Total Variation-based Algorithm For Multi-source Image Fusion

In this era of information technology,the demand for information are getting higher and higher. The information obtained by a single sensor is difficult to satisfy people’s all sorts of reality, must rely on the information provided by various sensors to better solve the problems faced by people. Multi-source image fusion is the comprehensive analysis and processing of image coming from different sensors on a particular scene or target, to produce a new image more conducive to the human eyes observation or further processing. Multi-source image fusion have a wide range of applications in remote sensing, medical images and other fields, both at home and abroad, people put more and more attention on it.This article first introduced the related background of image fusion, and then systematic introduce the basic theoretical knowledge of image fusion, including: classic methods of image fusion at pixel level, image fusion based on the variational, and image fusion evaluation system, etc.In pixel domain, the image can be regarded as the function of BV space, and then the image fusion problem can be regarded as the energy functional problem, then by solving the energy functional minimization problem can get the fusion result. To build the image fusion model, we must first to determine the regularization and fidelity term.In the actual cases, the fusion of the source image with noise, often so the structure of the fusion model should not only consider model of information fusion ability, but also the image denoising problem. Often in actual cases, the source image with noise, so the structure of the fusion model should not only consider the ability of information fusion, but also consider the ability of image denoising. Split Bregman iterative algorithm is a quick convergence algorithm for the convex optimization problem with constraints, it is especially effective for the optimization problem which containing the L1 regularization.In this paper, we use the Split Bregman iteration method to solve the typical TV-L2 model and give its algorithm. By improving the TV-L2 model, we propose two image fusion models: the TV-L1 model and the hybrid TV model, and gives their Split Bregman iteration algorithm respectively. In this paper,the three kinds of algorithms and TV-L2 algorithm based on variational PDE are compared by a large number of experiments, through the analysis of the fusion image and a variety of quantitative evaluation index, we prove the effectiveness of the proposed algorithm.