Unsupervised selection of HMMs architectures for handwritten text/word recognition

Abstract

The selection of the classifier architecture is a very important step in the recognition process. This paper presents a new algorithm for the HMMs architectures optimization: Multi-Models Evolvement using PSO (MME-PSO). The proposed algorithm is applied to an Arabic handwriting recognition system. The recognizer is based on character Hidden Markov Models which can have different architectures. This algorithm is evaluated using the IfN/ENIT database. The performance of the resulting system is compared to the participating systems at the 2005 competition organized on Arabic handwriting recognition on the International Conference on Document Analysis and Recognition (ICDAR). The final system (IfN-REGIM-2) performs an absolute improvement of 6% of word recognition rate with about 81% comparing to the baseline system (ARAB-IfN). The proposed recognizer outperforms also most of the known state-of-the-art systems.