face recognition

from Wikipedia, the free encyclopedia
Automatic face recognition

Face recognition describes the analysis of the characteristics of visible features in the area of ​​the frontal head , given by the geometrical arrangement and texture properties of the surface.

Differentiation of terms

A distinction must be made between the localization of a face in the image and the assignment of the face to a specific person. In the first case it is checked whether and where a face can be seen, in the second, who it is.

When it comes to face recognition in the sense of recognizing which face it is, one can differentiate between two cases: If this is done by people, the English-speaking world speaks of face perception , while face recognition by machines is called face recognition . Some animal species can distinguish faces in animals of the same species or in animals of another species, including humans, and recognize them in that sense. The terms face perception and face recognition ( conspecific / heterospecific ) are used here.

Ability of living beings to recognize human faces

The ability to recognize and differentiate between faces ( face perception ) is normally acquired by the human brain within the first few months of life. It is linked to the functions of the cerebrum, more precisely the occipital lobe . A partial or total lack of this ability is called prosopagnosia .

Newborns are more interested in faces or face-like objects than anything else. In the last third of pregnancy, embryos open their eyes and develop a sense of sight. In 2017, tests on unborn babies in the 34th week of pregnancy showed that they turn to a light pattern projected onto the mother's abdominal wall in a statistically significant manner if this has the basic pattern of a face.

Technical face recognition

In a technical context, face recognition is one of the biometric processes. It is used for security, criminal and forensic purposes for the purpose of identifying or verifying ( authenticating ) natural persons. The technical, computer-aided facial recognition is typically used to control access to security-sensitive areas and to search for duplicates in databases, for example in registration registers to prevent identity theft .

Decisive for the acquisition and digital representation of facial images for interoperable purposes, in particular for use in electronic passports and criminology , is the international standard ISO / IEC 19794-5. Its detailed specifications with regard to image content and recording technology aim to achieve high recognition quality.

2D process

Simple face recognition processes use a two-dimensional (2D) geometric measurement of special features (e.g. eyes, nose, mouth). Their position, distance and position to one another is determined here. However, today's methods mostly rely on complex calculations such as wavelet analysis (e.g. using Gabor transformation ) or principal component analysis . The National Institute of Standards and Technology (NIST) has repeatedly conducted comparative studies of various commercial and university processes. The results show a significant increase in recognition performance within approx. 10 years. While the false rejection rate with a false acceptance rate of 0.1% set in 1993 was still 79%, which is unfit for practice (i.e. almost four out of five people were not recognized at the time), this error rate is now (as of mid-2006) from the most efficient methods to only 1 % reduced (ie about one in a hundred people is not recognized). This rate is in the same order of magnitude as current fingerprint or iris recognition methods and exceeds the capabilities of human face recognition.

In 2001, two computer scientists developed the Viola Jones method for facial recognition, named after them . The process is based on machine learning and also recognizes other types of structures, such as traffic signs for autonomous driving . Comparable to this is the histogram of oriented gradients (HOG) method , which is also based on training data.

3D process

In addition to two-dimensional biometric face recognition, which uses commercially available cameras for recording , a new branch has been developed that relies on three-dimensional (3D) recording (e.g. by means of strip projection ) of the face. The additional information is intended to achieve greater recognition accuracy, better independence from poses and security against overcoming. NIST test results show that as of mid-2006 the 2D methods were still superior to the 3D methods in terms of recognition performance.

application

In Germany, during the Rhine Culture Festival in Bonn in autumn 2011, a project was started by the public broadcaster WDR , which enables the face recognition of festival visitors. This works via (high-resolution) photos of the festival visitors, which are “tagged” with the help of Facebook's face recognition . The faces are identified and linked to Facebook profiles.

The FindFace app produced by Russian software developers only requires a photo of a person, e.g. B. recorded on the street in order to then find this person on the Internet in social networks (as of May 2016). The system should z. B. can be used to help people who are in real shops such. B. watch a stereo system to be able to deliver targeted advertising on the Internet later. But private individuals can also take photos of strangers whom they want to data on the street so that they can later use the app to find their profiles on the Internet and contact the person. Moscow's city administration plans to use the software in the future to compare photos from surveillance cameras with mug shots. The security company Kaspersky has tested the system and attested the system a detection rate of 90 percent. The system is based on FaceN and uses machine learning techniques to recognize faces. Structures of the face are analyzed that do not change even when z. B. glasses or make-up are worn.

In October 2016, it was revealed that there were 117 million Americans in the FBI's facial recognition database.

On August 1, 2017, the Federal Ministry of the Interior and Deutsche Bahn started a facial recognition project at Berlin's Südkreuz train station. Voluntary test persons who passed the train station more often were to be recognized by means of three cameras. The test subjects had previously stored their names and two photos of their faces. 275 people volunteered for the project. The pilot project should initially run for six months. Data protection activists and civil rights activists reacted to the test run with some harsh criticism. Maja Smoltczyk, the data protection officer of the state of Berlin, told Rundfunk Berlin-Brandenburg rbb that this was "a very, very far-reaching encroachment on fundamental rights, especially the right to informational self-determination," that is, the constitutionally guaranteed right "to be unobserved and anonymous move the public ".

While this first test run was only supposed to deal with facial recognition, another was initially planned, which, according to the Federal Ministry of the Interior, can be used to automatically detect and report helpless people lying down or suspicious objects using software.

In August 2017, researchers at Stanford University presented an AI that reads the shape of the face, the facial expression and the way in which the person was dressed up from more than 35,000 photos from a dating platform and assigned the sexual preferences of the person. Then they had their program examine random photos of heterosexual and homosexual people and assign them to a sexual orientation. For men, the CI was correct in 81% of the cases, for women it was 71%. Human estimators, on the other hand, were around 20% worse. Activists from HRC and GLAAD questioned the methodology of the study, they had only sorted according to beauty standards. The researchers themselves determined that they could also be wrong, but pointed out the dangers for those affected if such a technology is misused.

Researchers from the University of Maryland and Dartmouth College have developed an AI that has a 92 percent chance of knowing if someone is lying by looking at someone's voice and facial expression. The AI ​​was trained on 104 videos showing people telling both the truth and the untruth in court. The AI ​​learned to recognize minimal changes in facial expression and voice in order to interpret the truth of what was being said. The system could be greatly improved with better audio data and more video material.

At the beginning of 2018 it was announced that in a pilot project at the Zhengzhou train station in China, the police are using sunglasses with facial recognition software. Within seconds, faces are compared with a criminal record and suspects are displayed on a mobile, tablet-like computer . Seven criminals have already been arrested this way.

In November 2018, the Chinese company Watrix presented an AI that can identify people from 50 meters away from the recordings of a surveillance camera with a hit rate of 94 percent based on their gait alone. Thus, it is no longer necessary to see a person's face in order to identify the person. The technology is already in use in Shanghai and Beijing.

The Swedish company "Visage Technologies AB" offers software for face recognition as SDK .

In her article "We are hurtling towards a surveillance state ': the rise of facial recognition technology" , journalist Hannah Devlin describes how a London restaurant owner uses Facewatch , a company for quick face recognition for business, hotel or casino customers against thieves. Thanks to an installed camera system, faces are compared with suspects in the shortest possible time. In a cloud of company these records are stored. The business owner is informed immediately via an app on his smartphone when a suspect enters his business premises. If real-time facial recognition is combined with the largest Aadhaar biometric database in existence , it could create the perfect Orwellian state , says Hannah Devlin in her article. On the company website, Facewatch advertises that for the Summer Olympics in Tokyo 2020, the faces of athletes, sponsors, volunteers and other accredited people will be scanned on hundreds of Intel and NEC terminals .

In January 2020, the New York Times reported that Clearview AI was being used by the FBI, Homeland Security, and numerous small, local law enforcement agencies.

Authentication

A recognized face can be used as a biometric factor for authentication . In China , such AI- assisted facial recognition is being used more and more. Many of these applications, which are particularly at home in the banking and finance sector , are based on the Face ++ software from Megvii . This is a biometric application based on the deep learning framework Brain ++. The algorithms behind it are trained with the help of large data sets, i.e. a large number of images. This machine learning process uses a large, multi-layered neural network that adjusts its parameters in the training phase until a person's face is reliably recognized. Apple markets face recognition for user identification under the name " Face ID ".

criticism

Critics of the technology point out the strong intrusions into privacy and warn against abuse for mass surveillance. The US civil rights organization American Civil Liberties Union cites the People's Republic of China as a chilling example of a surveillance state , whose authorities monitor members of the Muslim Uyghur minority nationwide with around 200 million surveillance cameras and face recognition .

In May 2019 , San Francisco became the first city in the world to prohibit its authorities and the city police from using facial recognition technology. The city council justifies this with the protection of civil rights. San Francisco as the “tech headquarters” would have to take responsibility here and regulate new technologies precisely.

The anti-racism demonstrations in America in the spring of 2020 have intensified the discussion about facial recognition programs, writes the FAZ . Therefore, IBM is withdrawing from the business. The newspaper quotes from a letter from CEO Arvind Krishna to several members of the American Congress that IBM no longer offers such software and is generally against this technology if its use leads to mass surveillance , discrimination and the violation of human rights . He believes that now is the time to start a national dialogue on whether and how facial recognition technology should be used by law enforcement agencies , Krishna said . The anti-racism demonstrations after the violent death of the African American George Floyd had fueled the discussion about facial recognition. The FAZ quotes from a blog entry by the organization " Algorithmic Justice League " after the police used facial recognition technology during the protests, and warned that this would "once again cause disproportionate harm to black people".

Even Amazon has the US police prohibited the use of its facial recognition software. The temporary ban is valid for one year and should give the Congress the time to adopt "appropriate rules" for the use of such technologies, reported ZEIT on June 11, 2020.

See also

literature

  • Claus-Christian Carbon: Face processing. Early facial recognition processes . Dissertation, FU Berlin 2003 ( full text )
  • Evgenij W. Dikich: Method for automatic face recognition . Logos, Berlin 2003, ISBN 3-8325-0428-1 (also dissertation, University of Karlsruhe 2003)
  • Claudia Freitag: Face processing in preschool age. Recognition of new faces depending on the expression of emotions and neurophysiological correlates of learning new faces . Dissertation, University of Giessen 2007 ( full text )
  • Roland Meyer: Operative Portraits. A picture story of identifiability from Lavater to Facebook . Konstanz University Press 2019
  • Doris Y. Tsao: How the brain reads faces. (Cover story) In: Scientific American , February 2019, pp. 18-25

Web links

Wiktionary: Face recognition  - explanations of meanings, word origins, synonyms, translations

Individual evidence

  1. ↑ Even unborn babies react to faces orf.at, June 9, 2017, accessed June 9, 2017.
  2. ↑ Results of the Face Recognition Vendor Test 2006, page 5 (Engl.)
  3. ↑ Technology Review: Computers outperform humans at recognizing faces (engl.)
  4. ↑ FRVT2006, page 15 (Engl.) ( Memento of 6 July 2007 at the Internet Archive ).
  5. ↑ http://rheinkulturpanorama.de/
  6. ↑ Shaun Walker: Face recognition app taking Russia by storm may bring end to public anonymity . In: The Guardian . May 17, 2016 ( theguardian.com ).
  7. ^ Lily Hay Newman: Cops have a database of 117M faces. You're probably in it . In: Wired . ( wired.com ).
  8. ↑ https://www.morgenpost.de/bezirke/tempelhof-schoeneberg/article211395129/Im-Bahnhof-Suedkreuz-startet-Test-zur-Gesichtsgewinn.html/
  9. ↑ https://www.rbb-online.de/politik/beitrag/2017/08/berlin-bahnhof-suedkreuz-gesichtsicherung-buergerrechtler-ueben-scharfe-kritik.html/
  10. ↑ http://bmi.bund.de/SharedDocs/Pressemitteilungen/DE/2017/08/gesichtsgewinnstechnik-bahnhof-suedkreuz.html/
  11. ↑ Derek Hawkins: "Researchers use facial recognition tools to predict sexual orientation. LGBT groups aren't happy." from September 12, 2017
  12. ↑ In court: Artificial intelligence is supposed to expose liars , futurezone.de of December 25, 2017
  13. ↑ "Face recognition through sunglasses: China's police on a futuristic hunt for criminals" In: Stern.de from February 11, 2018, accessed on March 1.
  14. ↑ " La policía china usa gafas con reconocimiento facial para identificar a sospechosos " In: elpais.com from February 8, 2018 on March 1.
  15. ↑ China's surveillance AI now recognizes people on the wired.de corridor on November 8, 2018
  16. ↑ Face Recognition. Visage Technologies AB, accessed January 4, 2019 .
  17. ↑ [1] The Guardian, October 5, 2019, accessed October 5, 2019
  18. ↑ Barbara Wimmer: Secret app scanned faces for police authorities. In: futurezone.at . January 20, 2020, accessed January 20, 2020 .
  19. ^ Will Knight: In China, you can pay for goods just by showing your face . In: MIT Technology Review . ( technologyreview.com [accessed August 25, 2017]).
  20. ↑ Yiting Sun: China: The face as a universal key . (No longer available online.) In: Technology Review. Heise, August 25, 2017, archived from the original ; Retrieved August 25, 2017 .
  21. ↑ Eike Kühl: Trust us (your data)! How manipulable is Face ID? In: zeit.de . September 27, 2017. Retrieved November 9, 2018 .
  22. ↑ a b "Civil Rights Endangered": San Francisco bans facial recognition. In: orf.at . May 15, 2019, accessed May 15, 2019 .
  23. ↑ Roland Lindner: IBM gives up facial recognition business. FAZ.NET from June 10, 2020 (Economy Department)
  24. ↑ Alex Hern: IBM quits facial-recognition market over police racial-profiling concerns. The Guardian, June 9, 2020 (accessed June 10, 2020)
  25. ↑ ZEIT ONLINE: Amazon suspends police cooperation in facial recognition. June 11, 2020