Small eyes provoke who? Man New Zealand face-recognition systems make Halo-IT information
Small eyes well outside the spotlight or have any special skills? New Zealand a Chinese student who used personal experience to prove small eyes can confuse facial recognition systems.
22 Lee (Richard Lee) was born in Taiwan and New Zealand grew up, are New Zealand citizens. He is a DJ, Melbourne, is currently studying Aerospace Engineering and business administration.
In order to be able after the Christmas holidays to Australia to go to school, Lee prepared to update their earlier New Zealand passports. 6th, his New Zealand passport photos check the website of the Ministry of the Interior submitted his photo, but does suggest that he submitted the photo does not meet the requirements of the system, because "eyes open".
But in the photo, Lee's eyes are open.
Screenshot submission. Photo: Facebook
In a CNN interview, Lee said he later tried a few pictures to no avail, so calling in New Zealand the Ministry of the Interior. Staff told him that because of his "eye shadow and face uneven lighting, lead to recognition software."
Lee had to go again some pictures eventually managed to have a system to identify and successfully submitted.
The Black Dragon event, Lee said "get down from the Chair. I know their eyes are small, it doesn't matter to me, but it seems the computer is aware of this problem. Too funny. ”
Later, Lee's own story on Facebook, triggering laughter and discussions of netizens. Have good friends to help build a new picture, I hope this recognition through the system.
There is one think technology has become racially.
But Lee doesn't think so. He said his eye is very small, the facial recognition system is not so advanced, so "I'm not unhappy."
Problems faced by Lee, New Zealand responded, a spokesman for the Ministry of the Interior, the Department uses recognition software is "one of the most advanced technology in the world," but "more than 20% photos submitted online will be rejected due to various reasons." Lee received tips are just "General error".
The spokesman also said, the website provides all kinds of help information to the user, to facilitate submission meets the required passport photos.
The daily mail after tests found that eyes received information that the applicant did not open help tips are: "photograph, make sure that the eyes are open again. ”
For a facial recognition system for racial differences in the issues, there have been discussions.
Early in 2009, HP's dynamic tracking camera is being questioned. From the United States of two colleagues-a white woman and a black man, testing dynamic tracking cameras, found that cameras can only track lock white women, and in the face of black men, they were out of.
The test immediately after the videos were put on Youtube has become a hot video, many users are starting to talk about whether there are racial differences of this technology. HP responded that this may be related to light, "in the foreground in low light situations, camera added to the contrast there will be difficulties".
Other bodies were later tested the camera, found under normal lighting conditions, camera in when chasing black men do not sensitive, normal operation only in the case of additional lighting.
In 2010, an Asian woman using a Nikon camera and found each other or laugh when her mother took her, the camera will prompt "If someone blinked." Only when her brother try to open your eyes, "bug-eyed a photograph, will prompt stop."
Photo: network
In April this year in the Atlantic Monthly published an article on the facial recognition systems is discussed. Points out that in 2011, a study found that China, Japan, and Korea developed algorithms are easier to identify East Asian faces, and France, and Germany and the United States it is more effective to develop algorithms to identify people of Caucasian facial features.
In 2012, the United States suspects in Pinellas County, Florida researchers used photos of three algorithms were tested facial recognition systems company. Study finds, the three companies of the algorithm to identify photos of suspects of African origin white suspects than recognition accuracy difference between 5% and 10%.
But too little as the number of related studies, the paper speculated that this racial difference is not deliberate.
Article speculated that engineers in the design of algorithms focused on facial features may be more easy to identify in a group, but in another group will have a different situation. For example, the width of the shape of the eyes, nose, mouth, and so on. Historical data collection and related errors may exist.
Another possibility is that the engineer in designing algorithm draws on his experience, which in itself would be an engineer my race ...
眼睛小招惹谁了？华裔男子让新西兰脸部识别系统犯晕 - IT资讯
小眼睛除了好聚光之外还有什么特长？新西兰的一名华裔学生就用亲身经历证明了小眼睛还能迷惑面部识别系统。
22岁的李（Richard Lee）出生于台湾，在新西兰长大，是新西兰公民。他是一名DJ，目前正在墨尔本学习航天工程和工商管理。
为了能在圣诞假期后回澳大利亚上学，李准备提前更新自己的新西兰护照。本月6日，他在新西兰内政部的护照照片检查网站上提交了自己的照片，然而系统却提示他提交的照片不符合要求，因为“眼睛没有睁开”。
但在照片上，李的眼睛确实是睁开的。
提交照片截图。图片来源：Facebook
在接受CNN采访时，李说他后来又试了几张照片都没有用，于是就打电话询问新西兰内政部。工作人员告诉他因为他“眼部的阴影和脸部的打光不均匀，导致软件出现识别困难”。
此后李只得重新去照了几张照片，最终总算有一张被系统识别，成功提交。
对于这起乌龙事件，李说自己“快从椅子上笑倒了。我知道自己眼睛小，我无所谓，但是看来电脑也意识到了这个问题。太搞笑了。”
事后，李将自己的遭遇发到了Facebook上，引发了网友们的一片笑声和讨论。有热心网友帮李修了一张新照片，希望这张能通过系统识别。
还有网友认为技术也变得带有种族歧视了。
但李自己并不这么认为。他称自己眼睛确实很小，而面部识别系统还没有那么先进，所以“我没有感到不高兴”。
对于李碰到的问题，新西兰内政部的一名发言人回应称，该部门使用的识别软件是“世界上最先进的技术之一”，但“网上提交的照片中有20%以上都会由于各种原因遭到拒绝”。李收到的提示只是“一般性错误信息”。
这名发言人还表示，网站为用户提供了各类帮助信息，以方便大家提交符合要求的护照照片。
而《每日邮报》试验后发现，对于收到信息称申请人眼睛没睁开的帮助提示是：“重新拍照、确保眼睛是睁开的。”
对于面部识别系统是否存在种族差别这个问题，此前也曾有过相关讨论。
早在2009年，惠普公司的动态跟踪摄像头就遭到了质疑。来自美国的两位同事——一位白人女性和一位黑人男性，测试了动态跟踪摄像头，结果发现摄像头只能跟踪锁定白人女性，而在面对黑人男性的时候就失灵了。
这个测试视频被放上Youtube以后立马成为了热门视频，很多网友开始议论这项技术是否存在种族差别。而惠普公司回应称出现这种情况可能和光线有关，“在前景光不足的情况下，摄像头在识别反差上存在困难”。
后来有其他机构对摄像头进行了测试，发现在正常照明情况下，摄像头在追踪黑人男性时确实不灵敏，只有在增加照明的情况下才能正常运作。
2010年，一名亚裔女性在使用尼康相机时发现，每次别人拍她或者她母亲笑的时候，相机都会提示“是否有人眨眼了”。只有当她哥哥努力睁大眼睛，“暴眼照相时，提示才会停止”。
图片来源：网络
《大西洋月刊》在今年4月发表的一篇文章里就对面部识别系统进行了讨论。文章指出2011年的一项研究发现，中国、日本、韩国开发的算法更容易识别东亚人的面孔；而法国、德国和美国开发的算法则更能有效识别白种人的面部特征。
2012年，美国的研究人员使用佛罗里达州皮内拉斯县的嫌疑犯照片对三家面部识别系统公司的算法进行了测试。研究发现，这三家公司的算法在识别非洲裔嫌犯照片时的准确度比识别白人嫌犯时差5%到10%。
但由于相关的研究数量太少，文章推测这种种族差别并不是有意造成的。
文章推测，工程师在设计算法时注重的面部特征可能在某个族群中更容易被识别，但是在另一个族群里会出现不同的情况。比如眼睛的形状、鼻子的宽度、嘴巴的大小等。而相关的历史数据在收集时可能就存在偏差。
另一种可能是工程师在设计算法时参照了自己的经验，而这本身就会受到工程师本人的种族特征影响。