Multi‐Lingual Handwritten Character Recognition Using Deep Learning
Giriraj Parihar, R. Rajalakshmi, J Bhuvana · 2021
Handwritten character recognition (HCR) is the most challenging task because it is a repeated work which is done by humans, and because of that, error can occur. We can solve this problem with image classification and deep learning algorithms, but in any language, because of similar type of character and different type of writing styles, it is the most challenging task. There are many single language character recognition models available, but there is no model available for multiple language character recognition. When number of classes increases, model performance will decrease. There are many techniques available for HCR. For this problem, Convolutional Neural Network (CNN) is used widely because it is a state-of-the-art model for image classification. In this, a new architecture is proposed to recognize any character, independent of language. The proposed architecture was fine-tuned by changing the hyper-parameters and choosing the appropriate activation function. The proposed system was evaluated on three different publicly available datasets, that includes English, Hindi, Bengali characters, and also on mathematical symbols. We have achieved an accuracy of 99.54%, 98.60%, 87.10%, and 95.77, respectively, for uni-lingual models. For the proposed multi-lingual model, we have achieved an accuracy of 95.73%. By comparative analysis with the existing methods, it is shown that there is a significant improvement in the performance of the proposed multi-lingual model.