Gorokhovatskyi V. Structural-hierarchical methods of analysis and image recognition under the conditions of the influence of spatial distortions

Research object - processes of recognition, transformation and interpretation of the videodata in computer vision intellectual systems. Research target is the synthesis of theoretical basic of construction and an experimental research of structural - hierarchical methods of image analysis as means for the decision of a video-objects recognition problem at influence of geometrical transforms and the structural and spatial distortions. Methods of research - formation of invariant characteristic attributes based on positions of the registration theory, processing and the analysis of images, the theory of probability, the functional analysis and statistical data processing; the synthesis of structural descriptions comparison methods base on the theory of pattern recognition, clusterization, theories of sets; the parameters optimization of structural - hierarchical methods is based on the theory of optimum statistical decision-making; imitating modelling is applied for the verification of theoretical positions and an efficiency of the suggested methods estimation using real-images databases. Theoretical and practical results - a general approach to the decision of image videoobjects recognition problem in conditions of their change under action of geometrical transformations and spatial disturbances is developed on the basis of structural - hierarchical methods which using local - spatial similarity by descriptions comparison provide efficiency and an opportunity to make a decision on a videoobjects class in conditions of incomplete and false data presence; hierarchical methods of videoobject recognition in conditions of a complex background using integrated description and an estimation of videoobjects descriptions similarity for system of attributes of the stratified representation which as a whole raises reliability of recognition concerning action of additive noise are developed; the principle of the developed hierarchical correlation methods of recognition optimization on the basis of analytically set statistical criteria is realized and decision-making methods on a class of object based on clusterization, compression and hierarchy descriptions comparisons using multilevel attributes that reduces time of recognition with reliability preservation are suggested; the developed methods are the methodological basis of scientifically proved decision of the broad audience of applied problems of computer vision: localizations and recognition of spatial - lingering objects in videodata databases, decodings of space shootings results, quality equipment control by the analysis of magnetic characteristics. Scientific novelty - the principle of structural - hierarchical methods construction for videoobjects recognition for the first time is suggested and theoretically proved by an estimation of local - spatial similarity of their structural descriptions as vectors or sets of characteristic attributes and their combinations that enables to carry out recognition procedure preserving high reliability in real conditions of an influence environment, background and local distortions; the model of local - spatial similarity is suggested and generalized for the first time at videoobject recognition by comparison of their structural descriptions using voting which takes into account a way of reception, the measure type for comparison and structure of description elements that provides efficiency and an opportunity of decision-making on a class of videoobjects at presence of the incomplete and false data; the hierarchical method of videoobject recognition in conditions of complex background is synthesized for the first time by construction of their integrated description which provides the one-dimensional integral-differential coordinated image processing and group attributes definition for subsets of elements description and provides recognition reliability in conditions of spatial disturbances; the method of construction and an estimation of descriptions videoobjects similarity for system of attributes of the stratified descriptions representation which includes definition and the analysis of component attributes by application of attributive and spatial relations for the first time is suggested, that as a whole raises reliability of recognition concerning action of additive noise; the principle of parameters optimization of the developed hierarchical correlation recognition methods which consists in optimization of submitted analytically statistical criteria in view of fragments of images similarity measure, and also distributions of additive and local disturbances that has enabled to confirm advantages and universality of structural - hierarchical methods application in sense of controllability and flexibility for achievement of optimum values of criterion is offered and proved for the first time; the method of unequivocal conformity set of descriptions videoobjects elements filtration which provides transition to space of values of geometrical transformations and realization of clusterization procedure which provides increase of noise immunity due to elimination of erroneous conformity of characteristic attributes is developed for the first time; the models of structural videoobjects descriptions compression using spatial characteristics and constructions of the list of similar descriptors has received the further development which differ from traditional methods that coordinates of characteristic attributes take into account and determine similarity of descriptions based on a subset of the most significant conformity of elements that reduces recognition time with sufficient reliability preservation; the method of hierarchy videoobjects descriptions construction and comparison using spatial multilevel attributes is improved which comparing with existing techniques is based on a rule of estimation concurrence classes voting with representation of characteristic attributes in two different affine bases that provides reduction of recognition videoobjects time. Degree of implementation - the research results are used in the Center of radiophysical Earth sounding if National Agency of Science and National space agency of Ukraine (the act from 07.07.2009); in Northeast branch of the State research center "Nature" (the act from 10.08.2009); in the Kharkov National University of an Agriculture (the act from 05.09.2009); in research firm "Special scientific development" (the act from 01.07.2009); theoretical results are used in educational process of the Kharkiv National University of Radio Electronics within the framework of disciplines "Algorithms and the software of images synthesis", "Modelling of economic, ecological and social processes", "Methods and systems of image recognition", "Computer vision", "Mathematical and information supply in systems of complex signals processing" and at performance of attestation of the bachelor, specialist and master degree works (the act from 12.09.2009). The scope of use - the organizations which dealt with problems of intellectual systems and development methods during visual information processing; in information technologies application branches of processing and video information synthesis for medicine, biology, ecology, transport; in educational process by experts preparation in branches of intellectual data processing systems.