Query Word Image based Retrieval Scheme for Handwritten Tamil Documents

This paper brings out an autoassociative neural network (AANN) based information retrieval mechanism to locate handwritten documents from a literary collection in Tamil language corresponding to query word images. The strategy extends to create models for the chosen search word images, evolve a methodology to identify the search word and subsequently retrieve the relevant documents. AANN emphasises a training procedure through an appropriate combination of units in the layers of the network to arrive at a suitable model for each word in the vocabulary. The training phase orients to segment the digitized text documents into lines and words, extract profile and moment based features from the words and articulate an index of words. The features computed based on the intensity values of the pixels cater to accrue the nuances of the strokes in the characters. The experimental results obtained for an index of words elaborate the astuteness of the scheme and its retrieval accuracy.