Deep neural network for recognizing digit images

Abstract

Nowadays, we live in an information era, it is no doubt that the data expanding rate keep continuously expanding, therefore, data needed to be processed to become meaningful, manageable and even can find knowledge from it. Neural network is a kind of machine learning algorithm that worked as a human brain and are used to estimate some unknown based on training large amount of data sets. The human brain recognizes objects by structuring multiple layers of visual representations from objects. Inspired by the brain, deep neural network has excellent performance in many classification tasks including speech recognition, hand-written language processing and object recognition. The network can approximately simulate the human brain ability to train and study image patterns or features. Therefore, the project is aimed at developing a deep neural network in recognizing handwritten digit images. Digits are selected from MNIST database for training and testing the accuracy of the network. The network was designed and structured with layers including input layer, hidden layers and output layer. After iteratively training with large amount of digits, the network will confidently classify digit as different classes that represent digits 0 to 9.