Colloquium Speaker

Speaker: Carola Wenk
Freie Universitat Berlin
Topic:Applications of geometric shape matching
Date:Tuesday, April 17, 2001
Time:11:00 AM
Place:Gould-Simpson, Room 701


Refreshments will be served in the 7th-floor lobby of Gould-Simpson at 10:45 AM


ABSTRACT


The task of comparing two geometric shapes ('shape matching') arises naturally in many applications. We will survey in the talk algorithms for the following shape matching problems.

- Brain-matching: In brain surgery the surgeon has to be careful not to destroy important parts of the brain. Some tools allow the surgeon during the surgery to locate (using a special pointer) a position in the patient's brain. The location of this point is then displayed in a prerecorded functional MR image. This requires a registration between the 3D image and the brain during the surgery. We will show in the talk how this problem was solved by attaching markers to the patient's head which are present in the MR scan as well as during the surgery, such that the registration procedure amounts to robustly match the two sets of markers. This is a collaboration with neurosurgeons from the medical school at FU Berlin.

- Gel-matching: Identifying the functionality of a given protein is a hot topic in molecular biology. In order to get hold of certain proteins a protein probe is usually separated by 2D gel electrophoresis yielding a 2D image of axis-parallel ellipses ('spots') each representing a single protein. By comparing two 2D electrophoresis gels it is possible to identify variations of the amount of certain proteins present in a probe. In the talk we will consider an algorithm that computes the correspondence of protein spots in two gel images. In our approach we represent each spot as a two dimensional point with an intensity describing the size of the spot. After a non-trivial phase of assigning point coordinates to each spot ('spot detection') we apply a point pattern matching algorithm in order to identify corresponding spots.

- Star-matching: A satellite orbiting the earth has to know in which direction the ground station is located in order to receive commands and to send back data. A way of self-orientation of the satellite is to take pictures of the starry sky and to locate the picture in a known star atlas. We will consider a point pattern matching algorithm that makes use of the different intensities of the stars in order to speed up computation time.