Yegorova O. Relational clustering segmentation of gray-scale images

The dissertation work is devoted to developing intellectual seg-mentation methods for context image interpretation and evaluating of obtained results via introducing a new metric on partitions. Spatial layout relationships formalization which allows considering all possible variants of regions positioning for effective partitions and coverings processing is given and using tolerance matrixes for region merging analysis is introduced. First a model for comparing images, using a new metric on finite partitions, a tool for analysis the partitions obtained on the earlier stages of image processing and for combining segmentations results on interpretation stage, is introduced, grounded and studied. Under that, from one side, a possibility for algorithms objective integral comparison arises, from the other side, introducing operations on partitions creates premises for obtaining the best in the meaning of given criterion partition, which allows reliable detection of objects or regions of interest. Experimental analysis which showed that the metric has number of advantages compared with traditional ones. Simulation of the developed method of image relational clustering segmentation has been carried out. Its advantages before known segmentation methods due to the analysis of regions spatial layout and their merging is shown.