This paper proposes a new non-rigid image registration model based on the combination of statistical and variational methods. Assuming that the residual image obeys a local Gaussian distribution with different means and variances, a dual energy functional is obtained. Combined with the variational regularization method, a new registration model is obtained in this paper. The novelty of this method lies in the introduction of weighting functions and some control parameters in the fidelity term. The weighting functions can automatically and effectively distinguish regions with different grayscale contrasts in residual image, and the control parameters improve the robustness of the algorithm. The registration results of synthetic images, two-dimensional lung CT, and three-dimensional brain MRI images demonstrate the effectiveness and accuracy of this method.