摘要： 人类视觉皮层采用分层结构，其中双目融合与双目竞争首先发生在低级视觉区域，但当前基于深度学习的立体图像质量评价模型普遍采用在网络的不同层次上融合左右视点图像特征来估计立体图像质量值，对人类低级视觉区域感知的模拟程度存在不足。鉴于此，本文提出了一种仿双目竞争的立体图像质量评价方法。首先，模拟双目视觉竞争现象，构建了一个基于无监督方法的双目图像融合模型。通过左右视点图像的梯度幅值响应来评估图像降质程度，确定左右视点图像的融合权重。并利用深度卷积神经网络对输入图像先验知识的获取能力，建立基于编码器-解码器架构的无监督图像生成网络，以左右视点两幅图像作为学习对象，实现左右视点图像的融合。其次，利用在大规模图像数据库上预训练的ResNet50模型从融合图像中提取质量感知特征，并构建了一个基于支持向量回归的特征质量映射模型来估计立体图像的质量值。实验结果显示，在四个经典立体图像基准测试数据库上，所提出方法在PLCC（Pearson linear correlation coefficient）和SROCC（Spearman rank order correlation coefficient）两个评价指标上均超过了0.96，并且均方根误差均优于对比方法。这表明，所提出的基于无监督双目图像融合的方法能够有效模拟双目视觉效应，从而显著提高立体图像质量评价的准确性。

Abstract: The human visual cortex has a hierarchical structure, in which binocular fusion and binocular rivalry first occur in the low-level visual areas. However, current deep learning-based stereoscopic image quality assessment (SIQA) models generally estimate the quality values of stereoscopic images by fusing the features of left and right view images at different levels of the network, resulting in insufficient simulation of the perception in the low-level visual areas of human visual cortex. To address this issue, this paper proposes a SIQA method that simulates binocular rivalry to further enhance the evaluation accuracy. First, we leverage the ability of deep convolutional neural networks to acquire prior knowledge of input image and build a binocular image fusion model based on an unsupervised approach. This model takes the left and right views as learning targets to simulate the binocular fusion process in the human visual system. The gradient magnitude responses of the left and right images are utilized to calculate the image degradation coefficient, which is then used to obtain the fusion weights of the left and right views, simulating the binocular rivalry phenomenon. Then, we utilize a pre-trained ResNet50 model to extract quality-aware features from the fused image and establish a feature-quality mapping model based on support vector regression to estimate the quality value of the stereoscopic image. Experimental results demonstrate that our proposed SIQA method achieves over 0.96 on both Pearson linear correlation coefficient (PLCC) and Spearman The human visual cortex has a hierarchical structure, in which binocular fusion and binocular rivalry first occur in the low-level visual areas. However, current deep learning-based stereoscopic image quality assessment (SIQA) models generally estimate the quality values of stereoscopic images by fusing the features of left and right view images at different levels of the network, resulting in insufficient simulation of the perception in the low-level visual areas of human visual cortex. To address this issue, this paper proposes a SIQA method that simulates binocular rivalry to further enhance the evaluation accuracy. First, we leverage the ability of deep convolutional neural networks to acquire prior knowledge of input image and build a binocular image fusion model based on an unsupervised approach. This model takes the left and right views as learning targets to simulate the binocular fusion process in the human visual system. The gradient magnitude responses of the left and right images are utilized to calculate the image degradation coefficient, which is then used to obtain the fusion weights of the left and right views, simulating the binocular rivalry phenomenon. Then, we utilize a pre-trained ResNet50 model to extract quality-aware features from the fused image and establish a feature-quality mapping model based on support vector regression to estimate the quality value of the stereoscopic image. Experimental results demonstrate that our proposed SIQA method achieves over 0.96 on both Pearson linear correlation coefficient (PLCC) and Spearman