Discrimination Based Gabor Pyramid Model for Image Fidelity Assessment

In this paper we describe the Image Fidelity Assessor (IFA), a model of the human visual system designed to evaluate perceived fidelity. In particular, we describe the structure of the model, present results from a contrast discrimination experiment that are used within the model, and demonstrate preliminary results of the model. The IFA is designed to be both physiologically and psychophysically plausible. It consists of a multichannel Gabor pyramid decomposition. The fidelity at each spatial location within each channel is evaluated based on psychometric functions. The psychometric functions were determined experimentally and describe the discrimination ability of the visual system as a function of spatial frequency, orientation, and adaptation level. Limited memory probability summation across the visual channels is used to obtain a measure of perceived image fidelity. Preliminary results of the IFA are consistent with human evaluation of image fidelity.