Uighur Character Recognition Based on Adaptive Models

In this paper, we propose improve the Uygur character recognition performance using model adaptation. We introduce the MLLR (maximum likelihood linear regression), iterative MLLR and gradual MLLR, maximum a posteriori probability (MAP) and the algorithm combined with the MLLR and MAP, and we also estimate the recognition performance of those methods. According to a series of experiments on choosing parameters and data, taking advantages of the differences between variant fonts of Uighur character, we compare the effect of each method. These conclusions will provide basis for further research.