Volkova N. Models and methods for spectral-statistical texture segmentation in the systems of computer recognition of visual patterns

The dissertation is devoted to solving the scientific and practical problem of developing models and methods for segmentation of combined spectral-statistical textures in the systems of computer recognition of visual patterns to improve the quality of segmentation of texture images in intelligent systems of medical and technical diagnostics. In the research, the existing texture segmentation methods were analyzed. A mathematical model of the combined spectral-statistical texture was developed. A method for identifying spectral-statistical textures was developed and investigated. A vector-difference segmentation method for combined spectral-statistical textures was developed and investigated. A texture identification method for spectral-statistical textures based on multifractal features and a spectral texture feature. The contour segmentation method with use of improved wavelets by transforming the graph of a power function was researched and implemented. A texture identification method for an ordered texture based on multifractal features and the response of a generalized comb filter was executed and implemented into the systems of computer recognition of visual patterns. A software module for spectral-statistical textures segmentation has been developed. This segmentation module has been tested in the systems of computer recognition of visual patterns of intelligent systems of medical and technical diagnostics.