Summary: We present numerical implementation of a new variational approach to non-rigid registration of images. Our method is based on adjusting divergence and curl of image displacement field, which determines pixel correspondences. An image registration procedure might be decomposed into three major components: the problem statement, the registration paradigm and the optimization procedure. Our method minimizes a similarity metric such as the sum of squared differences and uses Lagrange multipliers method to solve the optimization problem governed by the divergence and curl equations. In our implementation, the minimization problem is reduced to solving Poisson equations. A finite-difference multigrid strategy is used to solve these Poisson equations. Computational experiments demonstrate the promising potential of our registration method.