A new method for surface identification is introduced in this paper, and it is applied to object matching and object recognition. The first component of the method associates a surface-point P with a surface point signature (SPS), encoding “the surface curvature seen from P.” Since the mathematical definition of SPS (in section 2) is not similar to the known definitions of curvature, the above claim does not seem well founded.

The second component of the method (section 3) deals with selecting “important points” on the given surface: these are the points with a sufficiently high curvature. In section 4, the new SPS representation is used as a “knowledge representation,” for solving the classical registration problem using neural networks. Different types of networks are described, for both the complete object matching and the partial object matching problems. The final section of the paper discusses some examples demonstrating the efficiency of the proposed method.

I cannot say that this is a well-written paper: for example, initially, it is not clear if the subject of this research is exact analytical surfaces or approximate triangulated surfaces. It is also not clear how the SPS model is stored in a three-dimensional (3D) database. Finally, the reported experimental tests (section 5) are not fully documented; for example, we do not know the number of points/triangles in the scanned objects employed. Surely, the reported research is important and useful, however, since it correctly places emphasis on important surface points and curvatures for dealing with major geometry-processing problems.