A High Visibility and SNR Image From One Single-Shot Low-Light Image

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Low-light imaging presents challenges in achieving high visibility and signal-to-noise ratio (SNR) due to under-constrained problems.
• Existing methods struggle to effectively enhance single-shot images captured in suboptimal lighting conditions.

Purpose of the Study:

• To develop a novel algorithm for enhancing single-shot low-light images.
• To improve both image visibility and signal-to-noise ratio (SNR) for better perceptual quality.

Main Methods:

• Established intrinsic relationships between image and radiance domains using human visual, atmospheric scattering, and camera imaging models to derive ideal exposure.
• Introduced a convex optimization approach with gradient constraint and the Krisch operator, utilizing illumination-reflection-noise priors to estimate noise-reduced components.
• Generated a high-SNR image in the radiance domain at optimal exposure and inversely mapped it to the image domain.

Main Results:

• The proposed algorithm successfully generates a high-SNR image in the image domain from low-light conditions.
• Experimental results demonstrate superior performance in subjective and objective tests compared to state-of-the-art methods.
• The enhanced images exhibit improved visibility and a pleasant perceptual quality.

Conclusions:

• The developed method effectively addresses the under-constrained problem of low-light image enhancement.
• The algorithm achieves a high signal-to-noise ratio (SNR) and enhanced visual perception.
• This approach offers a significant improvement over existing techniques for single-shot low-light image enhancement.