Mutual information aspects of scale space images

摘要

In image registration, mutual information is a well-performing measure based on principles of uncertainty. Similarly, in image analysis the Gaussian scale space, based on minimal assumptions of the image, is used to derive intrinsic properties of an image. This paper starts an investigation of a combination of both methods. This combination results in a double parameterized mutual information measure using local information of the image. For single modality matching best response is found for coinciding parameters. Then critical values are found for which the parameterized mutual information has extrema. First results on multi-modality matching show that different parameter values instead of coinciding values yield the best response for the parameterized mutual information.