An automatic registration technique for Cultural Heritage images

Very often, diagnoses in cultural heritage field, as well as in the medical field, have to be accomplished by comparing images obtained with different techniques and sensors. However, acquired images are often slightly misaligned, and in order to exactly compare them, they need to be correctly registered among each other. We present in this paper an automatic registration technique to determine the correct displacement (that is a geometrical transformation including sub-pixel translation, rotation and scaling) to align points from one image with corresponding points coming from another one of the same object or scene. The proposed registration technique is based on the computation of the mutual information, which is a similarity measure coming from the information theory. Mutual information is a measure of the amount of information one image contains about the other one. It is a highly performing similarity measure, when compared to previously proposed ones, such as crosscorrelation, which can often fail when dealing with multi-source images (i.e. images coming from different sensors or regarding different frequency bands), for the inherently difference of the image structures and tone dynamics.