Classification of Handwritten Symbols using Deep Convolutional Neural Network
Handwritten character, digit, or symbol classification is one of the most studied research areas in image processing and machine learning. Automatic recognition of handwritten symbols from captured images in adverse practical scenarios is significant. This work also deals with a task where 25 different handwritten symbols are detected. We propose a well-known deep neural network, ResNet-18, to classify these symbols with the help of proper data preprocessing and extensive data augmentation. The proposed methodology reaches 99.71 % classification accuracy in the test set. This work also shows that data augmentation techniques improve the classification accuracy by 9.67%.