Recognition of hexagonally sampled printed Arabic characters

Abstract EN

Considerable progress in the recognition techniques for Latin and Chinese characters has been achieved.

By contrast, Arabic optical character recognition has started recently, with relatively slow advances.

In this paper, an Arabic character recognition system for hexagonally sampled data is presented.

Hexagonal sampling of images is found to be the optimal sampling scheme in terms of storage requirements, computational cost, connectivity, adjacency, and angular resolution.

The presented system is mainly composed of three stages: viz.

the preprocessing stage; the feature extractor; and the classification stage.

The preprocessing stage includes data acquisition, hexagonal sampling, and noise elimination via a statistical smoothing algorithm.

The Arabic character is then described by a set of features obtained from Fourier descriptors, boundary line encoding features, in addition to the presence of holes and upper or lower dots.

Finally, in the classification stage, recognition is achieved by comparing the features of the unknown character to the prestored prototypes generated in a training phase.

Experimental results indicate that the system is efficient in terms of correct recognition rates, speed, and space requirements.