Beyond Pixel Fidelity: Minimizing Perceptual Distortion and Color Bias in Night Photography Rendering

Problem
Night Photography Rendering (NPR) presents a significant challenge due to the extreme contrast between dark and illuminated areas, which complicates the rendering of images captured in low-light conditions. Existing methods primarily focus on pixel fidelity metrics, leading to perceptual gaps that detract from visual quality. This paper addresses these limitations by proposing a novel approach that minimizes perceptual distortion and color bias in NPR. The work is presented as a preprint and has not yet undergone peer review.

Method
The authors introduce pHVI-ISPNet, a RAW-to-RGB framework leveraging the HVI color space. The architecture incorporates four key refinements:

1. RAW-domain Feature Processing: This technique enhances the extraction of features from RAW images, preserving high-frequency details that are often lost in traditional processing.
2. Wavelet-based Feature Propagation: This method aids in maintaining detail across varying scales, ensuring that fine details are not compromised during the rendering process.
3. Sample-based Dynamic Loss Coefficients: This innovation allows for adaptive learning rates that stabilize training across different exposure levels, addressing the variability inherent in night photography.
4. Feature Distribution-based Loss Term: This loss function is designed to enforce color constancy by aligning the feature distributions of rendered images with those of ground truth images.

The training process and compute requirements are not explicitly disclosed, but the architecture is designed to optimize performance on the NTIRE 2025 challenge dataset.

Results
The proposed method achieves competitive fidelity while setting new state-of-the-art results in perceptual metrics. Specifically, it outperforms existing baselines in CIE2000 color difference and LPIPS (Learned Perceptual Image Patch Similarity). The paper reports a significant reduction in perceptual distortion compared to traditional methods, although exact numerical improvements over specific baselines are not detailed in the abstract.

Limitations
The authors acknowledge that their approach, while effective, may still struggle with extreme lighting conditions that exceed the training dataset’s variability. Additionally, the reliance on a specific color space (HVI) may limit generalizability to other imaging contexts. The paper does not address potential computational overhead introduced by the dynamic loss coefficients and wavelet processing, which could impact real-time applications.

Why it matters
This work has significant implications for the field of computational photography, particularly in enhancing the quality of images captured in low-light environments. By prioritizing perceptual quality over mere pixel fidelity, the proposed framework could lead to advancements in various applications, including mobile photography, surveillance, and artistic imaging. The integration of perceptually-driven metrics into the rendering pipeline may inspire further research into optimizing visual quality in other challenging imaging scenarios.

Authors: Furkan Kınlı
Source: arXiv:2604.28136
URL: https://arxiv.org/abs/2604.28136v1