Image Quality Assessment Using Gradient-weighted Structural Similarity

Digital images are subject to a wide variety of distortions during image processing application, and it is necessary to develop objective image quality metric to evaluate the degradation automatically. Images are prepared for human eyes so that the assessment result must be consistent with human visual effect. Structure Similarity (SSIM), a well-known objective image quality assessment, is proposed by Zhou Wang. SSIM assumes that human visual perception is highly adapted to extracting structure information from a scene. Compared with PSNR or MSE,SSIM has a stronger advantage which has been proved in many different image quality assessments. However, due to the HVS characteristics of the underlying visual are neglected, SSIM has some drawbacks such as failing in blurred image measurement. In this paper, an improved SSIM method which we call it Gradient-Weighted SSIM (GWSSIM) is proposed based on the visual masking effect. GWSSIM performs in different regions of images which are weighted with different values based on their weight values. Experimental results show that GWSSIM has a better performance than both PSNR and SSIM.