Feature Extractor Based Deep Method To Enhance Online Arabic Handwritten Recognition System

To enhance Arabic handwritten recognition (AHR) performance, a combination between online and offline features is investigated. In this paper we exploit handcrafted features based on beta-elliptic model and automatic features using deep classifier called Convolutional Deep Belief Network (CDBN). The experiments are conducted on two different Arabic databases: LMCA and ADAB databases which including respectively isolated characters and Tunisian names towns handwritten by several different writers. The advantage of the both databases was the offline images had built at the same time as the online trajectory. The test results show a significant improvement in recognition rate.