Improving persian digit recognition by combining data augmentation and AlexNet
Ebrahim Farahbakhsh, Ehsan Kozegar, Mohsen Soryani · 2017
The present paper, has used deep learning for Persian handwritten digit recognition by which valuable discriminative features are extracted from Persian digits using a deep convolutional neural network. Afterwards, these features are fed to a linear support vector machine for classification. Hoda dataset, which is the largest dataset for Persian handwritten digit classification, was used for validation of the proposed method. In this paper, first the related works were investigated which used Hoda and then the proposed deep network was introduced. Because of the high number of parameters that should be trained in deep neural networks, a high number of training images should be used for training. To increase the number of training images, augmented images have been prepared by rotating the original images with 15, 30, 45, -15, -30 and -45 degrees. Based on the experiments, the proposed method outperformed other competing methods in terms of accuracy measure.