Determining salient feature points of 3-D curved surfaces from range images

A new method for extracting salient feature points from range images of model surfaces is proposed. The feature points are characterized by their saliency for distinguishing the model object they belong to from the others. The saliency is measured by local shape similarity and the similarity coefficients of surface points are calculated by parameterizing surface curvatures into the local orientation coordinate systems. An algorithm for performing the Hough transformation weighted by the similarity coefficients is developed to optimally determine the saliency coefficients of the feature points. It is shown that the method is especially useful for dealing with the occlusion problem.<>