Pattern Recognition with Localized Gabor Wavelet Grids

This paper introduces Gabor wavelet grids localized within the region of importance (ROI) for a robust pattern recognition system that is capable of recognizing various image classes. The representation of the patterns is based on Gabor wavelet transforms. The features are extracted as a vector of values using an optimally chosen multiple frequency, multiple orientation 2D symmetrical Gabor wavelet grid that spans the region of importance (ROI). This technique of feature extraction is biologically motivated. The purpose of this study is to demonstrate that localized Gabor wavelet grids with a MLP classifier can be effectively used for recognition of various types of patterns. Recognition of close to 100% is demonstrated for some real world patterns like faces, currency notes, postage stamps and typed characters.