Abstract: Lens flare is a phenomenon that occurs when light scatters or reflects within an optical system. Existing methodologies primarily concentrate on spatial information, but they still face limitations in effectively addressing pronounced blooms and diversely shaped light spots while preserving image details. We propose a dual-branch flare removal network based on interactive frequency restoration. Specifically, this architecture comprises a frequency domain branch to capture global information and a spatial domain branch to capture local information. We initially observed that lens flare can be more effectively represented in terms of phase and amplitude, with the phase encompassing milder flare patterns and the amplitude being brighter yet substantially immune to flare effects. This insight allows us to design interactive frequency restoration modules in the frequency branch. Guided by the restored global frequency information, our spatial branch strives to reconstruct flare-free images by incorporating global and local fusion modules. Extensive experiments conducted on both synthetic and real-world datasets demonstrate that our method achieves remarkable performance improvements in comparison to state-of-the-art methods.
Loading