The images can be rotated and translated and sometimes also scaled. The recognition of objects in color images can be performed by Gaussian-Hermite moment Invariants of color images with respect to rotation. We extend the existing Gaussian-Hermite moment Invariants for graylevel images to be suitable for color ones. We propose two different approaches how to accomplish that: low-order invariants and joint invariants. The obtained invariants for color images significantly improve the success rate, as is demonstrated in three experiments with real data.