City University of Hong Kong

CityU Institutional Repository >
3_CityU Electronic Theses and Dissertations >
ETD - Dept. of Computer Engineering and Information Technology  >
IT - Master of Philosophy  >

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

Title: Facial features location and eye modeling using fuzzy clustering analysis
Other Titles: Ji yu mo hu ju lei fen xi de lian bu te zheng ding wei ji yan bu jian mo
Authors: Tse, Ka Wing (謝嘉榮)
Department: Dept. of Computer Engineering and Information Technology
Degree: Master of Philosophy
Issue Date: 2004
Publisher: City University of Hong Kong
Subjects: Cluster analysis
Human face recognition (Computer science)
Notes: CityU Call Number: TA1650.X54 2004
Includes bibliographical references (leaves 82-86)
Thesis (M.Phil.)--City University of Hong Kong, 2004
xi, 86 leaves : ill. (some col.) ; 30 cm.
Type: Thesis
Abstract: Face region and facial features extraction are important areas of research. An automatic process to achieve these tasks is crucial to many applications such as face recognition, model-based video coding and human based animation. This thesis will present the study on (i) face region and facial features location; and (ii) eye modeling. Extraction algorithms have been developed to process frontal view head-and- shoulder color images with plain background to achieve these two tasks. For locating the face region and facial features, the RGB image is first analyzed by the spatial fuzzy clustering method. Three clusters will be obtained to generally represent the face region, background, and hair. With the aid of an ellipse, the cluster containing the face region can be identified. Other facial features including eyes, eyebrows, nose and mouth can then be located from the face cluster with further fuzzy clustering analysis and a geometric face model. This approach has advantages that priori color information about the face is not required and it does not require edge information as in other approaches, hence the results are more robust. To model the eye features, two approaches have been studied and the spatial fuzzy clustering analysis is again the main technique for identifying the region of interest. A circular template is used to locate and model the eyeball from a cluster which contains the shadow, eyeball and eyebrow. With the use of sclera cluster, six control points for outlining the eye shape can be determined. A control point based eye shape method has been developed to model the shape. However, this approach will not deliver correct results if the control points are wrongly estimated due to unsatisfactory segmentation results from the clustering analysis. The second approach is a region based method. It has been observed that the eye shape becomes apparent if the eyeball and sclera clusters are combined. A cost function based on an eye shape template has been established for optimizing the model. The performance of this region based approach has been shown to outperform the control point based and other well developed deformable template methods. Based on the high successful rate of eyeball extraction, a real-time eyeball tracking system has also been implemented in a 1.9 GHz PC. The system can process the images with a rate higher than 25 frames/second.
Online Catalog Link:
Appears in Collections:IT - Master of Philosophy

Files in This Item:

File Description SizeFormat
fulltext.html159 BHTMLView/Open
abstract.html159 BHTMLView/Open

Items in CityU IR are protected by copyright, with all rights reserved, unless otherwise indicated.


Valid XHTML 1.0!
DSpace Software © 2013 CityU Library - Send feedback to Library Systems
Privacy Policy · Copyright · Disclaimer