Repository logo
 

Shoe Print Identification From Images With Convolutional Neural Network

Files

Thesis(3.66 MB)

Date

2019

Authors

Supervisor

Narayanan, Ajit

Item type

Thesis

Degree name

Master of Computer and Information Sciences

Journal Title

Journal ISSN

Volume Title

Publisher

Auckland University of Technology

Abstract

Shoe print identification/classification technology is a great significance in crime scene investigation. Traditional shoe print identification and classification technology relied on the experience of investigators. More recently, machine learning technology has brought new research direction and impetus to shoe print identification technology. This thesis presents a method for classifying shoe print images based on convolutional neural networks (CNN). The main contributions of this thesis are listed as follows. (1) The research includes surveys and summarises existing classification methods of shoe print and proposes: the implementation of a CNN in the classification of shoe prints. (2) The traditional machine learning method artificial neural network (ANN) is used as the classification accuracy benchmark for CNN. (3) The traditional machine learning method support vector machine (SVM) provides a second classification accuracy benchmark for CNN. (4) This research uses CNN to establish an image classification model, optimising the neural network to have higher classification accuracy. The results show that CNN has an outstanding performance in binary classification: the accuracy of classification reaches 99.91%; the sensitivity of CNN reaches 100%; and the specificity reaches 97.05%. These results surpass the other approaches (ANN and SVM). (5) The research successfully visualized the CNN model, which included features extracted from different layers, the kernel visualization, and the kernel heat map.

Description

Keywords

Pattern Identification, Shoe print identification, Convolutional neural network, Neural network Visualisation

Source

DOI

Publisher's version

Rights statement

Permanent link

https://hdl.handle.net/10292/12680

Collections

Masters Theses