A priori knowledge about objects of interest allows design of efficient segmentation procedures that are robust against occlusion and deformation. Detection of circular objects in the image is based on the hypothesis generation and hypothesis testing principle. Hypotheses are generated by fitting an integrated parametric circular-symmetric model. Redundant model suppression is based on a goodness of fit measure and the amount overlap with other models.

The procedure was developed for counting cell colonies in petri-dishes and gives good results in comparison to manual counting method.