BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:"Egalitarian Gradient Descent": A General Preconditioning Scheme f
 or Deformable Image Registration - Darko Zikic\, TU Munich
DTSTART:20110809T090000Z
DTEND:20110809T100000Z
UID:TALK32214@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:In this talk\, I will discuss a simple\, yet theoretically jus
 tified approach for improving the efficiency of optimization in deformable
  registration problems\, for arbitrary difference measures.\n\nGiven the r
 egistration problem as optimization of an energy containing a difference m
 easure\, the key step of our approach is extremely simple: We modify the v
 ector field corresponding to the gradient of the difference measure\, by n
 ormalizing the single point-wise vectors of the field to approximately uni
 t length. We show that this simple scheme improves the convergence speed o
 f optimization\, while being 1) theoretically justified\, 2) computational
 ly efficient\, and 3) applicable to arbitrary difference measures. Since t
 he above scheme incorporates our idea to give roughly the same influence t
 o all relevant points in the image domain\, we dub the resulting approach 
 “egalitarian".\n\n1) The above scheme is theoretically justified\, as we
  demonstrate that it corresponds to a preconditioning of the difference me
 asure. Actually\, the scheme approximates the optimal preconditioning with
  respect to the analyzed model.\n\n2) Due to the simplicity of the gradien
 t modification and the resulting computational efficiency\, the improvemen
 t in convergence speed is directly translated to shorter effective runtime
 s.\n\n3) Finally\, a very important aspect of our approach is that it is a
 pplicable to arbitrary difference measures. This is of high practical valu
 e for problems in which the range of applicable optimization methods is li
 mited\, such as deformable registration tasks in multi-modal settings\, wh
 ich utilize statistical difference measures such as Mutual \nInformation.\
 n\nBesides discussing the actual method\, the talk will focus on providing
  an intuitive motivation for the approach\, highlighting its relations to 
 other optimization schemes\, and demonstrating its application in popular 
 registration frameworks. Also\, the talk will contain a brief overview of 
 some of my other image registration projects\, including deformable 2D-3D 
 registration\, and linear registration by MRFs and discrete optimization.\
 n
LOCATION:Small lecture theatre\, Microsoft Research Ltd\, 7 J J Thomson Av
 enue (Off Madingley Road)\, Cambridge
END:VEVENT
END:VCALENDAR