Event-Assisted Blurriness Representation Learning for Blurry Image Unfolding

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Image deblurring aims to restore sharp images from blurry ones.
• Event cameras offer improved performance for deblurring tasks in dynamic scenes.
• Current methods often overlook effective blur modeling for diverse blur types.

Purpose of the Study:

• To propose an event-assisted approach for implicit blur modeling in image deblurring.
• To enhance image unfolding by integrating learned blurriness representations.
• To improve the recovery of sharp frames from blurry images, especially in dynamic scenes.

Main Methods:

• Developed an event-assisted blurriness encoder to compute blurriness representations.
• Formulated blurriness representation learning as a ranking problem using synthesized data pairs.
• Integrated blur information into a base unfolding network via blurriness-guided modulation and multi-scale aggregation.

Main Results:

• Achieved favorable performance against state-of-the-art methods on GOPRO and HQF datasets.
• Demonstrated effectiveness in recovering latent sharp frames from real-world blurry images.
• Validated the proposed method's ability to handle various blur types in dynamic scenes.

Conclusions:

• The proposed event-assisted method enhances image deblurring by incorporating explicit blur modeling.
• Implicit blur representation learning and integration modules improve the recovery of sharp sequences.
• The approach shows significant potential for real-world applications requiring high-quality image restoration.