An embedded system of a digital camera or printer has limited available memory resources. Under such environment, devices are required to detect a red-eye region from a high-resolution image. To solve this problem, an input image is divided into band images each including a predetermined number of lines, so that an overlap area exists between neighboring band images, and candidate pixels which form an image region exhibiting poor color tone are extracted for each band image. The band image which is to undergo characteristic amount determination is determined based on the position of the candidate region which includes the candidate pixels in the overlap area.