Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Face recognition method and system based on anti-noise meta-learning

A face recognition and meta-learning technology, applied in the field of face recognition, can solve problems such as poor recognition effect, unstable face recognition model, and many wrong labels in large face recognition data sets, and achieve good face recognition performance. , Accurate identification, time-saving effect of training process

Active Publication Date: 2022-08-02
GUANGDONG UNIV OF TECH
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The present invention provides a face recognition method and system based on anti-noise element learning to solve the problem that the trained face recognition model is unstable and the recognition effect is not good due to the existence of many wrong labels in a large face recognition data set

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Face recognition method and system based on anti-noise meta-learning
  • Face recognition method and system based on anti-noise meta-learning
  • Face recognition method and system based on anti-noise meta-learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0050] Face recognition methods based on anti-noise meta-learning, such as figure 1 shown, including the following steps:

[0051] S1. Obtain a face dataset and its corresponding labels, where the labels include noise labels and clean labels;

[0052] S2. Preprocess the face data set: sample small batches of face data (X, Y) from the face data set, where X={x 1 ,…,x k } is k face data, Y={y 1 ,…,y k } is the corresponding noise label of each face data; for each noise label, multiple noise labels {Y′ are generated 1, …,Y′ M }, whose distribution is similar to Y, so that a set of generated noise labels Y′ m ={y' m1 ,…,y′ mk };

[0053] At the beginning of the training process, the face recognition model is not capable of face recognition. Therefore, all face data should be initialized with the same weight parameters for subsequent training.

[0054] S3. construct a face recognition model and its meta-learning model learning strategy, the input of the face recognition ...

Embodiment 2

[0069] Face recognition systems based on anti-noise meta-learning, such as figure 2 shown, including:

[0070] A data acquisition module 1 is used to acquire a face dataset and its corresponding labels, wherein the labels include noise labels and clean labels;

[0071] Data initialization module 2, for preprocessing the face data set, generating synthetic noise labels for each face data therein, and initializing the same weight parameters for each face data;

[0072] A face recognition model and its meta-learning model learning strategy building module 3 is used to construct a face recognition model and its meta-learning model learning strategy, the input of the face recognition model is the face data to be recognized, and the output of the The recognition result of the recognized face data; the meta-learning model learning strategy adopts the gradient descent algorithm to update the weight parameter of the face recognition model;

[0073] The meta-learning model training m...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a face recognition method and system based on anti-noise meta-learning. First, a face recognition data set is prepared, and there are a large number of noise labels in the data set, then the data is preprocessed to generate the noise label, and the weight is initialized according to a specific method parameters; then build the face recognition model and its meta-learning model learning strategy and start training and update the weight parameters. In this process, the meta-learning model is trained and its weight parameters are updated; finally, the meta-learning model learning strategy is used to optimize the parameters of the face recognition model, and the face data to be recognized is input into the optimized face recognition model to obtain face recognition results. The invention can be widely applied to face recognition based on various large-scale face recognition data sets, especially data sets that are easy to collect, cheap, and have irregular label noise on the network.

Description

technical field [0001] The present invention relates to the technical field of face recognition, in particular to a face recognition method and system based on anti-noise meta-learning. Background technique [0002] The application scope of face recognition is very broad, such as video surveillance, check-in system, human-computer interaction, etc. Because of its non-mandatory and non-contact, intuitive and simple, face recognition has a good development prospect. The traditional method of face recognition is to collect images of faces and extract face features from the images for recognition. In the subsequent development, face recognition based on support vector machine, wavelet transform and neural network has appeared continuously. With the development of deep learning, the use of large-sample face image datasets to train deep neural network models for face recognition has gradually become a mainstream method. These face recognitions require huge datasets for deep neur...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06V40/16G06V10/764G06K9/62G06N20/00
CPCG06N20/00G06V40/172G06F18/241
Inventor 郑晨何昭水吕俊黄德添白玉磊谭北海
Owner GUANGDONG UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products