Restoring extremely low-light light fields via decomposition-enhancement and frequency-pixel fusion

Area of Science:

• Computer Vision
• Image Processing
• Optics

Background:

• Light field (LF) technology captures spatial and angular light information.
• Extreme low-light conditions pose significant challenges for LF image restoration.
• Simultaneous processing of complex degradations in low-light LF images is difficult.

Purpose of the Study:

• To propose a novel decomposition-enhancement method for low-light light field restoration.
• To address the difficulties in processing complex degradations under extreme low-light conditions.
• To achieve effective frequency-pixel domain information fusion for enhanced LF images.

Main Methods:

• A decomposition-enhancement method leveraging frequency space properties.
• Partial decomposition of luminance and noise for frequency-pixel domain fusion.
• Decoupling the task into brightness adjustment, noise suppression, and detail refinement.
• Utilizing distinct frequency-stage branches for brightness and noise.
• Employing cross-attention and spatial-angular attention modules for feature and detail enhancement.

Main Results:

• The proposed method achieves competitive performance against state-of-the-art techniques.
• Demonstrated effectiveness in enhancing light field images under challenging low-light conditions.
• Successful integration of frequency-enhanced outputs with original low-light inputs.

Conclusions:

• The novel decomposition-enhancement method effectively restores low-light light field images.
• The frequency-pixel domain fusion approach offers a robust solution for complex degradations.
• The method shows significant promise for applications requiring high-quality LF imaging in low light.