Multi-Level Probabilistic Relaxation

M. Petrou, M. Mirmehdi, and M. Coors
Centre for Vision, Speech, and Signal Processing
University of Surrey
Guildford  GU2 5XH, United Kingdom
email: M.Mirmehdi@ee.surrey.ac.uk

Abstract:

We present a multi-level probabilistic relaxation scheme appropriate for image segmentation on the basis of features computed at various resolution levels. At each level, we incorporate contextual information computed from the features in the coarser scale representation of the scene, as well as information from the features computed in the finer scale representation, until a full segmentation is achieved at the highest level of resolution. The features used can be textural, colour, or the results of any other type of measurements performed on the image. We demonstrate our approach by the segmentation of colour textures.

Full postscript version

• 1 Introduction
• 2 Literature Survey
• 3 Multiscale Probabilistic Relaxation
• 4 Application
• 5 Conclusions
• References