Image quality assessment based on Structure Similarity

Image Structure Similarity (SSIM) and its extended versions have been successfully used in image quality assessment. In this paper, we propose a similarity metric to evaluate image quality by extracting image sparse structure from natural scene image. A sparse dictionary trained on the data contains the basic elements for representing sparse structures, and it is insensitive to different databases. The sparse structure similarity of testing image pairs is calculated with this dictionary. The final score of image quality is obtained by counting the changed number of elements in sparse structure vector between distorted image and reference image. Experiments demonstrate that the proposed method could assess image quality effectively and outperform existing SSIM based methods.