Hybrid Methods for Robust Irradiance Analysis and 3-D Shape Reconstruction from Images

The analysis of the differential structure of images is an interesting task in machine vision, among other reasons because it can provide relevant featural representation of images, suited for higher level information processing task like geometry reconstruction and object recognition. The importance of invariants of the field of isophotae on lambertian surfaces in shape perception by means of chiaroscuro is discussed in (Koenderink and Van Doom, 1980). In their approach to shape from shading, (Breton et al, 1992) represent the shading of the image by means of its shading flow field, i.e. by the first order differential structure of the image expressed as the isoluminance direction and gradient magnitude. The (Grossberg and Mingolla, 1985) model of low level visual processes uses input sensors approximately sensitive to brightness gradient, providing a map of segments oriented along constant brightness lines.