Skip to Main Content
Shape retrieval and object recognition are two important tasks in pattern recognition society. Whether they can be achieved or not owes a lot to the performance of a descriptor primarily. In this paper, a new generalized affine moment invariants called illumination invariant MSA moments is proposed. This method combines the traditional AMI, MSA, MSA moment and the basic ideas used to construct illumination invariants. The results suggest that the new technique offers well discriminative power equivalent to the MSA in shape retrieval task, but is much more robust under illumination distortion and Gaussian noise. In 3D object recognition task, the proposed method outperforms both comparison methods under strong non-affine transformation.