Rethinking Color Precision: Real-Time Perceptual Spatial Quantization for Foveated Rendering

Head-mounted displays demand extreme data bandwidth, yet traditional pipelines allocate uniform color precision across the entire frame, ignoring the rapid decline of chromatic sensitivity as retinal eccentricity increases. This paper proposes a real-time perceptual spatial quantization method that modulates bit depth based on retinal eccentricity rather than maintaining per-pixel precision. Through a psychophysical study, we model the detection threshold of our method across quantization levels ranging from 24 to 3 bits per pixel, the minimal level. Our method is the first to achieve this level of color reduction without detection, as it takes advantage of lowered peripheral color perception in tandem with the visual pooling effect. By integrating our model into the theoretical bandwidth measure, we demonstrate that our method can achieve a data reduction of up to 83.5% compared to uniform quantization. Our findings suggest potential for reducing bandwidth in wireless streaming to minimize latency and optimizing DRAM traffic to save power in mobile VR/AR architectures. In summary, our work provides a strategy for high-efficiency, perceptually lossless color reduction in foveated streaming systems.

Project page: https://zheningz.github.io/rethinking-color-precision-project-page/