Date of Publication :20th March 2024

Abstract:The handwritten alphabet recognition system is currently important in any language. In Myanmar, there is less research on the Shan handwritten alphabet recognition system because of the requirement of a standard dataset. This paper proposes a novel Shan handwritten alphabet recognition system using a fine-tuned VGG-16 model. The aim of this paper is to provide an effective and efficient method for classifying Shan handwritten alphabets. This system uses a self-constructed dataset that contains 19 alphabet classes and a total of 19,000 images that were collected by 1000 Shan native participants. This dataset was divided into 80% 10% 10% for training, validation, and testing. In this system, image augmentation, fine-tuning, and L2 regularization are applied for classifying unseen images and reducing the risk of overfitting and underfitting problems. On the fine-tuned VGG-16 model, the different hyper parameters are changed in order to achieve higher accuracy. After selecting the optimal parameter values, the experimental results show that the rate of accuracy is 98.03%. Our proposed model outperforms the VGG16 without using fine-tuning for this dataset. Seventeen of the nineteen alphabets can be correctly for real world image.

Reference :