As an important process for reverse engineering, automatic registration is usually divided in two steps, the coarse one and the fine one. In coarse step, an initial estimate for relative rigid-body transform can be brought to realization by searching some similar geometry, and guessing the corresponding region,. In fine step, highly accurate alignment can be achieved by searching an approximate correspondence repeatedly. For two steps, researching on the first one is more important, because the coarse step will requires more computation time and the accuracy will affect the next step. Various methods such as Genetic algorithms, geometric histogram, graduated assignment, feature matching were submitted or applied. However those methods either need large computing time or only can be used upon the satisfaction of some limited conditions. In this paper, a new method base on CDCP model, including location descriptors, classification descriptors and direction descriptor, was .location descriptors and direction descriptors can provide the necessary information for matching the corresponding points, and classification descriptors can make this process more accurate and efficiency. Experimental results of three 3D images taken by laser scanner are carried out to compare the registration results of proposed method with other twins, and proposed method can be confirmed make the coarse registration process be high speedily without any limited condition.