An Affine Invariant-based Match Propagation Method for Quasi-dense Image Registration

Non-rigid models are needed to correct the non-uniform deformation between two images; dense, uniformly distributed control points is the premise of constructing such non-rigid models. This paper puts forward an affine invariant-based match propagation method for quasi-dense image registration. Interest points are extracted from the input image and reference image, respectively; and sparsely matched. According to the sparsely matched points, the global offset between the two images is estimated and corrected. A corresponding adjacent triangle is established using the interest points in the two images interactively; and an affine invariant, the ratio of the triangle area, is employed to determine whether the vertices in the two triangles match or not. Through this method, the spare initial seed matching points propagate to other points of interest to obtain quasi-dense match. The HJ-Landsat image and ETM+ data of Antarctic data are taken in an experiment, the results show that our method can extract dense and uniformly distributed matching points while outperforming other state of art methods in true positive ratio.