An automatic registration algorithm for cultural heritage images

In art diagnostics it is often needed to compare and integrate different sets of information, coming from different sources, and stored in different images. In order to successfully integrate these data, images corresponding to the same areas need to be registered, that is a geometrical transformation that aligns points in one picture with corresponding points in another picture needs to be found and applied. In this paper we present an automatic registration technique for multispectral images, based on the computation of mutual information. We applied this method to obtain RGB images from multispectral data, and derive a (undersampled) spectral signature per pixel of the painting.