Image Quality Assessment with Saliency Map in Nonsubsampled Contourlet Transform Domain

Abstract

Many researchers evaluate images by objective image quality assessments instead of subjective ones. Objective image quality assessment sets up mathematical model according to the human visual system, and it evaluates the image quality through the reference image and the distorted image. The Structural Similarity Index (SSIM) is one of the most classical methods in image quality assessment. However, SSIM has several inherent shortcomings. First, SSIM does not take spatial position, spatial frequency, or direction into account. Second, SSIM considers that different regions in an image have equal importance for overall image quality assessment. Third, it is unreasonable to use fixed parameters for various images. To overcome these shortcomings, we propose a new method of image quality assessment based on Nonsubsampled Contourlet Transform (NSCT). Firstly, NSCT is performed to decompose the image into a low-pass map and high-pass ones. Then, low-pass and high-pass maps are respectively assessed with different strategies. In addition, saliency map is added to describe the importance of different regions in an image. Last, we proposed an approach to calculate the adaptive parameters for various images. Experimental comparisons among five public benchmark databases demonstrate that the proposed method is better than other competing methods.