Method for identifying human face under non-restricted condition based on LBP and deep learning

A face recognition, non-restricted technology, applied in the field of face recognition, can solve the problems of unfavorable feature expression, ignoring the local structure of the image, difficult to learn the local features of the face image, etc., and achieve high recognition rate and good recognition effect

Inactive Publication Date: 2017-05-31
NANJING LANTAI TRAFFIC FACILITIES CO LTD +1
View PDF0 Cites 11 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, DBN ignores the local structure of the image, and it is difficult to learn the local features of the face image; at the same time, when the pixel-level face features are used as the input of DBN, the network will learn unfavorable feature expressions due to factors such as illumination.

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
  • Method for identifying human face under non-restricted condition based on LBP and deep learning
  • Method for identifying human face under non-restricted condition based on LBP and deep learning
  • Method for identifying human face under non-restricted condition based on LBP and deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0023] LFW face database experiment: The face images in LFW are collected from the Internet with the standard face detector Viola-Jones, including 13233 images of 5749 people. Among them, the number of images for 1680 persons was greater than or equal to two, and the other 4069 persons had only one image. The image resolution is 250×250, and the face images are mainly in color and contain a small amount of grayscale images. LFW is mainly used for face recognition under unrestricted conditions. This library can fully express changes in face images under real conditions, such as changes in posture, illumination, occlusion, expression, background, race, gender, etc. This application selects people whose number of images is greater than or equal to 20 as the experimental subjects, including 62 people and a total of 3023 images. Each person randomly selects 5 images as training samples and the rest as testing samples.

[0024] 1) Algorithm performance when the number of hidden un...

Embodiment 2

[0038] Yale face database experiment: Yale face database has 15 people, each with 11 images, a total of 165 images, the image gray level is 256, and the resolution is 243×320. Each person has 6 different expressions, 3 different lighting, and the image has the difference between wearing glasses and not wearing glasses. In the experiment, each person randomly selects 5 images as training samples, and the rest as testing samples.

[0039] Table 4: Correct recognition rate for different hidden units on the Yale library

[0040]

[0041] It can be seen from Table 4 that when there are fewer hidden units, the same deep network cannot accurately learn the category information of face images under the restricted conditions. With the increase of the number of hidden units, the network learns more and more features, and the number of hidden units When it is 5000, the face image features learned by the network are more discriminative.

[0042] Table 5: Correct recognition rate of...

Embodiment 3

[0047] Table 6: Correct recognition rate for different hidden units on the Yale-B library

[0048]

[0049] It can be seen from Table 6 that the algorithm in this paper has strong robustness to illumination changes.

[0050] Table 7: Correct recognition rate of different algorithms on the Yale-B library

[0051]

[0052] It can be seen from Table 7 that the algorithm of this application is equivalent to PCA, SVM, MSR (muitiscale retinex), SQI (self-quotient image), LBP, DBN and other algorithms in the recognition rate of subset 2 with little change in illumination, but when the illumination becomes complex , as shown in the experimental results on subset 3, the algorithm of this application can maintain a relatively stable recognition rate, while the recognition rate of the traditional algorithm drops sharply, which further shows that the algorithm of this paper is more robust to illumination.

[0053] Summary: From the results of the above examples, it can be seen th...

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 method for identifying a human face under a non-restricted condition based on an LBP (local binary pattern) and deep learning. The method comprises the steps of testing a sample; training the sample; performing preprocessing; extracting LBP texture features; performing DBN (deep belief network) deep learning; and identifying a result. The method has the advantages that for the adopted identification of the human face under the non-restricted condition based on LBP and DBN combination, an experimental result in an LFW (labeled faces in the wild) library shows that effective features of a human face image under the non-restricted condition can be automatically extracted from bottom to top; the LBP is combined with a DBN, so that the shortcoming that the DBN cannot learn local structure features of the human face image is overcome, and abstract features learned by the DBN are slightly influenced by illumination, tiny translation and the like; and a relatively good identification effect is achieved under the non-restricted condition influenced by comprehensive factors such as poses, illumination, expressions, shielding and the like, and a relatively high identification rate is achieved in a Yale library influenced by various factors and a Yale-B library influenced by the illumination factor.

Description

technical field [0001] The invention relates to the technical field of face recognition, in particular to a face recognition method based on LBP and depth school under unrestricted conditions. Background technique [0002] As a non-invasive biometric identification method, face recognition has a wide range of applications in national defense security, video surveillance, and human-computer interaction. Traditional face recognition algorithms can achieve better results under the restricted conditions affected by one or several specific factors, but the recognition performance drops sharply under unrestricted conditions. At present, face recognition research under unrestricted conditions can be divided into two categories: face recognition methods based on 3-D models and 2-D models. Among them, the latter is a research hotspot. The face recognition method based on the 3-D model is very effective in overcoming the influence of posture and illumination in environmental factors...

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): G06K9/00G06N3/08
CPCG06N3/088G06V40/172
Inventor 徐海黎沈标刘熙田强韦勇
Owner NANJING LANTAI TRAFFIC FACILITIES CO LTD
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