Multi-view, high-resolution face image analysis

Pao Yue-kong Library Electronic Theses Database

Multi-view, high-resolution face image analysis

 

Author: Li, Dong
Title: Multi-view, high-resolution face image analysis
Degree: Ph.D.
Year: 2014
Subject: Human face recognition (Computer science)
Hong Kong Polytechnic University -- Dissertations
Department: Dept. of Electronic and Information Engineering
Pages: xv, 112 p. : col. ill. ; 30 cm.
Language: English
InnoPac Record: http://library.polyu.edu.hk/record=b2747241
URI: http://theses.lib.polyu.edu.hk/handle/200/7461
Abstract: With advances in digital photography, people can obtain large-scale and high-quality pictures more easily. How to understand this large-scale and high-quality information and how to make use of this information to recover distortions in other images are two fundamental and challenging problems in computer vision and image processing. In this thesis, we solve these problems for face images so that facial-image analysis and recognition can be performed more efficiently and accurately. In this thesis, we will mainly focus on the following three areas: face matching, and face verification/recognition, and color correction. Establishing correct correspondences between two faces with different viewpoints has played an important role in 3D face reconstruction and other computer-vision applications. Usually, face images are considered to lack sufficient distinctive features to track their geometry. Hence, existing methods have to rely on other man-made features such as structured lighting, special makeup, and markers. These active methods need specific devices to capture an object's structure. We investigate pore-scale facial features, which have many characteristics that make them suitable for matching face images under different variations. To alleviate the effect of changing skin conditions, a new framework is proposed as a trade-off between robustness and completeness. Based on this framework, a method adapted from scale-invariant feature transform (SIFT), namely pore-SIFT (PSIFT), is proposed, which is an automatic, passive approach for extracting distinctive pore-scale facial features for the reliable matching of uncalibrated face images.
To improve the performance of face verification/recognition using high-resolution (HR) information and the robustness to misalignment, we propose an alignment-free and pose-invariant face-verification method using the HR information based on pore-scale facial features. Most current face-verification/recognition methods represent face images mainly based on the holistic or local facial features. This makes these methods rely heavily on face alignment, so their performances degrade severely under variations in expression and/or pose, especially with only one gallery per subject. In this thesis, we have proposed a new keypoint descriptor called pore-PCASIFT, which is adapted from PCA-SIFT and is used for the extraction of compact, distinctive pore-scale facial features. Furthermore, a more effective feature matching scheme is proposed for face verification. As one of the most fundamental processing tasks for image analysis and understanding, color correction has received significant attention in a wide range of research fields, such as image stitching, street-view maps, 3D reconstruction, and multi-view image processing. Although many different color-correction approaches have been proposed in the past decade, it is still an active and challenging topic due to the fact that the appearance of the same object differs greatly with variations in illumination, view-point, optics/sensor characteristics and the hardware processing employed by cameras. For facial images, skin color is important for face detection and recognition. We have devised a method that can handle the color correction of multiple photographs of the same landmark scene or face subject with robust image restoration simultaneously and automatically. We have proved that the local colors in a set of images of the same scene exhibit the low-rank property locally both before and after a color-correction operation. This property allows us to correct all kinds of errors in an image under a low-rank matrix model without requiring any particular priors or assumptions. All the proposed methods in this thesis have been evaluated and compared to existing state-of-the-art methods. Experimental results show that our algorithms can achieve convincing and consistent performances.

Files in this item

Files Size Format
b27472413.pdf 38.39Mb PDF
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.

     

Quick Search

Browse

More Information