Conventional vs Neuro-Conventional Segmentation Techniques for Handwriting Recognition: A Comparison

Abstract

The success of Artificial Neural Networks (ANNs) has been prominent in many real-world applications including handwriting recognition. This paper compares two techniques for the task of segmenting touching and cursive handwriting. The first technique uses a conventional heuristic algorithm to detect prospective segmentation points in handwritten words. For each segmentation point a character matrix is extracted and fed into a trained ANN to verify whether an appropriate character has been located. The second technique also uses a conventional algorithm for the initial segmentation process, however two ANNs are used for the entire segmentation and recognition procedures. The first ANN verifies whether accurate segmentation points have been found by the algorithm and the second classifies the segmented characters. The C programming language, the SP2 supercomputer and a SUN workstation were used for the experiments. The techniques have been tested on real-world handwriting scanned from various staff at Griffith University, Gold Coast. Some preliminary experimental results are presented in this paper.