Enhancement of unsupervised segmentation using Gibb's random fields for microscopy image analysis

In this paper, tissue component separation in color microscopy images using unsupervised segmentation methods is discussed. Implementation of a competitive learning algorithm is described in particular. Spatial connectivity constraint is added to the segmentation using a Gibb's random field model. A computationally efficient iterative conditional modes (ICM) algorithm is used subsequently to find the segmentation with maximum probability of existence. Example scanned images of Movat and immune stained microscopy slides are used to illustrate the segmentation process. Finally, possible improvements to ICM algorithm for adaptation to the stain variation are discussed.