A novel infrared and visible image fusion method based on multi-level saliency integration

Infrared and visible image fusion makes full use of abundant detailed information of multi-sensor to help people better understand various scenarios. In this paper, a novel method of infrared and visible image fusion based on multi-level saliency integration is proposed. First, the background image of each sub-image is reconstructed by the means of Bessel interpolation after the quadtree decomposition on the infrared image, and the difference saliency is extracted by the difference between the source infrared image and the estimated background. Then, the sparse saliency is calculated from the infrared image using the sparsity of salient objects and the low rank of background. Third, the multi-scale saliency is obtained by Laplacian transformation between the visible image and infrared image to preserve the detailed information. At last, the fusion strategy based on the adaptive weighting coefficient is present to get more natural fusion results. Experimental results on 20 pairs of source images demonstrate that the proposed method outperforms the other state-of-the-art methods in terms of subjective vision and objective evaluation.