A new kind of geometric invariants is proposed in this paper, which is calledaffine weighted moment invariant (AWMI). By combination of local affinedifferential invariants and a framework of global integral, they can moreeffectively extract features of images and help to increase the number oflow-order invariants and to decrease the calculating cost. The experimentalresults show that AWMIs have good stability and distinguishability and achievebetter results in image retrieval than traditional moment invariants. Anextension to 3D is straightforward.
展开▼
