MLP-based ImageInterpolation UsingLocalCharacteristic of
Wavelet Coefficients · 2007
Wanatural imagepasses through adigital image acquisition system,itlosesitshigh-frequency components duetolow-pass filtering anddecimation. The loss ofhigh-frequency components brings ablurring effect to animagewhenitisenlarged. Themainissue ofimage interpolation isto recoverthelosthigh-frequency components precisely fromtheobserved spatial orspectral data ofthelow-resolution image. Imageinterpolation inthespatial domainusually uses upsampling, followed by a spatial filter. In spatial interpolation, thefilter coefficients should beappropriately selected inorder topreserve theedges. Themostwidely used spatial filters arethebilinear andbicubic interpolation filters. Theseclassic linear filtering methods havetheadvantage of being fast butsuffer fromserious jaggedness andblur dueto theassumption ofseverely band-limited 2-Dsignal. They alsoassumethatalow-resolution imageonlyconsists of samples fromitshigh-resolution version. A numberof techniques havebeenproposed toovercome these problems [3],[4],[l3]-[l5 ]. A second class ofimageinterpolation methods isthe wavelet-domain interpolators. Wavelet transforms offer the possibility ofpreserving edgesbyvirtue oftheHolder regularity, andarefreefromcontinuity constraint. Inthe wavelet domain, themulti-resolution analysis andstatistical