Context Sensitive Mathematical Character Recognition

This paper describes methods to increase the accuracy of mathematical handwriting analysis by using context information. Our approach is based on the assumption that likely expression continuations can be derived from a database of mathematical expressions and then can be used to rank the candidates of isolated symbol recognition. We present how predicted continuations for an expressions are derived, how they are combined with the recognition candidates, and the effectiveness of the results. We first review the techniques we have used to build and represent a mathematical context database. We then describe different strategies for combining context information with results obtained from the recognition of individual characters. Finally we present a summary of a case study, using a fixed dataset of common mathematical expressions to test the accuracy of on-line analysis.