MPhil Thesis Defence


"A NEW EXPECTATION-MAXIMIZATION ALGORITHM FOR SOFT COLOR
 SEGMENTATION AND ITS APPLICATION TO LOCAL COLOR TRANSFER"

By

Mr. Yu-Wing Tai


Abstract

This thesis defines and addresses the problem of soft color segmentation
(CVPR 2005) and proposes related applications. Unlike conventional hard
segmentation where a single label is assigned to each pixel (x, y), our
soft segmentation assigns a 2-tuple for each pixel: (P, C) where P(x, y) =
{P_i(x, y) | i = 1, 2, ..., N} is the probability distribution for the
pixel assigned to color label i, and C(x, y) = {C_i(x, y) | i = 1, 2, ...,
N} are the corresponding colors. Under this framework, a pixel observation
may be explained by more than one color labels. When N = 2, our output is
compatible with that of image matting. We propose to solve the soft
segmentation problem by a new Expectation-Maximization (EM) algorithm.
Our automatic EM algorithm for soft color segmentation is unique because
it effectively considers the coherence in the spatial domain, where the
necessary intra-region and inter-region smoothness are enforced without
losing original details. We propose a variety of applications: local color
transfer, image matting, image denoising, and image colorization, which
make use the advancement of our new EM algorithm to address the problem of
regional color transfer between two natural images.


Date:			Tuesday, 31 May 2005

Time:			4:00p.m.-6:00p.m.

Venue:			Room 2406
                   	Lifts 17-18

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


**** ALL are Welcome ****