Local joint entropy based non-rigid multimodality image registration

• We present a new variational model for image registration.
• The local joint entropy is used to measure the similarity of the images to be aligned.
• The weighted Horn-type regularization term is used to protect displacement fields from over-smoothing.
• The proposed model has the advantage of aligning local edges of images well.
• A fast iteration algorithm is designed to solve our model.

Variational-based image registration is an important research topic in the field of pattern recognition. Classical models for this task usually use mutual information to measure the similarity of the images to be aligned. Although these models can generate good registration results for rigid deformation, they do not perform well for non-rigid registration because of the complexity of local deformation. In this paper, we propose a novel model to solve the problem of non-rigid registration of multimodality images. In the model, the local joint entropy is introduced to measure the similarity of the images to be aligned, and the weighted Horn-type regularizer is used to protect the displacement field to be estimated from over-smoothing. The proposed model has the advantage of aligning local edges of noise-free images better than the model based on mutual information and total variation, and the free-form deformation model. Furthermore, the proposed weighted regularizer is more robust than the classical total variation regularizer and the Horn-type regularizer in the alignment of noisy images. By using the alternative minimization method, we design a fast iteration algorithm to solve our model. Numerical results show the promising performance of our registration method.