Bayesian fusion of multivariate image series to obtain depth information

This contribution presents a fusion method for multivariate stereo and spectral series with the purpose of obtaining 3D information. The image series are gained using a camera array with spectral filters. In order to register them, features that are invariant with respect to the intensity values in the images are extracted. The fusion approach is region based and uses characteristics like their size, position and form for registration. Regions are identified using the watershed transformation. The fusion problem is modeled by means of energy functionals and solved by applying a standard minimization algorithm. A generalization of the fusion problem is obtained by connecting it to the Bayesian fusion framework. An example of a reconstructed scene is given, showing the potential of the implemented algorithm.