Text-independent writer recognition using modified texture and microstructure features

In this paper, we propose effective techniques for automatic writer recognition that operate at different analysis levels. At the texture level, we present a novel 2-D wavelet-based method to replace the traditional 2-D Gabor filter. In our analysis at the microstructure level, we use a floating grid around every edge pixel to record orientation and curvature information to characterize writer individuality. The final unique distance between two handwritten samples is computed as the average of the distances between the individual features. The experiments on a database containing 203 writers of address images exhibit that proposed methods achieve promising results.