To alleviate the influence of the unstability of extracting some feature points caused by affine transforms, new affine invariants for target recognition were proposed, which are based on feature points and the boundary that is almost unaffected by affine transforms. New invariants depend on the current feature point and the whole boundary when the target is not occluded and on the current feature point, two directly neighboring ones and the boundary between them when the target is occluded. New invariants depend on the least feature points among the affine invariants based on feature points, which helps to apply them to the situation that the extraction of feature points is unstable.