To accurately extract even a character which sticks out from a frame on a document.

The frame is recognized (S6), and the ruled lines constituting the frame are erased (S7). Rectangles of black pixel connecting components in an area obtained by expanding the frame are extracted from the document image having the ruled lines erased (S8) and are integrated to extract a character string area (line) (S9). Only the character area, which can be considered to belong to the frame, is selected from the extracted character string area (S10), and its image is cut out of the document image having the ruled lines erased (S11).