ROBUST SHAPE FROM SHADING

Existing shape from shading algorithms are not robust enough to work r eliably with real images. We present a new paradigm for shape from sha ding based upon a scale space surface representation. Using this repre sentation, the shape from shading problem becomes one of forming a sur face from Gaussian basis functions in such a way that the shading in t he original image is explained. The algorithm is extremely robust, pro ducing convincing results even when there is considerable noise presen t. In its current form, the algorithm requires the reflectance functio n to be approximately circularly symmetric. We show results for synthe tic images, both with and without added noise, and for real scanning e lectron microscope images.