Registration of point clouds is a key process in creating a digital model in reverse engineering. Registration is complex and ill-conditioned problem and these impede to achieve fully automatic comprehensive algorithm.in this study a new method to improve automation level is proposed. In this method, at first surface features are extracted from point clouds and then these data use for detecting correspondence points between point clouds. Registration process accuracy depends on carefully selected corresponding point between point clouds. In present research surface curvature and local shape are used for finding correct correspondence points. For feature extraction, surface curvature for each point of point clouds is calculated by using umbrella curvature and also a new method for determining local shape is presented. For each point of point cloud a shape number is determined. Determination of shape number is done by neighbors’ coordinates of point of concern. In this method, the corrected corresponding points are points that have almost equal umbrella curvature and shape number. Rigidity helps algorithm to find pairwise points. Analyzing results shows that the proposed algorithm performs well and has appropriate abilities on fully automatic registration of point clouds.