Automatic Recognition of Handwritten Arabic Numerals under Constrained Training Conditions

This work presents a technique for the automatic recognition of off-line handwritten Arabic (Indian) numerals under constraint training condition. Handwritten digit recognitio...on has been an active research area for more than a decade. There are various techniques used for digit recognition such as Neural Networks, Support Vector Machines, Random Forests and Hidden Markov Models. Regardless of the technique used, an integral task has to be done, which is training the recognizer through handwritten training data. However, data collection is a very demanding and time consuming task. To the contrary of the current research work in the literature, where handwritten digits are used to train the recognizer, our work involves using only machine-generated digits for training. By doing so, we will investigate the impact of the amount of training data on the recognition performance. Furthermore, we will examine the possibility of artificial data injection. Our recognizer will be tested using handwritten digits from the CENPARMI Arabic (Indian) digit database for bank checks. The results will be reported and possible future enhancements to the technique will be mentioned.