Structural boundary feature extraction for printed character recognition

Abstract

In this paper, an algorithm is developed for printed character recognition based on boundary analysis. New boundary smoothing schemes have been proposed, which can reduce noise significantly. The extracted boundary features are invariant to character size and are convenient for recognition. The whole algorithm is based on the chain code of each boundary and no floating operation is involved. The algorithm is tested with sample pages of a phone directory. Experimental results show that this algorithm is highly reliable and very fast.