A MULTISTAGE APPROACH TO THE DENSE ESTIMATION OF DISPARITY FROM STEREO SEM IMAGES

We are currently involved in an industrial project to recover depth in formation from stereo image pairs retrieved using a scanning electron microscope (SEM). Feature-based approaches to stereo provide accurate disparity estimations, however the quantity of estimates recovered is small (typically 1-2% of the image). If a continuous approximation to the surface is to be reconstructed, as requested by potential customer s, more data has to be recovered. Our approach involves using the disp arity estimates from a feature-based stereo algorithm to constrain a f unction fitting process. Assuming the image may be represented by an i terated facet model, the algorithm attempts to fit piecewise polynomia ls between the feature disparity estimates, which describe the mapping of grey-levels from the left to right image along epi-polars. The pro blems of illumination variation between the left and right images have been addressed using a modification to rank-order filtering which we call 'soft' ranking. Using a 'B-fitting' algorithm enables both the mo del order as well as the model parameters to be optimized. This is sho wn to improve the stability of the fitting process, when compared with a least-squares algorithm, by reducing the effective number of model parameters down to the minimum necessary in order to describe the data . The fitted functions are then used to calculate intermediate dispari ties, augmenting those recovered using stretch correlation. Finally a Laplace filter is used to close the surface. (C) 1998 Elsevier Science B.V. All rights reserved.