Recognition of handwritten characters is crucial for enhancing human-computer interaction. Due to the vagueness and unevenness of each person's writing styles and strokes, the task could be difficult. Additionally, the potential for a poor source image quality makes the work more difficult.
To interpret the scanned or uploaded images of handwritten texts. We have proposed a CNN architecture that uses keras as an interface for tensorflow library. The model has been validated for english as well as devanagari scripts. Dataset put-to-use for english is ‘EMNIST_ByClass’ [1] and for devanagari is ‘devanagari handwritten character dataset - DHCD’ [2]. The model achieved 87.20% accuracy for EMNIST_ByClass and 98.19% accuracy for DHCD dataset. Train Dataset: https://drive.google.com/file/d/1egHJ3E6ivL5355OVJypYmAGvYATQyYHL/view?usp=sharingTest Dataset: https://drive.google.com/file/d/1N7R-S5B9RMdUYa0sWiIgRMScrcy9lb_B/view?usp=sharing
Train Dataset: https://drive.google.com/file/d/1kGfJJPWAi6L7sJgFjBiV-qoTkPPDv5-J/view?usp=sharing
Test Dataset: https://drive.google.com/file/d/1AlnTmYPT_13pkAR7H8gfliMJkbw9gShk/view?usp=sharing
[1] G. Cohen, S. Afshar, J. Tapson and A. van Schaik, "EMNIST: Extending MNIST to handwritten letters," 2017 International Joint Conference on Neural Networks (IJCNN), 2017, pp. 2921-2926, doi: 10.1109/IJCNN.2017.7966217.
[2] Acharya, Shailesh et al. “Deep learning based large scale handwritten Devanagari character recognition.” 2015 9th International Conference on Software, Knowledge, Information Management and Applications (SKIMA) (2015): 1-6.