Images of various scenes can be classified with high accuracy by considering a characteristic part corresponding to a desired one of the scenes in the images and a change in position or area proportion thereof caused by composition difference. A local characteristic quantity image generation portion generates local characteristic quantity images from a target image input from an image input reception portion. A representative characteristic quantity calculation portion calculates representative characteristic quantities for each of the local characteristic quantity images from a result of sum-of-product calculations obtained by shifting and/or distribution change of masks of a type previously related to a type of the local characteristic quantity images in the corresponding local characteristic quantity image. An identification portion compares values of the representative characteristic quantities to identification conditions stored in a memory and judges whether the target image is of the specific scene.