A GENERICBINARYTREE-BASEDPROGRESSIVEDEMOSAICKING METHOD FOR MULTISPECTRALFILTERARRAY

Thetechnique ofmultispectral filter array (MSFA)isamultispectral extension ofthewidely deployed color filter array (CFA), whichusessingle chipsensors andsubsequent interpolation strategies toproduce full color images. However, multispectral demosaicking presents unique challenges totraditional CFAdemosaicking algorithms whichcannot bedirectly extended todeal withvarious filter arrays withdifferent numbers ofspectral bands ordifferent spatial patterns within eachband.Inaddition, themorespectral bandsinvolved, themoresparse eachbandsamples theimageplane, andthe less information wecanutilize whenperforming theinterpolation. Inthis paper, westudy ageneric MSFAdemosaicking method, whichfollows abinary tree structure todetermine theorder according towhichdifferent spectral bands anddifferent pixel locations within eachbandareprogressively interpolated, making theedgeinformation utilized moreeffectively. Theproposed method isdemonstrated tooutperform three traditional interpolation techniques aswellasthree advanced CFAdemosaicking methods using twoperformance measures, theroot meansquare error (RMSE)forreconstruction fidelity andtheclassification accuracy fortarget recognition performance. IndexTerms-Imagereconstruction, filter array, interpolation, spectral correlation, binary tree