WO2019095997A1 - Procédé et dispositif de reconnaissance d'image, dispositif informatique et support d'informations lisible par ordinateur - Google Patents

Procédé et dispositif de reconnaissance d'image, dispositif informatique et support d'informations lisible par ordinateur Download PDF

Info

Publication number
WO2019095997A1
WO2019095997A1 PCT/CN2018/112759 CN2018112759W WO2019095997A1 WO 2019095997 A1 WO2019095997 A1 WO 2019095997A1 CN 2018112759 W CN2018112759 W CN 2018112759W WO 2019095997 A1 WO2019095997 A1 WO 2019095997A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
database
region
query
query image
Prior art date
Application number
PCT/CN2018/112759
Other languages
English (en)
Chinese (zh)
Inventor
杨茜
牟永强
Original Assignee
深圳云天励飞技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳云天励飞技术有限公司 filed Critical 深圳云天励飞技术有限公司
Publication of WO2019095997A1 publication Critical patent/WO2019095997A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour

Definitions

  • a first aspect of the present application provides an image recognition method, the method comprising:
  • Whether the query image matches the database image is determined according to the cluster center of each region of the query image and the database image.
  • the query image and the database image are character images
  • the zoning of the query image and the database image includes:
  • the query image and the database image are respectively divided into upper and lower areas according to the character image in the query image and the database image, wherein the upper area corresponds to the upper body of the character, and the lower area corresponds to the lower body of the character.
  • determining whether the query image matches the database image according to the cluster center of each region of the query image and the database image includes:
  • the Mahalanobis distance of the cluster center of each region corresponding to the database image is calculated.
  • Figure 2 is a logarithmic versus RGB coordinate distribution of an image.
  • the image recognition method specifically includes the following steps:
  • the query image is an image that needs to be identified or matched
  • the database image is an image in a pre-established image library.
  • the image recognition method compares the query image with the database image to determine whether the query image matches the database image to confirm whether the content in the query image is consistent with the content in the database image. For example, when performing pedestrian recognition, the pedestrian image captured by the camera on the road is a query image, and the portrait library image of the traffic management system is a database image, and whether the pedestrian image and the portrait library image match are determined according to the similarity coefficient between the pedestrian image and the portrait library image.
  • the person in the pedestrian image is considered to be a character in the portrait library image; otherwise, if it does not match, the person in the non-portrait library image in the pedestrian image is considered to be able to image the pedestrian and another portrait library image. Identify.
  • the image recognition method can be applied to various fields such as video surveillance, product detection, medical diagnosis, and the like.
  • the present invention can be utilized for pedestrian identification, driver identification, vehicle identification, and the like.
  • clustering pixel points of the upper region R1 and the lower region R2 of the query image to obtain a cluster center (x 1 , y 1 ) of the upper region R1 of the query image and a cluster center (x 2 of the lower region R2 of the query image) y 2 ); clustering the pixel points of the upper region R1' and the lower region R2' of the database image to obtain the cluster center (x 1 ', y 1 ') and the lower region R2' of the upper region R1' of the query image Clustering center (x 2 ', y 2 ').
  • the distance between the query center and the cluster center of each region corresponding to the database image may be calculated, and whether the query image matches the database image is determined according to the distance between the query image and the cluster center of each region corresponding to the database image.
  • the Euclidean distance of the cluster center of each region corresponding to the database image of the query image may be calculated, and the query image is determined according to the Euclidean distance of the cluster center of each region corresponding to the query image and the database image. Whether the database images match.
  • the query image includes a cluster center (x1, y1) of the upper region R1 and a cluster center (x2, y2) of the lower region R2
  • the database image includes a cluster center (x1', y1') of the upper region R1'
  • the cluster center (x2', y2') of the lower region R2' is calculated as the cluster center (x1, y1) of the upper region R1 of the query image and the cluster center (x1', y1 of the upper region R1' of the database image.
  • Euclidean distance And the Euclidean distance between the cluster center (x2, y2) of the lower region R2 of the query image and the cluster center (x2', y2') of the lower region R2' of the database image
  • the Manhattan distance (ie, the absolute value distance) of the cluster center of each region corresponding to the database image of the query image may be calculated, and the query image is determined according to the Manhattan distance of the cluster center of each region corresponding to the query image and the database image. Whether the database images match.
  • the query image includes a cluster center (x1, y1) of the upper region R1 and a cluster center (x2, y2) of the lower region R2, and the database image includes a cluster center (x1', y1') of the upper region R1' and
  • the cluster center (x2', y2') of the lower region R2' is calculated as the cluster center (x1, y1) of the upper region R1 of the query image and the cluster center (x1', y1 of the upper region R1' of the database image.
  • other distances such as Mahalanobis distances, of the cluster center of each region of the query image corresponding to the database image may be calculated.
  • the distance between the query image and the cluster center of each region corresponding to the database image may be preset, and whether the query image matches the database image is determined according to the operation result.
  • an average distance (eg, (d 1 +d 2 )/2) of the cluster center of each region of the query image corresponding to the database image may be calculated, and each region corresponding to the database image is determined by the query image. Whether the average distance of the cluster center is less than or equal to the preset distance, and if the average distance of the cluster center of each region corresponding to the database image is less than or equal to the preset distance, the query image is judged to match the database image; otherwise If the average distance of the cluster center of each region corresponding to the query image and the database image is greater than the preset distance, it is determined that the query image does not match the database image.
  • the logarithm of the different poses and shooting angles is similar to the RGB coordinate distribution, so the robustness to the pose and the shooting angle is good, and the attitude and the shooting angle are invariant.
  • the logarithmic relative RGB coordinates are the normalized coordinates of the red component Ri and the green component Gi versus the green component Gi, which reduces the influence of illumination on the recognition, and thus is more robust to illumination and has illumination intensity invariance.
  • the logarithmic relative RGB coordinates reduce the three-dimensional coordinates of each pixel to two dimensions, which reduces the computational complexity and improves the recognition speed.
  • the pixel points are clustered according to the logarithmic relative RGB coordinate sub-region, and the pixel points of each region are transformed into a cluster center, which reduces the influence of the accidental error (ie, eliminates the influence of the stray point) and improves The robustness and anti-interference ability of the identification.
  • whether the query image and the database image match are determined according to the cluster center of each region of the query image and the database image, and the calculation amount of the calculation data of the cluster center is small, the operation complexity is low, and the matching result can be quickly obtained. Therefore, the image recognition method of the first embodiment can realize image recognition with fast high robustness and high anti-interference.
  • the area dividing unit 301 is configured to perform area division on the query image and the database image.
  • the query image is an image that needs to be identified or matched
  • the database image is an image in a pre-established image library.
  • the image recognition method compares the query image with the database image to determine whether the query image matches the database image to confirm whether the content in the query image is consistent with the content in the database image. For example, when performing pedestrian recognition, the pedestrian image captured by the camera on the road is a query image, and the portrait library image of the traffic management system is a database image, and whether the pedestrian image and the portrait library image match are determined according to the similarity coefficient between the pedestrian image and the portrait library image.
  • the image recognition method can be applied to various fields such as video surveillance, product detection, medical diagnosis, and the like.
  • the present invention can be utilized for pedestrian identification, driver identification, vehicle identification, and the like.
  • the same division method is adopted.
  • the query image and the database image are each divided into two upper and lower regions or two left and right regions.
  • the image recognition method is used for character recognition (for example, pedestrian recognition), and the query image and the database image are character images, and the query image and the database image may be divided into upper and lower regions according to the character shapes in the image. .
  • the upper area corresponds to the upper body of the character
  • the lower area corresponds to the lower body of the character.
  • the query image is divided into an upper region R1 and a lower region R2
  • the database image is divided into an upper region R1' and a lower region R2'.
  • the characters in the image are upright characters, since the proportions of the upright characters are roughly similar but the postures and actions are different, the division of the upper and lower regions according to the shape of the characters in the image is more robust.
  • the most colorful character costume is usually a jacket, so the character image is divided into upper and lower regions.
  • the query image and the database image may be respectively divided into two regions, and the query image and the database image may each be divided into more than two regions, for example, divided into three regions or four regions.
  • the matching unit 304 is configured to determine whether the query image and the database image match according to the query center and the cluster center of each region of the database image.
  • the sum of the distances of the cluster centers of each region of the query image corresponding to the database image may be calculated, and the cluster center of each region corresponding to the database image of the query image is determined. Whether the sum of the distances is less than or equal to the preset distance, and if the sum of the distances of the cluster centers of the respective regions corresponding to the database image is less than or equal to the preset distance, the query image is judged to match the database image; otherwise, if If the sum of the distances of the cluster centers of the regions corresponding to the database image corresponding to the database image is greater than the preset distance, it is determined that the query image does not match the database image.
  • an average distance (eg, (d 1 +d 2 )/2) of the cluster center of each region of the query image corresponding to the database image may be calculated, and each region corresponding to the database image is determined by the query image. Whether the average distance of the cluster center is less than or equal to the preset distance, and if the average distance of the cluster center of each region corresponding to the database image is less than or equal to the preset distance, the query image is judged to match the database image; otherwise If the average distance of the cluster center of each region corresponding to the query image and the database image is greater than the preset distance, it is determined that the query image does not match the database image.
  • the pixel points are clustered according to the logarithmic relative RGB coordinate sub-region, and the pixel points of each region are transformed into a cluster center, which reduces the influence of the accidental error (ie, eliminates the influence of the spurious point) and improves The robustness and anti-interference ability of the identification.
  • the query image and the database image match are determined according to the cluster center of each region of the query image and the database image, and the calculation amount of the calculation data of the cluster center is small, the operation complexity is low, and the matching result can be quickly obtained. Therefore, the image recognition method of the second embodiment can realize image recognition with fast high robustness and high anti-interference.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Library & Information Science (AREA)
  • Multimedia (AREA)
  • Probability & Statistics with Applications (AREA)
  • Databases & Information Systems (AREA)
  • Image Analysis (AREA)

Abstract

L'invention concerne un procédé de reconnaissance d'image. Le procédé consiste : à effectuer une division de région sur une image d'interrogation et une image de base de données; à calculer des coordonnées RVB relatives logarithmiques de chaque point de pixel dans chaque région de l'image d'interrogation et de l'image de base de données; à regrouper des points de pixels dans chaque région de l'image d'interrogation et de l'image de base de données en fonction des coordonnées RVB relatives logarithmiques de chaque point de pixel dans chaque région de l'image d'interrogation et de l'image de base de données afin d'obtenir un centre de regroupement de chaque région de l'image d'interrogation et de l'image de base de données; et à déterminer, en fonction du centre de regroupement de chaque région de l'image d'interrogation et de l'image de base de données, si l'image d'interrogation correspond à l'image de base de données. L'invention concerne également un dispositif de reconnaissance d'image, un dispositif informatique et un support d'informations lisible. Grâce à la présente invention, une reconnaissance d'image à robustesse élevée et à performance anti-brouillage élevée peut être rapidement mise en œuvre.
PCT/CN2018/112759 2017-11-15 2018-10-30 Procédé et dispositif de reconnaissance d'image, dispositif informatique et support d'informations lisible par ordinateur WO2019095997A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711132067.0 2017-11-15
CN201711132067.0A CN107871143B (zh) 2017-11-15 2017-11-15 图像识别方法及装置、计算机装置和计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2019095997A1 true WO2019095997A1 (fr) 2019-05-23

Family

ID=61754080

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/112759 WO2019095997A1 (fr) 2017-11-15 2018-10-30 Procédé et dispositif de reconnaissance d'image, dispositif informatique et support d'informations lisible par ordinateur

Country Status (2)

Country Link
CN (1) CN107871143B (fr)
WO (1) WO2019095997A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111950629A (zh) * 2020-08-11 2020-11-17 精英数智科技股份有限公司 对抗样本的检测方法、装置及设备
CN112329660A (zh) * 2020-11-10 2021-02-05 浙江商汤科技开发有限公司 一种场景识别方法、装置、智能设备及存储介质

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107871143B (zh) * 2017-11-15 2019-06-28 深圳云天励飞技术有限公司 图像识别方法及装置、计算机装置和计算机可读存储介质
CN107895021B (zh) * 2017-11-15 2019-12-17 深圳云天励飞技术有限公司 图像识别方法及装置、计算机装置和计算机可读存储介质
CN110689046A (zh) * 2019-08-26 2020-01-14 深圳壹账通智能科技有限公司 图像识别方法、装置、计算机装置及存储介质
CN116340991B (zh) * 2023-02-02 2023-11-07 魔萌动漫文化传播(深圳)有限公司 Ip图库素材资源的大数据管理方法、装置以及电子设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101794442A (zh) * 2010-01-25 2010-08-04 哈尔滨工业大学 一种从可见光图像中提取光照不敏感信息的标定方法
US20160092736A1 (en) * 2014-09-30 2016-03-31 C/O Canon Kabushiki Kaisha System and method for object re-identification
CN105574515A (zh) * 2016-01-15 2016-05-11 南京邮电大学 一种无重叠视域下的行人再识别方法
CN107871143A (zh) * 2017-11-15 2018-04-03 深圳云天励飞技术有限公司 图像识别方法及装置、计算机装置和计算机可读存储介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8913829B2 (en) * 2012-06-05 2014-12-16 Tandent Vision Science, Inc. Automatic processing scale estimation for use in an image process
CN105139042A (zh) * 2015-09-08 2015-12-09 携程计算机技术（上海）有限公司 图像识别方法及系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101794442A (zh) * 2010-01-25 2010-08-04 哈尔滨工业大学 一种从可见光图像中提取光照不敏感信息的标定方法
US20160092736A1 (en) * 2014-09-30 2016-03-31 C/O Canon Kabushiki Kaisha System and method for object re-identification
CN105574515A (zh) * 2016-01-15 2016-05-11 南京邮电大学 一种无重叠视域下的行人再识别方法
CN107871143A (zh) * 2017-11-15 2018-04-03 深圳云天励飞技术有限公司 图像识别方法及装置、计算机装置和计算机可读存储介质

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111950629A (zh) * 2020-08-11 2020-11-17 精英数智科技股份有限公司 对抗样本的检测方法、装置及设备
CN112329660A (zh) * 2020-11-10 2021-02-05 浙江商汤科技开发有限公司 一种场景识别方法、装置、智能设备及存储介质
CN112329660B (zh) * 2020-11-10 2024-05-24 浙江商汤科技开发有限公司 一种场景识别方法、装置、智能设备及存储介质

Also Published As

Publication number Publication date
CN107871143B (zh) 2019-06-28
CN107871143A (zh) 2018-04-03

Similar Documents

Publication Publication Date Title
WO2019095997A1 (fr) Procédé et dispositif de reconnaissance d'image, dispositif informatique et support d'informations lisible par ordinateur
US11703951B1 (en) Gesture recognition systems
Sarma et al. Methods, databases and recent advancement of vision-based hand gesture recognition for hci systems: A review
WO2019095998A1 (fr) Procédé et dispositif de reconnaissance d'image, dispositif informatique et support de stockage lisible par ordinateur
CN106682598B (zh) 一种基于级联回归的多姿态的人脸特征点检测方法
WO2022027912A1 (fr) Procédé et appareil de reconnaissance de pose du visage, dispositif terminal et support de stockage
Ban et al. Face detection based on skin color likelihood
US20190108447A1 (en) Multifunction perceptrons in machine learning environments
US20220375236A1 (en) License plate identification method and system thereof
JP4217664B2 (ja) 画像処理方法、画像処理装置
WO2019114036A1 (fr) Procédé et dispositif de détection de visage, dispositif informatique et support d'informations lisible par ordinateur
US20090180671A1 (en) Multi-view face recognition method and system
JP2020515983A (ja) 対象人物の検索方法および装置、機器、プログラム製品ならびに媒体
CN102622589A (zh) 一种基于gpu的多光谱人脸检测方法
US9213897B2 (en) Image processing device and method
WO2021036309A1 (fr) Procédé et appareil de reconnaissance d'image, appareil informatique, et support de stockage
WO2020087922A1 (fr) Procédé d'identification d'attribut facial, dispositif, dispositif informatique et support d'informations
CN110706238B (zh) 对点云数据进行分割的方法及装置、存储介质和电子设备
González-Ortega et al. Real-time hands, face and facial features detection and tracking: Application to cognitive rehabilitation tests monitoring
CN112651321A (zh) 档案处理方法、装置及服务器
Nasri et al. A novel approach for dynamic hand gesture recognition using contour-based similarity images
Juang et al. Stereo-camera-based object detection using fuzzy color histograms and a fuzzy classifier with depth and shape estimations
WO2023279604A1 (fr) Procédé de ré-identification, procédé d'apprentissage pour un réseau de ré-identification de cible et produit associé
CN111797862A (zh) 任务处理方法、装置、存储介质和电子设备
Cai et al. Robust facial expression recognition using RGB-D images and multichannel features

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18879388

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 14/08/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18879388

Country of ref document: EP

Kind code of ref document: A1