WO2018019149A1 - 人体性别自动识别方法及装置 - Google Patents

人体性别自动识别方法及装置 Download PDF

Info

Publication number
WO2018019149A1
WO2018019149A1 PCT/CN2017/093238 CN2017093238W WO2018019149A1 WO 2018019149 A1 WO2018019149 A1 WO 2018019149A1 CN 2017093238 W CN2017093238 W CN 2017093238W WO 2018019149 A1 WO2018019149 A1 WO 2018019149A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
feature information
region
gender
millimeter wave
Prior art date
Application number
PCT/CN2017/093238
Other languages
English (en)
French (fr)
Inventor
陈寒江
刘艳丽
祁春超
吴光胜
赵术开
丁庆
Original Assignee
华讯方舟科技有限公司
深圳市无牙太赫兹科技有限公司
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 华讯方舟科技有限公司, 深圳市无牙太赫兹科技有限公司 filed Critical 华讯方舟科技有限公司
Priority to EP17833452.0A priority Critical patent/EP3493103A4/en
Priority to US16/321,807 priority patent/US11250249B2/en
Publication of WO2018019149A1 publication Critical patent/WO2018019149A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/11Region-based segmentation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/103Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30196Human being; Person

Definitions

  • the invention relates to the field of image processing and security technology, and in particular to a method and device for automatically identifying human gender.
  • Millimeter waves have a certain penetration during propagation. Through the millimeter wave imaging technology, it is possible to obtain an image of the image of the object blocked by the body of the scanned body; therefore, the gender of the subject can be identified based on the millimeter wave imaging result; if it is directly observed by the security inspection operator during the detection process To complete the detection of suspected dangerous objects, the consumption of manpower, financial resources and time is enormous.
  • the traditional gender recognition method on the one hand, mainly for visible light image data; on the other hand, for the millimeter wave image, the human gender recognition method uses a single feature information (such as gray variance information) for classification and recognition.
  • the millimeter wave image is essentially different from the visible light image imaging mechanism.
  • the millimeter wave image has low gray level, low definition, and is affected by the multiplicative noise of the coherent plaque.
  • the pattern recognition method in visible light images can not be directly applied in millimeter wave images, and the calculation efficiency is low.
  • the millimeter wave images of different genders have many forms in the feature representation.
  • the gray variance information is only one of the possible differences between different genders. The characteristic information, therefore, based on a single feature detection method can not meet the needs of automatic gender recognition in the millimeter wave security system, which leads to low recognition accuracy.
  • an embodiment of the technical solution of the present invention is:
  • a method for automatically identifying a human gender including the following steps:
  • Obtaining a millimeter-wave grayscale image to be identified determining a location of a gender portion of the human body in the millimeter-wave grayscale image according to a preset body proportion; the gender location region location includes a chest region location and a crotch region location;
  • the feature information is extracted from the normalized region sub-image to obtain the feature information of the normalized region sub-image;
  • the feature information includes the shape feature information of the chest region position and the chest region feature information, and the gray of the crotch region position Variance information and crotch area feature information;
  • the millimeter-wave grayscale images are identified according to the size of the feature distances by using the preset classifiers corresponding to the feature information, and the results are respectively output;
  • the feature distance is the distance between the feature information and the feature information of the corresponding preset classifier. ;
  • the output result is integrated to obtain the classification and recognition result of the millimeter wave gray image.
  • an automatic gender recognition device including:
  • a normalization processing unit for performing scale normalization on the region sub-image to obtain a normalized region sub-image
  • the feature information extracting unit is configured to perform feature information extraction on the normalized region sub-image to obtain feature information of the normalized region sub-image;
  • the feature information includes shape feature information of the chest region position and chest region feature information, Gray-scale variance information and crotch region feature information of the position of the crotch region;
  • the classification unit is configured to identify the millimeter wave grayscale image according to the size of the feature distance by using each preset classifier corresponding to the feature information, and output the result separately;
  • the feature distance is the feature information and the corresponding preset classifier Distance of feature information;
  • the identification unit is configured to integrate the output result to obtain a classification and recognition result of the millimeter wave gray image.
  • the method and device for automatically identifying human gender in the invention achieve high recognition accuracy of human gender automatic recognition in millimeter wave images, and different characteristics of millimeter wave image imaging results of human body millimeter wave images for different genders
  • the information is analyzed, and different feature information is extracted.
  • the multiple genders are used to identify the gender, and the human gender automatic recognition of the millimeter wave image is realized.
  • the invention can be effectively applied to the real millimeter wave security inspection system, has high recognition rate and calculation efficiency, and solves the problem that the millimeter wave security inspection system adopts different ways of privacy protection and detection methods for different genders.
  • Embodiment 1 is a schematic flow chart of Embodiment 1 of a method for automatically identifying a human gender according to the present invention
  • Embodiment 2 is a schematic flow chart of Embodiment 2 of a method for automatically identifying a human gender according to the present invention
  • FIG. 3 is a view showing an example of input images of male and female human body automatic recognition methods according to the present invention.
  • FIG. 4 is a view showing an example of an image obtained by automatically extracting a sub-image of a male chest and an ankle region and normalizing the scale;
  • FIG. 5 is a view showing an example of an image obtained by automatically extracting a sub-image of a female chest and an ankle region and normalizing the scale;
  • FIG. 6 is a schematic structural view of Embodiment 1 of a human gender automatic identification device according to the present invention.
  • Embodiment 1 of human body gender automatic identification method of the present invention is a human body gender automatic identification method of the present invention.
  • FIG. 1 is a schematic flowchart of the first embodiment of the human gender automatic identification method according to the present invention. As shown in Figure 1, the following steps can be included:
  • Step S110 Acquire a millimeter wave grayscale image to be identified currently, and determine a gender location region position of the human body in the millimeter wave grayscale image according to a preset body proportion; the gender location region position includes a chest region position and a crotch region position;
  • Step S120 extracting a region sub-image corresponding to the location of the gender region
  • Step S130 Performing scale normalization on the region sub-image to obtain a normalized region sub-image
  • Step S140 performing feature information extraction on the normalized region sub-image to obtain feature information of the normalized region sub-image;
  • the feature information includes shape feature information of the chest region position, chest region feature information, and crotch region location Gray-scale variance information and crotch region feature information;
  • Step S150 Identify, by using the preset classifiers corresponding to the feature information, the millimeter wave grayscale images according to the size of the feature distances, and respectively output the results;
  • the feature distance is the feature information and the characteristics of the corresponding preset classifiers. Distance of information;
  • Step S160 Integrating the output result to obtain a classification and recognition result of the millimeter wave grayscale image.
  • the inventors of the present invention have found through a large number of experiments that the human millimeter wave images of different genders have obvious shape characteristics in the chest region sub-image, while the male chest sub-images are relatively flat. Grayscale; in the temporal region sub-image, the male temporal image appears as a larger grayscale variance, while the female temporal image exhibits a smoother grayscale; in the chest and temporal region sub-images, Men and women present different grayscale information in different small areas.
  • the input millimeter wave image determines the position of the human chest and the crotch region according to the body proportion, and then extracts the chest region sub-image and the crotch region sub-image respectively and performs scale normalization separately.
  • the shape feature information and the haar-like feature information are extracted for the normalized chest sub-image, the haar-like feature information and the gray variance information are extracted for the partial image, and the shape feature is extracted for the chest region sub-image.
  • step S110 may include:
  • a binary image I b of the millimeter wave grayscale image is obtained based on the following formula:
  • X is the number of rows of the millimeter wave grayscale image
  • Y is the number of columns of the millimeter wave grayscale image
  • the spatial distribution histogram vector H y is constructed in the y direction for the human body region segmented in I b (x, y) based on the following formula:
  • represents a Decara ⁇ function
  • x represents a row coordinate
  • y represents a column coordinate
  • the histogram vector H y is smoothed based on the following formula:
  • the left and right pixels of the column of the bit line in the binary image I b are counted according to the middle behavioral benchmark of the vertical median line of the human body, and the line x t where the human head is located and the line x f where the human foot is located are obtained;
  • the height H e x f - x t is determined ;
  • the ratio of the upper chest to the height, R cr is the ratio of the ankle to the height in the statistical sense.
  • the step of performing the iterative operation on the y-direction histogram vector H y to obtain the column of the human vertical vertical bit line may include:
  • the initial center point position value is set in the y-direction spatial distribution histogram vector H y based on the following formula
  • step of acquiring shape feature information in step S130 may include:
  • an edge curve of the normalized region sub-image is obtained, and the shape feature information is acquired based on the curvature of the point on the edge curve;
  • the step of acquiring the grayscale variance information in step S130 may include:
  • the normalized sub-image is subjected to gray-scale variance calculation processing to obtain gray-scale variance information.
  • the chest region feature information includes chest region haar-like feature information; the crotch region feature information includes crotch region haar-like feature information.
  • the normalized chest sub-image edge curve e ch [e 1 ,...,e i ,...,e n ] can be obtained by using the Canny edge detection algorithm for the normalized chest sub-image I ch .
  • SVD matrix singular value decomposition
  • decomposition is performed on each edge e i to estimate the tangential direction.
  • the ratio of the change of the tangential direction of the adjacent point to the Euclidean distance between them can be approximated as the curvature of the point.
  • Shape feature information of the chest sub-image E ch [e 1 ,...,e i ,...,e n ] can be obtained by using the Canny edge detection algorithm for the normalized chest sub-image I ch .
  • SVD matrix singular value decomposition
  • step S160 may include:
  • the millimeter wave grayscale image is classified to obtain the classification recognition result.
  • each classifier sets a respective weight according to the classifier accuracy rate, and if the results recognized by the plurality of classifiers are closer to which class, the current classifier will be The recognition image belongs to this class.
  • Embodiment 2 of the method for automatically identifying human gender in the present invention is a diagrammatic representation of Embodiment 2 of the method for automatically identifying human gender in the present invention.
  • FIG. 2 is a schematic flowchart of the second embodiment of the human gender automatic identification method; As shown in Figure 2, the following steps can be included:
  • FIG. 3 is a diagram showing an example of an input image of a male (left) and a female (right) human gender automatic recognition method according to the present invention
  • Step b) may comprise the following substeps:
  • the number of rows and columns of the matrix I b are equal to the number of rows and columns of the millimeter wave grayscale image I, respectively, and X and Y are respectively the number of rows and columns of I; wherein, the image is binarized, so that the target is White (grayscale is 255) and background is black (grayscale is 0).
  • represents a Decara ⁇ function
  • x represents a row coordinate
  • y represents a column coordinate
  • Step b4) may include the following substeps:
  • Step c) may comprise the following substeps:
  • Step d) may comprise the following substeps:
  • FIG. 4 is an example of an image obtained by automatically extracting a sub-image of a male chest and an ankle region and normalizing the scale of the human body according to the present invention
  • FIG. 5 is a method for automatically extracting a female chest and ankle, respectively. The region sub-image and the image normalization image after the scale is normalized.
  • Step e) may comprise the following substeps:
  • Step f) may comprise the following substeps:
  • N is an integer smaller than the smaller of the number of male images and the number of female images in the sample library
  • the training classifier S1 the N distances of the image shape feature information to be recognized and the N-frame male image and the N-frame female image shape feature information are kept small
  • the current image to be identified belongs to the same class as the image closest to it;
  • the feature information is respectively sorted from the N-frame male image and the N-frame female image chest sub-image and the ⁇ -sub-image haar-like feature information, respectively, from small to large, and the current image to be recognized and the image closest to it belong to Same class
  • the training classifier S4 the gradation variance information of the image to be recognized and the N-frame male image and the N-frame female image gray-scale variance information N
  • the distances are sorted from small to large, and the current image to be identified belongs to the same category as the image closest to it;
  • Step g) includes the following substeps:
  • each classifier sets its own weight according to the classifier accuracy rate, and if the results recognized by the plurality of classifiers are closer to which class, then The current image to be identified is attributed to the class.
  • the gray-scale segmentation threshold algorithm involved in step b1) is a haar-like feature extraction algorithm involved in step e2) by an automatic threshold segmentation algorithm, and a weight is set according to the classifier accuracy in step g).
  • the method is well-known in the art and will not be described here.
  • the present invention presents different characteristics of the chest and crotch region sub-images in the millimeter-wave images of different genders, and adopts different feature extraction methods instead of simply relying on the gray-scale variance for recognition, thereby making the present invention effective.
  • the accuracy of the recognition is improved, and the real-time performance can be guaranteed.
  • the invention adopts a plurality of classifier integration methods, and reduces the false positive rate of the single classifier classification and recognition result.
  • Embodiment 1 of human gender automatic identification device of the present invention is a human gender automatic identification device of the present invention.
  • the present invention also provides a human body gender automatic identification device embodiment 1;
  • a schematic diagram of the structure of the automatic gender identification device embodiment 1; as shown in FIG. 6, the method may include:
  • the determining area location unit 610 is configured to acquire a millimeter wave grayscale image to be currently identified, and determine a gender location area of the human body in the millimeter wave grayscale image according to a preset body proportion; the gender location area location includes a chest region location and a location Regional location;
  • Extracting the sub-image unit 620 configured to extract a region sub-image corresponding to the location of the gender region
  • the normalization processing unit 630 is configured to perform normalization on the region sub-image to obtain a normalized region sub-image
  • the feature information extracting unit 640 is configured to perform feature information extraction on the normalized region sub-image to obtain feature information of the normalized region sub-image;
  • the feature information includes shape feature information of the chest region position and chest region feature information. Gray-scale variance information of the position of the crotch region and feature information of the crotch region;
  • the classifying unit 650 is configured to identify the millimeter wave grayscale image based on the corresponding feature information by using each preset classifier, and output the result separately; the classifier is trained based on the feature information corresponding to the training sample in the millimeter wave image database. of;
  • the identifying unit 660 is configured to integrate the output result to obtain a classification and recognition result of the millimeter wave grayscale image.
  • the feature information extracting unit 640 includes:
  • the chest feature extraction module 642 is configured to extract feature information of the chest region, and obtain an edge curve of the normalized region sub-image according to the edge detection algorithm, and acquire shape feature information based on the curvature of the point on the edge curve;
  • the crotch feature extraction module 644 is configured to extract the crotch region feature information, and perform a gray-scale variance calculation process on the normalized region sub-image to obtain the gray-scale variance information.
  • the chest region feature information includes chest region haar-like feature information; the crotch region feature information includes crotch region haar-like feature information.
  • the identifying unit 660 is configured to classify the millimeter wave grayscale image according to the output result and the weight of each classifier, and obtain the classification and recognition result.
  • Embodiment 1 of the automatic human gender automatic recognition device of the present invention achieves high recognition accuracy for realizing automatic gender recognition in millimeter wave images, and different characteristics of millimeter wave images for different genders of human body millimeter wave image imaging results
  • the content information is analyzed, and different feature information is extracted.
  • the gender is recognized by multiple classifiers, and the human gender automatic recognition of the millimeter wave image is realized.
  • the invention can be effectively applied to the real millimeter wave security inspection system, has high recognition rate and calculation efficiency, and solves the problem that the millimeter wave security inspection system adopts different ways of privacy protection and detection methods for different genders.

Abstract

一种人体性别自动识别方法及装置,人体性别自动识别方法,包括以下步骤:获取当前待识别的毫米波灰度图像,根据预设的身体比例,确定毫米波灰度图像中人体的性别部位区域位置;提取性别部位区域位置对应的区域子图像;对区域子图像进行尺度归一化,得到归一化后的区域子图像;对归一化后的区域子图像进行特征信息提取,得到归一化后的区域子图像的特征信息;通过与特征信息对应的各预设分类器,根据特征距离的大小排序,对毫米波灰度图像进行识别,并分别输出结果;对输出结果进行集成,得到毫米波灰度图像的分类识别结果。上述方法具有较高的识别率和计算效率,解决了毫米波安检系统中的隐私保护和检测方法的问题。

Description

人体性别自动识别方法及装置 技术领域
本发明涉及图像处理与安检技术领域,特别是涉及一种人体性别自动识别方法及装置。
背景技术
毫米波在传播过程中具有一定穿透性。通过毫米波成像技术,能够对被扫描人体获得衣物遮挡下物体成像结果图像;因此可以基于毫米波成像结果对被检测者性别进行识别;如果在检测过程中,直接通过安检操作人员肉眼观察的方式来完成可疑危险物体的检测,在人力、财力及时间上的消耗是巨大的。而传统的性别识别方法:一方面,主要是针对可见光图像数据;另一方面,针对毫米波图像中人体性别识别方法多采用单一的特征信息(如灰度方差信息)进行分类识别。
在实现过程中,发明人发现传统技术中至少存在如下问题:
1)毫米波图像与可见光图像成像机理有着本质的区别,毫米波图像灰度层次低,清晰度低,且受相干斑乘性噪声的影响。可见光图像中的模式识别方法在毫米波图像中不能直接适用,计算效率低;2)不同性别毫米波图像在特征表现上存在多种形式,灰度方差信息仅仅为其中一种可能的区别不同性别的特征信息,因此,基于单一的特征检测方法并不能满足毫米波安检系统中性别自动识别的需求,易导致识别准确率低。
发明内容
基于此,有必要针对传统性别识别方法计算效率和识别准确率低的问题,提供一种人体性别自动识别方法及装置。
为了实现上述目的,本发明技术方案的实施例为:
一方面,提供了一种人体性别自动识别方法,包括以下步骤:
获取当前待识别的毫米波灰度图像,根据预设的身体比例,确定毫米波灰度图像中人体的性别部位区域位置;性别部位区域位置包括胸部区域位置和裆部区域位置;
提取性别部位区域位置对应的区域子图像;
对区域子图像进行尺度归一化,得到归一化后的区域子图像;
对归一化后的区域子图像进行特征信息提取,得到归一化后的区域子图像的特征信息;特征信息包括胸部区域位置的形状特征信息和胸部区域特征信息、裆部区域位置的灰度方差信息和裆部区域特征信息;
通过与特征信息对应的各预设分类器,根据特征距离的大小排序,对毫米波灰度图像进行识别,并分别输出结果;特征距离为特征信息与对应的预设分类器的特征信息的距离;
对输出结果进行集成,得到毫米波灰度图像的分类识别结果。
另一方面,提供了一种人体性别自动识别装置,包括:
确定区域位置单元,用于获取当前待识别的毫米波灰度图像,根据预设的身体比例,确定毫米波灰度图像中人体的性别部位区域位置;性别部位区域位置包括胸部区域位置和裆部区域位置;
提取子图像单元,用于提取性别部位区域位置对应的区域子图像;
归一化处理单元,用于对区域子图像进行尺度归一化,得到归一化后的区域子图像;
特征信息提取单元,用于对归一化后的区域子图像进行特征信息提取,得到归一化后的区域子图像的特征信息;特征信息包括胸部区域位置的形状特征信息和胸部区域特征信息、裆部区域位置的灰度方差信息和裆部区域特征信息;
分类单元,用于通过与特征信息对应的各预设分类器,根据特征距离的大小排序,对毫米波灰度图像进行识别,并分别输出结果;特征距离为特征信息与对应的预设分类器的特征信息的距离;
识别单元,用于对输出结果进行集成,得到毫米波灰度图像的分类识别结果。
上述技术方案具有如下有益效果:
本发明人体性别自动识别方法及装置,为实现毫米波图像中人体性别自动识别达到较高的识别准确率,针对不同性别的人体毫米波图像成像结果表现出的不同特点,对毫米波图像的内容信息进行分析,提取不同的特征信息采用多个分类器集成对性别进行识别,实现了毫米波图像的人体性别自动识别。本发明能有效适用于真实的毫米波安检系统中,具有较高的识别率和计算效率,解决了毫米波安检系统中对针对不同性别的人体采取不同方式的隐私保护和检测方法的问题。
附图说明
图1为本发明人体性别自动识别方法实施例1的流程示意图;
图2为本发明人体性别自动识别方法实施例2的流程示意图;
图3为本发明人体性别自动识别方法男性和女性输入图像示例图;
图4为本发明人体性别自动识别方法分别提取男性胸部和裆部区域子图像并进行尺度归一化后图像示例图;
图5为本发明人体性别自动识别方法分别提取女性胸部和裆部区域子图像并进行尺度归一化后图像示例图;
图6为本发明人体性别自动识别装置实施例1的结构示意图。
具体实施方式
为了便于理解本发明,下面将参照相关附图对本发明进行更全面的描述。附图中给出了本发明的首选实施例。但是,本发明可以以许多不同的形式来实现,并不限于本文所描述的实施例。相反地,提供这些实施例的目的是使对本发明的公开内容更加透彻全面。
除非另有定义,本文所使用的所有的技术和科学术语与属于本发明的技术领域的技术人员通常理解的含义相同。本文中在本发明的说明书中所使用的术语只是为了描述具体的实施例的目的,不是旨在于限制本发明。本文所使用的术语“及/或”包括一个或多个相关的所列项目的任意的和所有的组合。
本发明人体性别自动识别方法实施例1:
为了解决传统性别识别方法计算效率和识别准确率低的问题,本发明提供了一种人体性别自动识别方法实施例1;图1为本发明人体性别自动识别方法实施例1的流程示意图;如图1所示,可以包括以下步骤:
步骤S110:获取当前待识别的毫米波灰度图像,根据预设的身体比例,确定毫米波灰度图像中人体的性别部位区域位置;性别部位区域位置包括胸部区域位置和裆部区域位置;
步骤S120:提取性别部位区域位置对应的区域子图像;
步骤S130:对区域子图像进行尺度归一化,得到归一化后的区域子图像;
步骤S140:对归一化后的区域子图像进行特征信息提取,得到归一化后的区域子图像的特征信息;特征信息包括胸部区域位置的形状特征信息和胸部区域特征信息、裆部区域位置的灰度方差信息和裆部区域特征信息;
步骤S150:通过与特征信息对应的各预设分类器,根据特征距离的大小排序,对毫米波灰度图像进行识别,并分别输出结果;特征距离为特征信息与对应的预设分类器的特征信息的距离;
步骤S160:对输出结果进行集成,得到毫米波灰度图像的分类识别结果。
具体而言,本发明的发明人经过大量试验发现:不同性别的人体毫米波图像,在胸部区域子图像中,女性胸部子图像表现出明显的形状特征,而男性胸部子图像表现为较平缓的灰度;在裆部区域子图像中,男性裆部子图像表现为更大的灰度方差,而女性裆部子图像表现出更为平滑的灰度;在胸部和裆部区域子图像中,男性和女性在不同的小区域内表现为不同的灰度信息。
本发明为了提高人体性别识别的准确性,对输入毫米波图像根据身体比例确定人体胸部和裆部区域的位置,然后分别提取出胸部区域子图像和裆部区域子图像并分别进行尺度归一化;针对归一化后的胸部子图像进行形状特征信息和haar-like特征信息的提取,裆部子图像进行haar-like特征信息和灰度方差信息的提取;针对胸部区域子图像提取其形状特征信息,训练一个分类器;针对裆部区域子图像提取其灰度方差信息,训练一个分类器;针对胸部区域子图像和裆部区域子图像分别提取其haar-like特征信息,训练一个分类器;最后将三个分类器进行集成判断,完成人体毫米波图像人体性别的自动识别。
在一个具体的实施例中,步骤S110可以包括:
根据分割阈值T,基于以下公式获得毫米波灰度图像的二值图像Ib
Figure PCTCN2017093238-appb-000001
X为毫米波灰度图像的行数目;Y为毫米波灰度图像的列数目;
基于以下公式对Ib(x,y)中分割出的人体区域在y方向构建空间分布直方图向量Hy
Figure PCTCN2017093238-appb-000002
其中,δ表示迪克拉δ函数;x表示行坐标;y表示列坐标;
基于以下公式对直方图向量Hy进行平滑处理:
Figure PCTCN2017093238-appb-000003
对y方向直方图向量Hy进行迭代运算,得到人体垂直中位线所在列;
以人体垂直中位线所在列的中间行为基准,对二值图像Ib中位线所在列的左右像素进行统计,获取人体头顶所在的行xt和人体脚底所在的行xf
根据人体头顶所在的行xt和人体脚底所在的行xf,确定人体身高He=xf-xt
根据预设的身体比例,基于以下公式计算胸部起始行xch=xf-He*Rch,裆部起始行xcr=xf-He*Rcr;其中Rch为统计意义上胸部与身高的比例,Rcr为统计意义上裆部与身高的比例。
在一个具体的实施例中,对y方向直方图向量Hy进行迭代运算,得到人体垂直中位线所在列的步骤可以包括:
基于以下公式,在y方向空间分布直方图向量Hy中设定初始中心点位置值
Figure PCTCN2017093238-appb-000004
Figure PCTCN2017093238-appb-000005
基于以下公式,分别计算空间分布位置低于
Figure PCTCN2017093238-appb-000006
的集合Y1与空间分布位置高于
Figure PCTCN2017093238-appb-000007
的集合Y2的空间分布均值
Figure PCTCN2017093238-appb-000008
Figure PCTCN2017093238-appb-000009
Figure PCTCN2017093238-appb-000010
Figure PCTCN2017093238-appb-000011
Figure PCTCN2017093238-appb-000012
时停止迭代,获取人体中位线所在列的位置
Figure PCTCN2017093238-appb-000013
在一个具体的实施例中,步骤S130中获取形状特征信息的步骤可以包括:
根据边缘检测算法,得到提取归一化后的区域子图像的边缘曲线,基于边缘曲线上点的曲率,获取形状特征信息;
步骤S130中获取灰度方差信息的步骤可以包括:
对归一化后的区域子图像进行灰度方差计算处理,获取灰度方差信息。
在一个具体的实施例中,胸部区域特征信息包括胸部区域haar-like特征信息;裆部区域特征信息包括裆部区域haar-like特征信息。
具体而言,即可以针对归一化后胸部子图像Ich使用Canny边缘检测算法得到归一化后胸部子图像边缘曲线ech=[e1,…,ei,…,en],再对边缘线各点ei进行SVD(矩阵奇异值分解)分解估计出其切线方向,则邻近点切线方向的改变量对其之间欧式距离的比值可近似认为是该点的曲率大小,进一步构成胸部子图像的形状特征信息Ech
针对归一化后胸部子图像Ich和裆部子图像Icr分别提取其haar-like特征信息Fch和Fcr;针对归一化后裆部子图像Icr计算其灰度方差Tcr
在一个具体的实施例中,步骤S160可以包括:
根据输出结果和各分类器的权重,对毫米波灰度图像进行归类,获取分类识别结果。
具体而言,针对上述步骤训练出的分类器,对每个分类器根据分类器准确率来设定一个各自的权重,若多个分类器识别的结果更接近于哪一类,则将当前待识别图像归属于该类。
本发明人体性别自动识别方法实施例2:
为了解决传统性别识别方法计算效率和识别准确率低的问题,本发明还提供了一种人体性别自动识别方法实施例2;图2为本发明人体性别自动识别方法实施例2的流程示意图;如图2所示,可以包括以下步骤:
a)输入包含人体毫米波探测结果的毫米波灰度图像I;图3为本发明人体性别自动识别方法男性(图左)和女性(图右)输入图像示例图;
b)对毫米波灰度图像I根据统计数据的身体比例确定胸部和裆部区域位置;
步骤b)可以包括如下子步骤:
b1)通过分割阈值T,获得二值化图像Ib:将毫米波灰度图像I中灰度值高于T的像素区域被认为人体区域,灰度值设为255,其余位置被认为背景区域,设置为0:
Figure PCTCN2017093238-appb-000014
矩阵Ib的行数、列数分别与毫米波灰度图像I的行、列数目相等,且X和Y分别为I的行和列数目;其中,对图像进行二值化,就是使得目标为白色(灰度表现为255),背景为黑色(灰度表现为0)。
b2)对Ib(x,y)中分割出的人体区域在y方向构建空间分布直方图向量Hy
Figure PCTCN2017093238-appb-000015
其中,δ表示迪克拉δ函数;x表示行坐标;y表示列坐标;
b3)对直方图向量Hy进行平滑处理,平滑尺度为3(平滑尺度即一个平滑窗口大小):
Figure PCTCN2017093238-appb-000016
b4)对y方向直方图向量Hy进行迭代运算,求取人体垂直中位线所在列;
b5)以步骤b4)获得的人体垂直中位线所在列的中间行为基准,对二值图像Ib自下而上对中位线所在列的左右像素进行统计,若灰度值为255的像素数小于设定的阈值Tt,则将该行标记为头顶所在的行xt
b6)根据设定的搜索范围x∈[X-TF,X],y∈[0,Y],其中,TF为根据经验设定的一个脚底行与图像行数距离范围,对二值图像Ib自上而下进行像素统计,若灰度值为255的像素数小于设定的阈值Tf,则将该行标记为脚底所在的行xf
b7)根据步骤b5)和步骤b6)确定的头顶所在行和脚底所在行确定人体身高He=xf-xt,并进一步由人体比例分别计算胸部起始行为xch=xf-He*Rch,裆部起始行为xcr=xf-He*Rcr,其中Rch为统计意义上胸部与身高的比例,Rcr为统计意义上裆部与身高的比例。
步骤b4)可以包括如下子步骤:
b4-1)在y方向空间分布直方图向量Hy中设定初始中心点位置值
Figure PCTCN2017093238-appb-000017
Figure PCTCN2017093238-appb-000018
b4-2)分别计算空间分布位置低于
Figure PCTCN2017093238-appb-000019
的集合Y1与空间分布位置高于
Figure PCTCN2017093238-appb-000020
的集合Y2的空间分布均值
Figure PCTCN2017093238-appb-000021
Figure PCTCN2017093238-appb-000022
Figure PCTCN2017093238-appb-000023
b4-3)令
Figure PCTCN2017093238-appb-000024
Figure PCTCN2017093238-appb-000025
则停止迭代,此时
Figure PCTCN2017093238-appb-000026
即为人体中位线所在位置,否则返回步骤b4-2)继续迭代;
c)对毫米波灰度图像I,分别提取其胸部和裆部区域子图像Ich和Icr
步骤c)可以包括如下子步骤:
c1)设定胸部搜索范围x∈[xch,He*Rch],y∈[yt-He*Wch,yt+He*Wch],对胸部区域图像进行提取,获得胸部区域子图像Ich,其中Wch为统计意义上胸部宽度与身高的比例;
c2)设定裆部搜索范围x∈[xcr,He*Rcr],y∈[yt-He*Wcr,yt+He*Wcr],对裆部区域图像进行提取,获得裆部区域子图像Icr,其中Wcr为统计意义上裆部宽度与身高的比例。
d)分别对胸部和裆部区域子图像Ich和Icr进行尺度归一化,分别获得归一化后的胸部和裆部子图像Ich′和Icr′;
步骤d)可以包括如下子步骤:
d1)将胸部子图像Ich映射到一个大小为Xch*Ych的独立空间,获得新的胸部子图像Ich′;
d2)将裆部子图像Icr映射到一个大小为Xcr*Ycr的独立空间,获得新的裆部子图像Icr′;
其中,图4为本发明人体性别自动识别方法分别提取男性胸部和裆部区域子图像并进行尺度归一化后图像示例图;图5为本发明人体性别自动识别方法分别提取女性胸部和裆部区域子图像并进行尺度归一化后图像示例图。
e)针对归一化后的胸部子图像Ich′提取其形状特征信息和haar-like特征信息,针对归一化后的裆部子图像Icr′提取其灰度方差信息和haar-like特征信息;
步骤e)可以包括如下子步骤:
e1)针对归一化后胸部子图像Ich使用Canny边缘检测算法得到归一化后胸部子图像边缘曲线ech=[e1,…,ei,…,en],再对边缘线各点ei进行SVD分解估计出其切线方向,则邻近点切线方向的改变量对其之间欧式距离的比值可近似认为 是该点的曲率大小,进一步构成胸部子图像的形状特征信息Ech
e2)针对归一化后胸部子图像Ich和裆部子图像Icr分别提取其haar-like特征信息Fch和Fcr
e3)针对归一化后裆部子图像Icr计算其灰度方差Tcr
f)针对归一化后的胸部子图像Ich′提取的形状特征信息训练分类器S1,针对归一化后的胸部子图像Ich′提取的haar-like特征信息训练分类器S2,针对归一化后的裆部子图像Icr′提取的haar-like特征信息训练分类器S3,针对归一化后的裆部子图像Icr′提取的灰度方差信息训练分类器S4;
步骤f)可以包括如下子步骤:
f1)在毫米波图像库中分别随机选取N帧男性图像和N帧女性图像作为训练样本,N为小于样本库中男性图像数量和女性图像数量中的较小者的整数;
f2)分别对N帧男性图像和N帧女性图像按照步骤b)-e)提取其胸部子图像的形状特征信息,胸部子图像和裆部子图像的haar-like特征信息,裆部子图像的方差信息;
f3)针对N帧男性图像和N帧女性图像胸部子图像形状特征信息,训练分类器S1,将待识别图像形状特征信息与N帧男性图像和N帧女性图像形状特征信息的N个距离按从小到大进行排序,则当前待识别图像与距离其最近的图像属于同一类;
f4)针对N帧男性图像和N帧女性图像胸部子图像和裆部子图像haar-like特征信息,分别训练分类器S2和S3,将待识别图像胸部子图像和裆部子图像的haar-like特征信息分别与N帧男性图像和N帧女性图像胸部子图像和裆部子图像haar-like特征信息的N个距离分别按从小到大进行排序,则当前待识别图像与距离其最近的图像属于同一类;
f5)针对N帧男性图像和N帧女性图像裆部子图像灰度方差信息,训练分类器S4,将待识别图像灰度方差信息与N帧男性图像和N帧女性图像灰度方差信息的N个距离按从小到大进行排序,则当前待识别图像与距离其最近的图像属于同一类;
步骤g)包括如下子步骤:
针对步骤f)训练出的分类器S1,S2,S3和S4,对每个分类器根据分类器准确率来设定各自的权重,若多个分类器识别的结果更接近于哪一类,则将当前待识别图像归属于该类。
其中,步骤b1)中涉及的灰度分割阈值算法为通过自动阈值分割算法,步骤e2)中涉及的haar-like特征提取算法,以及步骤g)中涉及的根据分类器准确率来设定权重的方法,属于本领域当中公知技术,在此不再赘述。
本发明相对传统人体性别自动识别方法具有以下突出的优点:
(1)本发明针对不同性别人体毫米波图像中胸部和裆部区域子图像呈现出不同的差异特征,采取不同的特征提取方式,而非单纯地依靠灰度方差进行识别,从而使得本发明有效地提高了识别的准确度,同时在实时性上也能得到保证。
(2)本发明采用多个分类器集成的方式,降低了单一分类器分类识别结果存在一定的误判率。
本发明人体性别自动识别装置实施例1:
基于以上人体性别自动识别方法的技术思想,同时为了解决传统性别识别方法计算效率和识别准确率低的问题,本发明还提供了一种人体性别自动识别装置实施例1;图6为本发明人体性别自动识别装置实施例1的结构示意图;如图6所示,可以包括:
确定区域位置单元610,用于获取当前待识别的毫米波灰度图像,根据预设的身体比例,确定毫米波灰度图像中人体的性别部位区域位置;性别部位区域位置包括胸部区域位置和裆部区域位置;
提取子图像单元620,用于提取性别部位区域位置对应的区域子图像;
归一化处理单元630,用于对区域子图像进行尺度归一化,得到归一化后的区域子图像;
特征信息提取单元640,用于对归一化后的区域子图像进行特征信息提取,得到归一化后的区域子图像的特征信息;特征信息包括胸部区域位置的形状特征信息和胸部区域特征信息、裆部区域位置的灰度方差信息和裆部区域特征信息;
分类单元650,用于通过预设的各分类器基于对应的特征信息对毫米波灰度图像进行识别,并分别输出结果;分类器为基于毫米波图像数据库中训练样本对应的特征信息经训练得到的;
识别单元660,用于对输出结果进行集成,得到毫米波灰度图像的分类识别结果。
在一个具体的实施例中,特征信息提取单元640包括:
胸部特征提取模块642,用于提取胸部区域特征信息,并根据边缘检测算法,得到提取归一化后的区域子图像的边缘曲线,基于边缘曲线上点的曲率,获取形状特征信息;
裆部特征提取模块644,用于提取裆部区域特征信息,并对归一化后的区域子图像进行灰度方差计算处理,获取灰度方差信息。
在一个具体的实施例中,胸部区域特征信息包括胸部区域haar-like特征信息;裆部区域特征信息包括裆部区域haar-like特征信息。
在一个具体的实施例中,识别单元660,用于根据输出结果和各分类器的权重,对毫米波灰度图像进行归类,获取分类识别结果。
本发明人体性别自动识别装置实施例1,为实现毫米波图像中人体性别自动识别达到较高的识别准确率,针对不同性别的人体毫米波图像成像结果表现出的不同特点,对毫米波图像的内容信息进行分析,提取不同的特征信息采用多个分类器集成对性别进行识别,实现了毫米波图像的人体性别自动识别。本发明能有效适用于真实的毫米波安检系统中,具有较高的识别率和计算效率,解决了毫米波安检系统中对针对不同性别的人体采取不同方式的隐私保护和检测方法的问题。
以上所述实施例的各技术特征可以进行任意的组合,为使描述简洁,未对上述实施例中的各个技术特征所有可能的组合都进行描述,然而,只要这些技术特征的组合不存在矛盾,都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本发明的几种实施方式,其描述较为具体和详细,但并不能因此而理解为对发明专利范围的限制。应当指出的是,对于本领域的普通技术人员来说,在不脱离本发明构思的前提下,还可以做出若干变形和改进,这些都属于本发明的保护范围。因此,本发明专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种人体性别自动识别方法,其特征在于,包括以下步骤:
    获取当前待识别的毫米波灰度图像,根据预设的身体比例,确定所述毫米波灰度图像中人体的性别部位区域位置;所述性别部位区域位置包括胸部区域位置和裆部区域位置;
    提取所述性别部位区域位置对应的区域子图像;
    对所述区域子图像进行尺度归一化,得到归一化后的区域子图像;
    对所述归一化后的区域子图像进行特征信息提取,得到所述归一化后的区域子图像的特征信息;所述特征信息包括所述胸部区域位置的形状特征信息和胸部区域特征信息、所述裆部区域位置的灰度方差信息和裆部区域特征信息;
    通过与所述特征信息对应的各预设分类器,根据特征距离的大小排序,对所述毫米波灰度图像进行识别,并分别输出结果;所述特征距离为所述特征信息与对应的所述预设分类器的特征信息的距离;
    对所述输出结果进行集成,得到所述毫米波灰度图像的分类识别结果。
  2. 根据权利要求1所述的人体性别自动识别方法,其特征在于,根据预设的身体比例,确定所述毫米波灰度图像中人体的性别部位区域位置的步骤包括:
    根据分割阈值T,基于以下公式获得所述毫米波灰度图像的二值图像Ib
    Figure PCTCN2017093238-appb-100001
    X为所述毫米波灰度图像的行数目;Y为所述毫米波灰度图像的列数目;
    基于以下公式对Ib(x,y)中分割出的人体区域在y方向构建空间分布直方图向量Hy
    Figure PCTCN2017093238-appb-100002
    其中,δ表示迪克拉δ函数;x表示行坐标;y表示列坐标;
    基于以下公式对所述直方图向量Hy进行平滑处理:
    Figure PCTCN2017093238-appb-100003
    对y方向直方图向量Hy进行迭代运算,得到人体垂直中位线所在列;
    以所述人体垂直中位线所在列的中间行为基准,对所述二值图像Ib中位线所在列的左右像素进行统计,获取人体头顶所在的行xt和人体脚底所在的行xf
    根据所述人体头顶所在的行xt和所述人体脚底所在的行xf,确定人体身高He=xf-xt
    根据所述预设的身体比例,基于以下公式计算胸部起始行xch=xf-He*Rch,裆部起始行xcr=xf-He*Rcr;其中Rch为统计意义上胸部与身高的比例,Rcr为统计意义上裆部与身高的比例。
  3. 根据权利要求2所述的人体性别自动识别方法,其特征在于,对y方向直方图向量Hy进行迭代运算,得到人体垂直中位线所在列的步骤包括:
    基于以下公式,在y方向空间分布直方图向量Hy中设定初始中心点位置值
    Figure PCTCN2017093238-appb-100004
    Figure PCTCN2017093238-appb-100005
    基于以下公式,分别计算空间分布位置低于
    Figure PCTCN2017093238-appb-100006
    的集合Y1与空间分布位置高于
    Figure PCTCN2017093238-appb-100007
    的集合Y2的空间分布均值
    Figure PCTCN2017093238-appb-100008
    Figure PCTCN2017093238-appb-100009
    Figure PCTCN2017093238-appb-100010
    Figure PCTCN2017093238-appb-100011
    Figure PCTCN2017093238-appb-100012
    时停止迭代,获取所述人体中位线所在列的位置
    Figure PCTCN2017093238-appb-100013
  4. 根据权利要求1至3任意一项所述的人体性别自动识别方法,其特征在于,获取所述形状特征信息的步骤包括:
    根据边缘检测算法,得到提取所述归一化后的区域子图像的边缘曲线,基 于所述边缘曲线上点的曲率,获取所述形状特征信息;
    获取所述灰度方差信息的步骤包括:
    对所述归一化后的区域子图像进行灰度方差计算处理,获取所述灰度方差信息。
  5. 根据权利要求1至3任意一项所述的人体性别自动识别方法,其特征在于,所述胸部区域特征信息包括胸部区域haar-like特征信息;所述裆部区域特征信息包括裆部区域haar-like特征信息。
  6. 根据权利要求1至3任意一项所述的人体性别自动识别方法,其特征在于,对所述输出结果进行集成,得到所述待识别图像的分类识别结果的步骤包括以下步骤:
    根据所述输出结果和各所述分类器的权重,对所述毫米波灰度图像进行归类,获取所述分类识别结果。
  7. 一种人体性别自动识别装置,其特征在于,包括:
    确定区域位置单元,用于获取当前待识别的毫米波灰度图像,根据预设的身体比例,确定所述毫米波灰度图像中人体的性别部位区域位置;所述性别部位区域位置包括胸部区域位置和裆部区域位置;
    提取子图像单元,用于提取所述性别部位区域位置对应的区域子图像;
    归一化处理单元,用于对所述区域子图像进行尺度归一化,得到归一化后的区域子图像;
    特征信息提取单元,用于对所述归一化后的区域子图像进行特征信息提取,得到所述归一化后的区域子图像的特征信息;所述特征信息包括所述胸部区域位置的形状特征信息和胸部区域特征信息、所述裆部区域位置的灰度方差信息和裆部区域特征信息;
    分类单元,用于通过与所述特征信息对应的各预设分类器,根据特征距离的大小排序,对所述毫米波灰度图像进行识别,并分别输出结果;所述特征距离为所述特征信息与对应的所述预设分类器的特征信息的距离;
    识别单元,用于对所述输出结果进行集成,得到所述毫米波灰度图像的分类识别结果。
  8. 根据权利要求7所述的人体性别自动识别装置,其特征在于,
    所述特征信息提取单元包括:
    胸部特征提取模块,用于提取所述胸部区域特征信息,并根据边缘检测算法,得到提取所述归一化后的区域子图像的边缘曲线,基于所述边缘曲线上点的曲率,获取所述形状特征信息;
    裆部特征提取模块,用于提取所述裆部区域特征信息,并对所述归一化后的区域子图像进行灰度方差计算处理,获取所述灰度方差信息。
  9. 根据权利要求7或8所述的人体性别自动识别装置,其特征在于,所述胸部区域特征信息包括胸部区域haar-like特征信息;所述裆部区域特征信息包括裆部区域haar-like特征信息。
  10. 根据权利要求7或8所述的人体性别自动识别装置,其特征在于,
    所述识别单元,用于根据所述输出结果和各所述分类器的权重,对所述毫米波灰度图像进行归类,获取所述分类识别结果。
PCT/CN2017/093238 2016-07-29 2017-07-17 人体性别自动识别方法及装置 WO2018019149A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP17833452.0A EP3493103A4 (en) 2016-07-29 2017-07-17 METHOD AND DEVICE FOR AUTOMATICALLY DETECTING THE GENDER OF A HUMAN BODY
US16/321,807 US11250249B2 (en) 2016-07-29 2017-07-17 Human body gender automatic recognition method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610619371.7 2016-07-29
CN201610619371.7A CN106326834B (zh) 2016-07-29 2016-07-29 人体性别自动识别方法及装置

Publications (1)

Publication Number Publication Date
WO2018019149A1 true WO2018019149A1 (zh) 2018-02-01

Family

ID=57739800

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/093238 WO2018019149A1 (zh) 2016-07-29 2017-07-17 人体性别自动识别方法及装置

Country Status (4)

Country Link
US (1) US11250249B2 (zh)
EP (1) EP3493103A4 (zh)
CN (1) CN106326834B (zh)
WO (1) WO2018019149A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111103629A (zh) * 2018-10-25 2020-05-05 杭州海康威视数字技术股份有限公司 一种目标检测方法、装置、nvr设备及安检系统

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106326834B (zh) 2016-07-29 2019-12-10 华讯方舟科技有限公司 人体性别自动识别方法及装置
CN106971190A (zh) * 2017-03-07 2017-07-21 上海优裁信息技术有限公司 基于人体体型的性别判断方法
CN108108740B (zh) * 2017-12-18 2022-01-04 中国科学院上海微系统与信息技术研究所 一种主动毫米波人体图像性别识别方法
CN109886157A (zh) * 2019-01-30 2019-06-14 杭州芯影科技有限公司 一种基于毫米波图像的人脸识别方法及系统
CN110245675B (zh) * 2019-04-03 2023-02-10 复旦大学 一种基于毫米波图像人体上下文信息的危险物体检测方法
CN111209793A (zh) * 2019-12-05 2020-05-29 重庆特斯联智慧科技股份有限公司 一种基于人工智能的区域屏蔽人体安检方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102540264A (zh) * 2011-12-30 2012-07-04 北京华航无线电测量研究所 一种人体隐私部位自动检测与遮挡的微波安检系统
CN102708560A (zh) * 2012-02-29 2012-10-03 北京无线电计量测试研究所 一种基于毫米波成像的隐私保护方法
CN103198303A (zh) * 2013-04-12 2013-07-10 南京邮电大学 一种基于人脸图像的性别识别方法
CN106326834A (zh) * 2016-07-29 2017-01-11 华讯方舟科技有限公司 人体性别自动识别方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120119216A1 (en) * 2009-07-31 2012-05-17 Tadahiro Ohmi Semiconductor Device, Method of Manufacturing A Semiconductor Device, and Display Device
WO2012119216A1 (en) * 2011-03-09 2012-09-13 Optosecurity Inc. Method and apparatus for performing a security scan on a person
US20130121529A1 (en) * 2011-11-15 2013-05-16 L-3 Communications Security And Detection Systems, Inc. Millimeter-wave subject surveillance with body characterization for object detection
CN102542570B (zh) * 2011-12-30 2014-05-07 北京华航无线电测量研究所 一种微波图像中人体隐藏危险物体自动检测方法
CN102567733B (zh) * 2011-12-30 2014-05-07 北京华航无线电测量研究所 一种微波图像非合作情况下人体隐私部位检测与遮挡方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102540264A (zh) * 2011-12-30 2012-07-04 北京华航无线电测量研究所 一种人体隐私部位自动检测与遮挡的微波安检系统
CN102708560A (zh) * 2012-02-29 2012-10-03 北京无线电计量测试研究所 一种基于毫米波成像的隐私保护方法
CN103198303A (zh) * 2013-04-12 2013-07-10 南京邮电大学 一种基于人脸图像的性别识别方法
CN106326834A (zh) * 2016-07-29 2017-01-11 华讯方舟科技有限公司 人体性别自动识别方法及装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3493103A4 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111103629A (zh) * 2018-10-25 2020-05-05 杭州海康威视数字技术股份有限公司 一种目标检测方法、装置、nvr设备及安检系统

Also Published As

Publication number Publication date
EP3493103A4 (en) 2020-03-18
US11250249B2 (en) 2022-02-15
EP3493103A1 (en) 2019-06-05
CN106326834B (zh) 2019-12-10
US20200042779A1 (en) 2020-02-06
CN106326834A (zh) 2017-01-11

Similar Documents

Publication Publication Date Title
WO2018019149A1 (zh) 人体性别自动识别方法及装置
US10839510B2 (en) Methods and systems for human tissue analysis using shearlet transforms
CN107610114B (zh) 基于支持向量机的光学卫星遥感影像云雪雾检测方法
EP1229493B1 (en) Multi-mode digital image processing method for detecting eyes
He et al. A method of green litchi recognition in natural environment based on improved LDA classifier
WO2016150240A1 (zh) 身份认证方法和装置
RU2018142757A (ru) Система и способ обнаружения болезней растений
Abate et al. BIRD: Watershed based iris detection for mobile devices
US20120213440A1 (en) Systems and Methods for Automatically Identifying Shadows in Images
Pundlik et al. Iris segmentation in non-ideal images using graph cuts
CN109583452B (zh) 一种基于赤足足迹的人身同一认定方法和系统
Raffei et al. Feature extraction for different distances of visible reflection iris using multiscale sparse representation of local Radon transform
WO2013075295A1 (zh) 低分辨率视频的服装识别方法及系统
Gómez-Flores et al. Detection of Huanglongbing disease based on intensity-invariant texture analysis of images in the visible spectrum
Trivedi et al. Automatic segmentation of plant leaves disease using min-max hue histogram and k-mean clustering
Sajid et al. The role of facial asymmetry in recognizing age-separated face images
Gonzalez-Sosa et al. Image-based gender estimation from body and face across distances
Lian et al. Matching of tracked pedestrians across disjoint camera views using CI-DLBP
CN115953824A (zh) 一种人脸皮肤图像处理方法及系统
WO2016192213A1 (zh) 一种图像特征提取方法和装置、存储介质
Yang et al. Detection and segmentation of latent fingerprints
Abdu et al. An investigation into the effect of disease symptoms segmentation boundary limit on classifier performance in application of machine learning for plant disease detection
CN105701499B (zh) 一种用于脑部mri图像分类的图像处理方法
Jassim Hybrid image segmentation using discerner cluster in FCM and histogram thresholding
Rehman et al. Modified texture features from histogram and gray level co-occurence matrix of facial data for ethnicity detection

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: 17833452

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2017833452

Country of ref document: EP

Effective date: 20190228