When classifying objects in images of biological specimens, it is usual for there to be some dependence among neighbouring objects. This can in theory be used to augment the information available for classifying each object. However, much of the methodology developed for this type of contextual classification assumes a fixed number of neighbours, such as is found on a regular grid. In this paper, we show how Markov random fields can be used in the case where the number of neighbours varies, and we illustrate this with an application in the classification of cells types in microscope images of plant stems.
- Discriminant analysis