BMVC 2004, Kingston, 7th-9th Sept, 2004

Affine Invariant Image Segmentation
A. Bhalerao and R. Wilson (Warwick University)

This paper introduces a method of image segmentation that partitions an image
into regions exhibiting self-similarity. The new method is well suited
to the description and segmentation of images containing textures, such as
fingerprints and medical images. An affine texture description which models
an image as a collection of self-similar image blocks is developed and it is
shown how textural regions can be approximated by a single prototype block
together with a set of affine transformation coefficients. The transformations
are estimated by Gaussian mixture modelling of the prototype block spectrum
and a least-squares estimation to fit the model to the target block. An
appropriate set of prototype blocks is determined for a given an image by
clustering in a feature subspace of discrete samples from an affine symmetry
group. We consider how the texture synthesis outlined in the first stage can
be combined with the prototype clustering in the second, to build a novel image
segmentation method. Experimental results demonstrate the potential of
the method on widely varying images.
(pdf article)