Centre for Mathematical Imaging and Vision

Workshop on Mathematical and Computational Imaging

Date: 10 July 2009
Venue: WLB207, The Wing Lung Bank Building, Shaw Campus, Hong Kong Baptist 
University

Programme
10:00 a.m. - 10:45 a.m. Xiaojun Chen (PolyU)
10:45 a.m. - 11:30 a.m. Pierre Weiss (HKBU)
11:30 a.m. - 12:15 p.m. Tieyong Zeng (HKBU)
2:00 p.m. - 2:45 p.m.   Junfeng Yang (Nanjing U)
2:45 p.m. - 3:30 p.m.   Michael Ng (HKBU)

Talk information:

Speaker: Xiaojun Chen (PolyU)
Title: Efficient methods of  nonsmooth nonconvex optimization for
image restoration
Abstract: Image restoration problems are often converted into
large-scale, nonsmooth and nonconvex optimization problems.
Most existing optimization methods are not efficient for solving such 
problems.
In this talk, we consider three methods:
1. Smoothing projected gradient(SPG) method for nonsmooth and
nonconvex optimization on a closed convex set. (Joint work with
Chao Zhang)
2. Smoothing conjugate gradient (SCG) method for unconstrained
nonsmooth and nonconvex optimization. (Joint work with Weijun Zhou).
3. Orthogonal matching pursuit-smoothing conjugate gradient
(OMP-SCG) method for non-Lipschitz and nonconvex optimization.
(Joint work with Fengmin Xu, Yinyu Ye).
These methods are easy to implement without adding any new variable,
and globally converge to a Clarke stationary point. Moreover, we
show that using non-Lipschitz and nonconvex regularization can
provide sparse reconstruction.

Speaker: Pierre Weiss (HKBU)
Title: A dual approach for minimizing strongly convex functions with 
applications

Speaker: Tieyong Zeng (HKBU)
Title: Discussion on compressive sensing
Abstract: The compressive sensing problem over dictionary has revoked 
remarkable interests. In this talk, we present some recent developments on 
this topic in three parts. First, a Matching Pursuit shrinkage method and 
a proximal
point algorithm for variant of Basis Pursuit model will be discussed. 
Second, we
will illustrate two total variation models which are related to 
compressive sensing.
Finally, we will propose a statistical approach for dictionary learning.

Speaker: Junfeng Yang (Nanjing U)
Title: Alternating direction method and its applications in total 
variation image reconstruction

Speaker: Michael Ng (HKBU)
Title: A New Total Variation Method for Multiplicative Noise Removal