Engineering and Computingng

Computers traditionally find it very hard to identify objects in images it is presented with, to solve this problem much research into machine learning has been produced since computers were invented. The goal of this paper is to introduce the idea of neural networks and investigate their effectiveness in recognizing objects such as cars in a set of images. 

Tests were performed on a number of neural networks to obtain the best classifier, and were then compared to classifiers produced by KNN and Classification Trees.

Tests showed that a neural network was 92\% accurate at identifying cars in a test set of images. This is a acceptable rate of recognition for multiple applications and was better than the result of the Classification Tree which obtained only 81% accuracy. The network was however outperformed by the KNN classifier which obtained 98% accuracy. Regardless of this the neural network showed promise in computer vision, and has the capability to scale much better than the other classifiers tested here.

Recognising images using computers is a traditionally hard problem in computing, and one that becomes particularly difficult when these images are from the real world due to the large variations in them. This paper investigates the problem of recognising digits and characters in natural images using a deep neural network approach. The experiments explore the utilisation of a recently introduced dropout method which reduces overfitting. A number of different configuration networks are trained. It is found that the majority of networks give better accuracy when trained using the dropout method. This indicates that dropout is an effective method to improve training of deep neural networks on the application of recognising natural images of digits and characters.

Object recognition is a complex problem in the field of computer vision and artificial intelligence with many different solutions including the use of Neural Networks and Deep Learning. This thesis investigates the use of Deep learning for object recognition and improvements to the deep learning methodology using methods such as dropout to reduce overfitting. Improvements to the dropout method are made using selective neuron dropping instead of random neuron dropping used in traditional uniform dropout to improve results up to 1.17% on the CHARS74K dataset using neuron output variance to select neurons for dropping. This thesis also explores the use of Deep Convolutional Neural Networks to create a rotation invariant neural network by using parallel convolutional layers with tied weights with filters presented at different rotations, combining results using a winner takes all pooling method to produce rotation invariance. Results of this method showed great improvement over benchmarks on rotated test data showing a 52.32% increase in accuracy on the MNIST dataset and a 36.44% accuracy increase on the CHARS74K dataset. This work was furthered by attempting to introduce a scaling method into the filter rotation to reduce data loss and blurring by the rotation method, however it was found that results using this method worsened the accuracy of the network compared to benchmarks not using the method, showing a decrease in accuracy of 0.17% on the MNIST dataset and a decrease of 3.68% on the CHARS74K dataset.

I dedicate my Thesis to my family who have supported me throughout my studies. Firstly to my Grandmother Jean Barrow who passed away from cancer before I started University, but always encouraged me to do more and to do my best while not giving up on things I find difficult. Secondly I also dedicate my thesis to my remaining Grandparents Dudley Barrow, Ron Goldfinch, and Margaret Goldfinch, who where always available for a phone call if work was getting tough. Thirdly and finally, I also dedicate my thesis to my parents, Chris Barrow, and Susan Barrow, as well as my aunt, Mandy Barrow, and my great aunt, Ruth Barrow, who have all supported me in undertaking my undergraduate degree and my postgraduate studies. I would also like to acknowledge the rest of my supervisory team, Dr Chrisina Jayne and Dr Vasile Palade, who also provided me with support during my project. I would also like to thank my fellow postgraduate researchers, Luke Hicks, Ariel Ruiz-Garcia, and Ibrahim Almakky, for their support and helping in keeping me sane throughout my postgraduate project, as well as being there to help bounce around new ideas.