Multi-View Face Detection

Tan, Ai Sha (2021) Multi-View Face Detection. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Tan Aisha.pdf Restricted to Registered users only Download (2MB)

Abstract

Although face detection and face recognition technology is getting more common, it is still not yet stabilized as there are many conditions that may affect the accuracy of it, ranging from lighting issues, irregular face structures, or even views from different angles of the face. This project aims to look further into the issue of face detection when viewed from various angles, and in turn improve the accuracy of face detection under this condition. Various methods such as the Viola-Jones algorithm, part mixture model, facial landmark localization with MTCNN, and convolutional neural network (CNN) have been studied to observe which method is able to produce the highest accuracy. For this project, a deep learning model using CNN architecture will be used to develop a multi-view face detection system, as it has been proven in previous research papers that it is able to achieve the highest accuracy at about 90%, and is feasible to be developed within the time and resource constraint. However, due to the same constraints, the target achievement for this project will be for at least 75% accuracy.