Smelyakova A. Models and methods of irregular object boundary approximation for vision systems

The thesis is devoted to development of models and methods for polygonal and elliptic approximation and normalization of irregular object boundaries where the constraints and precision criteria reflect typical peculiarities of application problems of contour approximation that arise in vision systems. A generalized contour approximation problem for various precision criteria and two basic classes of lines (polygons and ellipses) is stated as the problem of conditional optimization. For the broken lines, the methods of polygonal approximation in a non-simply connected domain are developed which minimize the number of vertices of approximant provided that the auxiliary vertices either can, or cannot be used. For the second class of lines the model and method of elliptic approximation are developed which allow us to improve the adequacy of approximation and normalization on a basis of joint consideration of both the bound, and the radially-oriented elements of the analyzed object. The theoretical estimates of time complexity and the results of computer simulation confirm a computational efficacy of the proposed methods.