Human visual system based similarity metrics

Abstract

Objective assessment of image quality is important for a number of image processing applications. Similarity metrics have been used for methods such as automating compression, automating watermarking, and benchmarking algorithm success. The goal of objective quality assessment is to quantify the quality of images in a manner consistent with human perception. For this reason, we introduce a novel image similarity metric based on the human visual system. The measures of enhancement (EME, AME, and LogAME) have been successfully used to quantify human quality perception for image enhancement. In this paper, we present a modified version of the Logarithmic AME which can successfully be used to quantify image similarity. We compare the quantitative assessments of this algorithm with those of the well known Mean Squared Error (MSE), Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM) on the basis of correlation with subjective human evaluations for a number of images.