Likelihood - based classification of high resolution images to generate the initial topology and geometry of land cover segments : a hypothesis generator for model based image analysis
Ali Abkar, Nanno J. Mulder · University of Twente Research Information · 1998
This paper’s origin is to reach for a way for automatic initiation of shape hypothesis for Model Based Image Analysis (MBIA) in the specific case of agricultural fields. A solution is to start with local (topological) hypotheses. The topological data are integrated from local topology to the level of real 2-dimensional objects. The method requires radiometric model to generate the normalized class membership probabilities (likelihood vectors) and a minimum-size-of-object parameter. The paper gives a detailed description of the analysis approach for initiating the shape hypothesis for MBIA and besides resolves problems related to classical approaches such as per pixel maximum likelihood classification. An experiment is presented about its application in a case of RGB-CCD image of agricultural fields’ model. We obtained an overall accuracy of 97% in comparison with 83% in improved maximum likelihood classification.