Combining SVM Classifiers for Handwritten Digit Recognition
Dejan Gorgevik, Dusan Cakmakov
Abstract: In this paper, we investigate the advantages and weak-nesses of various decision fusion schemes using statistical and rule-based reasoning. The cooperation schemes are applied on two SVM (Support Vector Machine) classifiers performing classification task on two feature families referenced as structural and statistical features. The ob-tained results show that it is difficult to exceed the recog-nition rate of a single classifier applied straightforwardly on both feature families as one set. The rule based coop-eration schemes enable an easy and efficient implementa-tion of various rejection criteria. On the other hand, the statistical cooperation schemes provide higher recogni-tion rates and offer possibility for fine-tuning of the rec-ognition versus the reliability tradeoff.