精华内容
下载资源
问答
  • 讯飞脸部识别

    2016-05-11 16:56:43
    集成讯飞脸部识别
  • 写真美女套图:爬虫 美女脸部识别 DCGAN脸部自动生成
  • 人脸识别的JavaScript程序包是Face ...它是一个标准的jQuery插件,通过对提供的图片进行分析,返回所有找到的脸部图像的坐标,感兴趣的朋友跟着小编一起学习js人脸识别技术及脸部识别JavaScript类库Tracking.js吧
  • jQuery图片人物脸部识别插件是一款支持人物图片视频上的人脸识别,并且给出准确的标记。
  • 脸部识别SeetaFace介绍

    千次阅读 2016-09-16 16:19:32
    SeetaFace Engine是一款无需任何第三方库就能在CPU上运行的开源C++脸部识别引擎,一共分为三个关键的部分:SeetaFace Detection(脸部检测),SeetaFace Alignment(脸部校准)和SeetaFace Identification(脸部识别...

    https://github.com/seetaface/SeetaFaceEngine
    一、概述
    SeetaFace Engine是一款无需任何第三方库就能在CPU上运行的开源C++脸部识别引擎,一共分为三个关键的部分:SeetaFace Detection(脸部检测),SeetaFace Alignment(脸部校准)和SeetaFace Identification(脸部识别),这三个部分对于搭建一个实用的脸部识别应用系统是非常必要的。

    • SeetaFace Detection实现了一个funnel-structured (FuSt) cascade用来针对现实中多视角的脸部检测,并在检测准确性和速度之前达到了很好的平衡。在公开的数据集FDDB上可以达到很高的速度,更多细节见SeetaFace Detection部分
    • SeetaFace Alignment在实际中级联了多个栈式自编码网络去检测关键点(在单个i7台式机CPU中超过200fps),并且在一些公开的数据集上实现了最先进的精度,比如AFLW更多细节见SeetaFace Alignment部分
    • SeetaFace Identification是一个用于脸部识别的AlexNet CNN的改进,在准确性(使用LFW为97.1%)和速度(使用一个i7台式机处理器约120ms一张)上有更好的表现。更多细节见SeetaFace Identification部分。

    这个脸部识别引擎是由中国科学院计算所VIPL组研发的,代码使用C++编写且不依赖于任何第三方库,开源代码在BSD-2(详见LICENSE)许可下发布,代码可以自由的在学术界和工业界使用。有任何问题可以联系SeetaFace@vipl.ict.ac.cn

    二、SeetaFace Detection
    SeetaFace Detection用级联的漏斗结构(Funnel-Structured cascade,FuSt)实现,设计用来做现实中多视角脸部检测。FuSt的目的是使用“由粗到细的结构”从而在准确性和速度之间取得一个很好的平衡。在前几层它包含了多个针对视角的快速LAB级联分类器,后几层是粗多层感知器(coarse Multilayer Perceptron,coarse MLP)级联结构,最后由一个统一的MLP级联结构来处理所有姿态的候选窗口。FuSt级联结构
    开源的FuSt包括一个用20万脸部图片训练的模型用来检测近正面脸部(也能够检测一部分非正面脸部),值得注意的是(1)MLP级联结构使用了SURF检测而不是SIFT(2)加入了NMS(Non-Maximal Suprresion)(3)边界盒回归代替了关键点预测
    SeetaFace Detector在FDDB上的离散型得分ROC曲线如下图所示,最小的人脸大小设置为20,滑动窗口步长设置为2或者4,采样比例设置为0.8这里写图片描述
    速度比较见如下图,使用640x480的VGA图片,滑动窗口的大小设置为4,采样比例设置为0.8,Cascade CNN中图片金字塔的采样比例为0.7这里写图片描述
    SeetaFace Detector是在一个3.40GHz i7-3770 CPU上的测试速度,其他方法则是直接从相关论文中引用( Cascade CNN的CPU速度是在一个2GHz的CPU核心上检测的,GPU速度是在NVIDIA Titan Black GPU上检测的)

    未完待续

    展开全文
  • 爬虫+脸部识别+DCGAN脸部自动生成

    千次阅读 2017-12-16 10:48:59
    写真美女套图:爬虫+美女脸部识别+DCGAN脸部自动生成所有代码请到我的github上下载,请star一下,谢谢大家了。https://github.com/sileixinhua/BeautifulGirls第一部分:爬虫 抓美女套图(Python+BeautifulSoup+...

    爬虫+脸部识别+DCGAN脸部自动生成

    所有代码请到我的github上下载,请star一下,谢谢大家了。

    https://github.com/sileixinhua/BeautifulGirls

    第一部分:爬虫 抓图片(Python+BeautifulSoup+requests)

    前言

    本文主要是以爬虫爬取下来的图片为数据,做一个只针对脸部识别,和一个DCGAN合成脸的模型。

    第一部分:写爬虫主要看需求来决定工具的使用,python无非是众多语言中比较成熟的一个,如果要分析json,要分布式,就用scrapy,如果功能要求简单的就用BeautifulSoup+requests就可以了。requests用于和服务器的交互,BeautifulSoup解析HTML页面格式数据,并提取想要的信息。

    第二部分:现在脸部,物体的识别多是用tensorflow等机器学习框架来做,但是其实在很早的时候用opencv就可以做了,现在opencv也有DNN等功能,这里脸部识别主要是用了opencv的cascades识别功能,这部分基本不要写什么代码,但是过程会比较繁琐。

    第三部分:有人说近十年深度学习的重要发现就是GAN,第三部分就是用了GAN+CNN的变种DCGAN,GAN主要是用来根据现有的数据发现其中的模式生成数据,图像,语音等,主要组成部分为生成器和判别器,生成器是原始数据加噪声来合成新的数据,判别器主要是根据原始数据判别生成数据的相似度,准确度。

    开发环境

    windows10

    Python3.5

    https://www.python.org/downloads/

    这里写图片描述

    BeautifulSoup

    https://www.crummy.com/software/BeautifulSoup/bs4/doc/index.zh.html

    这里写图片描述

    Requests

    http://docs.python-requests.org/en/master/#

    这里写图片描述

    可能需要的python包安装(Python3环境)

    
    pip3 install BeautifulSoup
    
    
    pip3 install requests
    
    
    
    pip3 install lxml
    

    这里还是推荐使用Python3,但是用Python2的同学,把上述命令的“pip3”改成“pip”就可以了。

    爬虫目标网页结构分析

    目标网址:http://www.xingmeng365.com/

    爬虫需要抓取的页面:

    爬虫需要抓取的页面地址:

    这里写图片描述

    爬虫获取HTML页面信息的地址:

    这里写图片描述

    代码分析

    SpiderDownloadImages.py

    # 2017年11月10日 19点24分
    # 作者:橘子派_司磊
    # 爬虫
    # 目标网址:http://www.xingmeng365.com/
    
    from bs4 import BeautifulSoup
    import requests
    import os
    import urllib.request
    
    # 在mian.py当前位置创建图片收集的文件夹Photos
    if not os.path.exists('Photos'):
    		os.makedirs('Photos')
    
    num = 67
    image_list = []
    id = 7
    
    while(id <= 559):
        headers = {'User-Agent':'Mozilla/5.0 (Windows NT 6.1; WOW64; rv:23.0) Gecko/20100101 Firefox/23.0'}
        url = requests.get('http://www.xingmeng365.com/articles.asp?id='+str(id), headers=headers)
        # 此处用 “,str(id)” 的话,逗号打印出来会变成 “id=&2”
        print("当前爬取的网址为:"+url.url)
        html_doc = url.text
        # 此处用url不带".text"的话报错,Python: object of type 'Response' has no len()
        # 错误解决
        # https://stackoverflow.com/questions/36709165/python-object-of-type-response-has-no-len
        soup = BeautifulSoup(html_doc,"lxml")
    
        for link in soup.find_all('img'):
            if "/upload" in link.get('src'):
                # id=7以后,"../../"改为"/upload"
                image_url = link.get('src')
                # 获得的图片地址有错,需要改成
                # http://www.xingmeng365.com/upload/image/20170811/20170811203590079007.jpg
                # 即把 “../../” 改为 “http://www.xingmeng365.com/”
                # id=7 以后为/upload/image/20170811/20170811210596789678.jpg
                # 即http://www.xingmeng365.com/upload/image/20170811/20170811210545754575.jpg
                image_url = "http://www.xingmeng365.com/" + image_url[1:]
                # id=7以后,[6:]改为[1:]
                print("开始下载第"+str(num+1)+"张图片:"+image_url)
                file = open('Photos/'+str(num)+'.jpg',"wb")
                req = urllib.request.Request(url=image_url, headers=headers) 
                try:
                    image = urllib.request.urlopen(req, timeout=10)
                    pic = image.read()
                except Exception as e:
                    print("第"+str(num+1)+"张图片访问超时,下载失败:"+image_url)
                    continue
                # 遇到错误,网站反爬虫
                # urllib.error.HTTPError: HTTP Error 403: Forbidden
                # 原因是这里urllib.request方法还需要加入“, headers=headers”
                # 头文件来欺骗,以为我们是客户端访问
                file.write(pic)
                print("第"+str(num+1)+"张图片下载成功")
                file.close()
                num = num + 1
        id = id + 1
    

    实验结果

    视网络情况而定,一共花费7293.5秒,爬取15684张图。

    这里写图片描述

    这里写图片描述

    这里文件夹下所有的图像数据就保存在Photos文件夹里,一共有15684张图。

    第二部分脸部识别(Python+Opencv的Cascades)

    前言

    现在大家都在用TensorFlow等神经网络框架做识别,过程繁琐,有些功能可以直接用OpenCV做到,而且封装好的开发工具包可以节省很多时间,效果也还可以。

    这里解释一下OpenCV自带的cascades识别,官网信息地址。

    https://docs.opencv.org/3.0.0/d7/d8b/tutorial_py_face_detection.html

    首先安装OpenCV后,源码目录下\opencv\sources\data\haarcascades,就有很多自带的人脸识别.xml
    文件,这个文件里包含的就是要识别出物体的信息。

    开发环境

    Python3 + OpenCV

    OpenCV的window安装直接官网https://opencv.org/下载源码,把bin路径添加到系如变量即可

    在Ubuntu上的安装比较繁琐,我找到的最简单的方式是:

    https://www.youtube.com/watch?v=2Pboq2LFoaI

    http://www.daslhub.org/unlv/wiki/doku.php?id=opencv_install_ubuntu

    整个过程安装比较耗时,大概一刻钟左右。

    在Python中安装OpenCV开发包需要如下命令:

    pip3 install opencv-python
    

    这里如果是Python2就把“pip3”改成“pip”即可。

    实验分析与步骤设计

    本实验步骤有点繁琐,请仔细查阅,经过第一部分,文件下已经有Photos文件夹,这里是所有图的数据集。

    1. 我们需要用OpenCV自带的脸部识别把所有脸部截取下来,存放进Faces文件夹里。
    2. 然后用OpenCV自己的方法创建我们自己的cascades的识别器,用来识别脸部,丑的不识别,这一步主要就是生成.xml文件,文件里包含的就是脸部的信息。Negative文件夹里是背景,即负面Negative数据,用来和真实的脸部数据做对比,让训练器知道哪些是脸,哪些不是。结果会生成进data文件夹里,结果是一个.xml类型的文件,具体步骤如下代码分析所示。
    3. 用我们自己生成的.xml文件来识别脸部,丑的不识别。

    这一部分的实验主要是OpenCV的脸部识别器我用错了,所以从截取的脸部信息就有噪声数据,即不是脸部的图也被截取下来混进去了,所以效果不是很好,如果要提高效果,可以用别的脸部识别分类器,或者手动删除非脸部图片,并加大Negative文件夹里的图片即可。

    代码分析

    首先是上面实验分析的第一步,截图脸部图像

    TakeImgFace.py

    import cv2
    import sys
    import os.path
    from glob import glob
    
    # C:\Code\BeautifulGirls\Faces
    # C:\Code\BeautifulGirls\Photos
    
    # 一共有15683张写真美图
    
    # 本文件是用来从Photos图文件夹中,用opencv自带的人脸识cascade别出脸部并截图保存到Faces文件夹中
    
    # 在opencv的自带人脸检测中,haarcascade_frontalface_alt效果最好,缺点是时间长
    
    def detect(filename, cascade_file="C:\\OpenCV\\opencv\\build\\etc\\haarcascades\\haarcascade_frontalface_alt.xml"):
        if not os.path.isfile(cascade_file):
            raise RuntimeError("%s: not found" % cascade_file)
            # 这里确认找到cascades识别器,找不到显示not found,地址请根据你的自己安装位置修改一下
    
        cascade = cv2.CascadeClassifier(cascade_file)
        # 导入识别器
        image = cv2.imread(filename)
        # 读取图片
        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        # 获取图片的灰度图
        gray = cv2.equalizeHist(gray)
    
        faces = cascade.detectMultiScale(gray,
                                         scaleFactor=1.1,
                                         minNeighbors=5,
                                         minSize=(48, 48))
        # 识别脸部
        for i, (x, y, w, h) in enumerate(faces):
        	# 定义脸部在图像上的坐标
            face = image[y: y + h, x:x + w, :]
            # 获取坐标位置的图
            face = cv2.resize(face, (96, 96))
            # 重新定义大小
            save_filename = '%s-%d.jpg' % (os.path.basename(filename).split('.')[0], i)
            # 定义保存图片的地址
            cv2.imwrite("Faces/" + save_filename, face)
            # 保存图片
    
    
    if __name__ == '__main__':
        if os.path.exists('Faces') is False:
            os.makedirs('Faces')
        # 检查Faces文件夹,没有就创建一个
        file_list = glob('Photos/*.jpg')
        for filename in file_list:
            detect(filename)
    

    这步完成之后,在faces文件夹里会有很多很多的脸部的截图,但是识别器可以用更好的,我这里只做示范,有些噪音数据就先放着不管了,你们自己做的时候想提高效果可以手动去除噪音数据,或者用更好的分类器。

    \opencv\sources\data\haarcascades文件下有很多识别器,你们可以自己试试效果如果。用效果最好的那个即可。

    接下来就是重难点了,用OpenCV的cascades创建我们自己的识别器。

    首先需要获取正面和负面数据的数据列表信息。

    CreateInfoTxt.py

    import os
    
    # 创建positive.txt和negative.txt文件
    # 文件内容是数据集的list
    
    def Create_faces_info_lst():
        for file_type in ['Faces']:
        
            for img in os.listdir(file_type):
    
                if file_type == 'Faces':
                    line = file_type+'/'+img+' 1 0 0 48 48\n'
                    with open('info.lst','a') as f:
                        f.write(line)
    
    if __name__=="__main__":
        Create_faces_info_lst()
    

    上面的代码运行之后会生成一个info.lst文件,里面会有Faces正面数据里的数据列表(让OpenCV知道你一共有多少正面数据有哪些),然后下载去我的github里下载背景即负面数据的数据集,https://github.com/sileixinhua

    info.lst文件如下所示:

    这里写图片描述

    负面Negative文件夹数据如下所示:

    这里写图片描述

    这个时候有会用到Faces文件夹下脸部数据,info.lst正面数据列表信息(让OpenCV知道你一共有多少正面数据有哪些),从我github上下载的Negative文件夹,接下在本地项目文件夹下打开cmd。linux的打开terminal。运行如下命令

    opencv_createsamples -info info.lst -num 14229 -w 48 -h 48 -vec positive.vec
    

    这一行命令是根据正面数据的信息创建positive.vec文件,用来告诉opencv正面数据的特征。

    如果报错"Parameters can not be written, because file data/params.xml can not be opened.",请在项目文件夹里创建一个名字为“data”的文件夹,

    opencv_traincascade -data data -vec positive.vec -bg bg.txt -numPos 12000 -numNeg 202 -numStages 20 -w 48 -h 48
    

    这一行命令就是训练我们的cascades识别器了,存放地址为data文件夹,vec就是我们上一步创建的positive.vec,bg就是Negative文件夹里负面数据的列表,训练区块为20个,根据数据集的大小可以调整,输入为高48,宽48。这里我做错了,应该填96。(这里一错又导致我效果不好),这里我一共花了2天采用CPU计算完毕。

    全部计算结束后请再次输入一次上面的命令

    opencv_traincascade -data data -vec positive.vec -bg bg.txt -numPos 12000 -numNeg 202 -numStages 20 -w 48 -h 48
    

    用来把每一个区块的.xml结果信息合成为一个.xml文件

    然后把data文件夹里生成的.xml文件改为你想要的名字,我就改为BeautifulFacaCascade.xml。

    下一步就是第二部分的最后一部,识别脸部,我记录了视频和图片两种,代码如下所示:

    CascadaBeautifulFace.py

    import cv2
    import os
    import numpy as np
    
    # opencv_createsamples -info info.lst -num 14229 -w 48 -h 48 -vec positive.vec
    
    # create data file 
    # or will error "Parameters can not be written, because file data/params.xml can not be opened."
    
    # opencv_traincascade -data data -vec positive.vec -bg bg.txt -numPos 12000 -numNeg 202 -numStages 20 -w 48 -h 48
    
    # opencv_traincascade -data data -vec positive.vec -bg bg.txt -numPos 12000 -numNeg 202 -numStages 20 -w 48 -h 48
    
    # ----------------------------------------------------------------------------------------------
    # use video
    
    # beautiful_face_cascade = cv2.CascadeClassifier('C:\\Code\\BeautifulGirls\\BeautifulFacaCascade.xml')
    
    # cap = cv2.VideoCapture(0)
    
    # while 1:
    #     ret, img = cap.read()
    
    #     gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        
    #     beautiful_face = beautiful_face_cascade.detectMultiScale(gray, 1.3, 5)
    #     # 这里参数可改成 5
    #     # detectMultiScale()
    #     # https://docs.opencv.org/2.4/modules/objdetect/doc/cascade_classification.html
    #     # minSize – Minimum possible object size. Objects smaller than that are ignored.
    #     # maxSize – Maximum possible object size. Objects larger than that are ignored.
        
    #     for (x,y,w,h) in beautiful_face:
    #         cv2.rectangle(img,(x,y),(x+w,y+h),(255,255,0),2)
    
    #     cv2.imshow('img',img)
    #     k = cv2.waitKey(30) & 0xff
    #     if k == 27:
    #         break
    
    # cap.release()
    # cv2.destroyAllWindows()
    
    # ---------------------------------------------------------------------------------------
    # use image
    
    BeautifulFacaCascade = cv2.CascadeClassifier('C:\\Code\\BeautifulGirls\\BeautifulFacaCascade.xml')
    img = cv2.imread('0.jpg')
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    
    BeautifulFaca = BeautifulFacaCascade.detectMultiScale(gray, 500, 500)
    for (x,y,w,h) in BeautifulFaca:
        cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
        
    cv2.imshow('img',img)
    cv2.waitKey(0)
    cv2.destroyAllWindows()
    

    实验结果

    其实这一部分结果不是很好,原因有两个,一是脸部截取的数据有噪音,二是负面数据太少,但是我训练这些用了32G8核的7700K用了2天,应该改用GPU运算,最近事情比较多,我也就没再次训练了,有兴趣的同学可以自己再试一试。

    第二部分十分繁琐,如果描述的不清楚,大家可以去网上搜索一下OpenCV Cascades,中文信息不多,最好用Google搜索。

    第三部分 DCGAN脸部自动生成(Python+Tensorflow +DCGAN)

    前言

    开发环境

    Python3 + Tensorflow

    Tensorflow的安装在windows上十分繁琐,linux也一样。主要是cudnn和cuda的安装麻烦。

    去YouTuBe上去找视频看的话,但是绝大部分都是一年两年以前的视频。

    安装请按照官方网站的来https://www.tensorflow.org/versions/master/install/install_linux

    但是windows和linux上不用gpu运算的话可以用cpu运算,安装就十分简单了,本文第三部分实验我就是在自己笔记本上cpu运算的,耗时3,4小时而已。

    安装命令如下:

    pip3 install tensorflow
    

    Python2的改“pip3”为“pip”。

    上周我实验室的电脑系统崩了,还好我都有备份。但是我重装ubuntu之后,发现现在tensorflow不支持最新的cuda9,结果现在cuda官网只有9,测试失败之后又重装的系统装cuda8配cudnn,大家以后装tensorflow的时候可以注意,网上的教程一天一遍,过几个月基本都没用,还是要按照官方的安装指导来。

    实验分析与步骤设计

    这部分实验都在文件夹DCGAN里。

    首先在DCGAN文件夹下创建data文件夹,并把之前Faces文件夹复制到data文件夹里。

    这里的代码我是直接在github上找到的代码,可以直接用。

    https://github.com/carpedm20/DCGAN-tensorflow

    代码分析

    在DCGAN文件下打开cmd或者terminal,运行如下命令。

    python main.py --input_height 96 --input_width 96 --output_height 48 --output_width 48 --dataset Faces --crop --train --epoch 300 --input_fname_pattern "*.jpg"
    

    这个人的代码十分有用,推荐有兴趣的同学好好看看分析一下。但是看之前还是建议把“花书”看了。

    实验结果

    运行的效果如下图所示:

    这里写图片描述
    实验结果如下所示:

    再训练下去只会更加清晰,有条件的同学可以把结果传群里,谢谢。

    -------------------------------------------------------------------------------------------------------

    有学习机器学习相关同学可以加群,交流,学习,不定期更新最新的机器学习pdf书籍等资源。

    QQ群号: 657119450

    这里写图片描述

    展开全文
  • 在OpenCV基础上,完成了Unity脸部识别及自动选取最佳截图
  • 人脸脸部识别技术Facial Recognition Technology is a developing technology that shines a light on a new aspect of security and identity. How does it work and what are some of its practical application ...

    人脸脸部识别技术

    Facial Recognition Technology is a developing technology that shines a light on a new aspect of security and identity. How does it work and what are some of its practical application successes and failures in our modern-day society?

    面部识别技术是一项正在发展的技术,可以揭示安全性和身份识别的新方面。 它是如何工作的,在现代社会中它在实际应用中有哪些成功和失败之处?

    什么是面部识别技术? (What is Facial Recognition Technology?)

    Facial recognition technology, or otherwise commonly known as facial recognition systems, are computer-dependent security systems that can automatically detect and identify human faces. In simple terms, it is a way of recognizing a human face through technology. These machines rely on algorithms to do basic tasks such as identifying a person from a crowd. This technology uses facial features from photos and/or videos to better compare the information with a secured database of known faces to find a match.

    面部识别技术或其他通常称为面部识别系统的技术是计算机相关的安全系统,可以自动检测和识别人脸。 简单来说,它是通过技术识别人脸的一种方式。 这些机器依靠算法来执行基本任务,例如从人群中识别一个人。 这项技术使用照片和/或视频中的面部特征来更好地将信息与已知面Kong的安全数据库进行比较,以找到匹配项。

    面部识别如何工作? (How does Facial Recognition work?)

    From a human’s point-of-view, we know people by their faces and their respective facial features because after repeatedly seeing a person, it will automatically get stored into your brain. This is how facial recognition works in technology with the usage of algorithms and databases instead of the eyes and brain. According to a Georgetown University study, half of all American adults have their images stored in one or more facial recognition databases that law enforcement agencies can search and verify.

    从人的角度来看,我们通过其面Kong和各自的面部特征来了解人们,因为在反复看到一个人之后,它会自动存储到您的大脑中。 这就是面部识别技术在技术中如何使用算法和数据库而不是眼睛和大脑的方式。 根据乔治敦大学的一项研究,所有美国成年人中有一半的图像存储在一个或多个面部识别数据库中,执法机构可以对其进行搜索和验证。

    Here is the process as to how this technology works:

    这是有关这项技术工作方式的过程:

    (Sources: Electronic Privacy Information Center (EPIC) and Norton Security)

    (来源:电子隐私信息中心(EPIC)和Norton Security)

    1. A picture of one’s face is captured from a photo or video. It needs to be able to identify a human face and extract it from other people and the background (environment, buildings, cars, etc.).

      一个人的脸部照片是从照片或视频中捕获的。 它需要能够识别人脸并从其他人和背景(环境,建筑物,汽车等)中提取出来。
    2. The system and its technology will then analyze and measure nodal points on the face, such as the distance between the eyes, distance from the forehead to the chin, the shape of the cheekbones, and many other key distinguishable features that separate you from other human faces. All of these examined facial features equate to a personal facial signature.

      然后,该系统及其技术将分析和测量面部的结点,例如眼睛之间的距离,前额到下巴的距离,the骨的形状以及许多其他可与您分开的关键特征面Kong。 所有这些检查过的面部特征都等同于个人面部特征。
    3. Your facial signature is compared to a database of known faces. According to the Federal Bureau of Investigation’s (FBI) May 2018 report, the FBI has had access to 412 million facial images for searches.

      将您的面部特征与已知面Kong的数据库进行比较。 根据联邦调查局(FBI)2018年5月的报告,联邦调查局已经获得了4.12亿张面部图像进行搜索。
    4. The system will then determine whether your face matches with a face in a database using nodal points/key facial features that distinguish you from the environment and other human beings.

      然后,系统将使用节点/关键面部特征(将您与环境和其他人类区分开来)来确定您的面部是否与数据库中的面部匹配。

    面部识别技术在现实生活中有哪些应用? (What Are Some Real-Life Applications of Facial Recognition Technology?)

    Social Media: Platforms like Facebook use an algorithm to spot faces when a person uploads a photo to their network. When they ask if you want to tag people in your photos, you are given the option to either tag or not tag anyone in this picture. If you say yes, it will create a link to their profiles. Facebook can recognize faces with a 98% accuracy.

    社交媒体:当人们将照片上传到网络时,Facebook之类的平台使用一种算法来识别人脸。 当他们询问您是否要标记照片中的人物时,您可以选择是否标记此照片中的任何人。 如果您选择是,它将创建一个指向其个人资料的链接。 Facebook可以以98%的准确度识别人脸。

    Phone Manufacturers: Companies like Apple use facial recognition technology to unlock their phones. It started with the iPhone X/XS/XS Max/XR and it is now on higher generations including the iPhone 11, iPhone 11 Pro, and iPhone 11 Pro Max. Using this technology, it makes sure you’re the right person accessing the phone. According to Apple, the chance of unlocking your phone is about one in one million.

    手机制造商:像Apple这样的公司都使用面部识别技术来解锁手机。 它始于iPhone X / XS / XS Max / XR,现在已经出现在更高的世代中,包括iPhone 11,iPhone 11 Pro和iPhone 11 Pro Max。 使用这项技术,可以确保您是正确的访问电话的人。 根据苹果公司的说法,解锁手机的机会约为百万分之一。

    The United States Government (Department of Homeland Security/Transportation Security Administration): This technology can monitor people coming in and going out of airports. The Department of Homeland Security has used this technology to identify people who have overstayed their visas or under criminal investigation.

    美国政府(国土安全部/运输安全管理局):该技术可以监视进出机场的人员。 国土安全部已使用该技术来识别签证过期或正在接受刑事调查的人员。

    Retailers: Shops can now use surveillance cameras with this technology to scan the faces of every shopper. The primary goal is to fish out suspicious characters and potential shoplifters.

    零售商:商店现在可以使用具有此技术的监控摄像头扫描每个购物者的脸部。 主要目标是找出可疑人物和潜在的扒手。

    Businesses with Restricted Areas and Secured Entrances: Many companies have traded in security badges that granted them access into certain parts of the building for facial recognition systems. Instead of having to scan your card, you can walk up to the restricted area/secured entrance and it will open up after scanning your face and confirming it is an authorized person.

    具有禁区和安全入口的企业许多公司都在交换安全徽章,以使他们能够进入建筑物的某些部分以使用面部识别系统。 无需扫描您的卡,您可以走到限制区域/安全入口,并且它会在扫描您的脸部并确认其为授权人员后才会打开。

    Marketers and Advertisers: Marketers use one’s gender, age, ethnicity, and many other traits when targeting groups for a product or an idea. With facial recognition, they will be able to figure out the pattern of audiences that attend conferences, concerts, campaign events, etc.

    营销人员和广告商:营销人员在针对产品或想法的群体时使用性别,年龄,种族和许多其他特征。 通过面部识别,他们将能够确定参加会议,音乐会,竞选活动等的观众的模式。

    面部识别技术使用成功 (Successes of Facial Recognition Technology Usage)

    Within forty days, the United States Customs and Border Protection (CBP) officials along with the facial recognition technology at the Washington Dulles International Airport in Northern Virginia have caught three imposters.

    在四十天内,美国海关与边境保护局(CBP)官员以及北弗吉尼亚州华盛顿杜勒斯国际机场的面部识别技术抓获了三名冒名顶替者。

    August 22rd, 2018: Three Days After Facial Recognition Technology System Installation atWashington Dulles International Airport in Northern Virginia.

    2018年8月22日:在弗吉尼亚北部的华盛顿杜勒斯国际机场安装面部识别技术系统三天后。

    (Directly from the United States Customs and Border Protection Website)

    (直接来自美国海关和边境保护局网站)

    A 26-year-old man traveling from San Paulo, Brazil presented a French passport to the CBP officer conducting primary inspections. The officer utilized CBP’s new facial comparison biometric technology which confirmed the man was not a match to the passport he presented. The CBP officer referred the traveler to secondary inspection for a comprehensive examination. During this thorough check, CBP officers noted the traveler’s behavior changed and he became visibly nervous. A search revealed the man’s authentic Republic of Congo identification card concealed in his shoe.

    一名来自巴西圣保罗的26岁男子向法国海关和边境保护局官员出示了法国护照,以进行初步检查。 该官员利用了CBP的新面部比较生物识别技术,该技术证实该人与他出示的护照不符。 CBP官员将旅客转介至第二次检查以进行全面检查。 在进行彻底检查期间,CBP官员注意到旅行者的行为发生了变化,并且明显变得紧张。 搜查发现该男子的鞋中藏有该男子的真实刚果共和国身份证。

    September 8th, 2018: Second Imposter Within Three Weeks of Facial Recognition Technology System Installation at Washington Dulles International Airport in Northern Virginia.

    2018年9月8日:在弗吉尼亚北部的华盛顿杜勒斯国际机场安装面部识别技术系统三周之内的第二个冒名顶替者。

    (Directly from the United States Customs and Border Protection Website)

    (直接来自美国海关和边境保护局网站)

    A 26-year-old woman, who arrived on a flight from Accra, Ghana, presented a United States Passport to a CBP officer for admission as a returning citizen. Utilizing the new facial comparison technology, the CBP officer established that the traveler was not a match to the passport and referred her for further examination. A secondary examination confirmed that the traveler was a Ghanaian citizen and an imposter to the United States passport.

    一名来自加纳阿克拉的航班上的26岁妇女向美国海关和边境保护局官员出示了美国护照,以作为回国公民。 CBP官员利用新的​​面部比较技术,确定旅行者与护照不符,并转介她进行进一步检查。 二次检查证实该旅客是加纳公民,是美国护照的冒名顶替者。

    October 2nd, 2018: Third Imposter Within Two Months of Facial Recognition Technology System Installation at Washington Dulles International Airport in Northern Virginia.

    2018年10月2日:在弗吉尼亚北部的华盛顿杜勒斯国际机场安装了两个月的面部识别技术系统后,第三个冒名顶替者。

    (Directly from the United States Customs and Border Protection Website)

    (直接来自美国海关和边境保护局网站)

    A 26-year-old Cameroonian woman arrived on a flight from Accra, Ghana, which had originated from Johannesburg, South Africa. The women presented a United States Passport in the name of a 31-year-old United States citizen to the present CBP officer. The facial recognition technology used by the officer reported a mismatch to the photo in the passport. CBP officers confirmed her true identity during a secondary inspection and biometric examination. She was arrested for misuse of a passport (19 USC 1544).

    一名26岁的喀麦隆妇女从加纳阿克拉起飞,该飞机起源于南非约翰内斯堡。 这些妇女以现年31岁的美国公民的名义向现任CBP官员出示了美国护照。 警察使用的面部识别技术报告护照上的照片不匹配。 CBP官员在二次检查和生物特征检查期间确认了她的真实身份。 她因滥用护照被捕(19 USC 1544)。

    为什么要在面部识别方面关注其隐私? (Why Should One Be Concerned About Their Privacy In Regards To Facial Recognition?)

    With some modern-day examples of facial recognition listed above, the respective owners of these tools and databases don’t obtain permission from a person to capture their facial details. One will not have control over their personal information and how it is used.

    上面列出了一些现代人脸识别示例,这些工具和数据库的各自所有者没有获得任何人的许可来捕获其面部细节。 一个人将无法控制他们的个人信息及其使用方式。

    Here are some examples of how your privacy could be violated:

    以下是一些有关如何侵犯您的隐私的示例:

    (Source: Norton Security)

    (来源:诺顿安全)

    • Security: One’s facial data can be collected and stored, often without your permission. There is a possibility that hackers could access and steal that data.

      安全性:通常无需您的许可即可收集和存储一个人的面部数据。 黑客有可能访问和窃取该数据。

    • Prevalence: Facial recognition technology has started to become more widespread amongst the entirety of the United States and possibly the world. This means that your facial signature could end up in a lot of places without your knowledge. There is little to no chance that a person would know who has access to their faceprint.

      流行:面部识别技术已开始在美国乃至整个世界范围内变得越来越普遍。 这意味着您的面部特征可能会在您不知情的情况下出现在很多地方。 一个人几乎不可能知道谁有权使用自己的面部印记。

    • Ownership: Without any doubt, a person owns their face which is located above their neck. However, digital images are different. A person could have given up their right to ownership when you signed up on a social media network. A third-party could even track down images of you online and sell that data.

      拥有权:毫无疑问,一个人拥有自己脖子上方的脸。 但是,数字图像是不同的。 当您在社交媒体网络上注册时,一个人可能已经放弃了所有权。 第三方甚至可以在线追踪您的图像并出售这些数据。

    • Safety: Facial recognition could lead to online harassment and stalking. The way it could potentially work is when someone takes a picture of you without your knowledge and uses facial recognition software to find out exactly who you are.

      安全性:面部识别可能会导致网上骚扰和跟踪。 当某人在您不知情的情况下给您拍照并使用面部识别软件准确地找出您的身份时,它可能会起作用。

    • Mistaken Identity: Facial recognition systems may not be fully accurate. Check out the challenges part of the article to learn more about a real-life example.

      错误身份:面部识别系统可能不完全准确。 查看本文的挑战部分,以了解有关真实示例的更多信息。

    • Basic Freedoms: Government agencies and other authorized groups have the ability to track you in regards to what you do and where you go.

      基本自由:政府机构和其他授权团体可以根据您的工作和去向追踪您。

    面部识别技术使用失败 (Failures of Facial Recognition Technology Usage)

    Facial recognition systems have been used by police forces for over two decades. Studies by Massachusetts Institute of Technology (MIT) and the National Institute of Standards and Technology (NIST) have found that the technology is more accurate on white men compared to the other demographics/races. This is most likely because of a lack of diversity in the images used to develop the established databases.

    面部识别系统已被警察部队使用了二十多年。 麻省理工学院(MIT)和美国国家标准与技术研究院(NIST)的研究发现,与其他人口统计/种族相比,该技术在白人男性上更为准确。 这很可能是由于用于开发已建立的数据库的图像缺乏多样性。

    Here is an example of a fault in facial recognition technology.

    这是面部识别技术故障的一个示例。

    On a Thursday afternoon in January, Robert Williams was in his office when he got a call from the Detroit Police Department telling him to come to the station to be arrested. Thinking that it was a prank, he ignored the entire conversation. An hour later, he pulled into his driveway in Farmington Hills, Michigan when a cop car pulled up behind him. Two officers got out and handcuffed Williams on his front lawn in front of his wife and two daughters.

    1月的一个星期四下午,罗伯特·威廉姆斯(Robert Williams)在底特律警察局打来电话,告诉他要去警察局逮捕。 认为这是一个恶作剧,他忽略了整个对话。 一个小时后,当一辆警车从他身后驶过时,他驶入密歇根州法明顿希尔斯的车道。 两名警官下车,将威廉姆斯铐在妻子和两个女儿面前的前草坪上。

    The police drove Mr. Williams to a detention center. He had his mugshot, fingerprints, and DNA taken. He was held overnight in jail. The next day, two detectives took him to an interrogation room. The interrogation happened in this format below.

    警察把威廉姆斯先生赶到拘留所。 他拍摄了面部照片,指纹和DNA。 他被关在监狱里过夜。 第二天,两名侦探将他带到讯问室。 审讯以下面的这种格式发生。

    Detective 1: When was the last time you went to a Shinola store?

    侦探1:您上一次去Shinola商店是什么时候?

    • Shinola is an upscale boutique that sells watches, bicycles, and leather goods in Detroit, Michigan.

      Shinola是一家位于密歇根州底特律的高档精品店,出售手表,自行车和皮革制品。

    Robert Williams: My wife and I checked it out when the store first opened in 2014.

    罗伯特·威廉姆斯(Robert Williams):我和我的妻子在商店于2014年首次开业时就检查了一下。

    • One of the detectives then turned over the first piece of paper. It was a still image from a surveillance video that showed a heavily built man, dressed in black, and wearing a St. Louis Cardinals cap standing in front of a watch display. Five timepieces, in total worth $3,800.00, were shoplifted.

      然后,一名侦探把第一张纸翻了过来。 这是来自监视视频的静止图像,显示了一个身材魁梧的男子,身穿黑色衣服,戴着圣路易斯红雀队的帽子站在手表显示器前。 窃取了五枚总价值为$ 3,800.00的时计。

    One of the Detectives: Is this you?

    侦探之一:这是你吗?

    • The second piece of paper that was shown was a close-up. The photo was blurry, but it was distinctive enough to tell that it wasn’t Robert Williams.

      显示的第二张纸是特写镜头。 这张照片很模糊,但是足以证明不是罗伯特·威廉姆斯。

    Robert Williams: No, this is not me. Do you think all black men look alike?

    罗伯特·威廉姆斯:不,这不是我。 您认为所有黑人男子看起来都一样吗?

    • Robert Williams knew that he didn’t commit the crime. What he didn’t know was that his case is the first publically known account of an American being wrongfully accused and arrested based on a flawed match from a facial recognition algorithm.

      罗伯特·威廉姆斯(Robert Williams)知道他没有犯罪。 他所不知道的是,他的案子是第一个公开的说法,即基于面部识别算法中的错误匹配,美国人被错误地指控和逮捕。

    The charges against Robert Williams were dropped soon after they discovered it was a mistake made by the facial recognition technology itself.

    罗伯特·威廉姆斯(Robert Williams)被发现是面部识别技术本身犯了一个错误后,很快就撤销了指控。

    After this incident occurred, Amazon, Microsoft, and IBM announced that they would stop or pause their facial recognition offerings for law enforcement. The facial recognition technology used by police departments across the country is supplied by companies that aren’t known to many people. Companies include Cognitec, NEC, Vigilant Solutions, Rank One Computing, and Clearview AI.

    发生此事件后,亚马逊,微软和IBM宣布将停止或暂停其面部识别产品以进行执法。 全国各地的警察部门使用的面部识别技术由许多人所不了解的公司提供。 公司包括Cognitec,NEC,Vigilant Solutions,Rank One Computing和Clearview AI。

    结论 (Conclusion)

    Overall, facial recognition technology is extremely beneficial when it comes to matching a high-resolution picture to a database of faces. It shouldn’t be a primary source of evidence to accuse and arrest an individual because of the race boundaries and the fluctuation of accuracy levels between the demographics.

    总体而言,当将高分辨率图片与人脸数据库进行匹配时,人脸识别技术是极其有益的。 由于种族界限和人口统计数据之间准确性水平的波动,这不应成为指控和逮捕个人的主要证据来源。

    引文 (Citations)

    Brewster, Thomas. “The Wrongful Arrest Of A Black Man Provides More Proof Facial Recognition Is Racist.” Forbes, Forbes Magazine, 24 June 2020, www.forbes.com/sites/thomasbrewster/2020/06/24/a-wrongful-arrest-of-a-black-man-provides-more-proof-facial-recognition-is-racist/#314af3385deb.

    布鲁斯特,托马斯。 “对黑人的不法逮捕提供了更多证据,这是种族主义。” 福布斯》 ,《福布斯》杂志,2020年6月24日, www.forbes.com / sites / thomasbrewster / 2020/06/24 / a-wrongful-arrest-of-a-black-man-provides-more-proof-facial-recognition-is -种族主义者/#314af3385deb。

    “CBP at Washington Dulles International Airport Intercepted an Imposter Using New Cutting-Edge Facial Comparison Biometrics Technology.” CBP at Washington Dulles International Airport Intercepted an Imposter Using New Cutting-Edge Facial Comparison Biometrics Technology | U.S. Customs and Border Protection, www.cbp.gov/newsroom/local-media-release/cbp-washington-dulles-international-airport-intercepted-imposter-using.

    “华盛顿杜勒斯国际机场的CBP使用新的尖端面部比较生物识别技术拦截了冒名顶替者。” 华盛顿杜勒斯国际机场的CBP使用新的尖端面部比较生物识别技术拦截了冒名顶替者| 美国海关与边境保护局www.cbp.gov / newsroom / local-media-release / cbp-washington-dulles-international-airport-intercepted-imposter-using。

    Center, Electronic Privacy Information. “EPIC — Facial Recognition.” Electronic Privacy Information Center, epic.org/privacy/facerecognition/.

    中心,电子隐私信息。 “ EPIC-人脸识别。” 电子隐私信息中心 ,epic.org/privacy/facerecognition/。

    “Dulles CBP’s New Biometric Verification Technology Catches Third Impostor in 40 Days.” Dulles CBP’s New Biometric Verification Technology Catches Third Impostor in 40 Days | U.S. Customs and Border Protection, www.cbp.gov/newsroom/national-media-release/dulles-cbp-s-new-biometric-verification-technology-catches-third.

    “杜勒斯CBP的新生物特征验证技术在40天内抓住了第三个冒名顶替者。” 杜勒斯CBP的新生物特征验证技术在40天内抓住了第三名冒名顶替者| Business Wire 美国海关与边境保护局 ( www.cbp.gov/newsroom/national-media-release/dulles-cbp-s-new-biometric-verification-technology-catching-third)。

    Garbage In. Garbage Out. Face Recognition on Flawed Data, www.flawedfacedata.com/.

    垃圾进入。 垃圾了。 人脸识别缺陷数据www.flawedfacedata.com /。

    Gervais, Joe. “How Does Facial Recognition Work?” Norton Security, us.norton.com/internetsecurity-iot-how-facial-recognition-software-works.html.

    Gervais,乔。 “面部识别如何工作?” 诺顿网络安全 ,us.norton.com / internetsecurity-iot-how-facial-recognition-software-works.html。

    Hill, Kashmir. “Wrongfully Accused by an Algorithm.” The New York Times, The New York Times, 24 June 2020, www.nytimes.com/2020/06/24/technology/facial-recognition-arrest.html.

    克什米尔山。 “被算法错误指责。” 纽约时报》 ,《纽约时报》,2020年6月24日, www.nytimes.com / 2020/06/24 / technology / facial-recognition-arrest.html。

    “History of Facial Recognition Technology.” Facial Recognition, www.doc.ic.ac.uk/~hh4017/History#:~:text=It%20started%20as%20a%20search,to%20detect%20faces%20within%20images.

    “面部识别技术的历史。” 面部识别www.doc.ic.ac.uk /〜hh4017 / History#:〜:text = It%20started%20as%20a%20search ,以% 20images%的速度检测到%20faces%20。

    Liao, Shannon. “New Facial Recognition System Catches First Imposter at US Airport.” The Verge, The Verge, 24 Aug. 2018, www.theverge.com/2018/8/24/17778736/facial-recognition-washington-airport-immigration-biometric-exit.

    廖香农 “新的面部识别系统在美国机场率先冒名顶替。” 濒临边缘,2018年8月24日, www.theverge.com / 2018/8/24/17778736 / facial-recognition-washington-airport-immigration-biometric-exit。

    “The Perpetual Line-Up.” Perpetual Line Up, www.perpetuallineup.org/.

    “永恒的阵容。” 永久排队www.perpetuallineup.org /。

    Raviv, Shaun. “The Secret History of Facial Recognition.” Wired, Conde Nast, www.wired.com/story/secret-history-facial-recognition/.

    肖恩·拉维夫。 “面部识别的秘密历史。” Wired ,康德纳斯特(Conde Nast), www.wired.com / story / secret-history-facial-recognition /。

    “Second Impostor in Three Weeks Caught by CBP Biometric Verification Technology at Washington Dulles Airport.” Second Impostor in Three Weeks Caught by CBP Biometric Verification Technology at Washington Dulles Airport | U.S. Customs and Border Protection, www.cbp.gov/newsroom/local-media-release/second-impostor-three-weeks-caught-cbp-biometric-verification.

    “华盛顿杜勒斯机场CBP生物识别验证技术在三周内第二次冒名顶替。” 华盛顿杜勒斯机场CBP生物识别验证技术在三周内第二次冒犯 美国海关与边境保护局www.cbp.gov / newsroom / local-media-release / second-impostor-weeks-caught-cbp-biometric-verification。

    Van Wert, John Paul. “Facial Recognition Technology at Gate (44275734970).Jpg.” Wikimedia Commons, 19 Nov. 2018, commons.wikimedia.org/wiki/File:Facial_recognition_technology_at_gate_(44275734970).jpg.

    范·沃特,约翰·保罗。 “ Gate的面部识别技术(44275734970).Jpg。” Wikimedia Commons ,2018年11月19日, commons.wikimedia.org/wiki/File:Facial_recognition_technology_at_gate_(44275734970).jpg

    翻译自: https://medium.com/the-scitech-scoop/the-fundamentals-of-facial-recognition-technology-b8a5d67ea558

    人脸脸部识别技术

    展开全文
  • 一种基于脸部识别的人工智能口罩发放箱.pdf
  • 人脸脸部识别技术 揭露 (Disclosure) The following introduction references existing technology and future advances in facial recognition. This scenario is fictional but should be considered technology ...

    人脸脸部识别技术

    揭露 (Disclosure)

    The following introduction references existing technology and future advances in facial recognition. This scenario is fictional but should be considered technology possible.

    以下介绍参考了现有技术以及面部识别的未来发展。 这种情况是虚构的,但应视为可能的技术。

    故事开始 (The Story Begins)

    On a warm south Texas fall afternoon, Chloe and Olivia left school for home. They caught bus #632 at the corner in front of the bank across the street from RHS. They’d only ride a mile, initially. Chloe wanted to stop off at the Big Box SuperStore on the way. She had been at the store the previous Saturday when one of the salesladies told her about the new line of fall sweaters due to arrive Wednesday. They’d be stocked on the shelves and hangers by Thursday afternoon. She was dying to see what Big Box had.

    在德克萨斯州南部一个温暖的秋天下午,克洛伊和奥利维亚离开学校回家。 他们在RHS对面的银行对面的拐角处乘坐#632号公共汽车。 最初,他们只会骑一英里。 克洛伊(Chloe)希望在此期间在Big Box SuperStore停留。 她曾在上个星期六去过商店,当时一位女售货员告诉她有关定于周三到达的新系列秋季毛衣的信息。 他们要在星期四下午之前放在架子和衣架上。 她渴望看到Big Box拥有什么。

    They left campus as soon as they could. School let out at 4:00 pm. The bus would be at their stop by 4:08. They didn’t want to miss the bus because the next one wouldn’t arrive until nearly 4:30. Catching their preferred bus would get them to Big Box by 4:12, a four-minute-mile ride. Although they both were on the track team and conceivably could run the distance in about the same amount of time as riding the bus, why be all hot and sweaty when they got there.

    他们尽快离开校园。 学校在下午4:00放学。 公共汽车将在4:08停下来。 他们不想错过公共汽车,因为下一辆要到近4:30才到。 赶上他们偏爱的公交车,可以在4:12(四分钟车程)之前将他们送至Big Box。 尽管他们俩都在田径队,并且可以想象到,他们可以在大约与乘坐公交车相同的时间内跑完距离,但是为什么他们到达那里时却又热又汗呢?

    Chloe and Olivia walked through the main doors a little after 4:20 pm. They often came to Big Box. After all, the store had most anything a person could want. Stuff wasn’t fancy, but then it wasn’t expensive either. Chloe could generally afford what she wanted as long as it wasn’t too close to the end of the month. For Olivia, it was always too close to the end of the month. Olivia usually shopped at Goodwill across the street from Big Box.

    下午4点20分过后,克洛伊(Chloe)和奥利维亚(Olivia)走进大门。 他们经常来大盒子。 毕竟,商店拥有人们想要的大部分东西。 东西不是花哨的,但是那也不贵。 只要离月底还不太近,Chloe通常就能负担得起。 对于Olivia来说,离月底总是太近了。 奥利维亚通常在Big Box街对面的Goodwill购物。

    大哥在看 (Big Brother Is Watching)

    As soon as they walked through the doors, BBFR recognized them. BBFR is an abbreviation for Big Box Facial Recognition. The loss prevention staff just called it Big Brother. Big Brother is the first line of shoplifting defense in the arsenal of anti-theft devices employed by Big Box. Every person entering the store, any store in the Big Box empire, is photographed at the doors. BBFR’s algorithms determine if this customer has been in a Big Box or any of its affiliates before. A new customer is assigned a new account. The customer is assigned an 8-digit hexadecimal number, and their visit to the store is logged. If it’s a return customer, their visit is recorded.

    他们一走进门,BBFR就认出了他们。 BBFR是“ Big Box人脸识别”的缩写。 防损人员将其称为“老大哥”。 “大哥大”是Big Box所采用的防盗设备库中的入店行窃防御的第一道防线。 每个进入商店的人,在Big Box帝国的任何商店,都会在门口被拍照。 BBFR的算法确定该客户之前是否曾在Big Box或其任何关联公司中。 为新客户分配了新帐户。 为客户分配了一个8位十六进制数字,并记录了他们对商店的访问。 如果是回头客,则会记录他们的访问。

    Chloe is known to Big Brother as 3d9619a0, and Olivia is known as cd7cc365. Eight digit places of hex numbers will store 4.29 billion unique customers. So far, Big Box has used only a fraction of that number. Certainly not anywhere close to 7.57 billion numbers needed for the entire world population, but Big Box is optimistic about eventually needing a ninth digit.

    克洛伊(Chloe)被老大哥(Big Brother)称为3d9619a0,奥利维亚(Olivia)被称为cd7cc365。 八位数的十六进制数字将存储42.9亿的唯一客户。 到目前为止,Big Box仅使用了该数字的一小部分。 当然,整个世界人口所需的数字并不接近75.7亿,但Big Box乐观地认为最终将需要第九个数字。

    From the store management’s point of view, there are two cool things about Facial Recognition (FR). The first is that things like sunglasses, facial hair, and makeup are mostly irrelevant to the Big Brother algorithm. FR uses permanent facial features of the shape of eyes and distance between the eyes, size, and shape of the nose, and length of the jawline. Facial data points, known as nodal points, create a three-dimensional facial construction. Such constructions even use skin texture.

    从商店管理层的角度来看,面部识别(FR)有两个很酷的东西。 首先是太阳镜,胡子和化妆等东西与Big Brother算法无关。 FR使用永久性的面部特征,包括眼睛形状,眼睛之间的距离,鼻子的大小和形状以及下巴的长度。 面部数据点(称为节点)可创建三维面部构造。 这样的构造甚至使用皮肤纹理。

    The second cool thing about FR is the general public doesn’t know it’s in use, and the company, depending on the store location, has no legal obligation to inform anyone. (Wernick, 2019)

    关于FR的第二个有趣的事情是,公众不知道它的使用情况,根据商店位置,该公司没有通知任何人的法律义务。 (韦尼克,2019)

    Two years earlier, when the girls were 16, Olivia had been window shopping and accidentally dropped some L’Oreal Paris True Match Super-Blendable Blush, retail price $7.97, into her bag. She swore up and down she didn’t mean to and hadn’t noticed it falling into her purse. But unfortunately for her, Ms. Rose, a plainclothes store detective, observed the incident. It was Ms. Rose who confronted her and detained her outside the store.

    两年前,当女孩16岁时,奥利维亚(Olivia)逛街购物,不小心把一些欧莱雅巴黎True Match超级混合腮红(零售价7.97美元)放进了她的包里。 她发誓不愿冒犯,也没有注意到它掉进了钱包。 但不幸的是,便衣商店探员罗斯女士发现了这起事件。 罗斯女士面对她并将她拘留在商店外。

    Given the relatively low price of the item taken, the police were not called. Olivia was photographed, her parents were contacted, and she received a “Notification of Restriction from Property (NRP).” This notice essentially says stay away from all Big Box stores and their affiliates forever. If she showed up on a Big Box property, she could be arrested for criminal trespassing.

    鉴于所取物品的价格相对较低,因此未召集警方。 奥利维亚(Olivia)被拍照,与她的父母联系,并收到“财产限制通知(NRP)”。 该通知实质上表示永远远离所有Big Box商店及其分支机构。 如果她出现在Big Box财产上,可能会因涉嫌犯罪罪名而被捕。

    Initially, Olivia heeded the no-trespass notice. But as time went on, Olivia started coming back to the store, usually with Chloe.

    最初,奥利维亚(Olivia)留意了禁止进入的通知。 但是随着时间的流逝,Olivia通常会和Chloe一起回到商店。

    Upon BBFR installation at the store, Olivia’s past transgression was uploaded into the system. Anytime Olivia came into Big Box, an alert was sent to loss-prevention. However, because Olivia was still considered a juvenile and no theft issues arose, loss prevention viewed her presence as something to be monitored but not acted on unless there was a problem. Also, profit from a sale was still profit, and management bonuses are dependent on profit. Olivia’s grace period would end today.

    在商店中安装BBFR后,Olivia的过往违法行为被上载到系统中。 每当Olivia进入Big Box时,都会发出警报以防止损失。 但是,由于Olivia仍被视为少年,并且没有出现盗窃问题,因此预防损失将她的存在视为需要监视的事物,除非有问题,否则不会采取任何行动。 同样,销售利润仍然是利润,管理奖金取决于利润。 奥利维亚的宽限期将在今天结束。

    When Olivia entered the store, an alert was sent immediately to the loss prevention team on their store smartphones. Big Brother started a real-time camera track on cd7cc365, as it had every other time she had come into the store. Conrad was working on the camera monitoring that day. Conrad would monitor the cameras that were watching the store and high-risk customers. Though Big Brother would alert the loss prevention team of identified risks, Conrad was particularly adept at identifying shady behavior.

    当Olivia进入商店时,会立即通过商店智能手机向防损小组发送警报。 “老大哥”在cd7cc365上开始了实时摄像头跟踪,因为这是她每隔一次进入商店。 那天,康拉德(Conrad)正在监控摄像机。 康拉德(Conrad)将监视正在监视商店和高风险客户的摄像机。 尽管“老大哥”会提醒损失预防小组注意已发现的风险,但康拉德尤其擅长识别可疑行为。

    There was no need for Conrad to radio Ms. Rose of cd7cc365’s location. Ms. Rose already had that information on her handheld store device. She had a map of the store, and a small photo identified Olivia’s place. At the moment, there was only one thumbnail picture on her screen.

    康拉德无需广播cd7cc365所在位置的Rose女士。 罗斯女士已经在手持式存储设备上获得了该信息。 她有商店的地图,还有一张小照片标识了Olivia的位置。 目前,她的屏幕上只有一张缩略图。

    Various times of the year, particularly beginning Black Friday, multiple pictures tended to show up on the loss prevention team’s devices. At these times, those risks with NRPs would be met just inside the doors. Big Box staff reminded them they were unwelcome and should take their business elsewhere.

    在一年中的不同时间,尤其是从黑色星期五开始,防损团队的设备上往往会出现多张照片。 在这些时候,使用NRP的那些风险将在门内得到满足。 Big Box员工提醒他们他们不受欢迎,应该将业务转移到其他地方。

    Ms. Rose would be the eyes on the sales floor, looking for the telltale signs of dishonest behavior. Good thieves, rather, thieves that are good at stealing, know the cameras are there and can disguise their actions enough that the overhead cameras can’t see the concealment. Good thieves can also block concealment from loss prevention officers on the floor, but it is impossible to conceal theft from both cameras and floor staff simultaneously.

    罗斯女士会盯着销售现场,寻找不诚实行为的明显迹象。 好的盗贼更擅长偷窃,他们知道相机在那里并且可以掩饰其行为,以至于高架摄像机看不到隐藏物。 好的盗贼还可以阻止地板上防损人员的掩盖,但是不可能同时掩盖摄像机和地板人员的盗窃行为。

    Olivia and Chloe made their way to the women’s clothing department, unaware their actions were observed in duality. Together, the girls eagerly went through the fall sweaters. Although both girls happily went through the collection of sweaters, it was Chloe that was shopping. Olivia was there to offer advice.

    奥利维亚和克洛伊(Olivia)和克洛伊(Chloe)进入女装部门,却没有意识到他们的行为是双重的。 女孩们在一起热切地穿过秋天的毛衣。 尽管两个女孩都愉快地浏览了毛衣系列,但Chloe在购物。 奥利维亚在那里提供建议。

    They had gone through the Time and Tru sweaters priced $9.99 to $18.99. Cloe had found a couple that could do for her, but the sweater she wanted was a Calvin Klein Cowlneck Knit Sweater priced $39.99. The $25 she had would amply cover any of the Time and Trus, but not the Calvin Klein. Chloe would not have money for the CK until after the first of the new month. What should she do?

    他们经历过Time and Tru毛衣的价格在9.99美元至18.99美元之间。 Cloe找到了一对可以为她做的夫妇,但她想要的毛衣是Calvin Klein Cowlneck针织毛衣,售价39.99美元。 她所拥有的25美元足以支付《时代周刊》中的任何内容,但不包括Calvin Klein。 直到新月的第一个月,Chloe才没有钱买CK。 她该干什么?

    As the girls exited the store, Ms. Rose walked up to them, saying to Chloe, “Miss, may I see your receipt for the sweater in your bag?”

    当女孩们离开商店时,罗斯女士走到她们面前,对克洛伊说:“小姐,我可以在包里看到你的毛衣收据吗?”

    本文研究 (Research For This Article)

    In researching this article, I found many commercial vendors offering facial recognition products. Here is what several have to say about their services.

    在研究本文时,我发现许多提供面部识别产品的商业供应商。 以下是关于他们的服务的几点看法。

    Paravision — “Paravision’s algorithms excel across a range of challenging scenarios including cooperative and non-cooperative faces, light and angle variability, blur and pixelation, occlusions, and diversity across phenotype, age, and gender.”

    Paravision –“ Paravision的算法在一系列具有挑战性的场景中表现出色,包括合作和非合作的面Kong,光线和角度的可变性,模糊和像素化,遮挡以及表型,年龄和性别的多样性。”

    FaceFirst — “FaceFirst notifies your loss prevention team when known criminals, stalkers or disgruntled former employees enter the workplace…”

    FaceFirst —“当已知的罪犯,缠扰者或不满的前雇员进入工作场所时,FaceFirst会通知您的损失预防团队……”

    Cognitec — “Face recognition technology can detect people’s faces in live video streams or video footage and store anonymous information for each appearance of a person in front of a camera. Analysis of this information over time allows the software to compute people count… and to detect frequent visitors and crowds.”

    Cognitec —人脸识别技术可以检测实时视频流或视频镜头中的人脸,并在摄像机前为人的每次出现存储匿名信息。 随着时间的流逝,对这些信息的分析使该软件能够计算人数……并检测经常来访的游客和人群。”

    商店会使用这样的技术吗? (Would Or Do Stores Use Technology Like This?)

    Yes, of course, they would and do. The largest retailer in the world, Walmart, has experimented with facial recognition (Roberts, 2015). Walmart also uses the Everseen system at some store locations to detect unscanned items placed into shopping bags (BBC News, 2019).

    是的,当然,他们会并且愿意。 世界上最大的零售商沃尔玛已经尝试了面部识别(Roberts,2015年)。 沃尔玛还在某些商店位置使用Everseen系统来检测放入购物袋的未扫描物品(BBC新闻,2019年)。

    China is a world leader in facial recognition technology and regularly reminds its citizens that such equipment will make it almost impossible to evade the authorities.” — BBC News, April 13, 2018

    中国是面部识别技术的世界领导者,并定期提醒其公民,这种设备将使其几乎不可能逃脱当局。” -BBC新闻,2018年4月13日

    对于监控摄像机的最坏情况,我们可能会有什么期望? (What Might We Expect For A Worst Case Scenario Of Surveillance Cameras?)

    Probably the most widely known user of FR technology is the Chinese government. The news media reports extensive use of FR. There was a recent report of a fugitive identified out of the crowd of 60,000 concert goers (BBC News, April 13, 2018). Another story tells of Chinese police officers wearing unique FR enhance glasses. “The glasses are connected to an internal database of suspects, meaning officers can quickly scan crowds while looking for fugitives.” (BBC News, Feb. 7, 2018)

    FR技术最广为人知的用户可能是中国政府。 新闻媒体报道了FR的广泛使用。 最近有报道称,在6万名演唱会人群中发现了逃犯(英国广播公司新闻,2018年4月13日)。 另一个故事讲述了中国警察戴着独特的FR增强眼镜。 “这些眼镜与犯罪嫌疑人的内部数据库相连,这意味着警官可以在寻找逃犯的同时Swift扫视人群。” (BBC新闻,2018年2月7日)

    BBC correspondent John Sudworth partnered with the Chinese police to experiment. His photograph and identifying information were entered into the national database as a person of interest. He left police headquarters with the intent of walking to an unknown destination. He was apprehended 7 minutes later at a bus station by five policemen (Lui, 2017).

    英国广播公司(BBC)记者约翰·萨德沃思(John Sudworth)与中国警察合作进行实验。 他的照片和身份信息作为感兴趣的人被输入到国家数据库中。 他出于步行目的离开警察总部。 7分钟后,他在一个汽车站被五名警察逮捕(Lui,2017)。

    An estimated 170 million CCTV cameras are in use in China, and at the end of 2021, there will be an additional 400 million cameras added to the system. (Lui, 2017)

    估计在中国使用的CCTV摄像机有1.7亿部,到2021年底,该系统将再增加4亿部摄像机。 (吕,2017)

    技术的使用如何影响我们的第四修正案权利? (How does the use of technology affect our 4th amendment rights?)

    There is a difference between government use of FR and private use of FR.

    政府使用FR和私人使用FR之间有区别。

    The consequences of infringement of civil liberties by the government are clearly outlined in our laws.

    我们的法律明确规定了政府侵犯公民自由的后果。

    The 4th amendment states, “The right of the people to be secure in their persons, houses, papers, and effects, against unreasonable searches and seizures, shall not be violated, and no Warrants shall issue, but upon probable cause, supported by Oath or affirmation, and particularly describing the place to be searched, and the persons or things to be seized.” — National Constitution Center

    第四修正案规定:“不得侵犯人民保护其人身,房屋,文件和财物的权利,以免进行不合理的搜查和扣押,也不得发出任何认股权证,但在可能的原因下,应由誓言支持或确认书,特别是描述要搜寻的地点以及要扣押的人或物。” —国家宪法中心

    Use of FR by private individuals, companies we choose to do business with, and private lawful use by others is not so clearly defined.

    私人,我们选择与之进行业务往来的公司对FR的使用以及他人对私人合法使用的定义尚不明确。

    “‘Facial recognition really doesn’t have a place in society,’ said Evan Greer, deputy director of Fight for the Future. ‘It’s deeply invasive, and from our perspective, the potential harm to society and human liberties far outweigh the potential benefits.’ “ — (Yamanouchi, 2019)

    “争取社会认可确实没有社会地位,”“争取未来”副主任埃文·格里尔(Evan Greer)说。 “这是一种深远的侵略,从我们的角度来看,对社会和人类自由的潜在危害远远大于潜在的利益。” -—(山之内,2019年)

    摘要 (Summary)

    The year 1984 was 35 years ago. However, I see 1984 (Orwell, 1949 ) as clearly possible in America’s future.

    1984年是35年前。 但是,我认为1984 (Orwell,1949)在美国的未来显然是可能的。

    Citations:

    引文:

    BBC News, Feb. 7, 2018, Chinese police spot suspects with surveillance sunglasses

    英国广播公司新闻,2018年2月7日, 中国警察通过监视太阳镜发现嫌疑犯

    BBC News, April 13, 2018, Chinese man caught by facial recognition at pop concert

    英国广播公司新闻,2018年4月13日, 中国人在流行音乐会上被面部识别抓住

    BBC News, June 21, 2019, Walmart uses AI cameras to spot thieves

    BBC新闻,2019年6月21日, 沃尔玛使用AI摄像头发现小偷

    Lui, Joyce, Dec. 10, 2017, In Your Face: China’s all-seeing state, BBC News

    Lui,Joyce,2017 12月10日, 在您的面前:中国无所不能的国家 ,BBC新闻

    National Constitution Center, 4th Amendment — Search and Seizure

    国家宪法中心, 第四修正案-搜查和扣押

    Orwell, George, 1984, Secker & Warburg Publishers, 1949

    奥威尔,乔治(George),1984年;塞克与华宝出版社(Secker&Warburg Publishers),1949年

    Roberts, Jeff John, Nov. 9, 2015, Walmart’s Use of Sci-fi Tech To Spot Shoplifters Raises Privacy Questions, Fortune.com

    罗伯茨(Roberts),杰夫·约翰(Jeff John),2015年11月9日, 沃尔玛(Walmart)利用科幻技术发现商店行窃者引发了隐私问题 ,Fortune.com

    Warnick, Alan S., July 02, 2019, Biometric Information — Permanent Personally Identifiable Information Risk, American Bar Association

    Warnick,Alan S.,2019年7月2日, 生物识别信息—永久性个人身份信息风险 ,美国律师协会

    Yamanouchi, Kelly, Sept. 18, 2019, As Delta Air Lines Expands Face Recognition, Criticism Grows, The Atlanta Journal-Constitution

    Yamanouchi,Kelly,2019年9月18日, 随着达美航空(Delta Air Lines)扩大人脸识别能力,批评的增长 ,《亚特兰大期刊社》

    Cognitec.com

    Cognitec.com

    FaceFirst.com

    FaceFirst.com

    Paravision.ai

    Paravision.ai

    © 2019 Randle B. Moore All Rights Reserved

    ©2019 Randle B.Moore保留所有权利

    If you found this story informative, you may also find these stories fascinating.

    如果您发现此故事内容丰富,那么您可能还会发现这些故事很有趣。

    翻译自: https://medium.com/the-partnered-pen/dangerous-consequences-of-facial-recognition-technology-a55c89345c26

    人脸脸部识别技术

    展开全文
  • 脸部识别是根据什么识别的 通用情报 (General Intelligence) Welcome to General Intelligence, OneZero’s weekly dive into the A.I. news and research that matters. 欢迎来到 General Intelligence , OneZero ...
  • 人脸脸部识别技术Regulation moves at a snail’s pace, so it’s up to CEOs, executives, and employees to reject projects that put profit over privacy. Clearview AI facial-recognition tech is just the ...
  • 人脸脸部识别技术Terminator and similar movies depict a world that is controlled by AI and robots. For the mainstream this may seem outlandish, but notable figures like Elon Musk have expressed ...
  • 脸部识别是根据什么识别的 通用情报 (General Intelligence) Welcome to General Intelligence, OneZero’s weekly dive into the A.I. news and research that matters. 欢迎来到 General Intelligence , OneZero ...
  • opencv 脸部识别An eazy code from eazy ciphers 易密码的易密码 Are you keen to learn about the implementation of Real-Time Face recognition? 您是否想了解实时人脸识别的实现? Here eazy ciphers came ...
  • 修复windows脸部识别If you have the right hardware, Windows 10 lets you unlock your computer with nothing but a smile. However, Microsoft’s facial recognition isn’t always spot-on. Here’s how tohelp...
  • CartoonEyes.swift - 前置摄像头捕获图像后,采用 Core Image 脸部识别 CIDetector 和漫画效果滤镜复合出卡通效果眼睛。
  • 加拿大多伦多大学(University of Toronto)的研究人员宣布开发出了人工智能驱动的程序,可干扰脸部识别系统。该程序设计用于在像素水平精细地改变图像,干扰数...
  • 脸部识别/追踪最快的实现,没有之一↓↓↓
  • 本指导致力于帮助开发者使用骁龙脸部识别软件开发工具包来编写安卓应用。 在编写应用前,建议您熟悉编写使用的Snapdragon™SDK的应用程序---脸部识别框架,因为面部识别是脸部识别软件工具包扩展附加的功能。   本...
  • Kinect for windows提供了脸部识别的功能,可以识出人脸。主要是通过FaceTrackFrame类的GetTriangles()来得到一个三角形数组,这个三角形数组就是给成人面部的基本形状,并且组成的效果是立体的(可以这样理解,可以...
  • DKSVD字典学习的脸部识别

    千次阅读 2017-08-13 09:29:58
    在读完论文代码后,对于“DKSVD字典学习的脸部识别”这篇论文中的数学模型理解如下: 这篇论文采用递进的方式从SRC算法到KSVD,Baseline,DKSVD算法依次介绍,以下也按照论文所描述的流程对论文中的算法公式进行逐步...
  • 脸部识别算法Does this man look familiar? This is Robert Williams who was misidentified by a police facial recognition system and had to spend a day under arrest. As this incident gets thrown around in...
  • 南京公安脸部识别锁定 虚拟助手项目 (Virtual Assistant Project) Security is one of the biggest concerns of modern-day. Making sure only the right person is granted access to a device becomes extremely ...
  • 脸部识别登录Ubuntu

    2010-10-17 08:47:00
    使用pam-face-authentication脸部识别工具, 登录到Unbuntu只需要你的一个优雅的傻笑。它是专门为日常使用的普通用户而设计的。 文件依赖关系要安装和运行这个功能需要先安装一些依赖关系文件。我们为了节省一些...
  • 断断续续地折腾了大半个月,终于把身份证照片脸部识别以及OCR字体识别功能用Java实现了,需求很简单:通过摄像头所照的一张放在黑色底板上的身份证照,识别照片上身份证里面的人名和地址(OCR中文),再截取身份证上的...
  • 程序员和数据确定了脸部识别等分类算法的工作原理。 要注意的是,数据和人类都存在偏差,分别具有分布不均衡的图像类型或个人偏好。还要注意的是,这些偏差会分散到算法创建中。对于脸部识别,想象一下以下情形:...

空空如也

空空如也

1 2 3 4 5 ... 20
收藏数 1,204
精华内容 481
关键字:

脸部识别