Complete object recognition under projection distortion

Abstract The paper first of all addresses the fact that the object on the image plane of a camera is affine invariant in accordance with the standard form when the scene image or the camera itself is undergoing slight displacement. Thus, boundary matching can be the first step to match the scene image with a known model by identifying three non-collinear points of correspondence. Furthermore, image recognition of the scene image is completed by evaluating the degree of matching with its gray level model. This new method is compared with the conventional way using moment invariance and the merits of higher computational speed and uniqueness are discussed.