Recognition of Unconstrained Handwritten Malayalam Characters Using Zero-crossing of Wavelet Coefficients

This work focuses on application of Wavelets in offline recognition of unconstrained isolated handwritten Malayalam characters. The data set consists of 30 samples of each 25 consonants (out of 36) in Malayalam (one of the South Indian Languages). All samples are 256 x 256 gray level images. No preprocessing (such as denoising and thinning) is performed. The images are converted to inverted binary images and wavelet transform is applied (using Db4 filter). For each image, count of zero-crossing in each of the ten subbands is found and is used as the feature for classification. From the analysis of the range of zero-crossing values in different subbands, the 25 characters could be classified into 11 sets. The result from this preliminary work is promising. Hence a detailed study on effect of preprocessing, use of different filters and application of Neuro-Fuzzy classifier is under investigation.