Image Embedding Trees for Efficient Facial Expression Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current graph-based facial expression recognition (FER) methods face inefficiencies due to overly complex training of fully connected graphs or suboptimal predefined graph topologies, leading to inefficient computing and subpar performance.

Innovation Solution

An end-to-end deep FER architecture that learns an optimal tree topology using a combinatorial approach to determine a minimum-cost spanning tree, which is traversed to form a sequence for input into a neural network, combining structure and texture streams with soft attention mechanisms for enhanced recognition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If fully connected graphs are used for facial region connections, then comprehensive feature representation is achieved, but training complexity and computational load increase significantly

Engineering Contradiction:
Improvefeature representation completenessVSAvoidtraining process complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the fully connected graph into multiple tree-based subgraphs, where each tree represents a specific facial region hierarchy. This segmentation reduces the overall complexity while maintaining comprehensive feature coverage through multiple specialized trees rather than one dense graph.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts only the essential hierarchical relationships needed for facial feature representation, removing redundant connections present in fully connected graphs. By taking out only the necessary parent-child and sibling relationships in tree structures, the system achieves efficient representation without unnecessary computational overhead.

Inventive Principle:
Principle #2Taking out (Extraction)

2Ease of manufacture

If pre-defined graph topologies are used for facial regions, then training simplicity is improved, but recognition performance deteriorates due to suboptimal structure

Engineering Contradiction:
Improvetraining simplicityVSAvoidrecognition performance
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent introduces dynamic tree construction where the graph topology is not fixed but adapts based on the specific facial features and expressions being analyzed. The system dynamically determines which tree structures to use and how to configure them, allowing optimal performance for different scenarios while maintaining relative training simplicity.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the parameters of the graph structure by using multiple tree configurations with different hierarchical relationships. Instead of a single fixed topology, the system varies structural parameters like tree depth, branching factors, and connection patterns to optimize performance for different facial expression recognition tasks.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If multiple tree structures are constructed for different facial regions, then recognition accuracy is improved, but computational complexity increases

Engineering Contradiction:
Improverecognition accuracyVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the facial image into distinct regions (e.g., eyes, nose, mouth) and constructs specialized trees for each region. This segmentation allows each tree to focus on specific facial features, improving recognition accuracy for particular expressions while keeping individual tree computations manageable and efficient.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by selectively constructing trees only for facial regions that are relevant to the current recognition task. Instead of building comprehensive trees for all possible facial features, the system constructs only the necessary partial structures needed for the specific expression being analyzed, reducing unnecessary computational overhead.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4060553B1Systems, methods, and storage media for creating image data embeddings to be used for image recognition
Publication Date: 2026.04.29 IRDETO BV
  • EP4060553B1 patent drawingFigure 1
  • EP4060553B1 patent drawingFigure 2
  • EP4060553B1 patent drawingFigure 3

AI summary

Disclosed implementations include a method, apparatus and computer media for learning an optimal graph in the form of a tree topology defining a sequence that can be used by a learning network for image recognition. Image data representing the image of an object is received and N landmarks are detected on the image using a deep regression algorithm, wherein N is an integer. A weighted, fully connected, graph is constructed from the landmarks by assigning initial weights for the landmarks randomly. An optimized tree structure is determined based on the initial weights. A sequence is generated by traversing nodes of the tree structure and a series of embeddings representing the object image are generated based on the sequence. The embeddings can be processed by a neural network to generate an image recognition signal based on the embeddings.