Learning Based Saliency Weighted Structural Similarity

摘要

Image quality assessment (IQA) is a critical issue in image processing applications, but commonly used criterions for image quality assessment do not map well with perceived quality. The recently proposed structural similarity (SSIM) is regarded as an excellent work in image quality assessment criterions, but it only consider local information and ignore some important global concepts. Based on the SSIM image quality assessment criterion and the detection of visual saliency in image, this paper proposes a learning based saliency weighted structural similarity IQA criterion. The algorithm combines the SSIM index and saliency map in a machine learning framework to learn a mapping from these features to perceived image quality. Experiments on a standard image quality assessment database show that our algorithm performs better than commonly used criterions, and our algorithm captures results which correlate well with subjective judgments of image quality.