The approach of mapping photographic images on 3D geometries that have been created from real world objects is attracting wide attentions. The registration of the images to the 3D geometries is the key technique. Since previous work used either 3D-2D point matching or silhouette matching for registration, it makes some special demands on the surface features or silhouette shapes of the geometries. In this paper, a new method is presented to overcome this problem. By matching the 3D points reconstructed from the images and the known geometric model in space, the topology and curvature information of the model surface can be fully utilized and all the images to the 3D model stitched at the same time. Experimental results show this method can account for some cases that are intractable for the traditional methods, and the results are quite satisfactory when the distribution of the reconstructed points is relatively symmetrical around the model surface.