Off-line Arabic Hand-Writing Recognition Using

Artificial Neural Network with Genetics Algorithm

Khalid Nahar

Computer Science Department, Yarmouk University, Jordan

Abstract: Artificial Neural Networks (ANN) were used in the recognition of the printed Arabic text with a high rate of success. In contrast, Arabic hand-writing recognition has many challenges, some were tackled in some research recently. In this paper we used ANN in recognizing Arabic hand-written characters with the Genetics Algorithm (GA). The GA was used to search for the best ANN structure. We consider Arabic off-line characters represented by a series of (x, y) coordinate. The dataset was gathered from a couple of volunteers, used the E-pen to write different Arabic letters. A Matrix Laboratory (Mat Lab) program was implemented to store the written characters and extracts their features. Features were determined based on the shape and number of segments that made up the characters. The recognition results were very promising when using ANN with the GA in comparison with other relevant approaches. On average more than 95% of accuracy was achieved when GA is used to adjust ANN structure in order to get the best recognition rate.

Keywords: ANN, GA, Feature vector, character recognition, arabic hand-written text, Hidden Markov Model (HMM).