ICIEA 2026 Workshop

Workshop 4: From Flare Generation to Removal: Restoring Low-Light Scenes from Non-Stationary Degradation

Speakers:

Speaker 1: Wei Wang

Title: Associate Professor

Affiliation: Institute of Advanced Displays and Imaging, Henan Academy of Sciences, Henan Academy of Sciences; Wuhan University of Science and Technology

Summary of workshop:

Lens flare and glow induced by strong light sources significantly degrade image quality in nighttime photography, posing severe challenges to computer vision systems like autonomous driving. Addressing the critical scarcity of paired real-world training data, this talk presents a physics-aware framework titled "From Generation to Removal." We first introduce a Generation module that leverages the Atmospheric Point Spread Function (APSF) and ray tracing to synthesize high-fidelity flare data, effectively overcoming the data bottleneck. Building on these physical priors, we then detail a Removal strategy based on self-supervised feature disentanglement. This approach enables the precise separation of lighting artifacts from background textures in a zero-shot manner, without relying on ground-truth supervision. Experimental results demonstrate that this framework significantly improves image contrast and detail recovery in complex nighttime environments, offering a robust solution for computational photography and intelligent visual perception.