MFIF-GAN: A new generative adversarial network for multi-focus image fusion

Multi-Focus Image Fusion (MFIF) is a promising image enhancement technique to generate all-in-focus images meeting visual needs, and it precondition for other computer vision tasks. One emergent research trend in MFIF involves approaches avoiding defocus spread effect (DSE) around focus/defocus boundary (FDB). This study proposes generative adversarial network tasks called MFIF-GAN, attenuate the DSE by generating focus maps which foreground region correctly larger than corresponding objects. A Squeeze Excitation residual module employed proposed network. By combining prior knowledge of training condition, trained on synthetic dataset based an ?-matte model. In addition, reconstruction gradient regularization terms are combined loss functions enhance details improve quality fused images. Extensive experiments demonstrate that MFIF-GAN outperforms eight state-of-the-art (SOTA) methods perception quantitative analysis, as well efficiency. Moreover, edge diffusion contraction verify generated our method accurate at pixel level.