Modified Bootstrap Approach with State Number Optimization for Hidden Markov Model Estimation in Small-Size Printed Arabic Text Line Recognition

摘要

In printed Arabic text line recognition, hidden Markov model brings a facility from no pre-segmentation but leaves a hard work to model estimation. Although bootstrap training can supply good initialization, the bad image quality of small-size samples may make it difficult to find accurate model boundary. This paper introduces a modified bootstrap approach with state number optimization to improve the accuracy of model estimation. Experiments on small-size samples from the APTI dataset show that the modified bootstrap approach in this paper can decrease 13.3% error rate of word recognition and 14% error rate of character recognition than the original one.