Handwritten Character and Digit Recognition with Deep Convolutional Neural Networks: A Comparative Study

Handwritten character or digit recognition involves automatically classifying handwritten characters or digits from images. Previous studies focused on specific datasets and did not thoroughly compare different CNN architectures. This paper addresses these limitations by presenting a comparative study of six popular CNN architectures (VGG16, Xception, ResNet152V2, InceptionResNetV2, MobileNetV2, and DenseNet201) on three diverse datasets: English Handwritten Characters, Handwritten Digits, and MNIST. The experimental results demonstrate that the InceptionResNetV2 model with data augmentation achieves the highest accuracy across all datasets, with accuracies of 93.26%, 97.16%, and 99.71% on the English Handwritten Characters, Handwritten Digits, and MNIST datasets, respectively.