Invariant object representation and recognition using Lie algebra of perceptual vector fields
This paper presents a global method to represent objects invariantly under Euclidean motions using Lie algebra of perceptional vector field of the objects. We focus on the linear Lie sub-algebra of the tangent or normal Lie algebra of objects and use pure local information in these Lie algebra to represent global shapes. It is shown that this simple subalgebra can represent algebraic shapes and a much wider class of non-algebraic shapes as well. In this way, an occlusion-robust and fast recognition method is derived.
KeywordsObject Representation Infinitesimal Generator Global Shape Normal Algebra Normal Vector Field
- 1.Mundy, J.L. and Zisserman A.(ed) Geometric invariance in computer vision MIT Press, Cambridge Ma. 1982.Google Scholar
- 2.D. Forsyth, J. L. Mundy, A. Zisserman “Invariant descriptors for 3-D object recognition and pose'd, IEEE Trans.. PAMI, vo113, No. 10, 971–991, October (1991)Google Scholar
- 4.E. J. Pauwels, T. Moons, L. J. Van Gool, P. Kempenaers and A. Oostserlinck, “Recognition of planar shapes under affine distortion” Int. Journal of Computer Vision, 14, 48–51 (1995)Google Scholar