A study for handwritten Devanagari word recognition

Devanagari script is being used in various languages, in South Asian subcontinent, such as Sanskrit, Rajasthan, Marathi and Nepali and it is also the script of Hindi, the mother tongue of majority of Indians. Recognition of handwritten words of Devanagari script is an important area of research. A person who knows the script of a language can easily read the hand-printed words pertaining to that script on the basis of his/her mental dictionary. In this research work, a practical study on isolated hand-printed words is conducted. Some issues involved with the recognition and segmentation of hand-printed Devanagari words are also discussed. There are three approaches for hand-printed word recognition i.e. segmentation based, holistic and hybrid. In this research work, a segmentation based approach is studied. To conduct experimentation, a database of more than 3500 hand-printed Devanagari words collected from more than 200 writers is developed. The 80% words of database are properly written whereas 20% words contain touching characters. Separate single stage classifiers are used for training and testing the characters of upper and lower region. To recognize middle region characters a three stage classifiers is trained and tested. To train classifier we used more than 30,000 characters including half characters/ a part of characters. The classification methods used is multi layer perceptron (MLP).