MPhil Thesis Defence


"A study on Photogrammetry in Computer Vision:
 Shadow, Stereo and Separation"

By

Mr. Tai-Pang Wu


Abstract

Scene understanding is one of the core subjects in computer vision. Given
a single image or a sequence of image, a computer vision algorithm should
analyze the pertinent data and translate them into meaningful entities for
subsequent processing. In this thesis, we study the role of photogrammetry
in computer vision. Photogrammetry is a class of measurement techniques
that performs measurement to real world objects using images only, where
the measurement technique should provide sufficient information to allow
high level visual processing.

To understand a scene, we have to understand the components or factors
that explain the observation or measurement. In computer vision, it is
known that surface reflectance gives the color of a pixel, and the surface
orientation affects the reflectance, which is related to the object
geometry. Shadow is also an important cue for indicating the relationship
among objects. This thesis consists of the in depth study of these three
factors. Their respective measurements give rise to the proper computer
vision algorithms for scene understanding: reflectance separation (ECCV
2004), photometric stereo (CVPR 2005) and shadow extraction (ToG 2005). In
this thesis, we propose new and effective methods to tackle the
well-defined problem in each area, and we make contributions in robust
scene understanding by exploiting photometric measurement.


Date:				Tuesday, 31 May 2005

Time:				10:00a.m.-12:00noon

Venue:				Room 4333
         			Lift 3

Committee Members:		Dr. Chi-Keung Tang (Supervisor)
                        	Dr. Long Quan (Chairperson)
                        	Dr. Jiaya Jia (CUHK)


**** ALL are Welcome ****