1D Wavelet Transform of Projection Profiles for Isolated Handwritten Malayalam Character Recognition

This paper presents a novel method based on wavelet transform for offline recognition of isolated handwritten Malayalam characters. Thirty three characters from Malayalam, one of the South Indian languages, are chosen for the study. The scanned character images are first cropped and resized to fit into 32 times 64 window. This is followed by binarization and thinning. The preprocessed character images are then modeled with projection profile. One dimensional wavelet transformation is applied on the projection profile. The feature vector is formed from the smooth components of the transform coefficients. A Multi Level Perceptron network is used for classification. The results obtained for relatively large number of characters are encouraging and motivated us to investigate with different classifiers as our further study.