Sofina O. Methods of textured images filtration in tasks of objects recognition and classification

Research objects are processes of the textured images filtration and classification, foreign objects determination. A research purpose is an efficiency increasing of textured images filtration using models which including dynamic properties of image signal. For the decision of the putted purpose the methods of computational modeling and statistical analysis using harmonic decomposition and spectral analysis were used. Scientific novelty are: the method of textured images filtration, which allows to create a simple form signal by definition and exception of texture eigen fluctuations; autoregression model with auto-actuation factor which in contrast to known, has heightened sensibility to the signals of texture foreign objects; two-dimensional autoregression model with symmetrical parameters vector, which including periodicity properties of the textured image signal and permits a possibility to filtrate a signal with the phase fault; new class of the adapted to the image signal function basis, which allows to perform filtration in spectral domain and to decrease operation amount in 4 and more times; new filtration method of the quasiperiodical textures on the basis of the approximation by principal harmonic components. The practical value consists in creation of the software for filtration, recognition and classification of textured images. Implementation degree - results of the thesis are implemented to the software on such companies as "Advanced Coherent Technologіes" (La Jolla, Calіfornіa, USA), "InnoVinn", "Spilna sprava" and also to the educational process of the automation and informational-measuring technology department of the VNTU. Application fields are image recognition systems, environment monitoring, noncontact inspection of the quasihomogeneous materials, protection signaling and video supervising.