Infrared and Visible Image Fusion Combining Information Perception and Multiscale Features

Existing image fusion algorithms based on deep learning are unable to satisfy the demands of computational efficiency and fusion effect. Most have also adopted a fusion strategy based on a single-scale model, which cannot effectively extract the contextual information in images. This study proposes an end-to-end image fusion network based on information perception and multiscale features. The network consists of an encoder, a fusion strategy, and decoder. Specifically, the multiscale features of the infrared and visible images were extracted by the encoder, and a feature complementary enhancement module was designed to fuse different modal multiscale features. Finally, the lightweight decoder was designed to combine the low-level details and high-level semantic information. In addition, the information entropy of the source image was used to construct an information-sensing loss function to train the fusion network and generate the fused image with rich information. The proposed fusion framework was evaluated on the TNO and MSRS datasets. The results show that compared with existing fusion methods, the proposed network was superior to other methods in terms of both subjective visual description and objective index evaluation, with higher computational efficiency.