FOR HANDWRITTEN DIGIT RECOGNITION TROUGH PARTITIONING OF THE FEATURE SET.pdfVIP

[esta nacionalna konferencija so me|unarodno u~estvo ETAI2003 Sixth National Conference With International Participation ETAI2003 Ohrid, Republika MAKEDONIJA ? Ohrid, Republic of MACEDONIA 17–20. IX 2003 I4–2 COOPERATION OF SUPPORT VECTOR MACHINES FOR HANDWRITTEN DIGIT RECOGNITION TROUGH PARTITIONING OF THE FEATURE SET Dejan Gorgevik1, Dusan Cakmakov2 1 University “Sv. Kiril i Metodij”, Faculty of Electrical Eng., Department of Computer Science and In- formation Technology, Karpos II bb, POBox 574, 1000 Skopje, Macedonia, dejan@etf.ukim.edu.mk 2 University “Sv. Kiril i Metodij”, Faculty of Mechanical Eng., Department of Mathematics and Com- puter Science, Karpos II bb, POBox 464, 1000 Skopje, Macedonia, dusan@mf.ukim.edu.mk Abstract – In this paper, various cooperation schemes of SVM (Support Vector Machine) classifiers applied on two feature sets for handwritten digit recognition are examined. We start with a feature set composed of structural and statistical features and corres- ponding SVM classifier applied on the comp- lete feature set. Later, we investigate the vari- ous partitions of the feature set as well as the advantages and weaknesses of various decisi- on fusion schemes applied on SVM classifiers designed for partitioned feature sets. The ob- tained results show that it is difficult to exce- ed the recognition rate of a single SVM classi- fier applied straightforwardly on the comple- te feature set. Additionally, we show that the partitioning of the feature set according to feature nature (structural and statistical fea- tures) is not always the best way for designing classifier cooperation schemes. These results impose need of special feature selection pro- cedures for optimal partitioning of the featu- re set for classifier cooperation schemes. Index terms – classification, committee, featu- res, rejection, reliability 1. INTRODUCTION The classical paradigm for character recognition is concentrated around two