Methodological Proposal for Multispectral Stereo Matching

Remote sensors have begun to capture digital stereoscopic data. Although still monospectral (usually panchromatic), the capture of multispectral or hyperspectral stereoscopic data is just a matter of time. Digital photogrammetric workstations use area-based stereo-matching techniques based on the Pearson (product-moment) correlation coefficient. This is a technique that is not intended to take advantage of the multispectral data. The authors propose a new method that 1) can handle this multispectral information and 2) can take advantage of the spatial relations between pixel locations. The method is based on multidimensional scaling and Procrustes analysis. Our results indicate that the proposed new technique renders more robust results than classical methodology when noise in the original data is introduced.