多特征融合的合成视点立体图像质量评价

Quality assessment of synthetic viewpoint stereo image with multi-feature fusion

提出一种多特征融合的方法,该方法融合了合成视点立体图像中失真区域、失真边缘和奇异值的特征。首先,利用视差和阈值得到符合人眼感知的失真区域,计算该区域的平均结构相似度;然后,提取合成视点边缘图像中符合人眼感知的失真边缘,计算在该区域的平均结构相似度;接着,计算原始图像与失真图像奇异值差异;最后,将上述特征值进行融合得到最终的客观质量分数。在合成视点立体图像库上进行测试,实验结果表明,Pearson线性相关系数和Spearman相关系数均高于0.86。与现有评价方法相比,能更准确地反映合成视点立体图像质量。

A multi-feature fusion method was proposed,the features of the distortion areas,the distortion edges and the singular value were fused.Firstly,the distortion areas corresponding to the human eyes were obtained by using the parallax and the threshold,and then the average structural similarity in these areas was calculated.Secondly,the distorted edges in the synthetic viewpoint edge image were extracted,and then the average structural similarity was calculated.Thirdly,the difference between the original image and the distorted image singular value was calculated.Finally,the three features were fused to obtain the final objective quality score.The experiment was carried out on the synthetic viewpoint stereo image library.The experimental results show that the Pearson linear correlation coefficient and the Spearman correlation coefficient are both higher than 0.86.Compared with the existing assessment methods,it can better reflect the quality of synthetic viewpoint stereo images.