Rolands's Bowling Green project page
This thesis is on
"Invariant pattern recognition by means of artificial neural networks."
and is not available due to my own sloppiness. Sorry!
- This paper is an in-depth discussion of the use of some Fourier-, and Fourier-Mellin Transformation models in translation-, rotation-, and scale-invariant pattern recognition. Some of these mathematical models, like the phase correlation technique are discussed in detail. To show the reader how the theoretical results could be used for ``real world'' applications the Fourier Mellin Transformation model is implemented. The implementation does not just cover the description task by geometrical moments, but also the recognition and categorization task. There are well known statistical methods to recognize and categorize patterns. In this paper the author concentrates on a different method, known as Artificial Neural Networks. For the reader who is not familiar with the theory of Artificial Neural Networks, a short theoretical introduction to Neural Networks is given.
Literature on Invariant Pattern Recognition
Software on Invariant Pattern Recognition
- >ftp://ftp.fct.unl.pt/pub/di/packages/UFF.README The software written in C with a graphical interface to X computes
invariant features of binary line-oriented patterns.
Limitations: No occlusion, only thin, binary patterns.
Besides, the software computes Fourier descriptors (shape features
of closed boundary patterns). (Source: Thomas W. Rauber,firstname.lastname@example.org)
Btw, if you are interested in the same topic send me some
last modified Saturday, 10-Feb-1996 21:02:03 CET by