Vovk O. Statistical Methods of the Images Clusterization for Content-based Image Retrieval Systems

The purpose of work is development of fast-acted method of images regions selection of unlimited classes on the basis of statistical methods of objects clusterization, creation on its base of the content-based image retrieval system. Object of research are digital images. Subject of research are the statistical methods of objects clusterization for construction of the content-based image retrieval system. Basic methods of researches: theory of the digital image processing and methods of statistical cluster analysis. Theoretical and practical results: a new approach to the selection of images regions is developed, on the base of which the system of context search of images is built. Scientific novelty: the statistical hierarchical aglomerative method of images clusterization on the base of bit mask of correlation and ranks is developed, the condition of "comparability" of clusters is offered, at the search is suggested to take into account priorities of clusters descriptions. The results of work are used in the DonNTU educational process in courses the "Computer synthesis and image processing" and "Machine graphic arts and interactive systems", and also for modernization of the digital system of the "Drakar" safety (NPP "AMI", Donetsk). The field of the use: systems of images search.