Infrared and visible image fusion methods based on feature decomposition are able to generate good fused images. However, most of them employ manually designed simple feature fusion strategies in the reconstruction stage, such as addition or concatenation fusion strategies. These strategies do not pay attention to the relative importance between different features and thus may suffer from issues such as low-contrast, blurring results or information loss. To address this problem, we designed an adaptive fusion network to synthesize decoupled common structural features and distinct modal features under an attention-based adaptive fusion (AAF) strategy. The AAF module adaptively computes different weights assigned to different features according to their relative importance. Moreover, the structural features from different sources are also synthesized under the AAF strategy before reconstruction, to provide a more entire structure information. More important features are thus paid more attention to automatically and advantageous information contained in these features manifests itself more reasonably in the final fused images. Experiments on several datasets demonstrated an obvious improvement of image fusion quality using our method.

Infrared and visible image fusion involves integrating complementary or critical information extracted from different source images into one image. Due to the significant differences between the two modality features and those across different scales, traditional fusion strategies, such as addition or concatenation, often result in information redundancy or the degradation of crucial information. This letter proposes a multilevel adaptive attention fusion network to adaptively fuse features extracted from different sources. Specifically, we introduced an Adaptive Scale Attention Fusion (ASAF) module that uses a soft selection mechanism to assess the relative importance of different modality features at the same scale and assign corresponding fusion weights. Additionally, a guided upsampling layer is utilized to integrate shallow and deep feature information at different scales in the multi-scale structure. Qualitative and quantitative results on public datasets validate the superior performance of our approach in both visual effects and quantitative metrics.

Dear Editor, This letter presents a normalization mechanism to effectively fuse infrared and visible images in an encoder-decoder network. Source images are decomposed into source-invariant structure and source-specific detail features. Then, the information of detail features is sufficiently incorporated into the structure features using this normalization mechanism in the decoder, which generates high-contrast fused images with highlighted targets and abundant texture information. Qualitative and quantitative experiments on two challenging datasets demonstrate the superiority of our method over current state-of-the-art methods.