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:Cambridge Analysts' Knowledge Exchange
SUMMARY:Fast noise learning via nonlinear PDE constrained
optimization - Luca Calatroni (CCA\, CIA)
DTSTART;TZID=Europe/London:20131127T160000
DTEND;TZID=Europe/London:20131127T170000
UID:TALK46772AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/46772
DESCRIPTION:In this talk\, we recap the framework of Bounded V
ariation functions and their main features relevan
t to Imaging. In the course of that discussion we
highlight some drawbacks of TV reconstruction and
introduce higher-order versions of TV which impro
ve upon them\, highlighting the numerical obstacle
s arising when solving such models and possible di
rections (such as ADI directional splitting scheme
s).\n\nThe rest of talk will focus on my current r
esearch on the optimal setup of these models by me
ans a nonlinear PDE constrained optimisation\, bas
ed on a training set of original and noisy images.
In such approach\, the optimal weights are comput
ed such that the corresponding total variation reg
ularised solutions (encoded in the constraints) 'b
est' fit the original images. To improve upon the
efficiency of the numerical solvers\, we use dynam
ical sampling schemes to compute the optimal param
eters.\n\nThis is a joint work with J. C. De Los R
eyes and C.-B. SchĂ¶nlieb.
LOCATION:MR14\, Centre for Mathematical Sciences
CONTACT:Marcus Webb
END:VEVENT
END:VCALENDAR