Selecting most efficient Arabic OCR features extraction methods using Key Performance Indicators

Many Arabic Optical Characters Recognition (AOCR) systems have reached advanced stage but the accuracy and performance of these systems are still unsatisfying in comparison with Latin OCR systems. This fact is due to many reasons such as: the decorative details of computer fonts, and the large numbers of characters shapes that change according to the location of letter in the word. This leads to a difficult problem, which is finding the most efficient feature extraction methods for recognition. This paper is enclosed a new methodology for selecting most efficient feature extraction methods. This methodology uses a segmentation method and passes the results to some feature extraction methods which are used in some previously built OCR systems, then some proposed Key Performance Indicators (KPI's) are used to evaluate the accuracy and time consumption of these methods. The best feature extraction methods are used to extract feature values and pass them to a recognition engine in order to build a new OCR which obtains a higher performance than the original ones.