Research on Tamil script has mostly dealt with the identification of isolated letters. Different classifiers such as, hidden Markov models (HMM), neural networks and support vector machines (SVM) have been used along with a selection of features such as Fourier, wavelet,
angular and directional features. The challenge of Tamil word recognition that has been
addressed in the literature so far can be divided broadly into two approaches: (i) Recognition of individual strokes and then enating them using appropriate models to detect a word or a compound character. (ii) Grouping of strokes to form symbols or stroke groups, which may or
may not constitute a compound character and recognition of these symbols to form words. The
second method needs a framework for efficient segmentation of strokes into stroke groups and post processing techniques based on statistical language models. Use of word lists have also been investigated to enhance Tamil word recognition rates.
We portray experiments with preprocessed (x,y) coordinates, Fourier and derivative features for recognizing individual Tamil characters and the integration of segmentation, feature extraction, classification and post‐processing steps into a dynamic link library for use on a Tablet PC.