Research On Image Quality Assessment Based On HVS

The digital images and videos that can be researched on the Internet had caused the people's attention. And the attention and research in the field of image quality assessment also shows the accelerating development trend.In recent years, the research on physiological and psychovisual features of human visual system (HVS) has made important progress. Owing to that, many image quality assessment methods based on human visual characteristics was proposed, and the evaluation result was very good. Based on previous research, this paper focused on human visual characteristics, and proposed two effective image quality assessment methods based on HVS.Image edge is important information for human eye to distinguish object. But the image edge and texture information was lost during compression. This paper uses the gradient feature vector to represent the image edge structure information, and proposed a compressed image quality assessment method based on gradient vector similarity. At the first, compute the similarity between reference image and test image using the gradient vector to generate image quality map. By weighting the image quality map with the interesting region and the multi-scale characteristics of the human eye, we can get the final image quality index. Simulation results upon the largest TID2013 and TID2008 database show that the evaluation method consists with subjective evaluation results better. In addition the method can be widely used to assess the quality of JPEG and JPEG2000 compressed images.Aiming at the database dependence problem existed in the existing generic no-reference image quality assessment methods, a scale invariance based no-reference image quality assessment method is proposed. The algorithm combined the Natural Scene Statistic (NSS) feature of images and edge characteristic that the human eye perceives as the valid feature for image quality. On the basis of the scale invariant feature, we compute the global difference across scales as the image quality score using the two feature vectors. The method is no need for external information. The results of the Cross validation experiments show that the proposed method has good evaluation for multi-distorted images with low computational complexity. Compared to the state-of-the-art no-reference image quality assessment models, the proposed method has better comprehensive performance, making it well suit for applications.