Digital Watermark Training With Distortion-Aware Decoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing watermarking technologies struggle to achieve robustness, blindness, and automated embedding and extraction simultaneously, especially in the presence of distortions, and lack efficient detection mechanisms to reduce computational resources.

Innovation Solution

A joint training method for encoder and decoder machine learning models that generate and decode digital watermarks, incorporating distortion detection to minimize errors and optimize training parameters, allowing for visually imperceptible watermarks to be detected and decoded efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional manually designed watermarking methods are used, then they can handle specific types of distortions, but they cannot simultaneously handle a broad spectrum of distortions

Engineering Contradiction:
Improveability to handle broad spectrum of distortionsVSAvoidrobustness to distortions
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent changes the fundamental parameters of watermarking from manual design to automated deep learning-based embedding. The encoder network learns optimal embedding patterns through training on distorted images, enabling the system to adapt to various distortion types automatically rather than requiring manual design for each distortion type

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces manual watermarking mechanisms with automated machine learning systems. Instead of manually designing watermarks that can withstand specific distortions, the system uses trained neural networks to automatically embed and extract watermarks robustly across multiple distortion types

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

2Extent of automation

If deep learning networks are incorporated with image watermarking, then automated embedding and extraction can be achieved, but existing systems cannot achieve robustness, blindness and automated embedding and extraction simultaneously

Engineering Contradiction:
Improveautomated embedding and extractionVSAvoidrobustness to distortions
Core Design Contradiction:
Extent of automationVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where the decoder network receives distorted watermarked images and outputs decoded watermarks, which are then fed back to the encoder for re-embedding. This feedback loop enables the system to learn and adapt to distortion patterns, achieving both automation and robustness simultaneously

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent creates a universal watermarking system where a single trained encoder-decoder pair can handle multiple distortion types and scenarios. The networks are trained to be blind to various distortions, making them multi-functional across different application contexts without requiring system redesign

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

3Reliability

If complex watermarking systems are used to ensure robustness, then watermark reliability improves, but computational resources and processing time increase

Engineering Contradiction:
Improvewatermark robustnessVSAvoidcomputational resources
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent optimizes computational parameters by training efficient neural network architectures with appropriate complexity. The networks are trained to achieve robustness with reasonable computational cost, balancing reliability and energy consumption through careful model design and training procedures

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4238312B1End-to-end watermarking system
Publication Date: 2025.11.19 GOOGLE LLC
  • EP4238312B1 patent drawingFigure 1
  • EP4238312B1 patent drawingFigure 2a~2b
  • EP4238312B1 patent drawingFigure 3A

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for jointly training an encoder that generates a watermark and a decoder that decodes a data item encoded within the watermark. The training comprises obtaining a plurality of training images and data items. For each training image, a first watermark is generated using an encoder and a subsequent second watermark is generated by tiling two or more first watermarks. The training image is watermarked using the second watermark to generate a first error value and distortions are added to the watermarked image. A distortion detector predicts the distortions based on which the distorted image is modified. The modified image is decoded by the decoder to generate a predicted data item and a second error value. The training parameters of the encoder and decoder are adjusted based on the first and the second error value.