Affine Invariant Image Comparison Under Repetitive Structures

We focus on the problem of affine invariant image comparison in the presence of noise and repetitive structures. The classic scheme of keypoints, descriptors and matcher is used. A local field of image gradient orientation is used as descriptor and two matchers are proposed, based on the a-contrario theory, for handling repetitive structures. The affine invariance is obtained by affine simulations. The proposed methods achieve state-of-the-art performances under repetitive structures.