A Dense Network Based Framework for the Fusion of Infrared and Visible Images with Edge Extraction

Image fusion of the infrared (IR) and visible images has become one of the essential pre-processing steps for image analysis in medical, defense, remote sensing, etc. Till now, the spatial and transform domain methods have produced fused images that are not of good quality. Also, deep learning-based techniques are mostly concerned with feature extraction from the source images before fusion, but the distinct features of the source images have been overlooked. In the proposed model an attempt has been made to extract the edges of the IR image followed by the feature extraction of the IR and visible images using a dense-network based convolutional neural network. The feature maps from the two pipelines are concatenated along with the edge-extracted IR image and the enhanced visible image to obtain the final fused image. The proposed model has been tested on the TNO image dataset. The quantitative and qualitative results show that the proposed method outperformed when compared with the state-of-the-art method in the entropy content and also produces fused images with better human perception quality.