Skip to Main Content
We consider the problem of image matching under the unknown statistical dependence of the signals, i.e. a signal in one image may correspond to one or more signals in the other image with different probabilities. This problem is widely known as multimodal image registration and is commonly solved by the maximization of the empirical mutual information between the images. The deformation is typically represented in a parametric form and optimization w.r.t. it is performed using gradient-based methods. In contrast, we represent the deformation as a field of discretized displacements and optimize w.r.t. it using pairwise Gibbs energy minimization technique. This has potential advantage of finding good solutions even for problems having many local minima. In experiments we demonstrate that the proposed method working on a single scale achieves comparable performance to a state-of-the-art multi-scale method.