An image processing method of finding a motion vector between an original image and a reference image, including dividing the original image to extract a block, searching for a similar region of the block from the reference image to detect a motion vector of the block, further dividing the block from which the motion vector has been detected to extract a lower block, and searching for a similar region to detect a motion vector of the lower block with a range in the vicinity of the motion vector found from the block before division as a start point. Thus, the method is characterized by reflecting a tree structure data of the motion vector obtained in this manner in the subsequent detection of motion vector.