3D Recognition Using Affine Invariants

The size of 3D data stored around the web has become bigger. Therefore the development of recognition applications and retrieval systems of 3D models is important. This paper deals with invariants for 3D models recognition. Thus under general affine transform we propose in a first time determinants of three points to realize invariance under affinity. To solve starting point problem we needs Fourier Series (FS) to extract affine invariant descriptors, called Fourier Series Descriptor (FSD). The difference between first and second approaches: in first approach determinants are computed on cartesian coordinates directly while in the second one determinants are computed from FS coefficients to eliminate dependency on starting point. The FS are also applied on 2D slices to generate affine invariants for 3D volume. FS can be computed based on hole points of volume, but this technique. The principal advantages of proposed approaches are the possibility to handle affine transform and 3D volume. Two types of 3D objects are used in the experimentations: mesh and volume, the Princeton Shape Benchmarek (PSB) is also used to test our descriptor based on FSD.