A novel approach based on fusion of three neural experts for handwriting recognition

This paper presents a novel approach based on fusion of three neural experts for handwriting recognition. The first expert provides a heuristic based binary segmentation of the handwritten word, and passes the best segmentation hypotheses to the second expert. The second expert is a neural character classifier for classifying each segment into a character representation. The outcomes are fed into the third expert, which is a neural word recognizer. The word recognizer is responsible for matching the given sequential characters to one of the words in the lexicon. The preliminary experiments were performed on CEDAR benchmark database. The performance of the proposed approach was measured using the segmentation accuracy, the character classification accuracy and the word recognition accuracy. The experimental results show an improvement in segmentation, character and word recognition accuracies compared to the published results.