Image Quality Assessment Integrating With Multiple Regression And Structural Similarity Algorithm

Image quality assessment becomes one important research of image information engineering field. Subjective method is laborious and time-consuming so that it hard to apply into real-time systems, while traditional objective method is simple but its result often inconsistent with human perception, such as PSNR or MSE. Structural similarity index theory is an novel idea of quality assessment, and simulates the whole function of HVS from top to down, namely the function is HVS extracts structure information from visual scenes, different from the idea that bottom-up simulating the low-level structure of HVS.Structural similarity index Algorithm is easy to realize, avoiding some defects existing in bottom-up method, and superior to other existing methods.Although SSIM is superior to many other methods, it has a few defects. This paper proposes a new algorithm that image gradient information is added into SSIM as a new component. Human eyes are extremely sensitive to image edge and texture information, while gradient can reflect these information well.The algorithm has good performance in blur images, besides other types distorted images. And the results are better consistent with human perception.In the case of single but fixed weights of SSIM components, this paper puts forward an idea that SSIM algorithm integrated with multiple regression analysis, specifically, weights of all the components would be adjusted through regression analysis. It shows each component is of unequal importance in distorted image quality assessment, and also improves accuracy of SSIM algorithm applied to different distortion type of image quality assessment. The experimental results match subjective perception well.Through extensive experiments on LIVE database, methods proposed this paper are superior to traditional objective methods in correlation coefficient after nonlinear regression(CC), Spearman rank correlation coefficient(SROCC), and outliner ratio(OR), much improved comparing to some other SSIM improved algorithm, and effectively increased the consistency of subjective and objective in image quality assessment.