Multiple Outputs Techniques Evaluation for Arabic Character Recognition

Different OCR systems generate multiple outputs for same input image. The difference between several outputs of OCR is used to select the best features from them. Multiple outputs techniques show a significant improvement in OCR accuracy for images that have low scanning resolution and noises. Arabic language often produces high OCR error rate compared to the Latin character-based language. The reason is that, the unique characteristics of Arabic language. For this reason, this paper evaluates the performance of main multiple outputs techniques. The goal of the evaluation process is to test and analysis these techniques on Arabic dataset, and chooses the best for this language. The testing dataset used in the evaluation process is large in order to make the reliability and the validity of the testing are higher. The result of experiments shows that the best performance obtained among comparative techniques is the one that used multiple scanning for same input image to produce several outputs of OCR.