REGISTRATION OF MULTIMODALITY MEDICAL IMAGES USING A REGION OVERLAP CRITERION

Matching images is an important process allowing one to relate pieces of information available in several images of a given scene. It involves two steps: registration, in which image differences induced by scene and detector positions are discarded and comparison, in which image differences induced by actual scene modifications are estimated. Registration and comparison techniques depend on image content; thus no general methodology for matching images has been proposed. We present a unified description of existing registration methods. A model of image information is proposed and applied to registration, leading to a formulation of the registration problem in agreement with the intuitive view of it. We define a new criterion based on the above formulation, which maximizes the overlap of registrable regions in the image pair. Finally, an experimental evaluation of the proposed criterion (for multimodality medical images) is reported. © 1992.