A Hierarchical Method for Kannada-MNIST Classification Based on Convolutional Neural Networks

Handwritten digit classification considers one of the crucial subjects in machine vision due to its numerous practical usages in many recognition systems. In this regard, Kannada-MNIST was introduced as a challenging dataset. On the other hand, deep neural networks, especially convolutional neural networks, give us an encouraging promise to solve such a problem. In this paper, as a result, we propose a new hierarchically combination method with the help of two CNN models designed from scratch. The results of this novel approach on the Kannada-MNIST dataset indicate its excellent performance because the accuracy on the training, validation, and test sets are 99.86%, 99.66%, and 99.80%, respectively. Fortunately, this proposed method has been able to overcome all the state-of-the-art solutions with the best performance on this dataset.