西北大学学报(自然科学版)

目的采用一种新颖的视觉信息处理模型模拟人眼视觉系统的功能。方法在两个独立的视觉信息处理通道分别测量基层与细节层的视觉相似度,然后通过空间域加权合并得到图像感知质量。结果提出一种基于频域相似性的图像质量评价算法。结论该算法在多个数据库上的整体性能以及针对单个失真类型的表现都要优于均方差(MSE),通用质量指标(UQI)、结构相似度(SSIM)和多尺度结构相似度(MS-SSIM)方法;针对失真类型的变化,它的性能表现也相对稳定,表现出较高的鲁棒性。

Aim To develop a novel visual information processing model to simulate the function of the human vision system.Methods First the visual similarity of base and detail layers of reference is measured and the images in two independent channels are tested,then the perceptual image quality is derivedvia the weighed combination in the spatial domain.Results An image quality assessment method based on the frequency-domain similarity is proposed.Conclusion Either the general performance of the proposed method on multiple image databases or its specific performance according to individual distortion types is superior to the traditional image quality assessment algorithms,including the mean square error(MSE),the universal quality index(SSIM) and the multi-scale SSIM(MS-SSIM);With respect to changes of distortion types,its performance is also relatively stable,and exhibits a high robustness.