Multi-sensor image fusion has its effective utilization for surveillance and navigation. It provides a way to merge multisensor imagery by combining the outputs of different imaging sensors. In this paper, we utilize a variational approach to fuse images from different sensors, in order to enhance visualization for surveillance. Energy functional is established in a contrast vector field and a successive over-relaxation method is utilized to solve a Poisson equation. Experimental results show that the variational approach is robust and effective, regardless of time-consuming.