An image processing apparatus generates a binary image from a multivalue input image, detects a character region and a non-character region from the binary image, and clips regions surrounding each of characters included in the detected character region as character clipping rectangles. The apparatus determines, on the basis of a black pixel density of the binary image included in the character clipping rectangle, whether or not the character clipping rectangle is a character region and determines, on the basis of a number of pixels included in the character clipping rectangle determined to be a non-character region, whether or not the character clipping rectangle determined to be a non-character region is a character region.