Verifying The ‘Consistency’ Of Shading Patterns And 3-D Structures

The problem of interpreting images in terms of their shading and reflectance components has traditionally been addressed as an early vision task in a simple 2D Mondrian domain. Recently it has been appreciated that in a 3D world, such conventional approaches are inadequate; more sophisticated strategies are required. One such strategy has been proposed by Sinha [22, 25], who has addressed the problem as a mid-level vision task rather than as a purely low-level one. Sinha suggested that a key computation that needs to be performed for interpreting images acquired in a 3D domain is the verification of the consistency of image shading patterns and the likely 3D structure of the scene. This is the problem we have addressed in the present paper. Considerations of robustness and generality have prompted us to discard available quantitative techniques in favor of a qualitative one. The two prime attributes of our technique are its use of qualitative comparisons of gray-levels instead of their precise absolute measurements and also its doing away with the need of an exact pre-specification of the surface reflectance function. We show that this idea lends itself naturally to a linear-programming solution technique and that results obtained with some sample images are in conformity with human perception.