In this paper we propose a method for correcting character recognition errors at the higher level recognition step in an understanding system for mathematical formulae. The system consists of two-level recognition steps: the low level recognition including character recognition, and the higher level recognition including layout recognition. We use the layout information recognized in the latter step to correct the character recognition errors by using two sources of information. One is based on some keywords such as mathematical function names, and the other is based on a cost tree and co-occurrence probabilities between symbols. The efficacy of the proposed method is verified by some experimental results, and the character recognition rate increased from 80.2% to 89.2%.
展开▼
