This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: METHODS OF COMBINING SVMs APPLIED TO HANDWRITTEN DIGIT RECOGNITION Dejan Gorgevik 1 , Dusan Cakmakov 2 1 Faculty of Electrical Engineering Skopje, email@example.com 2 Faculty of Mechanical Engineering Skopje, firstname.lastname@example.org Abstract In this paper, the cooperation of four feature families for handwritten digit recognition using SVM (Support Vector Machine) classifiers is examined. We investigate the advantages and weaknesses of various cooperation schemes based on classifier decision fusion using statistical reasoning. Although most of presented cooperation schemes are variations and adaptations of existing ones, such an extensive number of investigated classifier decision fusion schemes have not been presented in the literature until now. The obtained results show that it is difficult to exceed the recognition rate of a single, well-tuned SVM classifier applied straightforwardly on all feature families by combining the individual classifier decisions. In our experiments only one of the cooperation schemes managed to exceed the recognition rate of a single SVM classifier. However, using classifier cooperation reduces the classifier complexity and need for training samples, decreases classifier training time and sometimes improves the classifier performance. Keywords structural, statistical, features, classifier, decision fusion 1. INTRODUCTION Combining features of different nature and the corresponding classifiers has been shown to be a promising approach in many pattern recognition applications. Data from more than one source that are processed separately can often be profitably re- combined to produce more concise, more complete and/or more accurate situation description. In this paper, we discuss classification systems for handwritten digit recognition using four different feature families and SVM classifiers . We start with a SVM classifier applied on all feature families as one set. Further, we used four SVM classifiers each working on different feature family from the same digit image. As the feature sets see the same digit image from different points of view, we examined the possibility of decision fusion using statistical cooperation schemes. An extensive number of statistical cooperation schemes were examined and corresponding recognition results are presented. Our aim was not to compete with the recognition rates of the other handwritten digit recognition systems e.g. , , but to compare the qualities of different feature families, corresponding SVM classifiers and their combination based on different classifier decision fusion schemes. The presented results show that it is difficult to achieve the recognition rate of a single SVM classifier applied on the feature set that includes all feature families by combining the individual SVM decisions. On the other hand, the cooperation of individual classifiers designed for separate feature families reduces classifier complexity and need for training samples, offering better opportunity to...
View Full Document
This note was uploaded on 02/18/2010 for the course ITK ETF113L07 taught by Professor Popovskiborislav during the Spring '10 term at Pacific.
- Spring '10