Infrared and Visible Image Fusion Based on Mutual Structure Extraction

Most existing methods for fusing infrared and visible images do not take into account the complementary and redundant relationships between the source images, resulting in a situation that the common information is improperly retained and the unique information is not adequately retained in the resulting image. To address this problem, this paper introduced an infrared and visible image fusion method based on mutual structure extraction. The mutually guided image filter is utilized to separate the common and unique information between source infrared and visible images firstly. Secondly, the unique information layer is decomposed into multiple scales and information exchange is performed in corresponding infrared and visible detail layers. Then the common information layer is fused based on visual saliency; the unique base layer is fused based on infrared information injection; the unique detail layers are fused based on local visual saliency and weighted average fusion rules. The fusion result generated by image reconstruction not only reasonably selects redundant features and fully retains complementary features but also significantly weaken the impact of noise, which is more convenient for human eyes to observe. Experimental results show that the proposed method outperforms other methods.