A fast and robust image registration algorithm

An image registration algorithm combining with local regions and feature points is put forward. An improved zero-crossing edge detection replacing Laplace-of-Gaussian operator with Mexican-hat wavelet is carried out to achieve local regions of image and preliminary registration is conducted. Relying on the feature extraction method for Mexican-hat filter response in different scale space, feature point extraction is conducted. Finally, the transformation matrix is achieved by grouping these feature points according to regions for registration and choosing the best registration result from each group, the strategy of the matching process is named grouped sample consensus. Combined with the advantages of the two methods, registration efficiency is improved on the basis of guaranteeing the precision.