融合2D和3D卷积神经网络的无参考立体图像质量评价  

No-reference stereoscopic image quality assessment based on 2D and 3D convolutional neural network

作　　者：贠丽霞 李朝锋 YUN Lixia;LI Chaofeng(Institute of Logistics Science & Engineering, Shanghai Maritime University, Shanghai 201306, China)

机构地区：[1]上海海事大学物流科学与工程研究院,上海201306

出　　处：《上海海事大学学报》2022年第2期120-126,共7页Journal of Shanghai Maritime University

基　　金：国家自然科学基金(61771223)。

摘　　要：为将图像处理技术更好地应用在智能交通中,发挥立体图像质量评价方法的作用,提出一种融合2D和3D卷积神经网络(convolutional neural network,CNN)的立体图像质量评价方法。该模型结合2D-CNN与3D-CNN两个通道;将独眼图输入2D-CNN通道,提取双目竞争相关特征;将左视图、右视图、和图像和差图像输入3D-CNN通道,通过3D卷积提取双目之间联系的相关特征;应用全连接层,将两个通道提取的特征融合并进行回归分析构建关系模型。在公开的LIVE 3D PhaseⅠ和LIVE 3D PhaseⅡ上的实验结果表明,所提方法与人类的主观感知保持高度一致。In order to better apply the image processing technology to intelligent transportation and play the role of the stereoscopic image quality assessment method,a stereoscopic image quality assessment method integrating 2D and 3D convolutional neural network(CNN)is proposed.In the model,the two channels of 2D-CNN and 3D-CNN are combined;the cyclopean images are input into the 2D-CNN channel to extract the related binocular competition features;the left view,right view,summation and difference images are input into the 3D-CNN channel to extract the related binocular connection features through 3D convolution;the full connection layer is applied to fuse the features extracted from the two channels and conduct regression analysis to construct the relationship model.Experimental results on the open LIVE 3D Phase I and LIVE 3D Phase II show that the proposed method is highly consistent with human subjective perception.

关 键 词：无参考立体图像 质量评价 卷积神经网络(CNN) 和图像 差图像 独眼图 

分 类 号：U495[交通运输工程—交通运输规划与管理] TP319.41[交通运输工程—道路与铁道工程]