Random-noise Contamination Assessment of Images via Non-parametric Hypothesis Testing on Wavelet Coefficients

Images encountered in our daily lives typically contain random noise of some level. The human visual system (HVS) has the ability to see through such noises at low level. It is desirable to replicate this ability in an image quality assessment method. Extensive experiments found that well-established full-reference image quality assessment methods, such as mean square error (MSE), structure similarity index (SSIM) by Wang et al. [2004] and peak signalto-noise ratio (PSNR), may poorly mimic the human visual system in this regard. In this article we propose a new full-reference image quality assessment method, namely, wavelet-based non-parametric structure similarity (WNPSSIM) index, based on a non-parametric test of the hypothesis of identical images. The test is conducted on the wavelet domain to take advantage of the belief that wavelet transform provides an alternate representation of image data that is more in sync with the HVS (Field [1999]; Portilla et al. [2003]; Chandler and Hemami [2005]). Experimental comparisons demonstrate that WNPSSIM performs better than some well-known image quality assessment methods when the images involved are corrupted with random noises. AMS 2000 subject classifications: Primary 68U10, 97K80, 62H35; secondary 62G10.