Nguen

pp. 198-205

One of the most important and challenging tasks of constructing the effective recognition systems is to select a working dictionary containing the most informative features. The successful solution of this problem provides both a reduced dimensionality of the feature vector and an improved efficiency of the recognition system as a whole. The article concerns the problem of reducing the feature space in pattern recognition algorithms using the object contours and proposes a method for its solution based on the search of the most essential features. It follows from the experimental results that the proposed method enables to abridge substantially the dimension of features dictionary and improve the efficiency of pattern recognition algorithm using the object contours on the basis of Fourier-descriptors.

This article describes development of an algorithm for recognizing assembly parts which can be used in computer vision systems of robotized manufacturing units. The problem of development has several distinctive features. First of all, assembly units have a fairly simple shape. That means their contours characterize their shape in a unique fashion. Secondly, all assembly units to be recognized are placed on an open conveyer belt and are oriented randomly. This problem formulation allows to use various recognition algorithms based on the contour analysis. This approach eliminates all inner points of images from the analysis; it leads to reduction in the volume of processing information at the expense of transition from two-variable function analysis to one-variable function analysis.