On-line Malayalam handwritten character recognition using wavelet transform and SFAM

The aim of this paper is to present a new practically implementable method, based on wavelet transform, for the on-line recognition of Malayalam handwritten characters. Wavelet transform of the input coordinates and the angle features is used to form the feature vector in compressed form. The classifier used is Simplified Fuzzy ARTMAP network, which requires comparatively very less time for training also supports incremental learning making it ideal for practical implementation. We have analyzed the recognition accuracy of the system using `dbl' and `haar' wavelets. The system, when tested on 1279 character samples gives a maximum accuracy of 97.81%. The training time required is less than five minutes and the recognition time is about 0.5sec/symbol.