An image processing method and apparatus which efficiently detect an image input direction, from an image with much differential information to an image with little differential information. The direction of an image including a character area, inputted into a computer, is detected. First, a binary image of the input image is generated, and a tile image is generated by adding a predetermined value to respective tiles, each tile corresponding to a predetermined size area in the binary image. Next, an area of the binary image, corresponding to an area within a circumscribed rectangle surrounding connected pixels having the same value in the tile image, is extracted as a character area. Then, the direction of characters included in the character area is recognized and thereby the direction of the input image is detected.