TL;DR
This paper introduces a novel method for creating physical 3D reliefs that display two distinct grayscale appearances: one visible under normal lighting and a second, hidden appearance revealed only when illuminated by a specifically designed projector pattern. This technique allows for information hiding directly within the physical form of an object, opening up new possibilities for security and authentication.
What Problem Does This Paper Address?
The paper tackles the challenge of embedding hidden information within physical objects, specifically in the visual appearance of 3D reliefs. While digital information hiding (like watermarking) is well-established, applying it to the physical world presents unique difficulties. The goal is to design a single physical surface that can present two different images: a primary image visible under standard, direct illumination, and a secondary, hidden image that only appears when illuminated by a precisely controlled projector. This secondary image should be invisible under normal lighting conditions, making it a secure way to embed data like watermarks, logos, or even secret messages. Previous approaches often struggled to achieve arbitrary second appearances or required specific material properties, whereas this work aims for a general solution using a constant albedo (material/color).
Methods
The core of the method lies in a computational process that designs the geometry of a 3D relief surface. The approach formulates the problem as an optimization task to find a heightfield that satisfies two conditions simultaneously:
- Primary Appearance (Image A): The relief, when illuminated by a simple directional light (like a desk lamp), should produce a desired grayscale image (Image A). This is achieved by calculating surface normals that, under this lighting, result in the target intensity distribution.
- Hidden Appearance (Image B): The same relief, when illuminated by a digital projector emitting a specifically designed pattern, should produce a second, arbitrary grayscale image (Image B). This requires finding both the relief's geometry and the projector's illumination pattern.
The key insight is that there are multiple relief geometries that can produce the same shaded image under normal lighting. The authors exploit this "multiplicity of solutions" to find a geometry that not only produces Image A but also can be manipulated by a projector to produce Image B.
The computational process involves:
- Geometric and Radiometric Calibration: Calibrating the camera and projectors to accurately map 3D points to 2D image pixels.
- Appearance Formation Model: Using a diffuse reflectance model to describe how light interacts with the surface.
- Optimization: An iterative optimization process is employed. It alternates between solving for the projector pattern (given a fixed relief geometry) and solving for the relief's geometry (given the projector pattern). This process is simplified by making assumptions to linearize the equations and uses constrained linear optimization.
- Smoothness Constraint: Additional equations are incorporated to ensure that the height changes between neighboring vertices are small, leading to smoother surfaces that are easier to fabricate and more robust.
Datasets
Not discussed in the paper. The paper focuses on generating custom reliefs based on user-provided input images (Image A and Image B) rather than using pre-existing datasets.
Key Insights and Ideas
- Exploiting Non-Uniqueness: The fundamental idea is that many different surface shapes can produce the same visual appearance under a given lighting condition. This redundancy is key to embedding a second, hidden appearance.
- Projector as a "Dynamic Light Source": A digital projector isn't just a light source; it's a programmable one. By carefully designing its output pattern, it can "activate" a specific visual appearance encoded in the relief's geometry.
- Arbitrary Second Appearances: The method allows for virtually any two grayscale images to be encoded, even those that might be "physically impossible" under normal lighting (as described by Horn et al.).
- Constant Albedo is Sufficient: The technique works with a single, uniform material and color, making the hidden appearance truly invisible until the correct projector is used. This contrasts with methods that might rely on special paints or materials.
- Robustness to Contrast: The method is shown to be effective even with high-contrast images, where previous methods struggled. This is crucial for practical applications where images might have significant variations.
Key Results
- The method successfully designs and fabricates physical reliefs that exhibit two distinct visual appearances.
- Experiments demonstrate that the hidden appearance (Image B) is not perceivable under normal illumination.
- The approach is robust for various image pairs, including those with high contrast, where single-appearance reliefs fail to produce the desired second image.
- The system is implemented and tested with real-world 3D printing, showing good visual results.
- The authors found that a mesh resolution of 100x80 cells provides a good balance between visual quality and computation time.
Limitations and Future Work
- Security of the Second Image: The paper acknowledges that the security of the hidden image needs further analysis. It's not explicitly guaranteed how secure the second image is against reverse-engineering or if it can be retrieved using unintended illumination patterns.
- Perceptibility of the Hidden Image: While the second appearance is not recognizable under normal illumination, the paper notes that it's not explicitly guaranteed or maximized to be unrecognizable under all other lighting conditions or unintended projector patterns. Future work aims to model and minimize this perceptibility.
- Colored Images: The current method is limited to grayscale images. Extending it to handle colored images is a significant future direction.
- Multiple Hidden Appearances: The authors suggest exploring the possibility of hiding multiple distinct appearances using different projector patterns.
- Sharpness Effects: Sharpness in the input images can lead to artifacts, particularly in the hidden appearance, due to simplifications in the correspondence model. Smoother input images for the hidden appearance yield better results.
Summary updated