Regularization Methods In Multi-Source Remote Sensing Image Fusion

Multi-source remote sensing image fusion is a method to combine various features acquired through different sources into a single one, which contains all significant charac-teristics. It has been widely used in many fields, and become a hot research topic. The variational fusion model basically consists of the regularization and fidelity term, usually the regularization is based on certain assumption of smooth. In this thesis, we thorough discussed the regularization methods in variational fusion model and proposed three varia-tional fusion models. Main work includes the following aspects:1) As for reducing the edge over-smoothing, we propose a L1 regularized scheme for variational image fusion. First of all, we us a gradient field to describe the features of all input images, and construct the gradient of fused image by using a weighted sum of the input gradients. Secondly, we establish an variational fusion model, using the L1-norm as the regularization, combined with the gradient enhancement as well as taking into account the image properties such as brightness and uniformity, to improve the edge preserving ability. We implement the model using the augmented Lagrangian method. Experimental results show that the proposed algorithm obtains remarkable results.2) As for overcomimg the disadvantages produced by single L2 or L1 regularization, This thesis proposes an adaptive regularized scheme for variational image fusion. Firstly, we proposed an efficient algorithm of image edge detection to distinguish the edge and non-edge of an image; then we proposed an adaptive regularized scheme for remote sensing image fusion based on variational method, which can automatically choose the L2-term for non-edge and L1-term for edges. To implement the algorithm, we use the steepest descent method to obtain the solution. Experimental results show the efficiency of edge preserving, and the reduction of over-smoothing or staircase effects.3)As for reducing the staircase effect produced by one order smooth assumption, we propose a variational fusion scheme based on total generalized variation (TGV) and percep-tual enhancement. Thesis discusses the convergence of the model and applys the alternating direction method of multipliers (ADMM) to achieve the solution. Qualitative and quanti-tative analysis of experimental results show that, the proposed algorithm achieves better performance than the compared methods containing L1-or L2-norm regularization.