PDF version of this document
From a system that has a collection of conjectured points of interest,
a crude combination of two aspects of data collection and warping
that is based on a vague emerging technique, we may soon be able to
integrate some of the existing algorithms to form a powerful descriptor
of appearance that can automatically learn about the mixture
of elements in images and produce models that can be confidently labelled
optimal according to some criteria. This system will be more
efficient and more responsive, regardless of the type of data being
inspected and its results will make it the best choice off-the-shelf.
Acknowledgements. The author would like to thank Chris Taylor
for lengthy helpful discussions. This project is being funded by the
EPSRC.