BDNet: Bengali handwritten numeral digit recognition based on densely connected convolutional neural networks

Images of handwritten digits are different from natural images as the orientation of a digit, as well as similarity of features of different digits, make confusion. On the other hand, deep convolutional neural networks are achieving huge success in computer vision problems, especially in image classification. BDNet, the title of this paper, is a densely connected deep convolutional neural network model used to classify (recognize) Bengali handwritten numeral digits. The design of BDNet is inspired by the state-of-the-art algorithm DenseNet. BDNet is trained end-to-end using ISI Bengali handwritten numeral dataset with 10-fold cross-validation. The model has achieved a test accuracy of 99.775%(baseline was 99.40\%) on the test dataset of ISI Bengali handwritten numerals. The trained model has also given 98.80\% accuracy on our own created dataset (which is not used during training and validation). So, the BDNet model gives a 62.5% error reduction compared to previous state-of-the-art models. Codes, trained model and our own dataset are available at: https://github.com/Sufianlab/BDNet.