基于纹理与结构失真的无参考图像质量评价模型  

No-Reference Image Quality Assessment Model Based on Texture and Structure Distortion

作　　者：单月 万晓霞[1] SHAN Yue;WAN Xiao-xia(Research Center of Graphic Communication,Printing and Packaging,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉大学图像传播与印刷包装研究中心,武汉430079

出　　处：《数字印刷》2022年第4期163-173,共11页Digital Printing

摘　　要：为了解决数字图像难以准确进行客观评价的问题,本研究提出了一种基于纹理与结构失真的无参考图像质量评价(NR-IQA)模型。在纹理方面,提取图像的Tamura纹理特征、颜色共生矩阵特征;在结构特征方面提取三阶梯度特征、Hu不变矩特征,同时对亮度与色度信息进行建模,并提取颜色感知特征和最大局部变化值,构成113维图像特征向量。利用支持向量回归算法(SVR)与广义神经网络(GRNN)算法进行模型训练,并分别在LIVE数据库与TID2013数据库进行有效性与泛化性验证。结果表明,本研究算法与人眼视觉主观评价有较高的一致性,总体性能稳定且运行效率较高。In order to solve the problem that it is difficult to accurately and objectively evaluate digital images,a noreference image quality assessment model based on texture and structural distortion was proposed in this study.In terms of texture distortion,the Tamura texture feature and color co-occurrence matrix feature of the image were extracted.In terms of structural distortion,the third-order gradient feature,Hu invariant moment feature were extracted.At the same time,the luminance and chrominance information were modeled,and color perception features and maximum local change value were extracted,which together form a 113-dimensional image feature vector.Model training was performed using Support Vector Regression algorithm and Generalized Neural Network algorithm,and the validity and generalization of the model were verified in the LIVE database and TID2013 database.The results showed that the algorithm has high consistency with human subjective assessment,the overall performance is stable and the operation efficiency is high.

关 键 词：无参考图像质量评价 Tamura纹理 颜色共生矩阵 结构失真 支持向量回归 广义神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]