Abstract: Fusion of thermal and visible images has a large research and application value due to their complementary benefits, which can overcome shortcomings of single modality under certain conditions. This paper adopts a sparse representation based algorithm to integrate thermal and visible information. Firstly, a relative complete dictionary is learned by some image samples. Secondly, given an image pair, this paper represents them on the learned dictionary by the improved sparse representation model, in which the Laplacian constraints on reconstructed coefficients are employed to improve its robustness. Then, the two modal images are integrated based on the constructed coefficients. Finally, extensive experiments on the public images and the collected images suggest that the method proposed in this paper can effectively fuse the information of two modalities.

Key words: multi-modal fusion, sparse representation, Laplacian regularization

摘要： 融合热红外与可见光图像能够达到信息的互补，弥补单一模态在某些条件下的不足，因此具有较高的研究和应用价值。采用了一种基于稀疏表示模型的热红外与可见光图像融合算法。首先，根据一定量图像样本学习出较为完备的字典。其次，对于给定的两模态图像对，通过稀疏表示模型在学习出的字典上分别对其进行稀疏表示。同时，为了提高鲁棒性，使用了拉普拉斯约束对表示系数进行正则化。然后，根据融合算法对两模态图像进行有效融合。最后，在公共的图像以及收集的图像上进行了实验，实验结果表明，该算法能够有效地融合两模态图像的信息。

关键词: 多模态融合, 稀疏表示, 拉普拉斯正则化