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

Single-picture three-dimensional face reconstruction method based on cascade regression network

A three-dimensional face, cascade regression technology, applied in the field of three-dimensional face reconstruction, can solve the problems affecting the accuracy of the three-dimensional face deformation model parameter solution, poor robustness, etc., and achieve improved feature extraction methods, large pose robustness, elimination of confounding effect

Active Publication Date: 2018-08-14
SEETATECH BEIJING TECH CO LTD
View PDF8 Cites 39 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The disadvantages of this method are: some facial feature points are not visible in large poses, which affects the accuracy of the parameter solution of the 3D face deformation model; on the other hand, this method only uses the information of facial feature points and does not use the entire face face picture information
This method does not perform further regression on the three-dimensional coordinates of the face feature points, so the robustness is poor

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
  • Single-picture three-dimensional face reconstruction method based on cascade regression network
  • Single-picture three-dimensional face reconstruction method based on cascade regression network
  • Single-picture three-dimensional face reconstruction method based on cascade regression network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0034] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0035] A single image 3D face reconstruction method based on cascade regression network, the overall steps are:

[0036] Step 1. Data preparation stage:

[0037] a. Use a 3D scanner to collect 3D face data, including 2D face images from each perspective, face poses from each perspective, 3D face shape point clouds, 68 feature point annotations, and triangular relationships in point clouds, etc. ;

[0038] b. Perform face detection on the collected face image, crop the face image, and process the three-dimensional face shape point cloud according to the face pose to obtain a three-dimensional face shape point cloud corresponding to the cropped face image;

[0039] Step 2. Model design stage:

[0040] a. The model consists of a two-level multi-task convolutional neural network, which regresses three-dimensional face shape, face pose ...

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 single-picture three-dimensional face reconstruction method based on a cascade regression network. The whole single-picture three-dimensional face reconstruction method includes the steps of 1, data preparation, wherein three-dimensional face data is collected, and a three-dimensional face-shape point cloud of a face picture is obtained; 2, model designing, wherein a network model comprising two-stage multitask convolutional neural networks is established, face features are regressed step by step, and then modified shape features are extracted; 3, model training, wherein the two-stage multitask convolutional neural networks are trained respectively, and a multitask cascade regression convolutional neural network M is obtained; 4, model testing, wherein an image Iis input into the multitask cascade regression convolutional neural network M, and model testing is carried out. According to the single-picture three-dimensional face reconstruction method based on the cascade regression network, the multitask cascade regression convolutional neural network M is used, a three-dimensional face shape, face feature points and a face posture are regressed step by step, and meanwhile a traditional feature extracting method is improved; by computing visibility of the feature points, the confusion performance of the feature points on a two-dimensional face picture is eliminated, and the single-picture three-dimensional face reconstruction method is more robust for large gestures.

Description

technical field [0001] The invention relates to a three-dimensional human face reconstruction method, in particular to a single-picture three-dimensional human face reconstruction method based on a cascade regression network. Background technique [0002] The goal of 3D face reconstruction is to reconstruct the corresponding 3D face shape from the input 2D face image. This technology is widely used in the fields of facial animation, facial expression analysis and 3D face recognition. The current representative methods for 3D face reconstruction are: [0003] (1) A single image 3D face reconstruction method based on a 3D face model. The main technical means are: based on a 2D face database, the SDM (Supervised Descent Method) algorithm is used to train the 2D face feature point parameters. Model. When performing face reconstruction based on a single frontal photo, first use the two-dimensional picture feature point parameter model to extract face feature points; then, acco...

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
IPC IPC(8): G06T15/20G06T3/40
CPCG06T3/4007G06T3/4038G06T15/205G06T2207/10012G06T2207/20081G06T2207/30201
Inventor 张刚韩琥张杰山世光陈熙霖
Owner SEETATECH BEIJING TECH CO LTD
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