3D Image Message Embedding for Imperceptible Multi-View Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Embedding imperceptible or near-imperceptible messages in three-dimensional image data poses unique challenges, as existing techniques often result in perceptible differences and fail to ensure message extractability from various viewpoints.
Innovation Solution
A computing system utilizing machine-learned models for embedding and extracting messages in three-dimensional image data, where the message is imperceptibly embedded by modifying aspects like vertex positions, material properties, or point labels, and extracted using neural networks, with a loss function to minimize perceptual differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional message embedding techniques are used in three-dimensional image data, then message length can be increased, but perceptual quality deteriorates with significantly perceptible differences
Solution Approach 1:
The patent transforms the 3D image data into a latent representation space where message embedding can occur with minimal perceptual impact. By changing the parameter space from raw 3D data to learned latent features, the system enables longer message embedding while maintaining visual fidelity through the transformation bottleneck that constrains perceptible changes.
Solution Approach 2:
The patent introduces an intermediary latent representation layer between the 3D image data and the embedded message. This intermediary space acts as a mediator that allows message embedding while preserving the original visual characteristics, as the message is embedded in the latent space rather than directly in the perceptible 3D data.
2Productivity
If messages are embedded in three-dimensional image data, then data transmission efficiency improves, but message extractability from various viewpoints deteriorates
Solution Approach 1:
The patent creates a universal message embedding mechanism that functions across multiple viewpoints and rendering conditions. The latent space representation and learned embedding model are designed to be viewpoint-invariant, allowing the same embedded message to be extracted regardless of the observation angle or rendering parameters used.
Solution Approach 2:
The patent employs feedback mechanisms during training where the system learns to embed messages in a way that maximizes extractability across various viewpoints. The loss function incorporates viewpoint variations and the model adjusts its embedding strategy based on feedback from multiple rendering perspectives, ensuring robust message recovery.
Data Source
AI summary
Systems and methods are directed to a computing system. The computing system can include one or more processors, a message embedding model, a message extraction model, and a first set of instructions that cause the computing system to perform operations including obtaining the three-dimensional image data and the message vector. The operations can include inputting three-dimensional image data and a message vector into the message embedding model to obtain encoded three-dimensional image data. The operations can include using the message extraction model to extract an embedded message from the encoded three-dimensional image data to obtain a reconstructed message vector. The operations can include evaluating a loss function for a difference between the reconstructed message vector and the message vector and modifying values for parameters of at least the message embedding model based on the loss function.