SPECTRAL INFORMATION EXTRACTION BY MEANS OF MS+PAN FUSION

摘要

This paper deals with multi-spectral (MS) and panchromaticrn(Pan) image fusion based on redundant multiresolutionrnanalysis (MRA). Lower-resolution MS bandsrnare sharpened by injecting high-pass details takenrnfrom the higher-resolution Pan image. Crucial point isrnmodelling the relationships between detail coefficientsrnof a generic MS band and the Pan image at the samernresolution. Once calculated at the coarser resolution,rnwhere both types of data are available, such a modelrnshall be extended to the finer resolution to weight the Panrndetails to be injected. Two injection models embeddedrnin an "`a trous" wavelet decomposition will be describedrnand compared on a test set of very high resolutionrnQuickBird MS+Pan data. One works on approximationsrnand provides a partial unmixing of coarse MS pixelsrnvia high-resolution Pan. Another is based on spectralrnfidelity of the merged image to the original MS data.rnFusion comparisons on spatially degraded data, ofrnwhich higher-resolution true MS data are available forrnreference, show that the former yields better results thanrnthe latter, in terms of both spatial and spectral fidelity.