Image data fusion using projection over the set of solutions of an underdetermined least squares problem

This paper presents a new image data fusion method that aims to combine images with different spatial and spectral resolutions. It is based on a projection onto the subspace of the set of solutions of an underdetermined least squares problem. The data fusion problem may also involve an interpolation process, when images with different spatial resolutions are used. In our case, this interpolation was performed by using a Bayesian approach, with or without spatial adaptation. As an example, the method is applied to SPOT images over Sao Jose dos Campos, SP, Brazil.