基于视觉特性的无参考图像质量评价算法  

Reference-free Image Quality Assessment Algorithm Based on Visual Characteristics

作　　者：陈蓉 刘本永 CHEN Rong;LIU Benyong(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025

出　　处：《软件导刊》2024年第5期156-161,共6页Software Guide

基　　金：国家自然科学基金项目(60862003)。

摘　　要：针对传统无参考图像质量评价算法特征单一、需主观评价结果参与训练和对比度失真图像性能差等问题,提出一种结合颜色和纹理特征及视觉特性的改进算法。首先,为待评价图像构造一幅伪参考图像;其次,依据视觉特性将二者划分为主要与次要感兴趣区域,进一步将区域细化为更小的块并提取其中不同的颜色特征与纹理特征以建立多元高斯模型;最后,计算二者主要感兴趣区域之间、次要感兴趣区域之间的模型参数距离,作为质量分数评价图像的好坏。在公开数据集SPAQ和CSIQ上与传统算法QAC、NIQE、IL-NIQE等进行比较,计算线性皮尔森系数、斯皮尔曼秩相关系数、均方根误差等质量评价指标,实验结果表明所提算法优于传统算法。Concerning the problems in the traditional reference-free image quality assessment algorithms,such as only single features are used,subjective assessment results are needed in training,and poor performance for contrast distortion images,an improved algorithm is proposed based on color and texture features and visual characteristics.Firstly,a pseudo-reference image is constructed for an image to be evaluated,and both are divided into primary and secondary regions of interest according to the visual characteristics.Secondly,the regions are further divided into smaller blocks,and then different color and texture features are extracted for each block and multivariate Gaussian models are established.Finally,the model parameter distances between the main regions of interest and the secondary regions of interest are calculated as a quality score to evaluate the quality of image.Using the public datasets SPAQ and CSIQ and comparing with the traditional algorithms QAC,NIQE,IL-NIQE,etc.by calculating the values of evaluation indicators of linear Pearson coefficient,Spearman rank correlation coefficient,and root mean square error,the results show that the proposed algorithm outperforms the traditional ones.

关 键 词：无参考图像质量评价 视觉特性 伪参考图像 MVG模型 纹理特征 颜色特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]