Comparison of Genetic Algorithm and Sequential Search Methods for Classifier Subset Selection
ICDAR '03 Proceedings of the Seventh International Conference on Document Analysis and Recognition - Volume 2
Hi-index | 0.00 |
The combination of multiple recognizers shows various performances according to the selection of recognizers and the combination method of them. The selection of recognizers on how to select them or how many to select them still remains one of important research issues. In this paper, information-theoretic criteria for selecting the multiple numeral recognizers are evaluated together with the simple selection criteria in terms of recognition rates by the handwritten numerals from Concordia University and UCI, on the supposition that the number of selected recognizers is fixed in advance. The selection criteria are applied to the pool of numeral recognizers and examine the possible recognizer sets from the pool, and then select one of the recognizer sets as a multiple recognizer system candidate.