Context-driven fusion of high spatial and spectral resolution images based on oversampled multiresolution analysis

Abstract

This paper compares two general and formal solutions to the problem of fusion of multispectral images with high-resolution panchromatic observations. The former exploits the undecimated discrete wavelet transform, which is an octave bandpass representation achieved from a conventional discrete wavelet transform by omitting all decimators and upsampling the wavelet filter bank. The latter relies on the generalized Laplacian pyramid, which is another oversampled structure obtained by recursively subtracting from an image an expanded decimated lowpass version. Both the methods selectively perform spatial-frequencies spectrum substitution from an image to another. In both schemes, context dependency is exploited by thresholding the local correlation coefficient between the images to be merged, to avoid injection of spatial details that are not likely to occur in the target image. Unlike other multiscale fusion schemes, both the present decompositions are not critically subsampled, thus avoiding possible impairments in the fused images, due to missing cancellation of aliasing terms. Results are presented and discussed on SPOT data

Pansharpening by Convolutional Neural Networks

“…Therefore, to reduce the possible mismatches, we smooth data before downsampling, following [2,22], using a filter that matches the modulation transfer function of the sensor. Likewise, we upsample the MS component using the interpolation kernel proposed in [13]. In any case, great attention must be devoted to assess performance not only on the low-resolution data but also on the original images.…”

“…The smooth version can be computed by means of a simple low-pass filter on a single level decomposition [11] or through a multiresolution analysis (MRA) on several decomposition bands [12]. These last methods rely on redundant representations, characterized by the shift-invariance property, like the á Trous Wavelet transform (ATWT) or the Laplacian pyramid (LP) [13]. In order to better capture directional information, methods based on non-separable transforms, like curvelets, have also been developed [14].…”

Effects of the spectral properties of a panchromatic image on pan-sharpening simulated using hyperspectral data

“…The Block-SVR and GLP-SDM were similar but displayed a different trend from that of the GS transformation. This was because these two methods have a similar sharpening scheme based on the injection of spatial details, which are weighted by the ratio of PAN and MS bands (Aiazzi et al 2002, 2005, Zhang 2010a). The GS transformation had narrow dynamic ranges in ERGAS, WSNR and UIQI.…”

“…This method applies the GLP, one method of multi-resolution analysis, to extract the spatial details from PAN (Aiazzi et al 2002(Aiazzi et al , 2005. Subsequently, the spatial details are injected based on the SDM scheme.…”

Pansharpening using total variation regularization

“…Classical pansharpening techniques can be divided into two categories [5]: the component substitution methods [6,7,8] and the multiresolution-analysis methods [9,10]. The component substitution methods first make use of a projection of the upsampled MS images to obtain a better representation of which one component contains most of the image structures.…”