Naderan E. Recognition of online handwritten mathematical expressions using fuzzy neural network

The dissertation is devoted to the development of efficient methods of handwritten mathematical expressions recognition. The study proposes an approach to features extractions technique for handwritten symbols, which is based on simplifying a piecewise linear curve with the Ramer-Douglas-Peucker algorithm. The dissertation presents an efficient symbol recognition method based on handwritten symbols classification using NEFCLASS a neuro-fuzzy approach as a means for classification. Also a hybrid machine learning technique proposed by using genetic algorithm during the initial machine learning stage and the conjugate gradient method during the extended machine learning stage in order to increase the symbols recognition accuracy and decrease machine training time. A structural analysis approach proposed to determine spatial relations among symbols of mathematical expressions. It consists of the stage of position determining, reconstruction and symbols grouping. An efficient rule-based dynamic Heuristic approach presented based on the knowledge of writing sequence, semantic values and spatial relations between the symbols to increase accuracy of structural analysis.