Combining HMM-based two-pass classifiers for off-line word recognition

For off-line recognition of cursive handwritten word, the intersection between segmentation and recognition is complicated and makes the recognition problem still a challenging task. Hidden Markov models (HMMs) have the ability to perform segmentation and recognition in a single step. In this paper we present an HMM based unsymmetric two-pass modeling approach for recognizing cursive handwritten word. The two-pass recognition approach exploits the segmentation ability of the Viterbi algorithm and creates three different HMM sets and carries out two passes of recognition. A weighted voting approach is used to combine results of the two recognition passes. A high recognition rate was achieved for recognizing cursive handwritten words with a lexicon of 1120 words. An experiment on NIST sample hand print data of ten different writers was also carried out. The experimental results demonstrate that the two-pass approach can achieve better recognition performance and reduce the relative error rate significantly.