Stolbovyi M. Clustering-based video summarization technology for information retrieval

The subject of the research is the methods of temporal segmentation and video sequences clustering for video summarizing and information retrieval with queries ad exemplum. The object of the research is the video streams processing and analyzing in information technology for the visual information search. Methods of research: in the development and research of methods and models of fuzzy segmentation and video streams clustering, the main fundamentals of the pattern recognition and image processing theory, computing intelligence methods, time series analysis, as well as elements of mathematical statistics during conducting and analyzing the results of experimental research are used. The purpose of the thesis is to develop tools for intelligent information technology for video retrieval based on clustering. Results and their novelty: The methods and models for changes detection in the vector and matrix nonstationary noisy signals properties based on the adaptive models ensembles with own identification algorithms with different depth of memory, on-line procedures of fuzzy clustering, providing time segmentation for both slow and fast changes in the video content. On the basis of hybridization of hierarchical agglomeration and fuzzy clustering based on centers of weight approaches for dynamic video summarization, the multidimensional time series with different lengths clustering method with an unknown number of classes and the possibility of their mutual overlapping is proposed. For video summarization, the video data sequences clustering is proposed, which is based on the use of modified iterative dynamic time warping and sequential clustering of reduced time series based on matrix fuzzy clustering based on harmonic-means. The results of theoretical and experimental research are implemented in the form of ecology monitoring application, research and education in Kharkiv National University of Radio Electronics.