Variational Multiscale Image Decomposition Based SAR And Optical Image Fusion

Image fusion technique can combine images that describe the same place at different time or images from different sensors, then the fused image can be finer and more perfect, and it is very useful. In remote sensing area, with the development of remote sensing technology,more and more remote sensing images taken from different sensors or taken at different time are stored. By fusing these image, we can know these places and nature resources much better.Synthetic aperture radar has the all-weather and all-time ability and can penetrate clouds,fog area and ground surface. SAR image has abundant texture and structural information while the optical image is based on spectral information which is easily influenced by weather factors. Taking full use of image technologies, we can get multiple features of objects from source images so as to facilitate the processing of subsequent target recognition and understanding.We propose a variational multiscale image decomposition based SAR and optical image fusion algorithm. Variational multiscale image decomposition believes that an image contains structures with different morphologies and can be accordingly modeled as a superposition of multiple scales of cartoon and texture components. Cartoon part is the contour and edges of image, and texture component contains fine scale-details, usually with some periodicity and oscillatory nature. For this two completely different components,we take different fuse rules. For cartoon part, we take the curvelet transform, because it can describe cartoon, for low frequency coefficients, we take weighted average fusion rule, for high frequency coefficients, we take the maximum coefficients and the fused high frequency coefficients. For texture part, the finest scale of SAR image contains so much noise and contain little information, then we choose to abandon the least two scale texture when we fuse texture components. The fuse scheme of texture part is choosing maximum local energy part of source image texture. The effectiveness of our algorithm has been testified by several groups of real remote sensing images. Compared with traditional image fusion algorithm, the fused image of our algorithm has finer structure, while speckle noise has been suppressed obviously.In the first part of this paper, we introduce basic concepts, history and application area of image fusion, then describes objective evaluation indexes. In the second part, we introduce the classic multilscale image fusion algorithms, such as laplacian pyramid based image fusion and discrete wavelet transform based image fusion algorithm.Variation method and its numerical solution method are introduced in the third part, it is the basis of variational multiscale image decomposition model and its numerical solution, we deeply analysis the variational multicale image decomposition model, then we take the finite difference method to solve this model. In part four, we propose a new SAR and optical image fusion method, then take experiments on MATLAB, the experiment results are much better than traditional multiscale image fusion algorithm. Finally, we summarized the whole paper,and proposed future development possibilities.