Arabic handwriting recognition: Between handcrafted methods and deep learning techniques

Recently, the area of pattern recognition has attracted the attention of many researchers in various domain and applications such as biometric, classification problems, and object recognition in general. In the last two decades, handwriting recognition is considered as one of the most active topics in this research area. The researchers focus their efforts in order to recognize many language handwriting. Among the language that still a challenged task for researcher is the recognition of Arabic handwriting because of several inherent characteristics of Arabic script including cursiveness and the existence of dots and diacritics. etc. Since deep learning algorithms have become the main core of most proposed solutions, the main aim of this paper is to evaluate the performance of some handcrafted feature extraction methods against CNNs based extraction features on well representing Arabic handwriting. The experimental results have been done on the public AHDB benchmark database.