Unified Deep Network for Face Image Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing face image recognition methods lose important correlations between compared face images during feature extraction and cannot jointly optimize the feature extraction and recognition stages, leading to inefficiencies in handling large-scale data with complex distributions.

Innovation Solution

A unified deep network architecture that directly and jointly extracts relational features from face region pairs, using convolutional neural networks to generate feature modes, extract identity relational features hierarchically, and pool these features to improve accuracy and robustness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If separate feature extraction and recognition stages are used, then the process is simpler and more modular, but important correlations between face images are lost and the stages cannot be jointly optimized

Engineering Contradiction:
Improvesystem structureVSAvoidcorrelations between face images
Core Design Contradiction:
Device complexityVSLoss of information

Solution Approach 1:

The patent merges the separate feature extraction and recognition stages into a unified deep neural network architecture. The network simultaneously performs feature extraction and recognition in an integrated manner, allowing joint optimization of both stages while preserving important correlations between face images through the shared convolutional feature representations.

Inventive Principle:
Principle #5Merging (Combining)

2Device complexity

If shallow structure models are used for recognition, then the model is simpler and faster, but they cannot effectively handle large-scale data with complex distributions

Engineering Contradiction:
Improvemodel structureVSAvoidrecognition accuracy
Core Design Contradiction:
Device complexityVSReliability

Solution Approach 1:

The patent transitions from shallow structure models to a deep neural network architecture with multiple hierarchical layers. This dimensional expansion in network depth enables the model to capture complex distributions and patterns in large-scale face image data, significantly improving recognition accuracy while maintaining computational efficiency through the shared feature extraction mechanism.

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

Data Source

PatentEP3074918B1Method and system for face image recognition
Publication Date: 2019.04.03 BEIJING SENSETIME TECH DEV CO LTD
  • EP3074918B1 patent drawingFigure 1
  • EP3074918B1 patent drawingFigure 2
  • EP3074918B1 patent drawingFigure 3~4

AI summary

A method for face image recognition is disclosed. The method comprises generating one or more face region pairs of face images to be compared and recognized; forming a plurality of feature modes by exchanging the two face regions of each face region pair and horizontally flipping each face region of each face region pair; receiving, by one or more convolutional neural networks, the plurality of feature modes, each of which forms a plurality of input maps in the convolutional neural network; extracting, by the one or more convolutional neural networks, relational features from the input maps, which reflect identity similarities of the face images; and recognizing whether the compared face images belong to the same identity based on the extracted relational features of the face images. In addition, a system for face image recognition is also disclosed.