Seal Identification Using the Delaunay Tessellation

A new approach which differs from the conventional pattern recognition techniques in partitioning a seal imprint image into a triangular tessellation is proposed. The image plane is segmented into a mesh of triangles according to the Delaunay triangulation paradigm, and the distribution of the triangular area is used as the primary feature for matching unrecognized seal imprints with model images stored in a database. The Delaunay tessellation of a planar point pattern has two desirable properties, namely, arrangement invariance and local re-meshing. Arrangement invariance means that the mesh structure formed is invariant to the rigid motions of Euclidean geometry, including translation and rotation. The recognition process using an arrangement metric, e.g., nearest neighbor analysis, will be unaffected irrespective of differences in the location and orientation of imprints from the same seal. Also, addition or deletion of a point from the original point pattern affects only the local mesh structure without incurring global re-structure. This local re-meshing property enhances the robustness of the recognition algorithm in the presence of noises, a situation commonly found when imprints from the same seal using different inks and setting forces are compared. In this paper, an overview of the general pattern recognition principles is given first, with emphasis on characteristics specific to seal identification. Delaunay triangulation is described next, and the desirable properties relevant to the identification of seal imprints are stated. The proposed approach utilizing the triangular area as the major feature is detailed in the following section. The displaced, rotated, smeared and noise-corrupted imprints from the same model seal were used as test images in the experimental study using a database containing 30 registered seals. The results demonstrate that the proposed identification approach employing the Delaunay tessellation is invariant in displacement and orientation, and robust in a noisy environment.