Improved consistent elastic image registration based on landmark correspondence

Consistent image registration jointly estimates the forward and reverse transformation between two images while constraining these transforms to be inverses of one another. The consistent image registration produces mapping functions that have less pair-wise registration error and reflect the anatomical relationship of image transformation. This paper improves the consistent image algorithm based on landmark proposed by Johnson. After the convergence of consistent image registration proposed by Johnson, our method inserts new landmarks at the position where has large consistent error, in order to reduce the consistency error as a whole. Where and how to insert new landmarks are discussed in details in this paper. The Experimental results demonstrate that our algorithm is effective to reduce the consistent error and improve the registration accuracy concurrently.