Persian handwritten digit, character and word recognition using deep learning

In spite of various applications digit, letter and word recognition, only a few studies have dealt with Persian scripts. this paper, deep neural networks are utilized through different DenseNet Xception architectures, being further boosted by means data augmentation test time augmentation. Dividing the datasets to training, validation sets, utilizing k-fold cross-validation, comparison proposed method state-of-the-art alternatives is performed. Three datasets: HODA, Sadri Iranshahr used, which offer most comprehensive collections samples in terms handwriting styles forms each may take depending on its position within word. On HODA dataset, we achieve recognition rates 99.49% 98.10% for digits characters, 99.72%, 89.99% 98.82% digits, characters words from respectively, as well 98.99% outperforms performances achieved advanced alternative networks, namely ResNet50 VGG16. An additional contribution paper arises capability holistic image classification. This improves resulting speed versatility significantly, it does not require explicit character models, unlike earlier such hidden Markov models convolutional recursive networks. addition, computation times been compared better performance has observed.