双通道立体图像质量评价方法的研究

Dual Channel Stereoscopic Image Quality Assessment

针对立体图像质量评价问题,基于人眼观测图像的感知特性,提出一种双通道立体图像质量评价算法。首先,获取双目视图的拉普拉斯金字塔序列构建融合图,采用并行域分解多权重化策略提取双目局部质量感知特征;然后,结合视觉平衡特性引入语义特征通道提取双目高层次语义特征;最后,在支持向量回归的基础上得到双通道主客观图像质量评价值的关系映射。双通道网络集成了包含视差信息的多局部细节特征与全局语义特征,在LIVE 3D phaseⅠ与LIVE 3D phaseⅡ立体图像库进行性能测试。结果表明:所提算法所得预测值与主观评价值间具有良好的一致性。

Aiming at the problem of stereoscopic image quality assessment,based on the perceptual characteristics of images observed by human eyes,a dual channel stereoscopic image quality assessment algorithm was proposed.Firstly,the fusion map was constructed by obtaining the Laplace pyramid sequence of binocular views to extract the binocular quality-aware features.Binocular local quality perception features were extracted by employing a parallel domain decomposition multi-weighting strategy.Then,a semantic feature channel was introduced for the extraction of binocular high level semantic features based on the visual balance characteristics.Finally,the relationship mapping between the subjective and objective image quality evaluation values of the dual channel was obtained based on the support vector regression.The dual channel network integrated multi-local detail features with parallax information and global semantic features.The performance was tested in LIVE 3D phaseⅠand LIVE 3D phaseⅡstereoscopic image databases.The results show that the predicted values have better consistency with the subjective evaluation values.