摘要
针对畸变失真图像影响后续图像信息获取的问题,提出了一种基于双线性特征融合方法的畸变失真图像质量评价算法。首先基于人类视觉对图像结构特征敏感的特点提取结构图像,然后将原始图像和结构图像作为双流卷积网络的输入,将两支路网络输出的低层结构特征和高层语义特征通过双线性池化层进行特征融合,最后通过全连接层输出图像质量预测分数。为了验证算法的有效性,在4个公开的图像数据集LIVE、CSIQ、MLIVE与TID2013上进行实验。结果表明,所提算法与当前的失真图像质量评价算法相比,在LIVE、CSIQ和MLIVE数据集上斯皮尔曼等级相关系数(Spearman rank-order correlation coefficient,SROCC)和皮尔逊线性相关系数(Pearson linear correlation coefficient,PLCC)指标至少提升0.2%左右,在TID2013数据集上SROCC值至少提升0.5%左右。所提算法评价结果与人类视觉主观感知有较好的一致性,对多种失真图像质量评价可以取得不错的效果。
Aiming at the problem that the distorted image affects the subsequent image information acquisition,a distorted image quality assessment algorithm based on a bilinear feature fusion method was proposed.Firstly,the structural image was extracted based on the feature that human vision is sensitive to image structural features,then the original image and the structural image were used as the input of the two-stream convolutional network,and the low-level structural features and the high-level semantic features output from the two branch networks were fused through the bilinear pooling layer for feature fusion,and finally the image quality prediction score was output through the fully connected layer.To verify the effectiveness of the algorithm,experiments were conducted on four publicly available image datasets,LIVE,CSIQ,MLIVE and TID2013.The results show that the proposed algorithm improves the Spearman rank-order correlation coefficient(SROCC)and Pearson linear correlation coefficient(PLCC)metrics by at least 0.2%on the LIVE,CSIQ and MLIVE datasets,and SROCC by at least 0.5%on the TID2013 dataset compared to the current distorted image quality assessment algorithms.The assessment results of the proposed algorithm are in good agreement with the subjective perception of human vision,and good results can be achieved for a wide range of distorted image quality assessments.
作者
陆绮荣
丁昕
梁雅雯
LU Qirong;DING Xin;LIANG Yawen(College of Information Science and Engineering,Guilin University of Technology,Guilin,Guangxi 541000,China;Guangxi Key Laboratory of Embedded Technology and Intelligent System(Guilin Universityof Technology),Guilin,Guangxi 541000,China)
出处
《中国科技论文》
CAS
北大核心
2023年第3期259-264,291,共7页
China Sciencepaper
基金
国家自然科学基金资助项目(62166012)。
关键词
畸变失真图像
双流卷积网络
双线性池化
特征融合
distorted images
two-stream convolutional network
bilinear pooling
feature fusion
