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

Cross-model face feature vector conversion system and method

A face feature and conversion system technology, applied in the field of face recognition, can solve the problems of uncommon face feature vector, ungeneric feature vector, waste of human resources, etc., so as to improve performance and generalization, and improve feature expression ability. , the effect of improving the conversion success rate

Active Publication Date: 2021-03-16
成都东方天呈智能科技有限公司
View PDF18 Cites 9 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] At present, different face recognition devices use different face recognition algorithms to extract the feature vectors of face images, which will lead to uncommon feature vectors. Once the feature vectors are replaced, the corresponding face recognition devices also need to modify the deployment plan, resulting in Convenience is reduced and human resources are greatly wasted
Therefore, it is necessary to propose an easy-to-use and easy-to-operate face feature vector conversion method to solve the problem of non-universal use of face feature vectors across models

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
  • Cross-model face feature vector conversion system and method
  • Cross-model face feature vector conversion system and method
  • Cross-model face feature vector conversion system and method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0034] A conversion system for cross-model face feature vectors, including a data extraction module, a network training module, and a conversion module.

[0035] The data extraction module is used to collect face image data sets, source feature vectors and target feature vectors; mark the identity category of the face image data sets and form a training data set, and use the face image data and source feature vectors as training samples, And the labeled identity category is used as the real label value;

[0036] The network training module is used to input the input samples in the training data set in the data extraction module into the network model for training to obtain the trained network model;

[0037] The conversion module is used for inputting the human face feature vector to be converted into the network model trained in the network training module, and outputting the converted human face feature vector.

[0038] Such as figure 1As shown, the network model is compos...

Embodiment 2

[0041] This embodiment is optimized on the basis of Embodiment 1. The network part of the deformed attention mechanism first uses the convolution layer to down-sample the image to improve semantics; then uses the multi-head attention mechanism module to slice the convolution features to The relationship information between each feature is extracted; finally, the batch normalization layer, the fully connected layer, and the activation function layer are sequentially processed and then partially spliced ​​and fused with the convolutional network.

[0042] The present invention proposes to introduce the deformed attention mechanism into the conversion process of the face feature vector. By building the network part of the deformed attention mechanism, the local information and global information available in the face image can be fully extracted, and the feature expression ability of the model is greatly improved. , to improve the performance and generalization of the model.

[0...

Embodiment 3

[0045] This embodiment is optimized on the basis of embodiment 2, such as figure 2 As shown, the network part of the deformed attention mechanism consists of the first integration module, the position embedding vector layer, the second integration module, the batch normalization layer, the fully connected layer, the activation function layer, and the fully connected layer arranged in sequence from front to back , the activation function layer is encapsulated and obtained; the first integrated module and the second integrated module are respectively provided with several; the first integrated module and the position embedding vector layer are added and connected to the second integrated module; the first integrated module The module consists of a convolutional layer, a batch normalization layer, and an activation function layer arranged in sequence from front to back. The second integrated module consists of a reorganization vector layer, a multi-head attention mechanism module...

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 cross-model face feature vector conversion system and method. Through a network training module, input samples in a training data set are input into a network model for training to obtain the trained network model. The network model is composed of a convolution network part and a deformation attention mechanism network part. A source feature vector is input into the convolution network part, and the depth convolution feature is obtained; a face image is input into the deformation attention mechanism network part and the depth feature of the face image is extracted; and then, the depth convolution features are spliced and fused with the depth features of the face image, and a feature map with stronger expressive ability is obtained and classified and converted. According to the method, the conversion relationship between the source feature vector and the target feature vector is learned by establishing the double-branch deep neural network, and meanwhile, the human face image is used as assistance, so that the problem of information loss in the main task training process is improved, the generalization of the network model is improved, and the conversion success rate is effectively improved.

Description

technical field [0001] The invention belongs to the technical field of face recognition, and in particular relates to a conversion system and method of cross-model face feature vectors. Background technique [0002] In the era of big data, with the rapid rise of deep learning technology, amazing artificial intelligence equipment has been continuously derived, making people's lives more convenient. At the same time, there has been a lot of personal information exchange, which makes people focus on Security of Personal Information. With people's demand for information security, various authentication technologies have emerged, such as face recognition, pupil recognition, fingerprint recognition, etc., with a wide range of application scenarios. [0003] Among many biometric technologies, face recognition stands out due to its advantages of low cost and non-contact. Face recognition is a technology that uses human facial feature information for classification and recognition....

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 Applications(China)
IPC IPC(8): G06K9/00G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06V40/172G06V40/168G06N3/045G06F18/253G06F18/214
Inventor 闫超黄俊洁韩强
Owner 成都东方天呈智能科技有限公司
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

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com