Feature-Based Deformable Image Registration with RANSAC Based Search Correspondence

The paper presents an algorithm for deformable image registration based on point features extracted from input images using the Harris corner detector. The correspondence between the points extracted from the different images is established using RANdom SAmple Consensus (RANSAC) method with the affine and perspective global transformation used to model the deformations. The initial correspondence is established using an enumerative search with rotation invariant cross-correlation measure. Based on the estimated corresponding set of points the further refinement of the displacement field is achieved through application of the multilevel B-spline technique applied to the overlapping image area. The performance of this method is demonstrated using few pairs of remote sensing images. The method is also evaluated against the global polynomial transformation and the registration results are compared with the results achieved using ENVI software.