Figure 3.

The architecture of our proposed U-shape multi-focus image fusion network with a Siamese encoder. The Siamese encoder extracts the features of two input images simultaneously, and fuses them through the global perception fusion module in the bridge stage of the network. Each stage of the decoder cascades the feature maps from the corresponding stage of the Siamese encoder and outputs a decision map, which participates in the supervision of the training. ResBlocks are added in each stage of the encoder and decoder.