Mashtalir S. Models and methods of temporal video processing in information retrieval

In the thesis the scientific and technical problem of videostreams temporal processing is solved, that provides the structuring of date for informational search with query "ad exemplum" in big videocollections in images or features spaces. On account of very large information amount, elimination region constructions are introduced to delete information explicitly irrelevant to the query. To speedup information retrieval process, information search is divided into on line phase and off line one which makes it possible to vigorous data process before execution of a query. Methods of video sequence temporal segmentation on the base of induced multivariate time series analysis are proposed for video structuring, the segmentation allows to partition video into a set of segments with large degree of frames similarity. For this, construction of video frame feature vectors is also analyzed. To facilitate the video indexing in segments method of video key frame extraction was developed and modified, which can further simplify procedures for information retrieval "ad exemplum". Advanced intelligent video compression techniques for raster data have been proposed. The theoretical and experimental results are implemented as applied and research software systems which are used for information retrieval "ad exemplum" in large video archives