A New Method for Unconstrained Handwritten Numerals Recognition
一种非限制性的手写数字的识别方法

In this paper, a model-based finite state attribute automaton is proposed from syntatic and semantic recognition point view for the extraction of features. For each typical deformed model, a new finite state attribute grammar and its corresponding automaton is designed. This method combines Top-Down and Bottom-Up control strategy, introduces knowledge in low levels, this cuts down the amount of operations and reduces the uncertainty. An unconstrained handwritten numerals recognition system is realized based on this approach. By testing a set of 1, 100 samples, the average recognition ratte is 95.2%, rejection rate is 4.6%, substitution rate is 0.2%.