基于视觉显著性与对比度特性的图像质量评价  

Image Quality Assessment Based on Visual Saliency and Image Contrast

作　　者：李钊[1] 史再峰[1] 李斌桥[1] 徐江涛[1] 姚素英[1] 庞科[1] 

机构地区：[1]天津大学电子信息工程学院,天津300072

出　　处：《南开大学学报（自然科学版）》2015年第6期46-52,共7页Acta Scientiarum Naturalium Universitatis Nankaiensis

基　　金：国家高技术研究发展计划("863"计划)资助项目(2012AA012705);国家国际科技合作专项(2012DFB10170)

摘　　要：提出1种用于客观图像质量评价的新方法.分别对参考图像与失真图像进行高斯金字塔分解;对于分解后的每1层图像,采用基于相位谱的视觉显著图来计算2幅图像中每个对应像素点间的相似性,同时采用梯度图表示图像的对比度特性.视觉显著性和对比度特性互为补充,分别反映了人类视觉系统的不同特性,将2者加权相乘可获得单尺度的质量评价值;最后对各层结果进行综合可获得失真图像的客观质量评价值.在LIVE、TID 2008、CSIQ、IVC 4个公开的图像质量评价数据库上进行了大量的实验,结果表明,与其他相关的图像质量评价方法相比,提出的客观质量评价方法与人类主观评价的结果具有更好的一致性.A new objective image quality assessment method was proposed. Gaussian pyramid decomposition was conducted on the reference image and the distorted image, respectively. For each level of the images, the phase spectrum visual saliency map was used to weight the contribution of each point to the similarity of two images, and the image gradient map was computed to describe the information of image contrast. The phase spectrum visual saliency of the image and the image contrast were complementary and reflected different aspects of human visual system separately. The weighted product of the visual saliency and the gradient of all levels were synthesized to gain the objective quality evaluation value. Extensive ex- periments were performed on four publicly available IQA databases, including LIVE, TID 2008, CSIQ and IVC. The experimental results confirm that the method proposed can obtain greater consistency with human subjective evaluation than the other related image quality assessment indexes.

关 键 词：图像质量评价 人类视觉系统 视觉显著性 图像对比度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]