Cross-domain correspondence intensity modulation based on Bayesian-decision for remote sensing image pansharpening.

Pansharpening usually improves the resolution of low-resolution multispectral (LRMS) images with spatial information from corresponding high-resolution panchromatic (HRPAN) images to produce high-resolution MS (HRMS) images. Traditional pansharpening methods use various domain transformations to make the fused image suffer varying degrees of spatial or spectral distortion because the information in the LRMS and PAN images is heterogeneous and distributed in different domains. The motivation of our proposed work is to develop a balanced and robust pansharpening method named cross-domain correspondence intensity modulation, which is based on Bayesian decision-making for remote sensing image pansharpening. First, the intensity component of the MS image is obtained via the intensity hue saturation (IHS) transform. Second, a fusion rule based on the Bayesian probabilistic model is designed to fuse the intensity component and the corresponding PAN image to obtain an intermediate component. Third, a cross-domain correspondence intensity modulation algorithm is proposed to modulate the intensity information in the intermediate component to produce the desired intensity component. Finally, an inverse IHS transformation is performed to obtain the pansharpened MS image by replacing the original intensity component with the modulated intensity component. The results on different satellite datasets show that the proposed method can effectively enhance the spatial and spectral fidelity of the fused image.