Accurate multi-spectral image registration based on scale invariant feature

Aim at automatic registration of temporal images, multispectral images and multisensor images with rigid affine displacement, a new registration algorithm based on scale invariant local feature matching is presented in this paper. The proposed method build optimized local feature descriptors according to the character of remote sensing images and obtain the correct set of corresponding points while minimizing the points set feature distance on the basis of the relative main orientation and scale ratio; muliti-data sample consensus algorithm is adopted to attain final registration while the coordinates of the corners are further refined. Experimental results confirm that the correct rate and registration accuracy are substantially improved by the proposed algorithm.