Symbol Recognition in Handwritten Mathemati- Cal Formulas

In this paper an efficient on-line recognition system for handwritten mathematical formulas is proposed. After formula preprocessing a symbol hypotheses net is generated by extracting different features for stroke unity. Each element of the symbol hypotheses net represents a possible symbol. The data of each symbol hypotheses are preprocessed and an image is calculated. Features for symbol classification are extracted by moving a horizontal and vertical window along the image. The symbol recognition is based on a first-order, left-to-right Hidden Markov Model. The final classification of the handwritten input is done by calculating the best fitting path through the symbol hypotheses net under regard of the stroke group probabilities and the probabilities obtained by the HMM symbol recognizer.