Fusion of synthetic aperture radar and visible images based on variational multiscale image decomposition

Abstract

Multisensor image fusion can be used as an advanced technique for image enhancement. We propose a synthetic aperture radar (SAR) and visible image fusion method under the framework of a variational multiscale image decomposition (VMID) model. In the proposed method, two fusion rules are respectively designed to fuse the structure and texture components obtained with a VMID model of source images. A fusion rule based on the curvelet transform is employed for fusing the structure component and the local energy criterion is adopted to construct the texture component. Moreover, considering the influence of speckle noise present in an SAR image, its first two texture components are skipped. The final fused result is composed by the obtained structure and texture components. The experimental results on several pairs of SAR and visible images demonstrate the effectiveness of the proposed method. Compared with the conventional image fusion methods, our fused results have a finer structure and are more robust against speckle noise.