A new training-free method for enhancing low-light images has been developed, combining local illumination estimation, Retinex division, and edge-preserving denoising. This approach utilizes a conditional Negative-Binominal pseudo-count method to characterize heteroscedastic noise amplified by division, resulting in an unconstrained reflectance ratio estimate. The method achieves a mean PSNR/SSIM of 17.74dB/0.739 on the LOL-v1 dataset, outperforming conventional techniques. Furthermore, it can process a 400x600 image at approximately 43 FPS on an Apple M2 Pro CPU. AI
IMPACT This method offers a potential improvement for applications requiring high-quality image processing in low-light conditions.
RANK_REASON The item is an academic paper detailing a novel method for image enhancement. [lever_c_demoted from research: ic=1 ai=0.7]
- Apple M2 Pro
- arXiv
- LOL-v1
- Negative-Binominal
- peak signal-to-noise ratio
- Structural Similarity Index Measure
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
