On the performance of wavelets for handwritten numerals recognition

This paper validates a recognition system using bi-dimensional wavelet transforms as the feature extractor and investigates the relevance of each subband image in the recognition process. An experiment to verify the efficiency of the wavelets was performed omitting the feature extraction step. Results show that information about the relevant image features are evenly distributed in all subband images of wavelet coefficients and that wavelets are promising feature extractors. Numerals from the NIST database were used for evaluation of the systems tested.