Peredrii O. Normalization of projective image transformations using method of perspective influence compensation

Object of research - processes of perception, transformation and interpretation of images in the systems of computer vision. Purpose of research - development and analysis of the models and methods of compensation of the influence of perspective distortion during the normalization of the projective transformations of images in the systems of computer vision. Methods of research - analytical methods of normalization and image processing for the generalization of the criterion of presence of the projective transformations; elements of the functional analysis and group theory for the synthesis of the criterion of determination of the direction of perspective; statistical analysis, simulation, based on which systematization of the data was conducted for practical and theoretical conclusions. Equipment - personal computer. The result of theoretical and practical research is a solution of the problem of normalization of images under the influence of projective distortion obtained by the compensation of the perspective for reducing computational costs when implementing in the applied systems. The complexity of the proposed methods below traditional complexity that allows to use effectively developed methods to solve practical problems. The scientific novelty - for the first time criterion for assessing the direction of one-parameter perspective distortion on the basis of image projections was proposed, analysis of values of which makes it possible to determine the direction of the distortions and a sign of the normalization parameter; for the first time method of compensation of the perspective influence based on the application of the criterion of estimation of the direction of the perspective distortion and the correlation analysis, which significantly reduces time of the normalization, was presented; was further developed correlation method of normalization of the tracking type based on the computation of the functionals, which take into account redistribution of the image brightness due to the action of projective transformations, which, in contrast to the method which was used before, allows to perform normalization of the perspective along an arbitrary line; has been further developed model for consecutive normalization of the projective group of transformations, which is based on the application of the reflection of structural or integral type, invariant to the affine group, and on the normalizer for the perspective. The results of the thesis were used in the development of the software used in the surveillance system for the location of the number plates of cars in the company "Elite Promservis" in Kharkiv; for automatic analysis of micro-particles for the purpose of the quality control of drugs in the company "Farma Start" in Kiev; to determine mass fraction of the granulated slag in the cement in order to determine its quality in the company "Seprocem" in Kharkiv. The results of the thesis are used in the course material for the laboratory sessions and course project developed for the course named "Computer vision", given at the Computer science department in Kharkiv National University of Radio Electronics. Scientific theoretical and practical results of the thesis can be used: in organizations working with development, research, implementation and operation of the systems of computer vision, systems of image processing and processing of video data in real time; in video surveillance systems and identification; in the educational process during preparation of specialists in such areas as mathematical modeling and methods of image processing and computer science.