| As we know, Human vision system (HVS) is the vision analysis terminal, so subjective image quality assessment algorithm is most reasonable. But subjective methods is very time consuming and needs many professional persons to take part in, resulting in it has limitation in actual application. Although prior objective image quality assessment methods such as PSNR, which is based pixel-wise, is very simply for calculating, the assessment result is not very precise.Recently, an image quality assessment algorithm based on structural similarity (MSSIM) and improved version (MGSSIM) are proposed. They has been verified to be more correlation with HVS than PSNR through experimental result.However, they also have been found they are not universal in evaluating slight noise distorted image and blur image quality.This paper proposes two image quality assessment algorithms based on wavelet transform and HVS property: MWSSIM and MRWSSIM. First, we apply one level 2-D discrete wavelet transform to original image and distorted image and obtain four bands: LL, LH, HL, and HH. And then, we compare luminance information difference in LL band. At the same time, we compare contrast information difference and edge structure information difference in edge information map constructed by other three bands. Finally, we construct a normalized image assessment value (MWSSIM). We also propose the multi-resolutions version based on MWSSIM (MRWSSIM). We apply three levels 2-D discrete wavelet transform to original image and distorted image, and then, we compute the MWSSIM value at every level. Finally, we construct a normalized image assessment value (MRWSSIM) through giving corresponding weighs to the three MWSSIM values. Through non-linear regression analysis, the new algorithms can be verified more correlate with HVS than MSSIM and MGSSIM when assessing white noise image and blur image, at the same time, MRWSSIM can be verified more correlate with HVS than PSNR,MSSIM, MGSSIM when assessing blur image... |
PDF Full Text Request