An Effective and Practical Classifier Fusion Strategy for Improving Handwritten Character Recognition

In this paper, we propose a classifier fusion strategy which trains MQDF (modified quadratic discriminant functions) classifiers using cascade structure and combines classifiers on the measurement level to improve handwritten character recognition performance. The generalized confidence is introduced to compute recognition score, and the maximum rule based fusion is applied. The proposed fusion strategy is practical and effective. Its performance is evaluated by handwritten Chinese character recognition experiments on different databases. Experimental results show that the proposed algorithm achieves at least 10% reduction on classification error, and even higher 24% classification error reduction on bad quality samples.