Arabic Handwritten Alphanumeric Character Recognition Using Very Deep Neural Network

Abstract

Abstract:The traditional algorithms for recognizing handwritten alphanumeric characters are dependent on hand-designed features. In recent days, deep learning techniques have brought about new breakthrough technology for pattern recognition applications, especially for handwritten recognition. However, deeper networks are needed to deliver state-of-the-art results in this area. In this paper, inspired by the success of the very deep state-of-the-art VGGNet, we propose Alphanumeric VGG net for Arabic handwritten alphanumeric character recognition. Alphanumeric VGG net is constructed by thirteen convolutional layers, two max-pooling layers, and three fully-connected layers. The proposed model is fast and reliable, which improves the classification performance. Besides, this model has also reduced the overall complexity of VGGNet. We evaluated our approach on two benchmarking databases. We have achieved very promising results, with a validation accuracy of 99.66% for the ADBase database and 97.32% for the HACDB database.

Convolutional ensembles for Arabic Handwritten Character and Digit Recognition

“…Generally speaking there isn't a definitive answer as to which of these two methodologies is best for a given task, as this is highly dependent on the particularities of each dataset. Mudhsh & Almodfer (2017), for instance, have used 10-fold cross validation in their study of MADbase.…”

“…For the KCV, a 10-fold Cross-Validation was used to allow for direct comparison with the results of Mudhsh & Almodfer (2017), but it must be noted that dividing the original training set of 60,000 into 10 folds means each validation set has a size of 6,000. Since the size of the validation set can be adjusted by changing the value of K, and the test set size is fixed, a 6-fold validation was also performed (since this implies validation sets of size 10,000, the same as the provided test sets).…”

“…The method of data augmentation has been used previously in AHR (Mudhsh & Almodfer, 2017). In the present study, data augmentation was applied to the training sets of both MADbase and AHCD in some of the experiments.…”

“…Other parameters were also tested, but invariably led to a worse performance of the augmented classifiers. It is known that not all forms of augmentation are necessarily helpful for all tasks (Mudhsh & Almodfer, 2017), and the tree chosen yield the best results for these AHR architectures. The batch size of augmented images had a size of 128.…”

“…Loey, El-Sawy & EL-Bakry (2017) used Stacked Autoencoders on the MADbase dataset to obtain a classification accuracy of 98.5%. Mudhsh & Almodfer (2017) obtained a validation accuracy of up to 99.66% on the MADbase dataset by usage of dropout regularization and data augmentation, and an architecture inspired by the VGGNet Convolutional Neural Network (CNN) (Simonyan & Zisserman, 2014). Importantly, they mention in the text that this validation accuracy does not hold for the test set, without mentioning explicitly the test accuracy.…”