My supervisors and I submitted a journal paper to the International
Conference on Pattern Recognition in Quebec at the end of last year
and have just found out that it has been accepted. This means that with
a bit of luck I'll get to go over there and present it - not bad for an
honours project! Click here to check the
paper we wrote and here you'll find some extra
tests which we ran on the algorithm but couldn't include in the paper due
to space limitations (they only allowed us 4 pages - this required some
shrinkage of my honours dissertion, which was about
50 pages).
The background to this page is the shading pattern obtained from the
surface of the log of a distorted quadratic. The singular point is the
bright white point in the center. As you can see its kind of hard to pin
point the exact location of it - this was one of the aims of my honours
project. Of course this is a nice resoultion, the problem is even more
pronounced in a coarse grid.
This year we will be working on some new stuff, it will be in computer vision, although exactly what is as yet unknown. At the moment we are interested in space carving, which is a very neat method developed by Kiriakos N. Katulakos et. al for extracting 3D shapes from a set of N photographs. The method carves away voxels (volume elements) in the scene by means of a consistancy check across the set of photographs. One problem we see with this is the very large parameter set required (several hundred thousand voxels). If the three dimensional shape can be approximated adequately with a neural network (or perhaps a radial basis function approximation), then this number of parameters should be able to be reduced by several orders of magnitude. There are many problems associated with this concept, so it will be interesting to see how it all pans out. I'll keep you posted!