Automated Multisensor Registration: Requirements and Techniques

A necessary first step in the fusion of data from a number of different remote sensors is the correction of the systematic geometric distortion characteristic of each sensor followed by a precision registration to remove any residual random offsets. This paper describes our approach to automated multisensor registration. The effects of spatially, temporally and spectrally varying factors which influence image dynamics are reviewed. A specification of the requirements for an operational algorithm is formulated using these factors. Additionally, the structure of an efficient, automated system is defined. A number of candidate image processing techniques are evaluated within this structure using a multisensor test data set assembled from the Landsat TM, SEASAT and SPOT sensors. The results are presented and discussed.