A feature amount cluster holding unit extracts respective feature amounts from N images sensed under N respective types of image sensing conditions, and manages the feature amounts extracted from each sensed image as a feature amount cluster in association with corresponding one of the image sensing conditions. A feature space distance calculation unit specifies the second feature amount cluster containing a feature amount similar to the feature amount of interest in the first feature amount cluster. An image sensing condition setting unit specifies sets of feature amounts associated with the N respective types of image sensing conditions from the first and second feature amount clusters. The image sensing condition setting unit specifies a set having a largest distance between feature amounts in the set among the specified sets. An image sensing condition associated with the specified set is output as an image sensing condition for discriminating the feature amount clusters.