Kinoshenko D. Models of image retrieval and recognition from databases

The thesis is devoted to developing of methods and models of image segmentation results processing, analysis and indexing as quotient sets in order to increase efficiency and effectiveness of image retrieval and recognition systems from databases with query by example. The model of nested quotient set comparison is introduced, grounded and studied what allows fulfilling analysis of image segmentation results for adjustment of their refining or quantization level. Integral measures of quotient sets nesting are introduced on base of what the method of regions of interest searching which represent target object is developed. The method of indexation based on block-diagonal form of distance matrix which ensures guaranteed number of matches on the retrieval stage is provided and studied. Region-based image retrieval system which takes into account spatial layout of the scene as results of segmentation with queries by the whole domain or by regions of interest is developed and introduced. Comparative analysis has shown higher performance of developed software compared to analogs.