基于偏度和结构特征的无参考虚拟视点图像质量评价  

No-reference quality assessment for virtual view images based on skewness and structural feature

作　　者：王晨 彭宗举[1,2] 章联军 陈芬[1,2] 陆志华[1] WANG Chen;PENG Zongju;ZHANG Lianjun;CHEN Fen;LU Zhihua(Faculty of Information Science and Engineering,Ningbo University,Ningbo Zhejiang 315211,China;School of Electrical and Electronic Engineering,Chongqing University of Technology,Chongqing 400054,China)

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211 [2]重庆理工大学电气与电子工程学院,重庆400054

出　　处：《计算机应用》2021年第S02期226-233,共8页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61771269);浙江省自然科学基金资助项目(LY20F010005);重庆市教委科学技术研究项目(KJZDK202001105);宁波市自然科学基金资助项目(2019A610107);重庆理工大学科研启动基金资助项目(2020ZDZ029)。

摘　　要：针对虚拟视点图像存在与普通图像不同失真的问题,提出了一种基于偏度和结构特征的无参考虚拟视点图像质量评价方法。首先,将输入图像分为H、S、V三个通道,每个通道都均分为九个小块,对每个小块提取偏度特征;然后,利用局部二值模式算子对结构特征映射进行编码,计算质量感知分数;最后,将偏度特征和结构特征输入支持向量机进行训练,得到无参考的虚拟视点图像回归与视觉质量预测模型来预测图像质量。在IRCCyN/IVC和MCL-3D两个公共虚拟视点图像数据库上对提出方法进行了实验。实验结果表明,提出方法的皮尔逊线性相关系数(PLCC)分别为0.853 8和0.953 4,斯皮尔曼秩相关系数(SROCC)分别为0.796 6和0.915 9。提出方法的PLCC和SROCC均高于自回归加阈值(APT)等10个虚拟视点图像质量评价方法和BRISQUE等6个通用的无参考质量评价方法。该方法采用的偏度和结构特征能很好地评价虚拟视点图像的视觉质量,评价结果与主观感知有较好的一致性。In order to solve the problem that virtual viewpoint images have different distortions from ordinary images,a non-reference virtual viewpoint image quality assessment method based on skewness and structural characteristics was proposed. First,the input image was divided into three channels,H,S,and V,and each channel was divided into nine small blocks,and the skewness feature was extracted for each small block. Then,the local binary pattern operator was used to encode the structural feature map and calculate the quality perception score. Finally,the skewness features and the structural features were taken as the input to the support vector machine for training,and the reference-free virtual viewpoint image regression and visual quality prediction model was obtained to predict image quality. Experiments for the proposed method were carried out on two public virtual viewpoint image databases of IRCCyN/IVC and MCL-3 D. The experimental results show that the Pearson Linear Correlation Coefficients(PLCC)are 0. 853 8 and 0. 953 4,and the Spearman Rank Order Correlation Coefficients(SROCC)are 0. 796 6 and 0. 915 9,respectively. Compared with 10 virtual viewpoint image quality assessment methods such as APT(Autoregression-Plus Thresholding)and 6 general non-reference quality assessment methods such as BRISQUE,the PLCC and SROCC of the proposed method are higher. The skewness and structural features proposed in this paper can well evaluate the visual quality of virtual viewpoint images,and the result of the proposed prediction model is highly related to subjective visual perception.

关 键 词：虚拟视点 图像质量评价 偏度 局部二值模式算子 支持向量机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]