A Full-Reference Image Quality Assessment Algorithm Based on Haar Wavelet Transform

This paper proposes a new objective full-reference (FR) image quality assessment method with discrete Haar wavelet transform (HWSSIM). Here, both reference and test images are decomposed into four bands by 2-D Haar wavelet transform: LL, HL, LH and HH. The luminance and contrast information are extracted in the LL band, and the edge structure information is obtained with edge information map based on other three bands and the whole image quality is defined as MHWSSIM. We also develop our algorithm on the multi-resolutions case (MRHWSSIM). which is defined as the human visual system (HVS) contrast sensitivity based weighted mean of four MHWSSIM values evaluated at four Haar wavelet decomposition levels. Experiment results indicate that the proposed methods are better correlated with HVS when compared to PSNR, MSSIM and MGSSIM.