Abstract: In the fields of computer vision and pattern recognition,moments and moment invariants are commonly used invariant features of images.The existing moment invariants of color images to both geometric deformations and illumination changes are constructed based on geometric moments.Thus,they are sensitive to noise.To address this issue,we first propose a method to construct orthogonal Gaussian-Hermite moment invariants of color images to rotation-affine transform by means of fundamental differential operators and color geometric primitive.Then,we generate all possible invariants with low orders and low degrees,and derive thirteen linearly independent invariants from them.Finally,based on synthetic images and HPatches image database,the experiments are carried out to test the numerical stability of thirteen invariants,and to evaluate the performance of them in image classification and template matching.The results show that these invariants have good stability and robustness to noise.The classification and matching accuracy rates from using them are 10%and 30%higher than the same types of moment invariants,respectively.