Novel image quality metric based on similarity

In this paper, we present a novel approach to image quality metric taking into account degradation of contrast and brightness as well as block similarity. The metric is achieved by performing of the following steps: 1) reducing contrast and brightness in distorted image, 2) using block-matching (BM) to group similar 2D image fragments into 3D data arrays in original image and preprocessed distorted image separately, 3) analyzing of these blocks in DCT domain. The DCT coefficients differences are calculated between pixel values with contrast sensitivity function (CSF) and reduced by contrast masking according to Human Visual System (HVS). We validate the performance of our algorithms with five most popular quality image databases: TID, LIVE, CSIQ, IVC and Cornell-A57. The analysis of the results shows that the proposed quality metric provides better correlation to Mean Observer Score (MOS) than most of recent popular state-of-the-art metrics, e.g. MSSIM, SSIM. The average Spearman Correlation of proposed metric reaches 0.894.