ON BETWEEN-COEFFICIENT CONTRAST MASKING OF DCT BASIS FUNCTIONS

TLDR

The paper proposes a simple, HVS‑based model of between‑coefficient contrast masking for DCT basis functions and introduces a modified PSNR metric. The model operates on 8×8 DCT blocks, computing for each coefficient the maximum imperceptible distortion using the contrast sensitivity function. Experiments on 18 test images with 155 observers show that the new PSNR‑HVS‑M metric outperforms existing reference‑based metrics, achieving Spearman 0.984 and Kendall 0.948 correlations with subjective judgments.

Abstract

In this paper we propose a simple and effective model of visual between-coefficient contrast masking of DCT basis functions based on a human visual system (HVS). The model operates with the values of DCT coefficients of 8x8 pixel block of an image. For each DCT coefficient of the block the model allows to calculate its maximal distortion that is not visible due to the between-coefficient masking. A modification of the PSNR is also described in this paper. The proposed metric, PSNR-HVS-M, takes into account the proposed model and the contrast sensitivity function (CSF). For efficiency analysis of the proposed model, a set of 18 test images with different effects of noise masking has been used. During experiments, 155 observers have sorted this set of test images in the order of their visual appearance comparing them to undistorted original. The new metric, PSNRHVS-M has outperformed other well-known reference based quality metrics and demonstrated high correlation with the results of subjective experiments (Spearman correlation is 0.984, Kendall correlation is 0.948).