基于非局部结构张量的SSIM图像质量评价方法

SSIM image quality assessment based on nonlocal structure tensor

针对基于局部运算的图像质量评价方法的局限性,提出一种基于非局部结构张量的SSIM图像质量评价方法。图像在各像素点的非局部结构张量的主特征值大小很好地反映了该像素点的结构强度信息,特别是纹理结构等细节信息;主特征向量的方向反映了该像素点的结构方向信息。利用退化图像和参考图像的非局部结构张量的主特征值相似度刻画结构强度相似度,利用主特征向量夹角的余弦刻画结构方向相似度。数值实验结果显示,利用该方法对TID2008数据库中的图像进行评价的平均运算时间为778.43 s,且评价结果与主观评价接近。

For the weakness of the image quality assessment methods based on local computation, this paper proposed a nonlocal structure tensor based SSIM image quality assessment method. The primal eigenvalue of nonlocal structure tensor at each pixel well reflected the structural strength information, especially the details information such as texture and so on. And the orientation of the primal eignvector reflected the structural direction information. This paper used the similarity of the primal eigenvalues to describe the consistency of the structural strength between the reference image and the distorted image, and the cosine of the intersection angle between the primal eignvectors to characterize the similarity of the structural direction. Numerical experiment shows that the average computing time is 778.43 s by using this method to assess the images in TID2008 data- base. And the assessment result is close to the subjective assessment.