Fourier-Based Decoupling Network for Joint Low-Light Image Enhancement and Deblurring

Area of Science:

• Computer Vision
• Image Processing
• Digital Photography

Background:

• Nighttime handheld photography suffers from simultaneous low light and blur degradations.
• Existing spatial domain methods struggle to decouple these interdependent degradations, limiting performance.
• The Fourier domain offers a new perspective for independent degradation representation.

Purpose of the Study:

• To develop a novel method for joint low-light image enhancement and deblurring.
• To leverage the Fourier domain for effective decoupling of low light and blur.
• To improve the quality and sharpness of nighttime handheld photographs.

Main Methods:

• Analysis of low light and blur degradations in the Fourier domain (amplitude and phase).
• Derivation of a frequency attention mechanism and a filtering mechanism.
• Proposal of a Fourier-based Decoupling Network (FDN) for joint enhancement and deblurring.

Main Results:

• The proposed FDN method achieves state-of-the-art performance on synthetic and real-world datasets.
• Demonstrated effective decoupling of low light and blur degradations in the Fourier domain.
• Significantly sharper edges and improved overall image quality in enhanced nighttime photos.

Conclusions:

• The Fourier domain provides an effective representation for decoupling low light and blur.
• The proposed frequency attention and filtering mechanisms enable joint enhancement and deblurring.
• The Fourier-based Decoupling Network offers a superior solution for challenging nighttime photography conditions.