A Novel Multi-focus Fusion Network for Retinal Microsurgery

Retinal microsurgery requires high precision. Due to the limited depth of field (DOF) of the ophthalmic microscope and eyeball’s spherical construction, doctors observe the retina with partially in focus and partly out of focus. To solve this problem, we propose a deep-learning-based multi-focus fusion model to reconstruct an all-in-focus image. A focus mea-sure block (FMB) is proposed to obtain the focus area in an image, and a fusion network (FN) is adopted to fuse the selected focus areas to produce the all-in-focus image. Considering the characteristics of retinal images, we propose to adopt two new losses to constrain our network. Based on our in-house dataset, extensive experiments prove the effectiveness of our algorithm.