VLLCVD Subjective Image Fidelity Criterion and Its Application to Evaluation of Visual Models

Abstract

The emerging JPEG2000 image coding standard includes the requirements for visually lossless images, however, “visually lossless” is not clearly defined. Many researchers claim that their encoded/decoded images are visually lossless, yet if we examine those images closely under a strict viewing condition, such as in a dark room at night at very short distance, differences between the original and the decoded images can be detected. For this reason, serious image users, like physician and radiologist, are reluctant to use so call visually lossless compressed images. In this paper, we propose a more strict and precise subjective image fidelity criterion based on two techniques (1) the same position swapping (SPS) technique between the decoded image and the original and (2) the measurement of Visually Lossless Critical Viewing Distance (VLLCVD). The VLLCVD based SFC provides two scores: (1) the ratio of testers who perceive the decoded image as absolutely visually lossless (at any viewing distance) and (2) the average critical viewing distance that the image is perceived as visually lossless. The proposed VLLCVD SFC can be used to examine many visual models proposed for image coding as well as invisible image watermarking.