Date of Submission
Spring 5-4-2021
Degree Type
Thesis
Degree Name
Master of Science in Computer Science (MSCS)
Department
Computer Science
Committee Chair/First Advisor
Dr. Jing (Selena) He
Track
Others
Face Computation with Deep Learning
Chair
Dr. Jing (Selena) He
Committee Member
Dr. Meng Han
Committee Member
Dr. Yong Shi
Abstract
Despite significant advances in the field of face analysis over last decade, the current studies are still limited to specific face computation tasks using deep learning approaches. In this paper, we propose an end-to-end hierarchical deep learning structure, called Multi-Features Convolutional Neural Networks (MFCNN), which can comprehensively implement face analysis including age, gender, race and emotion. Moreover, we take the advantages of the mutual support among different facial features from individual tasks to improve the performance of our model. We also contribute one all-labeling dataset called Multiple Facial Features Computation (MFFC) based on Apparent-age-V2 dataset. Firstly, we train four different VGG-based classifiers to analyze facial features independently like age, gender, etc. using MFFC. Then, we systematically introduced cross-task verification approaches to demonstrate features extracted from different pre-trained models have mutual support to each other. After that, multi-task features extracted from pre-trained models are integrated to train MFCNN. Furthermore, feature fusion strategies also have been implemented to enhance our framework from accuracy and time complexity. Finally, the experimental results demonstrate that MFCNN outperforms state-of-the-art methods of face analysis by 10.3% in average and the best result improves up to 21.3% margin for emotion estimation.