BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Machine Learning Reading Group @ CUED
SUMMARY:Fast Fusion of Multi-band Images: A Powerful Tool
for Super-resolution - Qi Wei (University of Cambr
idge)
DTSTART;TZID=Europe/London:20160331T143000
DTEND;TZID=Europe/London:20160331T160000
UID:TALK65399AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/65399
DESCRIPTION:Hyperspectral (HS) imaging\, which consists of acq
uiring a same scene in several hundreds of\ncontig
uous spectral bands (a 3D data cube)\, has opened
a new range of relevant applications\, such\nas ta
rget detection [Manolakis and Shaw\, 2002]\, class
ification [C.-I Chang\, 2003] and spectral un-\nmi
xing [Bioucas-Dias et al.\, 2012]. However\, while
HS sensors provide abundant spectral informa-\nti
on\, their spatial resolution is generally more li
mited. Thus\, fusing the HS image with other highl
y\nresolved images of the same scene\, such as mul
tispectral (MS) or panchromatic (PAN) images is\na
n interesting problem\, also known as multi-resolu
tion image fusion [Amro et al.\, 2011] (Fig. 1).\n
From an application point of view\, this problem i
s also important as motivated by recent national\n
programs\, e.g.\, the Japanese next-generation spa
ce-borne hyperspectral image suite (HISUI)\, which
\nfuses co-registered MS and HS images acquired ov
er the same scene under the same conditions\n[Yoko
ya and Iwasaki\, 2013]. Bayesian fusion allows for
an intuitive interpretation of the fusion process
\nvia the posterior distribution. Since the fusion
problem is usually ill-posed\, the Bayesian metho
dology\noffers a convenient way to regularize the
problem by defining appropriate prior distribution
for the\nscene of interest.\n\nIn this work\, a n
ew multi-band image fusion algorithm to enhance th
e resolution of HS image\nhas been proposed. By ex
ploiting intrinsic properties of the blurring and
down-sampling matrices\,\na much more efficient fu
sion method has been developed thanks to a closed-
form solution for the\nSylvester matrix equation a
ssociated with maximizing the likelihood. The main
contribution of this\nfusion scheme is that it ge
ts rid of any simulation-based or optimization-bas
ed algorithms which\nare quite time consuming. Cou
pled with the alternating direction method of mult
ipliers and the block\ncoordinate descent\, the pr
oposed algorithm can be easily generalized to inco
rporate different priors or\nhyper-priors for the
fusion problem\, allowing for Bayesian estimators.
This method has been applied\nto both the fusion
of MS and HS images and to the fusion of PAN and H
S images. We have tested the\nproposed algorithm i
n both synthetic data and real data. Results show
that the proposed algorithm\ncompares competitivel
y with existing algorithms with the advantage of r
educing the computational\ncomplexity significantl
y.
LOCATION:Engineering Department\, CBL Room 438
CONTACT:Yingzhen Li
END:VEVENT
END:VCALENDAR