3D invariants and their application to object recognition
Gilles Burel, Hugues Hénocq · 1995
Although recognition of objects from 2D projections (i.e. images) has been widely studied among the image processing community, little research has been devoted to recognition using 3D information. A general approach for deriving 3D invariants is proposed in this paper. These invariants can be used as input to a statistical classi…er, such as a k-nearest-neighbours algorithm or a neural network. The approach consists of decomposing the object onto an orthonormal basis composed of the eigenvectors of the angular momentum operator from quantum mechanics. Then, using Clebsch-Gordan coe¢cients, contravariant tensors of order 1 are constructed, and 3D invariants are obtained by tensor contraction. The approach