An image processing apparatus with capabilities for determining the direction of image on an original document when said document is read by a reader. The direction of the image is automatically determined by performing character recognition of the characters associated with the image and thereby determining the direction of the characters. The image is then assumed to have the same direction as the characters, thus it is possible to read images of different directions and assembly them in a single direction for subsequent processing. The process can be adaptable to provide either processing speed or high productivity.