Kozina Y. Image Recognition Methods in Wavelet Transformation Space under Quality Control of Templates.

The thesis is devoted to development and research of images recognition methods in artificial intelligence systems based on wavelet transformation for increase of credibility of templates quality in noise conditions. For the first time method of low-sized objects localization on image was developed. It is embodied on basis of hyperbolic wavelet transformation (HWT). Further development got classification method, that unlike the existing method is embodied on basis of subgradient iteration search of extremums in HWT space. For the first time training method for multilayer perceprton was developed. It is based on subgradient iterative search of extremums in HWT space. Further development got method of building of separating function based on integer-valued optimization while searching for values of elastic variables.Developed image recognition methods are used in artificial intelligence system for templates quality control. Proved rise of credibility of templates quality control in noise conditions in range of quotient signal/noise 10..20 (in power) up to 1,2 times in comparison with existing complexes and also support of control credibility while using of the developed system in range of quotient signal/noise 7..10 (in power) on basis of classifiers: first – 0,95; second – 0,90; third– 0,80 .