MODELING UNCERTAINTY IN AGRICULTURAL IMAGE-ANALYSIS

No absolute certainty can be given for information derived from images . In most cases image analysis uses single algorithms, or multiple sin gle algorithms' results which are combined in an ad hoc manner, to der ive certain information (e.g. edges and textures) to segment images in to various regions of interest. However, more robust methods of data f usion can be developed which are based on mathematical foundations of probability theory. One such method combines results from single algor ithms using a Bayesian network. This should improve the confidence in the derived image segmentation and gives a direct measure of the proba bility of each region to be classified correctly. Specific agricultura l examples using a Bayesian data fusion approach are given. (C) 1997 E lsevier Science B.V.