Modeling is an important step in 3-D target recognition. Avoiding the recovery of 3-D obje- ct, multi-view modeling is used widely in 3-D object modeling. Multiple 2-D views are obtained by proje- cting a tessellated regular polyhedron onto the viewing sphere.In order to describe all shape and pose info- rmation, a fine partition of the viewing sphere is necessary,which leads to computation complexity. Zernike moments have a rotation invariant property. A method is used to decrease the number of 2-D projections view by extracting the Zernike moments of each 2-D view, whose max order is decided by the difference between the original image and the reconstruction one.Those partitions with Zernike moments of correspon- ding views are similar are merged and a characteristic view is obtained for a new partition. Experimental results show the validity of the method.