Dense Image Point Matching through Propagation of Local Constraints
We present a conceptually simple algorithm for dense image point matching between two multi-modal (e.g. color) images. The algorithm is based on the assumption that correct image point matches satisfy locally a particular statistical distribution. Through an iterative evaluation of a local probability measure, global constraints are taken into account and the most likely set of image point matches is found. An advantage of this approach is that no information about the camera geometries, as for example the epipoles, has to be known. Therefore, the algorithm may be used for stereo matching and optic flow.