Sabelnikov P. Hardware-algorithmic means of identification of differences between reference and partially distorted contours of objects in images

In the dissertation work new scientific and applied results are obtained that allow to expand the functional capabilities and increase the speed of computer vision systems by developing hardware-algorithmic tools for comparing standard and partially distorted contours of image objects. Developments will find application: in industrial systems of quality control of products; in diagnostic automation systems in medicine to identify and evaluate changes based on the results of computer and magnetic resonance imaging, X-ray examination; in special purpose systems, to identify, compare and track objects on television and thermal images. The main requirements for computer vision systems, the current trends in the development of hardware and software are considered, the existing methods and algorithms of comparison the contours of objects in images were analyzed. Conclusions are made regarding the directions of further research related to the organization of the computational process for solving the problems posed in the dissertation. As a method of comparing standard and partially distorted contours of objects in images, the method of direct comparison is adopted. The way to improve the speed of calculations is the use of processors with a set of vector operations or multiprocessors, which can be implemented with modern microelectronic technologies. Improved methods and algorithms for vectorization, the encoding of contours and their layering by levels, calculation of the geometric moments of objects and contours, the piecewise-linear approximation of contours using vector operations are proposed. They allow to accelerate the process of computing the primary characteristics of objects in proportion to the number of parallel data processed on a vector processors. For the first time, methods and devices for calculating integral vectors of partial sums and extreme values of a sequence of numbers with the use of vector operations are proposed, they provide the possibility to accelerate the solution of problems considered in this paper, as well as other similar problems. It is advisable to implement the proposed technical solutions in modern processors to expand the set of vector operations. The method is improved and algorithms for comparing contours based on the search for invariant parameters of probable identical segments of the contours of the analyzed object and the standard are proposed. In accordance with the characteristics of the segments parameters for geometric contour matching, detection and evaluation of their differences are calculated. All algorithms are oriented on the organization of computational process using vector operations, which reduces the time of contour comparison by 1-2 orders of magnitude. The obtained results made it possible to develop a video device for special purposes with the following basic functionality: image filtering, highlighting moving targets on a stationary background, recognizing targets by their silhouettes, tracking targets, detecting and comparing objects on television and thermal images etc. For the first time, a method is proposed (its correctness is proved) and a storage device for parallel multi-format access to fragments of two-dimensional data arrays. The storage device, in contrast to analogues, allows in parallel in one or two cycles to read or write permitted by the mask the fragment data on an arbitrary address with the same type of ordering at the input and output.