Local similarity measures for demons-like registration algorithms

The main limitation of the original demons image registration algorithm is the need to ensure that the grey level of corresponding voxels remains the same in both the images to register. This difficulty may be overcome by the use of more robust similarity measures, whenever they may be expressed as the sum over the entire image of a local energy. We demonstrate the convenience of a local optimisation strategy of this energy instead of the optimisation of the global metric, and suggest a methodology for the use of more general similarity measures.