A foreground layer generating unit generates a foreground layer on the basis of a foreground mask generated by a foreground mask generating unit and an original image. A binary image processing unit generates N binary images corresponding to N kinds of foreground identifiers on the basis of the foreground layer, and sets a plurality of tiles for each of the N binary images. Further, the binary image processing unit determines whether or not adjacent tiles are to be merged, and then sets tiles obtained by merging tiles determined to be merged, and leaving tiles determined not to be merged as they are. An image compressing unit cuts out rectangular regions from the individual set tiles by employing the binary images, and compresses the rectangular regions.