| Author: | Xue, Kaiwen |
| Title: | Enhanced joint multi-task learning for age-invariant recognition |
| Advisors: | Lam, Kin Man (EIE) |
| Degree: | M.Sc. |
| Year: | 2022 |
| Subject: | Face perception Human face recognition (Computer science) Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Electronic and Information Engineering |
| Pages: | [29] pages : color illustrations |
| Language: | English |
| Abstract: | Age-invariant face recognition has aroused great research interest in recent years. This task is challenging since the facial ageing process is very complex. Therefore, extracting the identity-sensitive feature, which is insensitive to age information, remains a great challenge. In this project, we propose a multi-task learning method with decorrelation of identity and age features to tackle this problem. We introduce the Inception-Resnet v1 as the backbone network to generate the primary facial feature containing both the identity and age information. After that, the primary feature is factorized into the identity and age components by the use of Residual Factorization Mapping (RFM). The identity feature and the age feature are then fed into two discriminators, aiming to enhance the discriminability of the two features for their respective tasks. The CosFace loss function is adopted in the discriminators. The correlation between the identity and age-sensitive features, measured by cosine similarity, is reduced to generate effective features for AIFR. The cosine similarity and the two CosFace loss functions are optimized simultaneously in the training process, ensuring that the output feature is sensitive to identity but insensitive to age information. Experiments on the well-known MORPH Album2 data set demonstrated the effectiveness and superiority of our proposed model. |
| Rights: | All rights reserved |
| Access: | restricted access |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 6527.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 1.07 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12068