Towards Imperceptible Watermarking Via Environment Illumination for Consumer Cameras

TL;DR

This work tackles embedding metadata into video without altering viewer perception by leveraging spectral modulation of ambient LED illumination. It optimizes two ambient spectra $\boldsymbol{L}_1$ and $\boldsymbol{L}_2$ to be imperceptible to humans while being detectable by RGB cameras, switching at $15$ fps to encode data at about $n$ bps (up to $128$ bits in $10$ seconds). The method enforces $D65$-style white illumination with CRI$\ge 60$, and balances human perceptibility $\mathcal{L}_h$, camera detectability $\mathcal{L}_c$, and white-light fidelity $\mathcal{L}_w$ via LED-intensity reparameterization. Experimental results demonstrate perceptual invisibility in human studies, camera-robust decoding across devices, and practical reliability in diverse scenes, supporting privacy protection and content verification use cases. The approach offers a non-intrusive, hardware-free way to embed verifiable metadata in video, while acknowledging a modest data rate and suggesting avenues to scale with parallel wavelength channels.

Abstract

This paper introduces a method for using LED-based environmental lighting to produce visually imperceptible watermarks for consumer cameras. Our approach optimizes an LED light source's spectral profile to be minimally visible to the human eye while remaining highly detectable by typical consumer cameras. The method jointly considers the human visual system's sensitivity to visible spectra, modern consumer camera sensors' spectral sensitivity, and narrowband LEDs' ability to generate broadband spectra perceived as "white light" (specifically, D65 illumination). To ensure imperceptibility, we employ spectral modulation rather than intensity modulation. Unlike conventional visible light communication, our approach enables watermark extraction at standard low frame rates (30-60 fps). While the information transfer rate is modest-embedding 128 bits within a 10-second video clip-this capacity is sufficient for essential metadata supporting privacy protection and content verification.

Figures (8)

• Figure 1: Overview of the proposed method. Using two optimized spectra as lighting and switching them at 15 fps, the flickering is completely imperceptible to humans but detectable by the camera, acting as a watermark.
• Figure 2: Configuration of a hyperspectral lighting system using narrowband LEDs. By controlling the emission intensity of each LED, a custom spectral output is generated through the weighted sum of individual LED spectra.
• Figure 3: Proposed optimization pipeline. The objective functions ensure human imperceptibility, camera detectability, and natural white light for everyday use. These functions guide the optimization of LED intensity variables to achieve the desired lighting characteristics. Dark blue and dark green arrows indicate the forward pass for the two illuminations, while red arrows denote the backpropagation path used to optimize the LED intensities.
• Figure 4: Lightbox using the prototype hyperspectral lighting. Each LED's intensity is adjusted using a constant-current chip and a Raspberry Pi. The spectral distribution of each LED is measured inside the light box by capturing the white patch of the color chart.
• Figure 5: (a) Plot of the two optimized spectra. (b) The CIE 1931 Standard Observer XYZ Color Matching Functions and an example from the camera spectral sensitivity dataset (Sony NEX-5N). (c) Synthetic rendered color checker and pixel value variations obtained using the optimized spectra.