Modified Gradient Similarity Vs Structural Similarity for Image Quality Assessment

Abstract

In this paper, we have proposed a Image Quality assessment (IQA) using gradient information which considers the image with edges to assess quality through contrast and structure comparison along with Luminance comparison. The existing structural similarity (SSIM) method measures structure loss based on statistical moments, i.e., the mean and variance, represents mainly the luminance change of pixels rather than describing the spatial distribution. However, the human visual system (HVS) easily identifies the quality when there is comparison of two images. Of course HVS is very sensitive to gradient changes. In this paper, we use the modified gradient similarity concept comparing with existing Structural similarity based IQA. Also we have shown here a better result Using modified gradient similarity when compared to simple gradient based IQA. Furthermore, considering the viewing condition, we extend the ISSIM metric to the multi-scale space. Experimental results demonstrate the proposed IQA method is more consistent with the human perception than the SSIM metric and normal Gradient metric as well.

Keywords: Image Quality Assessment (IQA); Peak Signal-To-Noise Ratio (PSNR); Mean Square Error (MSE).