In this paper we propose a probabilistic model and an expert system which constructs image processing procedures to detect specific figures automatically based on the model. An image processing procedure is constructed when user inputs a set of pictures consisting of normal samples and abnomal samples with sketches corresponding to abnormal areas and also inputs requirements on misclassification rate. We consider the image processing prosedure as a kind of description of knowledge which a user acquires from experience. We also show experimental results of applying the expert system to a real world problem.
