Author: Hong, Hang
Title: Low resolution face recognition
Advisors: Lam, Kin Man (EIE)
Degree: M.Sc.
Year: 2021
Subject: Human face recognition (Computer science)
Resolution (Optics)
Hong Kong Polytechnic University -- Dissertations
Department: Department of Electronic and Information Engineering
Pages: vi, 64 pages : color illustrations
Language: English
Abstract: In recent years, face recognition techniques have made significant progress and have been successfully applied to many areas, such as face unlock, attendance check, and entertainment. Most of the existing face recognition methods have been widely used in well-controlled scenarios and have achieved amazing performance. However, when face images are captured in uncontrolled surveillance scenarios, the face recognition methods will perform poorly. One main reason is that the face images, captured by surveillance cameras, are of low resolution (LR), because of the long distance between the faces and the cameras. This leads to limited discriminant information available from the face regions and the problem of resolution mismatch. This is called the low-resolution face recognition (LRFR) problem. In this dissertation, we have studied different methods for low-resolution face recognition, and proposed an algorithm, named "SR-MDS". In our method, super-resolution (SR) techniques are employed. We use the enhanced deep super-resolution network (EDSR), trained by a novel loss function, called identity-preserved loss, to perform face hallucination. Furthermore, multi-dimensional scaling (MDS), which projects features from different resolutions into a new common space, is used. First, we use EDSR[2] to super-resolve low-resolution face images. After generating a higher resolution face image, MDS is applied to learn coupled feature mapping for projecting the SR facial features and high-resolution (HR) facial features into a new common space. In this space, the distance between the two features of different resolutions is as close to the corresponding distance in the HR space as possible. The proposed method can generate representative and discriminative features from LR face images by using a SR model, and MDS can guarantee the structure consistency between the features in the transformed feature space and the HR face feature space. Furthermore, MDS takes inter-class variations into consideration, which further enhances the discriminability of the method. We conduct experiments on four databases: Yale, ARface, YALEB, and FERET. Experimental results have demonstrate the effectiveness of our proposed method.
Rights: All rights reserved
Access: restricted access

Files in This Item:
File Description SizeFormat 
5666.pdfFor All Users (off-campus access for PolyU Staff & Students only)6.6 MBAdobe PDFView/Open

Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. 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.
  3. 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.

Show full item record

Please use this identifier to cite or link to this item: