Study of Highly Efficient Algorithms for the Character Recognition System of Medicine Bottle Label

Abstract

This paper studied character recognition on medicine bottle label containing batch number, production date and expiry date. It mainly consisted of character location, character segmentation and character recognition. Firstly, it preprocessed the image of medicine bottle label for the following procedures; secondly, it obtained the general character region by contour tracing and realized the precise location by horizontal projection algorithm; then it located character adhesion region by vertical projection algorithm and segmented the merged characters by drop fall algorithm; finally, it adopted Karhunen-Loeve transform to extract character features from the subclasses of coarse classification and accurately recognized characters by the improved BP algorithm. Experiments show that the system can recognize the bottle label characters quickly and accurately, it is of great practical value.