PhD Qualifying Examination


"Correspondence Computation between Two Images in
Static Scene"

By Mr. Yichen Wei


Abstract:

Correspondence computation plays a fundamental role in low
level vision. It is critical for the success of many high level
vision applications, such as motion tracking, auto-calibration,
reconstruction and image-based rendering. In this survey, I
investigate the problem of computing correspondences between two
images in the static scene. According to the availability of
camera geometry, the involved algorithms fall into two groups:
(1)Robust sparse matching without camera geometry; (2)Dense stereo
matching with know camera geometry and short baseline. While (1)
is by and large a solved problem, (2) is still in development and
there remains many questions to be solved.

Firstly, the problem description and some basic issues in matching
are given, including some constraints, block matching and corner
detection. Then I discuss some sparse matching algorithms. Before
going into the discussion of individual dense stereo algorithm, a
framework is introduced. Afterwards, the investigated dense stereo
algorithms are classified as three sub categories according to the
property of being local or global: (a)local iterative methods;
(b)graph-cut based methods(global); (c)region-based
methods(global). Each sub-category has several algorithms involved
and some comparisons and discussions made.


Date:     		Friday, 24 January 2003

Time:                   11:00a.m.-1:00p.m.

Venue:                  Room 2302
			lifts 17-18

Committee Members:      Dr. Long Quan (Supervisor)
                        Dr. Chiew-Lan Tai (Chairman)
			Dr. Michael Brown
			Dr. Chi-Keung Tang


**** ALL are Welcome ****