3D Message Embedding Models for Imperceptible Viewpoint Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing techniques face challenges in embedding imperceptible or near-imperceptible messages in three-dimensional image data, particularly in maintaining message extractability from any viewpoint and minimizing perceptible differences in the image data.

Innovation Solution

A machine-learned message embedding model modifies aspects of three-dimensional image data, such as vertex positions, material properties, or point cloud attributes, to embed messages imperceptibly, using neural networks like conditional variational autoencoders, and a corresponding extraction model to recover the message, with training optimized through loss functions and differentiable rendering schemes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If conventional message embedding techniques are used in three-dimensional image data, then messages can be embedded, but the messages are significantly perceptible and cause noticeable differences in the image data

Engineering Contradiction:
Improvemessage embedding imperceptibilityVSAvoidmessage embedding complexity
Core Design Contradiction:
Loss of informationVSEase of manufacture

Solution Approach 1:

The patent replaces conventional mechanical or algorithmic message embedding methods with a machine-learned model (neural network) that automatically learns optimal embedding patterns. The model takes cover image data and message data as inputs and generates watermarked image data, substituting traditional embedding algorithms with a trained neural network that produces imperceptible modifications.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent modifies parameters of the cover image data (such as pixel values, vertex positions, or other numerical attributes) by small amounts determined by the machine-learned model. These parameter changes are optimized to be minimal yet sufficient to encode the message, thereby maintaining visual fidelity while achieving message embedding.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If messages are embedded in three-dimensional image data to be extractable from any viewpoint, then message extractability is improved, but the embedding technique causes significantly perceptible differences

Engineering Contradiction:
Improvemessage extractability from any viewpointVSAvoidimage data perceptibility
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The patent embeds messages in three-dimensional image data structures (such as 3D meshes, point clouds, or volumetric data) rather than two-dimensional images. By operating in 3D space, the embedded message can be extracted from multiple viewpoints through 3D rendering or projection, providing viewpoint-independent extractability while distributing the embedding modifications across the 3D structure to minimize perceptibility.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The machine-learned model is trained to produce embeddings that satisfy multiple requirements simultaneously: viewpoint-independent extractability, imperceptibility, and sufficient message capacity. The model learns to distribute message information across multiple 3D features that remain consistent under different viewing conditions, achieving universal extractability without compromising image quality.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Loss of information

If machine-learned models are used for message embedding, then imperceptible embedding is achieved, but the device complexity and training requirements increase

Engineering Contradiction:
Improvemessage embedding imperceptibilityVSAvoidmodel training and computation complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent performs message embedding training in advance using a trained machine-learned model before actual message embedding operations. The model is pre-trained on large datasets to learn optimal embedding patterns, and once trained, can efficiently embed messages without requiring complex real-time computation. This preliminary training phase separates the complex learning process from the operational phase, reducing real-time complexity.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses the trained machine-learned model to generate embeddings that can be replicated and applied to different cover images and messages. Once the model is trained, it serves as a reusable component that can embed messages across multiple 3D images without requiring retraining, effectively copying the learned embedding strategy to new situations and reducing overall system complexity.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4162475B1Systems and methods for message embedding in three-dimensional image data
Publication Date: 2025.12.17 GOOGLE LLC
  • EP4162475B1 patent drawingFigure 1A
  • EP4162475B1 patent drawingFigure 1B
  • EP4162475B1 patent drawingFigure 1C

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.