Intensity, visible-surface, and volumetric representations

摘要

One approach to studying human vision is to treat it as a computation that produces descriptions of the external world from retinal images, asking what can be said about this process based on an investigation of the information processing problems that it solves [11]. This paper examines the structure of the problem at a level of abstraction Marr and Poggio [16] call the computational theory. The type of information the vision process must make explicit at various stages is treated and used as a base for decomposing the process into subparts that can be studied independently. Examples are taken from ongoing research at M.I.T.