Impact of ISP Affine Re-quantization on Photon Transfer Characteristics and Perceptual Quality

This paper addresses the decoding and evaluation of consumer image sensors under ultra-low-noise conditions. It systematically investigates the effects of ISP affine requantization, black-level drift, and prescaling on the photon transfer curve (PTC), dynamic range (DR), and dark-region color cast. An exact quantization model for Gaussian inputs is established, the periodic noise oscillation and phase shift induced by non-integer requantization are analyzed, and a suppression strategy based on bit-depth expansion, realignment, and inverse normalization is proposed. Experimental and simulation results show that, without modifying the existing ISP firmware, the proposed method significantly improves linearity and perceived color fidelity in extremely dark regions, while enhancing the consistency and interpretability of measurements in the low noise regime.

本文面向消费级图像传感器在极低噪声条件下的解码与评测问题,系统研究了 ISP 仿射重量化、黑电平漂移与预缩放(prescaling)对光子转移曲线(PTC)、动态范围(DR)与暗部色偏的影响。文中建立了高斯输入下的精确量化模型,分析了非整数倍重量化引发的周期性噪声震荡与相位漂移,并提出“位深扩展--重对齐--逆归一化”的抑制策略。实验与仿真结果表明,该方法在不修改既有 ISP 固件的前提下,可显著改善极暗部线性度与感知色偏,并提升低噪时代测量结果的一致性与可解释性。