An unsupervised multi?focus image fusion method based on Transformer and U?Net

This work presents a multi-focus image fusion method based on Transformer and U-Net with an unsupervised training fashion. In this work, the authors introduce into because it has great ability to capture global dependencies low-frequency features. processing, convolutional neural network (CNN) good performance of detailed feature extraction but weakness for extraction, limited power in local or information strong capacity extraction. Thus, combines advantages CNN propose decision map making model joint U-Net. The construct including reconstruction modules which correspond encoder decoder U-Net, respectively. addition, perceptual loss is introduced basis structural similarity combination these two functions can achieve better lower cost. Experiments show that proposed performs compared existing methods.