Gorokhovatskyi O. Projective methods of image normalization and recognition in condition of geometrical distortions

Research object - processes of perception, processing and analysis of images in artificial intelligence systems. Research target is the development of methods and models of image normalization and recognition in computer vision systems applying projection analysis in order to increase recognition effectiveness and estimation precision of visual objects parameters undergoing geometrical transformations. Research methods: analytical methods of pattern recognition and image processing theory, integral transforms theory, normalization theory, elements of group theory, imitational modeling and theory of moment invariants. Research apparatus: personal computer, photographical apparatus, videocamera, scanning device. Theoretical and practical results - projection methods for estimation image geometric distortion parameters and for invariant features construction are introduced; main theoretical results are realized as applications, that gives an opportunity to solve practical problems of image processing. Novelty - firstly method of geometrical distortion parameters and its combinations estimation based on projections is introduced, validated and studied for normalization, which assumes functional influence division of different types of distortions action due to Radon transform properties, that enables to determine normalization parameters using separate projections, provides processing time decreasing; firstly method of construction and comparing of image invariant features on the base of Radon transform and moment invariant features is suggested and validated, which carries out transition from 2D-space into the space of 1D vectors-projections with the help of projecting operation, that gives an opportunity to decrease recognition time without reducing of probability parameter; method of invariant features construction attributes as result of application of the projective analysis based on complex of integral transforms has received the further development, which, as against existing approaches, improves accuracy of suggested invariant features calculation and provides high probability recognition by consecutive using of Radon and Fourier transforms. Research results as algorithms of interactive processing signals at identification of underlying objects are introduced in the Center of radiophysical sounding the Earth named by A.I.Kalmykov of the National academy of sciences of National space agency of Ukraine that has allowed to simplify the analysis and interpretation of radar-tracking images. Suggested method of hand-written signatures verification based on invariant Radon transform features is introduced in Kuryaz corrective colony of management of the State department of Ukraine on questions of punishments execution in Kharkov area for the decision of a problem of automatic authors identification by their static signatures on documents that is confirmed with acts of introduction. Scientific theoretical and practice results of dissertational work can be used: in the organizations which dealt with problems of processing, researches, introductions and operation of i computer vision ntellectual systems, in the real-time systems of image and videodata processing; in systems of video observation and identification; in educational process by preparation of experts in the field of computer sciences, processing of images and an artificial intellect.