Multi-scale image description with rotation invariants of Gaussian-Hermite moments

In this paper, the rotation invariants based on orthogonal Gaussian-Hermite moments are derived and a multi-scale image description by the invariants is proposed. An interesting fact is that these invariants can be obtained by substituting Gaussian-Hermite moments for the geometric ones in a rotation invariant derived from the geometric moments. So we could have a new set of rotation invariants from orthogonal Gaussian-Hermite moments. The experimental results show that they have a good numerical stability. Moreover, the Gaussian-Hermite rotation invariants can be used as efficient tools for multi-scale analysis.