Generative adversarial learning network-based domain learning method

A learning method and learning network technology, applied in the field of generative confrontation learning network, image processing and pattern recognition, can solve the problems of high data cost and poor adaptability, and achieve the effect of eliminating distribution differences and high similarity

Active Publication Date: 2018-10-26
ZHEJIANG UNIV OF TECH
View PDF5 Cites 14 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the deficiencies of poor adaptability and high data cost in the prior art, the present invention provides a domain learning method based on generative adversarial learning network with strong adaptability and low data cost

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
  • Generative adversarial learning network-based domain learning method
  • Generative adversarial learning network-based domain learning method
  • Generative adversarial learning network-based domain learning method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0049] Embodiment: Take the comparison between a person and an ID card as an example.

[0050] During the learning process of the generative neural network G: (1) collect more than 1,000 face images stored in ID documents, remove the image boundaries, and denote it as Si; (2) collect more than 100,000 face academic public datasets, denote it as Di, And use the Resnet50 network to train the face classifier C on this data set. If there is already a face classifier model, just collect 1000 training data sets of the classifier, which is recorded as Di; (3) the data set Si and Di Rotate the face image within 10°, scale within 0.2, use PCA for color transformation, horizontal mirroring, etc., generate 10 disturbed images for each face image, scale to 100x100 resolution, and process the data Sets are denoted as Si' and Di' respectively; (4) use figure 1 The network structure of G is a generative neural network G. The input and output of the network are 100x100 3-channel images. The ...

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 generative adversarial learning network-based domain learning method. The method comprises the following steps of 1) collecting a human face image set of a source domain, positioning human face positions by using a human face detector and extracting human face images; 2) constructing a generative neural network G, wherein input and output are the human face images with the same resolution, the input is the collected specific application scene human face image, and the output is the converted image; 3) constructing a classification neural network D, and applying neurons of a convolutional neural network, wherein the input is an output image of the generative neural network and a training image of a human face classifier, and the output is classification of two types of input images; 4) iteratively training the generative neural network G and the classification neural network D in an asynchronous mode; 5) when the human face recognition is carried out in a specified scene, performing conversion G(I) on the input human face image I, and inputting the converted human face image into a human face recognition module to obtain a human face recognition result. Themethod is relatively high in adaptability to a universal human face recognizer and relatively low in data cost.

Description

technical field [0001] The invention belongs to the technical field of image processing and pattern recognition, and relates to a domain learning method based on a generative confrontation learning network. Background technique [0002] Applying classification models across domains is a major problem that limits the application of pattern recognition. In practical applications based on face recognition technology, such as witness comparison, face attendance, financial authentication, etc., due to the difference in image quality between the training data set of the face classifier and the actual application scene, face recognition technology The effect becomes worse in practical application. For specific application scenarios, it is time-consuming and labor-intensive to re-collect face images and manually label enough data, especially at present, the face recognition method based on deep learning generally requires a large number of training samples, so re-collecting face im...

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): G06K9/62G06K9/00
CPCG06V40/172G06F18/214G06F18/24
Inventor 高华陈胜勇
Owner ZHEJIANG UNIV OF TECH
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