Recognition of Printed Farsi Characters Using Contour-Based Features

Recognition of Printed Farsi Characters Using Contour-Based Features

Reza Azmi, Ehsanollah Kabir, Kambiz Badie

Abstract

In this paper, two recognition algorithms for printed Farsi characters are proposed. In the first algorithm, the chain code of the contour is applied to a set of automata which serve as recognizers for different classes. The second algorithm used the Fuzzy Hough transform of the contour image for feature extraction, and a three-layer perception for classification. In the final stage, the number of dots and their position together with some structural features are taken into account to complete the recognition process. <br/>
Using a test set of 11500 characters in 10 different fonts, the recognition rates of 97.13% and 98.32% were achieved for the first and second algorithms, respectively.

Keywords

Farsi, Character recognition, Feature extraction, Contour, Fuzzy Hough transform, Neural network, Perception

References