D. Yu. Stepanov

There are a lot of problem domains of pattern recognition, they are bioinformatics, data mining, document classification, remote sensing, astronomy, biometrical identification. One can meet biometrical identification systems in bank, trade, insurance, transportation and other organizations. The importance of biometrical identification process is explained by increasing number of terrorist attacks, economical and legal crimes, level of which can be deceased by identification of suspect persons. Biometrical systems are developed last 50 years so much due to high level of computing and low price of it. It allows us to manage huge data arrays in a short time. There are a great number of statistical, wavelets, structural, neural and other methods of pattern recognition, however, each of them is restricted by kind of input data. In general, person identification can be done by recognition of face, finger print, signature, palm, speech wave, eye etc. Pattern recognition begins from XVIII century and strongly links with such Russian and foreign scientists as: Aizerman M., Bra-vermann E., Vapnik V., Chervoninkis A., Mazurov V., Zhuravlev Y., Rudakov K., Vorontsov K. and Pentland A., Turk M., Kirby М., Sirovich L., Wiskott L., Blanz V., Cootes T., Liu С., Zhao W., Phillips P., Jain A. Classification in a space of multilayer representations of object is one of most efficient pattern recognition method, which doesn-t depend on kind of input data. The method handles with grayscale images. For one part of biometrical sources such as finger prints, signatures, speech waves this limitation is not important, for others, such as faces, eyes is, due to all available information about object is not used. In the paper classifier creation in a space of multilayer representations is explored. In this case training means template selection and radius of covering spheres estimation. All these can be executed by developed tree structured covering method. TSC-classifier is built by found multilevel networks of templates and their spheres. In general, decision rule of such classifier can be nearest neighbor, voting etc. We carried out some experiments to recognize color face images and calculate recognition error rate. The paper consists 7 chapters. Chapter 1 describes statement of problem. Chapter 2 has a description of populatad color face database. In chapter 3 one can find the object of interest selection procedure. In chapter 4 the multilayer representation method of objects and distinct measure are described. Tree structured covering procedure at training stage is in chapter 5. Chapter 6 has a description of distinct measure parameters estimation. Last chapter has results of face recognition by TSC and SVM classifiers.