[BANANA] CHANGE IN TITLE: LA/Opt Seminar: Wed Sept 27

[Gene H Golub]

 		****  CHANGE IN TITLE AND ABSTRACT ****

  DATE:  Wed, Sept 27
  TIME:  4:15
  ROOM:  Gates 104
  SPEAKER:Ofer Levi
  FROM:  Ben-Gurion University
  E-MAIL: levio at bgu.ac.il
  TITLE: Super fast algorithm for effective multi-scale geometric
  decomposition of 3D images.

ABSTRACT: We propose a new method for piece-wise constant approximation of 
3D images using multi-scale geometric analysis approach. The building 
blocks in the decomposition are generated by intersection of various 
planes with cubes in different scales. The algorithm implements fast 
evaluations of plan integrals over the image in the frequency domain and 
has an overall complexity of o(Nlog(N)) when N is the number of voxels in 
the input image. The low complexity is achieved by facilitating an exact 
discrete projection slice relations using the so called Pseudo-Spherical 
grid in the frequency space. We use exact ANOVA tests for evaluating and 
comparing different possible decompositions and a single statistically 
meaningful global parameter for all tests, level of significance. This 
method is very robust with respect to the shape of the objects of interest 
and proves to be successful even in the case of extremely faint objects 
buried in a high level on noise.

Coming Attractions: October is Tensor Decomposition Month. Watch this space!
-++**==--++**==--++**==--++**==--++**==--++**==--++**==
This message was posted through the Stanford campus mailing list
server.  If you wish to unsubscribe from this mailing list, send the
message body of "unsubscribe linear_algebra_optimization" to 
majordomo at lists.stanford.edu