Recognizing Noisy Patterns via Iterative Optimization and Matching of Their Descriptions

This paper presents an approach to the recognition of noisy patterns. We assume that learned concept descriptions contain noisy components. These descriptions can be optlmizcd to gain bener performance through the elimination of some less signilicant noisy componenls. Parametric conlrol of such an oplimization. however, is very difficult and a parameter value has 10 be found Ihrough the training process. This value can change over concept variations and time. and in consequence, it can decrease the system recognition effectiveness. We found that the dassilkation decision can be based on the dynamic characteristics of the rccognition curve aC4uired ovcr increasing degree of optimization. In this approach. the optimization and mah;hing ~teps arc repeated iteratively. The dyn;Jrnic characteri~tics uf recognillon curves arc cOl1lpictetl and a p;Jllern of e;Jch recognition curve b l;Jbelcd. For a given set of test data. the system selects recognitIOn curves of uptrcntl p;JUerns {when the recognition rate increascs with the incrcase of thc optimizallon degree I. and the final d;Jssificaliun is made for the dass which h;JS the highest recognition rates alllon!:! sclcl:tetl uptrend rewgnllion curves. Thu!i. the method d;Jssifics d;Jt;J h;Jscd on a ,enes of IIl;JIl:hcs th;Jt determines the hchavior (If the recognJlion curve instead t,1 the Iradilional single match. The effL"Cllvcncss of the proposed methodology IS illustrated on Ihe cX;Jl11plc 01 '>JX:cl;Jlly prcp;Jred VCI;' nOIsy ICXlUre allnhul<!S.