The Devil is in the Dark Pixels: Toward Brightness Bias-Robust Denoising

2026-07-15

AI Summary

This work addresses the challenge of luminance bias in image denoising, where dark regions—due to their low signal-to-noise ratio—are particularly susceptible to signal-dependent noise, and conventional MSE-based training exacerbates brightness inaccuracies. To mitigate this, the authors propose Brightness Bias-Robust Denoising (BBRD), a novel approach that groups pixels by luminance, normalizes reconstruction errors according to noise variance, and employs Group Distributionally Robust Optimization (Group-DRO) to dynamically upweight the worst-performing luminance group during training. This strategy enables balanced optimization across all brightness levels without introducing additional parameters or inference overhead. Experiments across eight network architectures and two datasets demonstrate that BBRD is the only method capable of simultaneously improving performance in both bright and dark regions, achieving up to a 0.45 dB PSNR gain in dark areas and a 0.65 dB overall improvement on SIDD.

Abstract

In this paper, we reveal an important yet overlooked problem in image denoising: under signal-dependent camera noise models, dark regions suffer from inherently low Signal-to-Noise Ratio (SNR), as signal intensity decays far faster than noise variance diminishes, making detail recovery in dark areas fundamentally challenging. Yet rather than compensating for this difficulty, MSE-trained denoisers exacerbate it -- reconstructing dark pixels up to 6x worse relative to their per-band noise floor. This bias stems from two compounding factors: signal-dependent noise inflates bright-pixel residuals, and the network's Jacobian norm increases monotonically with brightness. Together, these cause bright regions to chronically dominate gradient updates at the expense of dark ones. To this end, we propose Brightness Bias-Robust Denoising (BBRD), a drop-in replacement for MSE loss that partitions pixels into brightness bands, normalizes per-band error by empirical noise variance, and applies Group Distributionally Robust Optimization (Group-DRO) to dynamically upweight whichever band is currently worst, with zero additional parameters or inference cost. Across 8 architectures and 2 datasets in our experiments, BBRD is the only method among 13 tested alternatives that improves each brightness band simultaneously, achieving up to +0.45 dB on dark bands, +0.32 dB on bright bands, and +0.65 dB aggregate Peak Signal-to-Noise Ratio (PSNR) on SIDD, with the largest per-band gains in the darkest regions where detail recovery matters most. Code is available at https://github.com/xmed-lab/BBRD

Problem

• image denoising
• brightness bias
• signal-dependent noise
• dark pixels
• SNR

Innovation

• Brightness Bias
• Signal-Dependent Noise
• Group-DRO
• MSE Loss
• Dark Pixel Denoising