基于特征解纠缠表示的无参考图像质量评价  

No-Reference Image Quality Assessment via Disentangled Representations

作　　者：廖东 孟浩 葛珊[2] 马希超 曹洪龙[1] 胡剑凌[1] LIAO Dong;MENG Hao;GE Shan;MA Xichao;CAO Honglong;HU Jianling(School of Electronic and Information Engineering,Soochow University,Suzhou 215006,China;The Third Research Institute of China Electronics Technology Group Corporation,Beijing 100015,China)

机构地区：[1]苏州大学电子信息学院,江苏苏州215006 [2]中国电子科技集团公司第三研究所,北京100015

出　　处：《电视技术》2020年第12期44-49,共6页Video Engineering

摘　　要：无参考图像质量评价是在不给定参考图像的情况下评估受损图像的质量。由于缺少无损的参考图像,无参考图像质量评价方法的表现与全参考图像质量评价方法相比差距较大。为了解决这一问题,提出一种基于特征解纠缠表示的无参考图像质量评价方法。该方法能分离出受损图像的内容特征和受损信息特征,通过内容特征对受损图像进行复原,使用孪生卷积网络从受损图像和复原图像中提取图像特征,并将其与受损信息特征进行融合,进而预测受损图像的质量。在TID 2013数据集上,提出的无参考图像评价方法的Spearman秩相关系数与Pearson线性相关系数分别为0.885和0.876,性能优良。No-Reference Image Quality Assessment(NR-IQA)is to evaluate the quality of distorted images without the reference images.Due to the lack of distortion-free reference images,the performance of the NR-IQA is worse than Full-Reference Image Quality Assessment(FR-IQA).In order to address this problem,this paper proposes a NR-IQA framework based on disentangled representations.This method can separate the content representation and the distortion information representation of the distorted image,and restore the distorted image through the content representation.Then a siamese convolutional network is designed to extract image features from the distorted image and the restored image,which are merged with the distortion information representation to predict the quality of the distorted image.On the TID 2013 dataset,The Spearman rank correlation coefficient and Pearson linear correlation coefficient of the non-reference image assessment method proposed in this paper are 0.885 and 0.876 respectively.

关 键 词：无参考图像质量评价 受损图像 图像特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]