Mathematical Formula Recognition Using Virtual Link Network

In this paper, we propose a new method of recognizing mathematical formulae. The method is robust against the recognition errors of characters and the variation of the printing styles of the documents. The outline is as follows: we first construct a network with vertices representing the characters (symbols), linked each other by several edges with labels and costs representing the possible relations of the pair of characters. The network has multiple edges with different labels and costs representing the ambiguity of the decision of the relation of character pairs. Then, we output the spanning tree of the network with minimum cost which corresponds to the recognition result of the structure of the mathematical formula, using not only the local costs initially attached to the network but the costs reflecting global structure of the formula. The advantage of this method is that local errors of the recognition are recovered automatically by the total cost of the recognition tree.