Neural Network Masked Face Recognition Training

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current face recognition technologies face limitations in accurately determining characteristic information, such as gender and age, for individuals wearing masks, which hampers their application in access control and other systems.

Innovation Solution

A method utilizing a pre-trained neural network to predict characteristic information from face images covered by masks, where the network is trained on both masked and unmasked face images, and additional synthetic images with masked features, allowing for accurate identification of gender and age.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If face recognition technology is applied to identify characteristic information such as gender and age, then the system can provide access control and customer information services, but the accuracy deteriorates when the target is wearing a mask

Engineering Contradiction:
Improveaccuracy of characteristic information predictionVSAvoidinformation loss due to mask coverage
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent generates synthetic masked face images by copying mask images and superimposing them onto original face images. This creates training data that simulates real masked face scenarios, enabling the neural network to learn characteristic information prediction from masked faces without requiring actual masked face data collection

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent performs preliminary training of the neural network using synthetic masked face images before deploying it for actual prediction tasks. This preliminary action with simulated data prepares the model to handle real masked face inputs effectively, improving reliability when actual masked face recognition is needed

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If a neural network is trained to predict characteristic information from masked face images, then the system can maintain functionality during mask-wearing periods, but the training complexity and data preparation requirements increase

Engineering Contradiction:
Improveability to recognize masked facesVSAvoidtraining process complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system generates its own training data by automatically synthesizing masked face images from existing unmasked face images and mask images. This self-service approach to data generation eliminates the need for external data collection and annotation processes, reducing training complexity while improving adaptability to masked face recognition

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent transforms the training problem by adding a synthetic dimension - creating masked face images as a new data dimension that combines original face images with mask overlays. This dimensional transformation allows the network to learn masked face recognition patterns without requiring physically different training conditions

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

Data Source

PatentUS11989972B2Method for predicting characteristic information of target to be recognized, method for training neural network predicting characteristic information of target to be recognized, and computer-readable storage medium storing instructions to perform neural network training method
Publication Date: 2024.05.21 SUPREMA INC
  • US11989972B2 patent drawing
  • US11989972B2 patent drawing

AI summary

There is provided a method for predicting characteristic information of a target to be recognized. The method comprises: acquiring a plurality of first face images for learning and characteristic information on each first face image; generating a plurality of second face images for learning obtained by synthesizing a mask image with the plurality of first face images for learning by a predetermined algorithm; and training a first neural network by using the plurality of second face images for learning as input data for learning and characteristic information as label data for each second face image corresponding to one of the first face images.