Overlapping Character Recognition for Handwritten Text Using Discriminant Hidden Semi-Markov Model

The field of handwritten character recognition has always attracted a large number of researchers. The proposed methodology uses Discriminant Hidden Semi-Markov Model for tackling the problem of recognition of handwritten characters. Preprocessing on the input image such as denoising and adaptive thresholding is done for input conditioning, followed by segmentation for finding the area which contains text. The text image is then passed through the second stage of segmentation, which separates overlapping characters. Then, these segmented characters are digitized using feature extraction. For feature extraction, Discriminant Hidden Semi-Markov Model is used. For feature matching and character extraction, the proposed methodology uses KNN Classifier. The training feature library, consisting 180 samples of each character in the capital and small, processed using training algorithm of Discriminant HsMM. Paragraphs of 80–120 characters are processed in recognition module. The 86% average accuracy rate is achieved for a large set of characters.