Provided is an image recognition method in which a first pattern area which is set inside a target to be recognized and a second pattern area which is set outside the target are used as a mask pattern. When a comparison circuit judges that a difference obtained by subtracting, by a differential circuit, the largest luminance value in the target image contained in the first pattern area, which is detected by a largest value detection circuit, from the smallest luminance value in the target image contained in the second pattern area, which is detected by the smallest value detection circuit, is larger than a certain offset amount, the matching judgement that a predetermined pattern is present is made. This recognition judgment is performed with the whole target image being scanned.