A large number of stable local regions can be set with low calculation cost. In a face recognition apparatus which discriminates similar face images using feature amounts extracted from local regions included in an image to be discriminated, a moving destination of a feature point extracted from the image to be discriminated, and the size of an image to be clipped at the moving destination are calculated based on a table which defines information required to designate a moving destination of each feature point, and information required to designate the size of an image to be clipped at the moving destination, and an image with the calculated size is clipped at the calculated moving destination as the local region.