LF-VGCNet: Deep Perceptual Gamma Correction with Low-Frequency Enhancement for Cross-Device Display Calibration

Perceptual consistency in display color calibration across heterogeneous devices is challenged by variations in hardware response, ambient lighting, and human visual perception. This paper proposes LF-VGCNet, a compact and hardware-efficient deep learning model that addresses these challenges by integrating four key innovations: (1) a low-frequency halftone encoder to enhance gamma sensitivity, (2) joint-channel gamma regression for inter-channel coherence, (3) a differentiable embedding of the CIECAM02 model to align with human color appearance perception, and (4) an observer-adaptive shift module to simulate perceptual diversity across users. Extensive experiments on 10 diverse displays — including LCD, OLED, AMOLED, and QLED panels — demonstrate that LF-VGCNet achieves an average [Formula: see text]00 of 1.97, significantly outperforming recent deep learning baselines by over 30%, while maintaining real-time performance with a model size under 7 MB. These results validate LF-VGCNet as a scalable and human-centric solution suitable for deployment in mobile, embedded, and professional imaging pipelines.