Support Contours for Image Feature Extraction in Object Classification

Abstract

A method of support contours for classifying images of objects with an a priori undefined shape is presented. The method is based on a two-level approximation of contours using support points, which form unique polyhedrons for each class. The first level approximates the contour with a support polyhedron, identifying key points, while the second level divides the contour into segments, which are approximated by segmental polyhedrons. This enables automatic object classification by comparing their contours within the corresponding classes. Key advantages of the method are highlighted: invariance to affine transformations, reduction in the number of polyhedron vertices, and improved classification speed due to natural parallelization of computations. The method also addresses the shortcomings of existing approaches, such as dependency on the starting point and algorithmic complexity. The application of the method is demonstrated in pattern recognition tasks where object shape plays a critical role, such as in robotics, technical diagnostics, and medical diagnostics. The research results show that the proposed approach is effective for classifying objects with arbitrary shapes and can be used in intelligent image processing systems.