评价结果.Image quality assessment （IQA） aims to use computational models to measure the image quality in consistency with subjective evaluation, and human visual perception characteristics play an important role in the design of IQA metrics. From many researches on human visual perception, it has been found that the cognitive manifolds and the topological continuity can be used to describe the human visual perception, that is, human perception lies on the low-dimensional manifold. With this inspiration and manifold analysis of image, a new IQA metric called manifold feature similarity （MFS） is proposed for full-reference image quality assessment. First, orthogonal locality preserving projection algorithm is used to simulate the brain＇s visual processing process to obtain the best projection matrix so that low-dimensional manifold features of images are obtained. And the similarity of the manifold features is used to measure the structure differences between the two images so as to reflect differences in perceived quality and get a manifold features-based image quality index. Then, to consider the impact of brightness on human visual perception, the block mean values of the image are used to calculate the distortion of the image＇s brightness and design a brightness-based image quality index. The final quality score is obtained by incorporating these two indices. Extensive experiments on four large scale benchmark databases demonstrate that the proposed IQA metric works better than all state-of-the-art IQA metrics in terms of prediction accuracy.

关 键 词：图像质量评价 流形特征相似度 正交局部保持投影 视觉感知 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]