A novel image quality metrics was proposed. Firstly, images were decomposed into subbands with different scales and directions by hybrid wavelets and directional filter banks (HWD). Secondly, in order to offer a uniform visual sensitivity to different scales and directions subbands for human perception, each subband was masked with contrast sensitivity function (CSF). A rational sensitivity threshold was defined, and then the proportion of sensitivity coefficients in each subband was computed. Variation of visual perception coefficient between the reference image and distorted one was used to evaluate image quality. Experimental results illustrate that the proposed metrics agrees quite well with human performance. Thus, it can be used to describe the visual perception of image effectively.