Image Quality Assessment by Considering Multiscale and Multidirectional Visibility Differences in Shearlet Domain

Abstract

Conventional objective image assessment metrics, such as mean squared error and peak signal-to-noise ratio, which only calculates pixel-based differences between the original and the degraded images, are not in agreement with the human vision. In this paper, we present an improved objective full-reference image quality assessment method, called the multiscale and multidirectional visibility differences (MMVD) predictor. The proposed MMVD metric considers multiscale and multidirectional visibility differences in the domain of the discrete nonseparable shearlet transform, which emulates the multichannel structure of information processing of the human vision system. In the process of constructing the visual just noticeable difference threshold in the shearlet domain, the contrast sensitivity function and the visual masking effect which are important properties of the human visual perception are considered simultaneously to approximate the sensitivities of human visual responses. Both contrast masking and entropy masking are considered to tackle the visual masking issue. All subbands of the shearlet transform are evaluated, and perceptual errors of subbands are pooled together to yield the objective quality index of a distorted image. The extensive validation experiments are conducted on five public image databases, namely, TID2008, TID2013, CSIQ, IVC, and LIVE. The experimental results demonstrate the proposed method is well coherent with human perception and has better performance compared to the several state-of-the-art image quality metrics.