Image registration method of dense and sparse feature matching binding

The present invention provides an image registration method of dense and sparse feature matching bonded, dense and sparse feature matching combination, to obtain a new mathematical model, the new model comprising two variables: the geometry transformation and discrete non-rigid displacement flow field, the former applies to sparse matching flow, we have introduced a partial constraint to adjust the linear transformation, so that the problem is posed, which is applicable to dense matching stream, and taking similar SIFT flow model, while using belief propagation algorithm for solving optimization , it is possible for matching an unknown remote sensing image comprising non-rigid motion is obtained between the pixel accurate comparison.