Multispectral image data fusion using projections onto convex sets techniques

The problem of image data fusion includes several techniques for data integration from different sensors about a given object. The process can be divided in two stages: (a) pixel interpolation on each observed image to equalize their resolutions; and (b) synthesis of the fused image using the interpolated images. This work proposes the use of projection onto convex sets (POCS) techniques in both stages of the image data fusion process: on the interpolation stage, by creating intermediate pixels more adapted to the local characteristics of the observed images; and on the synthesis stage, by generating a final multispectral image that incorporates the best spatial and spectral characteristics of the initial images. As an example, the algorithm is applied to Spot satellite images.