A novel approach to characterize of surface shape in local area is presented. The image model underling some image segmentation methods considers images as the sampling of a smooth manifold. It allows applying the apparatus of differential geometry to construction the points' attributes. The decision on belonging of an image's point to the object is made on the basis of these features. In contrast of well-known curvature settle features our classification based on analysis of normal vectors arrangement in point vicinity. The normal vector built for the tangent plane for every distinct pixel. In one's turn the tangent plane choosen from planes` set relying on extent of error between its and image plane. We established that the sum of corners wich form by normal vectors and another one from image plane could characterize the shape of surface. The application of proposed technicks for some syntetic and real images is shown.