Improving measurement bias of Structural Similarity Index (SSIM) using absolute difference equation

Abstract

Structural similarity index (SSIM) is a framework for assessing the perceptual quality from an image using the degrading and structural information of an image. It is an alternative method for quantifying visual image quality since subjective evaluation by humans, in practice, is too inconvenient, time-consuming and expensive. SSIM has been widely used for almost two decades in different research disciplines in quality assessment. However, there is a deficiency in some of the mathematical components in the SSIM that may lead to incorrect, impractical and unrealistic results. In this paper, we address the problems with the SSIM and propose replacing the luminance and contrast component with absolute difference equations that obeys Weber’s law of human perception of change. We compare the results of both SSIM and new proposed SSIM in the luminance, contrast and human perception test. The proposed SSIM seem to be more human-like in determining image quality compared to the previous version of SSIM.