[Colloquium] Reminder: Li/MS Presentation/Apr 19, 2016

[Margaret Jaffey]

This is a reminder about Kai Li's MS Presentation tomorrow.

------------------------------------------------------------------------------
Date:  Tuesday, April 19, 2016

Time:  1:30 PM

Place:  Ryerson 276

M.S. Candidate:  Kai Li

M.S. Paper Title: Connections Between Bandwidth Selection in Kernel
Density Estimation and Mode Stability in Scale Space

Abstract:
Technologies for scanned volumetric imaging (producing data on
three-dimensional spatial or four-dimensional spatio-temporal grids)
continue to grow in resolution and flexibility. As new imaging
modalities are applied in new domains, the utility of
application-specific and modality-specific tools decreases, and there
is greater need for general purpose methods of finding and
communicating structure in 3D/4D images. Scale-space is a conceptual
framework that was studied and applied extensively for 2D projective
images, but less so for 3D/4D data. We consider an image feature as a
mathematically defined set of points in the image domain, such as
critical points, edges, ridges, and valleys. We propose that the
spatial stability of an image feature, respect to changes in scale
(blurring, or diffusion), is an effective way to discern which
features at a given scale are significant or ``real'', and to choose
an optimum scale for that feature. This is conceptually rooted in the
``spatial coincidence assumption'' of Marr-Hildreth feature detection
in classical computer vision, which selects edges that are spatially
coincident over a range of scales as indicators of some significant
underlying physical feature. Even when restricted to characterizing
peaks (modes) in one-dimensional density distributions, there are
interesting connections to other considerings of multi-scale feature
selection, which we explore in this paper. Specifically, we identify
the theoretical connection between our scale-space stability criterion
and different approaches to bandwidth selection in kernel density
estimation, such as the mean-shift vector method of D. Comaniciu et
al. and the SiZer method of J. S. Marron et al.

Kai's advisor is Prof. Gordon Kindlmann

Login to the Computer Science Department website for details:
 https://www.cs.uchicago.edu/phd/ms_announcements#lik11

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Margaret P. Jaffey            margaret at cs.uchicago.edu
Department of Computer Science
Student Support Rep (Ry 156)               (773) 702-6011
The University of Chicago      http://www.cs.uchicago.edu
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=