DIRECT RECOVERY OF DEPTH - MAP - I: DIFFERENTIAL METHODS.

Three methods are described for recovering the depth-map (i. e. distance of visible surfaces from a camera) of a scene from images formed by a convex lens. The recovery is based on observing the change in the scene's image due to a small known change in one of the three intrinsic camera parameters: (i) distance between the lens and the image detector plane, (ii) focal length of the lens, and (iii) diameter of the lens aperture. No assumptions are made about the scene being analyzed. The recovery process is parallel, involving simple local computations. In comparison with some shape-recovery processes such as stereo vision and motion analysis, the methods are direct in the sense that three-dimensional scene geometry is recovered directly from intensity images of the scene; spatial properties of the intensity distribution are not computed as an intermediate step, and further the correspondence problem does not arise. These methods are relevant to both machine vision and human vision.