A task is to correctly classify an input image regardless of a fluctuation in illumination and a state of occlusion of the input image.Input image sub-region extraction means 2 extracts a sub-region of the input image. Inter-pattern distance calculation means 3 calculates an inter-pattern distance between this sub-region and a sub-region of a registration image pre-filed in dictionary filing means 5 for each sub-region. Region distance value integration means 10 integrates the inter-pattern distances obtained for each sub-region. This is conducted for the registration image of each category. Identification means 4 finds a minimum value out of its integrated inter-pattern distances, and in the event that its minimum value is smaller than a threshold, outputs a category having its minimum distance as a recognition result.