Infrared and Visible Image Fusion with Nuclear Norm Activity Level Measurement

Image fusion produces a single image from numerous images with complementary information. Infrared images collect information on the thermal distribution of the scene, whereas visible images collect textural information. The fusion of these images creates images with thermal and textural details suitable for night-vision cameras and surveillance applications. The proposed auto encoder network with selected residual paths extracts the salient features from the images and then combines them using the nuclear norm's optimization effectiveness. The combined images are created with 5 CNN layers with a 3 x 3 filter size, and the fused output retains more information from both inputs. The suggested algorithm generates images with improved objective evaluation metrics with values of 6.89971 for entropy, 0.76133 for structural similarity, 3.83682 for mutual information, and 0.91325 for feature mutual information. The model outper- forms similar models for the fusion, and the algorithm is suitable for other fusion problems.