To solve the problem of poor robustness of distance-based metrics in multi-sensor remote sensing image registration methods， an image registration algorithm based on Kullback-Leibler divergence using variational approximation was proposed. First， edge features were extracted from the infrared image and visible image， respectively. Second， the infrared image features were orthorectified using imaging poses， and Gaussian Mixture Models （GMMs） were constructed for the feature point sets of the infrared and visible images， respectively. Third， the Kullback-Leibler divergence between the two GMMs was calculated using the variational approximation method， in which variational parameters were introduced and the Lagrange multiplier was utilized. Finally， the Particle Swarm Optimization （PSO） algorithm was applied to search for the optimal registration parameters. In the remote sensing image registration experiments， the proposed method’s average Root Mean Square Error of registration parameters is 2.5， and the average runtime is 1.5 seconds. Additionally， the proposed method still achieves correct registration when the variance of Gaussian noise and the salt-and-pepper noise coefficient is 0.07， respectively. These results validate the robustness and high computational efficiency of our method.