A model describing the decomposition of imagery in the human retina is developed based on the retina's cellular structure. Two types of retinal cells, horizontals and amacrines, perform spatial averaging across the retina to form a low-pass image channel. This low spatial frequency information is fed back to the retina's receptor cells to form a difference channel of high-passed spatial frequencies. Such a model is suggested by electro-physiological as wll as psychophysical evidence. Analysis of the model characterizes the low-pass channel as a constrast channel and the difference channel as an edge detection channel. Application of the model to image quality assessment suggests a two factor approach involving metrics in the model's eye domain.