In dice recognition, there are several situations which may cause dice image identification problems including that the shape of the dice is round, or the dice may touch each other. Especially in camera zooming in or zooming out, the pip sizes become irregular. In this paper, we adopt an affine transform technique and a pattern comparison method to develop an intelligent dice pattern match ratio to achieve accurate dice recognition. A normalization technique solves image zooming in and zooming out problems and an affine transformation rotates the shape in various angular positions for object pattern comparison so that it greatly improves the dice image recognition successful. After over 200 test images that include different shapes, styles, sizes and colors used in simulations, this scheme is proven to achieve 100% dice image identification. (14 refs)
