Image Recognition Using New Set of Separable Three-Dimensional Discrete Orthogonal Moment Invariants

In this paper, we propose new sets of 3D separable discrete orthogonal moment invariants, named Racah-Tchebichef-Krawtchouk Moment Invariants (RTKMI), Racah-Krawtchouk-Krawtchouk Moment Invariants (RKKMI) and Racah-Racah-Kr-awtchouk Moment Invariants (RRKMI), for 3D image recognition. The basis functions of these new sets of moment invariants are represented by multivariate discrete orthogonal polynomials. We also present theoretical framework to derive their Rotation, Scaling and Translation (RST) invariants based on the 3D geometric moment invariants. Accordingly, the performance of these proposed separable moment invariants is evaluated under heterogeneous databases and through several appropriate experiments, including 3D image invariance against geometric deformations, local feature extraction, computation time and recognition accuracy, in comparison with the traditional moment invariants. The obtained results showed that our proposed separable moment invariant are very efficient in terms of object recognition, numerical stability and local feature extraction, and can be highly useful for computer vision applications.