| Abstract: | This paper presents a novel method for multi classifier dynamic combination based on minimum cost criterion.Differentfrom common combination methods,the proposed dynamic combinationselectsthe mostsuitable group ofclassifiersaccordingtothe Performance Predication Feature (PPF) extractedfrom theinputsample.PPFs arethefeatures ofsample thathave greatinfluence onthe performance ofclassifiers being studied.The decisionis made based on the criterion thatthe selected group ofclassifiers should minimize the cost caused by recognition errorand recognition time .Systematic methods for making this kind of combination is proposed and a practical example ofapplication is given.Because the adjustment ofcostfunction willresultin differenttrade off between recognitionrate and recognition speed,itis very convenientto satisfy different needs.The application in on line Chinese characterrecognitiontechnologyshowsthatthis kind ofcombination method hasthe merits ofhighflexibility and practicality,anditisindeed able toimprove the system performance . |
