Boundary is an important character of image. The boundary of an object was parameterized by differential area, and then the parameterized boundary was transformed by discrete cosine transformation. Based on this method, a set of descriptors were obtained, which are invariant under any affinetransformation (translation, rotation, scaling, and shearing). Especially, they are invariant as the change of viewport in recognition of 3D motion object. Through analysis of theory, the computational process of this method was given. Many planes of different poses in 3D space were simulated. The experiment results show that values of each kind of character are stable correspondingly.