On the Segmentation of Multiple Touched Cursive Characters: A Heuristic Approach

Abstract

Heuristics are based on the experiences and solves problems approximately that cannot be solved exactly. In handwritten documents recognition, the most difficult phase is touched character segmentation as incorrectly segmented characters cannot be recognized correctly. Accordingly, this paper presents a heuristic approach for multiple touched cursive characters. Initially, a possible segmentation zone is detected using peak to valley function. Next, greedy best search algorithm is implemented in the possible segmentation zone for touched characters segmentation. Experimental results on a test set extracted from the IAM benchmark database exhibit high segmentation accuracy up to 91.63%. Moreover, proposed approach is very fast and can handle multiple cursive touching characters. © Springer-Verlag Berlin Heidelberg 2010.