Rotation invariants from Gaussian-Hermite moments of color images

The topic of the paper is recognition of objects and patterns in color images regardless of their position, orientation, and scale. Gaussian-Hermite moment invariants designed especially for color images are introduced in this paper. We extend the existing invariants for graylevel images and show that in the case of color images there exist additional independent invariants, which can be constructed as joint invariants from cross-channel moments and/or from new non-trivial low-order moments. The experiments on real data confirmed that the new invariants improve the recognition rate.