Medical image fusion based on image pre-fusion network and multi-scale side window guided filter

This paper proposes a fusion algorithm based on image pre-fusion network (IPFNet) and multi-scale side window guided filter. The fusion algorithm is divided into two stages: pre-fusion and re-fusion. In the pre-fusion stage, IPFNet is constructed to generate pre-fusion image containing edge information and detail feature of the source image. In the re-fusion stage, a multi-scale side window guided filter is design to decompose the source image and pre-fused image, effectively solving the problem of edge blur. To enhance the brightness of the base layer, a coefficient selection strategy is designed to generate a fusion image of the base layer. In detail layer image fusion, a weight allocation strategy composed of structural similarity function and weighted local energy is designed. Weighted fusion of detail layer image effectively preserves detail information. Through experimental results, it has been verified that the algorithm proposed in this paper can better preserve texture detail and achieve high contrast in fused image compared to the comparative algorithms.