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

Face model construction method, face recognition method, device and equipment

A construction method and face model technology, applied in the field of face recognition, can solve the problems of 3D model difficulty, ignoring texture reconstruction, and inability to apply it to face recognition, so as to eliminate the difference in illumination and improve the accuracy rate

Active Publication Date: 2019-12-13
SICHUAN UNIV
View PDF9 Cites 14 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, it is much more difficult to reconstruct the 3D face shape based on the three-face face image taken under unconstrained conditions than the traditional calibrated multi-view face image.
At this stage, many teams at home and abroad have carried out research on the three-dimensional face reconstruction technology of three-sided photos, but most of the algorithms only focus on the reconstruction of three-dimensional face shapes, ignoring the reconstruction of textures and not making full use of the texture information of three-sided photos. Or the shape accuracy of the reconstructed 3D face is low, and it cannot be applied to face recognition, or it takes a long time to reconstruct a complete 3D face, which is not conducive to the application in real scenes

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 model construction method, face recognition method, device and equipment
  • Face model construction method, face recognition method, device and equipment
  • Face model construction method, face recognition method, device and equipment

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0067] As shown in Figure 1, the present invention provides a method for constructing a face model, and the subject of execution of the method may be a terminal device or a server. In order to improve the construction efficiency of the face recognition database and enrich the image information of the database, the present invention adopts the following steps:

[0068] Use the X-face photo of the same person to reconstruct the three-dimensional face shape, X is a positive integer, and X≥3;

[0069] This method can use photos of multiple face faces to reconstruct the face model, which can be face photos from multiple angles. This embodiment uses the three-face photo of each person in the registration database to reconstruct the three-dimensional face shape of each person. The registration database includes the three-face photo of each person. The photos are frontal photos, left face photos and right face photos. In this embodiment, three photos of N individuals are used as N se...

Embodiment 2

[0125] The difference between this embodiment and Embodiment 1 is that step S24 can be replaced by step A24, and step A24 is specifically:

[0126] After calculating the adjustment amount of the two-dimensional feature point and the adjustment amount of the three-dimensional face shape, the adjustment amount of the two-dimensional feature point is used as the input of the multi-layer perceptron network, and the adjustment amount of the three-dimensional face shape is used as the supervisory signal output by the network, Train the network until the network converges, output the update amount of the three-dimensional face shape, and use the three-dimensional face shape and the update amount of the three-dimensional face shape to calculate and update the three-dimensional face shape:

[0127] The update amount ΔS of the three-dimensional face shape output by the multi-layer perceptron network k+1 , calculate the 3D face shape S after the k+1th completion k+1 :

[0128] S k+1 =...

Embodiment 3

[0141] Use the three-dimensional photo and the reconstructed 3D face shape to reconstruct the texture, obtain the texture reconstructed 3D face shape, and complete the construction of the 3D face model;

[0142] In order to further reconstruct the complete texture information of the 3D face, this step assigns a corresponding texture value to each point in the 3D face shape obtained in the previous stage.

[0143] Such as Figure 4 Shown is the flow chart of 3D face texture reconstruction. Texture reconstruction is divided into two stages: texture mapping and texture fusion. The texture mapping stage roughly obtains the texture value of each point, and the texture fusion stage further optimizes the texture selected in the previous stage to eliminate splicing traces or texture inconsistencies caused by factors such as lighting.

[0144] B1, texture mapping

[0145] B11. Texture mapping based on three-dimensional vertices: based on the three-dimensional face photo and the recon...

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 model construction method, a face recognition method, a face model construction device and face recognition equipment. The face model construction method comprises the following steps: firstly, reconstructing a three-dimensional face shape through a group of cascaded regression models or nonlinear regression models obtained by training by taking feature points of a two-dimensional image as input; then, providing a three-dimensional face texture reconstruction algorithm based on a convolution pyramid, reconstructing texture on the reconstructed three-dimensional face shape, wherein an algorithm comprises two stages of texture mapping and texture fusion, so that the illumination difference between multi-surface illumination can be well eliminated, vivid three-dimensional face texture information can be recovered, and a three-dimensional face model with complete textures is obtained. The multi-view face image is obtained by rendering the three-dimensional face model, the face database can be constructed according to the multi-view face image, face recognition is achieved through the face database, and the recognition rate is greatly increased.

Description

technical field [0001] The invention relates to the technical field of face recognition, in particular to a method for constructing a face model, a face recognition method, a device and equipment. Background technique [0002] At this stage, due to the limited accuracy of face recognition technology under unconstrained conditions, the complementary information of three-face photos has not been effectively used. At present, some researchers have used the 3D face shape based on single reconstruction to assist the 2D face recognition of any pose and achieved good results. Three-sided photos can provide richer information than a single image because they correspond to different perspectives when collected. In particular, the introduction of side images can greatly improve the accuracy of 3D reconstruction and further improve the effect of face recognition. Although some researchers have explored the use of three-sided photos to better reconstruct the shape of three-dimensional ...

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): G06K9/00G06T17/00G06T15/00G06F16/51
CPCG06T17/00G06T15/00G06F16/51G06V40/168
Inventor 赵启军梁洁涂欢刘峰
Owner SICHUAN UNIV
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