Parallel model based segmentation using a 3rd order hidden Markov model

The work is based upon an algorithm which uses a third order hidden Markov mesh random field (HMMRF) for the segmentation of images. The main emphasis is to illustrate two important uses of the algorithm. Firstly, how the basic algorithm can be used to segment out urban and non-urban regions from airborne images. Secondly it is illustrated how the algorithm can be enhanced in order to model textures, the resulting `texture models´ can then be fused together and used to segment images which contain several Brodatz textures. Finally an apparently basic but very effective method of improving the performance by implementing the approach onto a multi-transputer array is described