BlindPSNR: A No-Reference Fidelity Predictor for Low-Light Image Enhancement

2026-07-29

AI Summary

This work addresses the lack of effective no-reference fidelity metrics for low-light image enhancement, which hinders automatic selection of optimal enhancement parameters. To this end, we propose the first lightweight no-reference PSNR prediction network tailored for low-light enhancement. Our method fuses the original low-light image with its enhanced counterpart and leverages a window-based cross-attention mechanism combined with heteroscedastic regression to achieve highly accurate fidelity estimation under paired-data supervision. Evaluated on benchmark datasets, the proposed approach attains a top-1 parameter selection accuracy of 89.5% with a remarkably low regret of only 0.026 dB. Furthermore, it demonstrates strong generalization capability on unseen datasets, achieving SRCC scores ranging from 0.61 to 0.67.

Abstract

Low-light image enhancement (LLIE) methods involve tunable parameters that are typically fixed, often leading to performance degradation when applied across scenes. Manually selecting the best configuration, however, can be time-consuming and not always practical. Peak signal-to-noise ratio (PSNR) is the natural fidelity criterion for automating parameter selection, yet it requires a ground-truth reference that is typically unavailable. To our knowledge, no learning-based method addresses no-reference PSNR prediction for low-light image enhancement; the natural surrogate, no-reference image quality assessment (NR-IQA), targets perceptual quality rather than signal fidelity, and all seven baselines we test achieve 0% top-1 selection accuracy on our benchmark. With paired training data, the ground-truth PSNR is analytically computable, providing exact supervision without a separate teacher network. Building on this, we propose BlindPSNR, a lightweight no-reference network that fuses the enhanced image with the degraded low-light input via windowed cross-attention and estimates PSNR through heteroscedastic regression. While a scalar-regression baseline achieves top-1 accuracy of 54.4%, BlindPSNR raises this to 89.5% with regret dropping from 1.62 dB to 0.026 dB, and generalizes to unseen datasets (SRCC = 0.61-0.67).

Problem

• Low-light image enhancement
• No-reference PSNR prediction
• Signal fidelity
• Parameter selection
• Blind quality assessment

Innovation

• BlindPSNR
• no-reference PSNR prediction
• low-light image enhancement
• heteroscedastic regression
• windowed cross-attention