基于失真模型的结构相似度图像质量评价

Structural similarity image quality assessment based on distortion model

提出了一种基于图像失真模型及失真视觉特性的图像质量评价方法,解决了传统结构相似度(SSIM)度量不能同时有效评价不同失真强度与不同失真类型图像质量的问题.将图像失真分解为局部线性模糊及加性噪声,通过质量敏感区域加权与噪声SSIM补偿,实现各种失真类型SSIM的聚合以提高综合评价性能.实验结果表明,这种基于失真模型的区域加权SSIM能够一致评价各种失真类型、各种失真强度的图像质量,在LIVE图库上与主观评价分回归后的相关系数达到0.946 7,优于其他SSIM算法.

This work proposed a new structural similarity （SSIM） method, which is adapted for assessing images of different distortion types and different distortion intensities. The new method models a distorted image as an original image that subjects to linear frequency distortion （LFD） and additive noise injection （NI）. LFD is local, and the SSlM of this type can be clustered by being weighted on quality sensitive regions. NI is uncorrelated with the original image, and the image quality of this type is underestimated by SSIM. So quality compensation is used to unify SSIM metric of these two types. Finally, the new method was validated with subjective quality scores on LIVE database which containing 982 images. Experimental results showed that the performance of the new method is comparable with the art-of-the-state objective methods.