Unlock instant, AI-driven research and patent intelligence for your innovation.

Three-dimensional face point cloud recognition method based on adversarial data enhancement

A technology of 3D face and recognition method, which is applied in the field of 3D face recognition, can solve problems such as limited identity and expression, non-adaptive model and data, and limited data enhancement methods, so as to improve recognition rate, generalization, The effect of improving recognition performance

Pending Publication Date: 2022-05-27
XI AN JIAOTONG UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Transformation-based data enhancement methods are limited, generative models generate limited identities and expressions, and the reconstructed 3D face identity is not highly discriminative, and the existing 3D face enhancement methods are not adaptive to models and data
Therefore, there is still room for improvement in 3D face recognition based on deep learning

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
  • Three-dimensional face point cloud recognition method based on adversarial data enhancement
  • Three-dimensional face point cloud recognition method based on adversarial data enhancement
  • Three-dimensional face point cloud recognition method based on adversarial data enhancement

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0049] The effectiveness of the method provided by the present invention is verified in the public 3D face dataset Bosphorus, and a total of 2902 3D face samples from 105 subjects in the Bosphorus dataset are used as a test set.

[0050] Table 1 compares the recognition rate of the method provided by the present invention with the existing method based on deep learning on the 3D face recognition data set Bosphoms, and MLAT-PointNet++ represents the recognition rate obtained by the present invention.

[0051] Table 1 Comparison of recognition rate of Bosphorus 3D face dataset

[0052]

Embodiment 2

[0054] In the public 3D face dataset BU-3DFE, the effectiveness of the method provided by the present invention is verified, and 100 subjects in the BU-3DFE dataset have a total of 2500 3D face samples as a test set. Table 1 compares the recognition rate of the method provided by the present invention with the existing deep learning-based method on the 3D face recognition data set BU-3DFE, and MLAT-PointNet++ represents the recognition rate obtained by the present invention.

[0055] Table 2 Comparison of recognition rate of BU-3DFE 3D face dataset

[0056]

[0057] It can be seen from Embodiment 1 and Embodiment 2 that the 3D human face based on adversarial data enhancement proposed by the present invention can achieve a competitive recognition effect, and the amount of training data used is relatively small.

Embodiment 3

[0059] In the public 3D face data set FRGC v2.0, the effectiveness of the adversarial samples provided by the present invention in improving the performance of 3D face recognition is verified respectively. There are 466 subjects in the FRGC v2.0 data set with a total of 4007 3D face samples as a test set.

[0060] PointNet and PointNet++ in Table 3 represent two types of 3D face point cloud recognition networks respectively, AT-PointNet and AT-PointNet++ respectively represent the addition of confrontational samples to the training set for training the 3D face point cloud recognition network, and Points represents the input of 3D face point cloud recognition networks. The coordinates of each vertex of the face sample are used as the input of the 3D face point cloud recognition network, while Points+Normals means that the coordinates of each vertex and the normal vector are concatenated as the input of the 3D face point cloud recognition network. It can be seen from Table 3 tha...

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 three-dimensional face point cloud recognition method based on adversarial data enhancement, and the method comprises the steps: carrying out the preprocessing of a collected three-dimensional face data set, and obtaining a training set and a test set after preprocessing; adding noise to the three-dimensional face point cloud in the preprocessed training set to obtain an optimized confrontation sample; using the preprocessed three-dimensional face point cloud in the training set and the optimized adversarial sample to optimize three-dimensional face point cloud recognition network parameters under a meta-learning-based training framework to obtain a trained three-dimensional face point cloud recognition network; and inputting the preprocessed test set into the trained three-dimensional face point cloud recognition network to complete three-dimensional face recognition. According to the method, the generalization of the three-dimensional face recognition model is improved by reasonably utilizing the generated three-dimensional face confrontation sample, so that the three-dimensional face recognition effect is improved. Experiments show that the most accurate recognition effect at present is achieved in the disclosed data sets BU-3DFE, BU-4DFE and Boscorus according to the method disclosed by the invention.

Description

technical field [0001] The invention belongs to the technical field of three-dimensional face recognition, and in particular relates to a three-dimensional face point cloud recognition method based on confrontational data enhancement. Background technique [0002] Face recognition has become one of the most widely used biometric identification methods due to its high accuracy and non-contact characteristics. In recent years, large-scale training data has promoted the development of deep learning-based 2D face recognition. 3D faces have real geometric information, so they have the potential to extract more discriminative face representations from them, leading to more accurate face recognition performance. However, the scarcity of 3D face samples limits the development of deep learning-based 3D face recognition. [0003] At present, 3D face recognition based on deep learning focuses on data enhancement. There are three main types of existing 3D face data enhancement: (1) ba...

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): G06V40/16G06K9/62G06V10/774
CPCG06F18/214
Inventor 余璀璨李慧斌孙剑徐宗本
Owner XI AN JIAOTONG UNIV