Monochromatic Event Guided Image Deblurring with Event-Triggering-Aware Decomposition

Event cameras have been applied to restore sharp images from blurry ones with high-temporal-resolution events. However, these event cameras typically record in monochromatic luminance space, leading to significant color information loss. Addressing this challenge, this pa-per introduces EvTraDe-Net, a novel framework for color image deblurring. It uniquely decomposes color images into luminance terms, aligned with event streams from the event-triggering process, and color terms that capture vari-ations across color channels. EvTraDe-Net employs the event fusion for luminance deblurring and an edge map guidance for color compensation, and mutually utilizes the luminance deblurring to optimize the image decomposition in a task-oriented manner. It effectively reduces shape dis-tortion and color bleeding, as validated on both synthetic data and our new EvRGB-Deblur dataset.