Combining SVM Classifiers for Handwritten Digit Recognition

Abstract

In this paper, we investigate the advantages and weaknesses 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 obtained results show that it is difficult to exceed the recognition rate of a single classifier applied straightforwardly on both feature families as one set. The rule based cooperation schemes enable an easy and efficient implementation of various rejection criteria. On the other hand, the statistical cooperation schemes provide higher recognition rates and offer possibility for fine-tuning of the recognition versus the reliability tradeoff.