Fast computation of orthogonal spectral descriptors for the recognition of patterns with external and internal contours

Abstract

The processing of images with well-defined edges can be carried out with a significant reduction of the data to be handled, when the boundaries of the shapes are appropriately coded. The authors use this method of data reduction to characterize multi-edge patterns. Recognition descriptors are obtained through orthogonal spectral expansions. Descriptor computation is optimized by line integrals on the pattern contours. Contours are internal as well as external. The numerical coefficients corresponding to selected descriptors for recognition are computed from the code contours. Experimental results show the efficiency of this method for the recognition of industrial parts.