Bangla Handwritten Digit Recognition Using an Improved Deep Convolutional Neural Network Architecture

Deep Convolutional Neural Network has recently gained popularity because of its improved performance over the typical machine learning algorithms. However, it has been very rarely used on recognition of Bangla handwritten digit. This paper proposes a Deep Convolutional Neural Network (DCNN) based Bangla handwritten digits recognition scheme. The proposed method applies a seven layered D-CNN containing three convolution layers, three average pool layers and one fully connected layer for recognizing Bangla handwritten digits. Rigorous experimentation on a relatively large Bangla digit dataset namely, CMATERdb 3.1.1 provides considerable recognition accuracies.