ICDAR '05 Proceedings of the Eighth International Conference on Document Analysis and Recognition
Segmentation-driven offline handwritten Chinese and Arabic script recognition
SACH'06 Proceedings of the 2006 conference on Arabic and Chinese handwriting recognition
| Hi-index | 0.00 |
Abstract: The statistical approach to offline handwritten character recognition has been used widely, in which classifier design is very important. For approximating the class conditional density more precisely, it can be represented by orthogonal Gaussian mixture model (OGMM). The parameters of OGMM are commonly estimated by expectation-maximization algorithm (EM), which converges to the maximum likelihood estimation (ML). Since ML cannot directly minimize the classification errors as the goal of classifier design, to solve this problem, the discriminative training method based on the minimum classification error criterion (MCE) is used to adjust the parameters of OGMM. In order to achieve a good generalization, the complexity of the classifier, namely the component number of OGMM, is determined following the structure risk minimization principle (SRM). Finally, the recognition performance is demonstrated by applying it to the handwritten numerals in the NIST database.