Color-based object indexing and matching is attractive since color is an efficient visual cue for characterizing an object. However, different illuminations will bring in color deformation of objects so as to degrade the performance of recognition. In order to circumvent this confounding influence, we present an effective color invariant descriptor that is considerably insensitive to the variations of light conditions, object geometric transformation, and image blur level. The descriptor is a feature vector consisting of several two dimensional normalized moment invariants of color pixels' distribution with different orders in a new color coordinate. The new color coordinate is introduced on the diagonal-offset model. The experiments on real image databases in terms of recognition rate show that our method is robust and performs very well under various situations.