A human vision system model for objective picture quality measurements

It is often useful to know the perceptual impact of distortions introduced at critical points in the production, distribution and display of video. Direct measurement using human observers is possible for some applications, but this is a time consuming operation that must be performed under carefully controlled conditions. A faster and more easily standardized approach is to use a vision model that provides accurate estimates of the visibility of differences between original and distorted image sequences. A model that fulfils this need is the Sarnoff just-noticeable difference (JND) model. This model is based on known physiological and psychophysical principles of human visual discrimination performance, rather than on particular classes of image distortions, and is therefore more robust across the often unpredictable range of distortions that can occur in modern digital video. The operation and general structure of the JND model are described, and its performance in a range of video applications is discussed