Recognizing Digits from Natural Images and handwritten Digits using Deep Convolutional Neural Networks

Abstract

Recognizing digits from Natural Images and Handwritten digits are one of the famous problems in Computer Vision Applications.Many Machine Learning Techniques have been employed to solve the problem of recognizing Digits from Natural Images and Handwritten Digits.The most three famous NN approaches are deep neural network (DNN), deep belief network (DBN) and convolutional neural network (CNN). This Paper focuses on Convolutional Neural Networks (CNN), also known as Deep Convolutional Neural Network (DCNN) that operates directly on image pixels, also the three NN approaches are compared and evaluated in terms of many factors such as accuracy and performance. Recognition accuracy rate and performance, however, is not the only criterion in the evaluation process, but there are interesting criteria such as execution time. In addition, DistBelief [15] implementation of deep neural network is employed to train large & distributed neural networks on high quality images. At the end, we find that the performance of this approach increases with the depth of the convolutional network. With the best performance occurring in the deepest architecture with eleven hidden layers. We evaluate this approach on the publically available SVHN dataset and achieve over 96% accuracy in recognizing complete street numbers. This paper also presents blocky artifact as an augmentation technique to increase the accuracy of DCNN for handwritten digit recognition i.e 0-9 and conducts experiments on MNIST dataset only. DCNNs with the proposed augmentation technique give better results than those without such augmentation.