Low-pixel multi-target face detection and key point positioning method and alignment method

A technology of face detection and positioning methods, applied in neural learning methods, character and pattern recognition, instruments, etc.

Inactive Publication Date: 2019-07-02
DONGHUA UNIV
View PDF5 Cites 40 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a low-pixel multi-target face detection and key point positioning method and an alignment method for the defect that it is difficult to perform face detection on low-pixel multi-target scenes in the prior art

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
  • Low-pixel multi-target face detection and key point positioning method and alignment method
  • Low-pixel multi-target face detection and key point positioning method and alignment method
  • Low-pixel multi-target face detection and key point positioning method and alignment method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0076] A low-pixel multi-target face detection and key point positioning method, the specific steps are as follows:

[0077] (1) Training the deep neural network based on the deep learning model, the training process is as follows image 3 shown;

[0078] (1.1) The face image is preprocessed to generate a mixed image pyramid, and the coordinates of the face regression frame and facial key points in the face image are known labels, and the preprocessing steps are as follows;

[0079] (1.1.1) Image scaling using a single template to generate an image pyramid;

[0080] (1.1.2) generate region proposals using fixed-size images of templates of different sizes;

[0081] (1.1.3) fuse (1.1.1) and (1.1.2) to generate as figure 1 Multi-scale hybrid image pyramid and region proposals shown;

[0082] (1.2) Training improved MTCNN, wherein the improvement of improved MTCNN relative to MTCNN lies in the classification loss function L of its R-Net and O-Net networks, and its expression i...

Embodiment 2

[0116] A low-pixel multi-target face alignment method based on low-pixel multi-target face detection and key point positioning method, specifically: using the coordinates of the face regression frame and facial key points obtained in step (4) to align the target pixel multi-target The faces in the face image are similarly transformed, and the standard face is aligned to the reference face;

[0117] The formula used for the similarity transformation is as follows:

[0118]

[0119] In the formula, z is the coordinate of the face to be transformed, and z * is the transformed face coordinates, s is the scaling scale, R is the rotation matrix, and a is the translation vector.

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 relates to a low-pixel multi-target face detection and key point positioning method and an alignment method. The positioning method comprises the steps that a low-pixel multi-target faceimage is preprocessed and then input into a trained deep neural network based on a deep learning model and a trained integrated regression tree based on a cascade shape regression framework, and output results of the low-pixel multi-target face image and the trained integrated regression tree are combined to obtain coordinates of a face regression frame and face key points; the alignment method comprises the following steps: carrying out similarity transformation on a human face by utilizing a human face regression frame determined by the low-pixel multi-target human face detection and key point positioning method and coordinates of human face key points, and aligning a standard human face to a reference human face. According to the low-pixel multi-target face detection and key point positioning method, the problem that a low-pixel multi-target face image is difficult to detect and carry out key point positioning is successfully solved; the low-pixel multi-target face alignment methodlays a foundation for face recognition of a low-pixel multi-target face image.

Description

technical field [0001] The invention belongs to the technical field of machine vision and pattern recognition, and relates to a low-pixel multi-target human face detection and key point positioning method and an alignment method, in particular to a method for detecting human faces in a scene with low pixels and many human face targets and Position the five key parts of the face (eyes, nose, left and right mouth corners), and finally perform facial alignment based on the detected key points. Background technique [0002] With the rapid development of science and technology, video is getting deeper and deeper into people's lives. Since the amount of information contained in video far exceeds that of pictures and text, it has increasingly become the main way to record and transmit information. At the same time, analyzing video can help people obtain video information in . Face detection and alignment technology is the hot research direction of current video analysis, which is ...

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): G06K9/00G06N3/04G06N3/08
CPCG06N3/084G06V40/172G06V40/168G06N3/045
Inventor 郝矿荣陈雨薇唐雪嵩赵栋祺
Owner DONGHUA UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap