The Analysis of Stereopsis

Our eyes capture different 2-D images of the objects around us and our brain uses these images to recover a description of the 3-D structure of the environ­ ment. Stereopsis is the process responsible for this reconstruction of the depth dimension in our visual world. Since Wheatstone invented the stereoscope in 1 838, the processes underlying primate stereo vision have been intensively studied, first with psychophysical techniques, and more recently in terms of the underlying physiological mechanisms in the visual cortex. During the last few years, the computational aspects of the problem of stereo vision have received increased attention. As with so many other visual tasks that humans perform easily and effortlessly, the development of automatic systems of stereoscopic vision, which would yield immediate and important applications, has proven surprisingly difficult. It is now clear that the problem of stereopsis is not only a problem in the area of psychophysics and physiology, but also a complex problem in information processing. Because of its knowledge-free, low-level character, the solution of the stereopsis problem may thus uncover some fundamental principles that apply equally well to artificial and natural vision systems. This somewhat new perspective is reflected in the organization of this paper. We first review the computational problems posed by stereopsis and describe briefly the main theories and models that have been proposed in the context of human and computer vision. The structure of these algorithms and their