Deepfake Spoofing & Liveness Detection | BioID Biometrics Presentation @CODE2021 | Deepfakes Attack
4.5 هزار بار بازدید -
3 سال پیش
-
UPCOMING WEBINAR ON DEEPFAKES, Feb.
UPCOMING WEBINAR ON DEEPFAKES, Feb. 22nd, 2022: https://eab.org/events/program/283#ea...
#Deepfakes are becoming a growing challenge to identity verification systems. #Deepfake faces and videos modified by artificial intelligence become more and more realistic and neither the human eye nor biometric systems can detect these with ease. Liveness detection is an effective means to counter deepfake presentation attacks. https://www.bioid.com/liveness-detect...
This presentation was held at CODE2021 - the annual conference by the University of the Bundeswehr Munich.
During CODE2021, Ann-Kathrin Freiberg, Manager Business Development & Marketing at BioID, was invited to speak at the workshop on "Secure Digital Identity Verification":
Taking new consumer behavior, globalization and COVID 19 into account, today there is a growing need for digital identity verification. While initially coping with this situation via video identification through agents, today even photo-based, fully automated processes are available. For such ‘selfie verifications’, as they are called, three major steps need to be carried out:
1) the authenticity of the captured ID document needs to be determined.
2) ID ownership by the submitting person needs to be validated.
3) Liveness detection needs to be carried out to determine the selfies were captured live from the subject.
For deriving reliable credentials through automated identity verification, ongoing development is required to cope with new challenges and adapt to technical advancements concerning the creation of attacks. The following will focus on step 3 in remote identity verification, which is liveness detection.
Face liveness detection is an anti-spoofing method for facial biometrics. Scientifically, it is called presentation attack detection (PAD). The core function of a PAD technology is to determine whether a biometric feature (e.g. a picture), was captured from a live person. As BioID's algorithm is software-based, this anti-spoofing is independent from special hardware and requires standard cameras only. This way it is ideal for end-user applications where the usage scenario and device is unknown, e.g. digital selfie verification. These unsupervised situations give way to fraud though, and thus need to incorporate sophisticated anti-spoofing, especially for critical processes like deriving trusted credentials about a person’s legal identity.
According to the international PAD standard ISO/IEC 30107, fraudulent biometric traits can for instance be derived from photo printouts, paper masks or video replays presented to a camera. Such attacks are carried out at the sensor level and referred to as replay or presentation attacks, as formerly captured facial traits are replayed to a sensor, e.g. a camera. Attacks at the application level, e.g. using a virtual camera driver to feed a video into the application directly, are not subject of the international PAD standard ISO/IEC 30107. Anyhow, they are becoming a growing challenge, especially due to the enhancing quality and availability of deepfake tools.
State-of-the art ISO/IEC 30107-3 compliant technologies prevent biometric fraud through printed photos, cutouts, prints on cloth, 3D paper masks, videos on displays, video projections and more. Deep fakes presented at the sensor level (e.g. on displays) can be rejected through the same methods, e.g. texture analysis and artificial intelligence. For counter fighting attacks at the application level, namely through the injection of modified camera streams, challenge-response mechanisms can be utilized to reject prerecorded videos/deep fakes.
For both, the agent-based as well as the fully remote identity verification processes, deepfakes (especially injected via virtual cameras) become a growing challenge. Deepfake faces and videos modified by artificial intelligence become more and more realistic and neither the human eye nor biometric systems can detect these with ease.
While ensuring secure applications which prevent virtual cameras and modified video streams as input, further research needs to be pursued for detecting deep fakes in the photo/video material directly.
For more information about BioID you can also subscribe to
LinkedIn: LinkedIn: bioid-gmbh
Facebook: Facebook: bioidgroup
Twitter: Twitter: BioID_GmbH
And our Newsletter: https://landing.mailerlite.com/webfor...
ABOUT BIOID
https://www.bioid.com
BioID is a German biometrics company. The team originated from the German research institute Fraunhofer IIS in 1998. The proprietary software for face liveness detection and facial recognition is used by companies worldwide for biometric authentication. The BioID Web Service also includes PhotoVerify, a software for verifying document ownership biometrically for remote identity proofing.
#biometrics
#Deepfakes are becoming a growing challenge to identity verification systems. #Deepfake faces and videos modified by artificial intelligence become more and more realistic and neither the human eye nor biometric systems can detect these with ease. Liveness detection is an effective means to counter deepfake presentation attacks. https://www.bioid.com/liveness-detect...
This presentation was held at CODE2021 - the annual conference by the University of the Bundeswehr Munich.
During CODE2021, Ann-Kathrin Freiberg, Manager Business Development & Marketing at BioID, was invited to speak at the workshop on "Secure Digital Identity Verification":
Taking new consumer behavior, globalization and COVID 19 into account, today there is a growing need for digital identity verification. While initially coping with this situation via video identification through agents, today even photo-based, fully automated processes are available. For such ‘selfie verifications’, as they are called, three major steps need to be carried out:
1) the authenticity of the captured ID document needs to be determined.
2) ID ownership by the submitting person needs to be validated.
3) Liveness detection needs to be carried out to determine the selfies were captured live from the subject.
For deriving reliable credentials through automated identity verification, ongoing development is required to cope with new challenges and adapt to technical advancements concerning the creation of attacks. The following will focus on step 3 in remote identity verification, which is liveness detection.
Face liveness detection is an anti-spoofing method for facial biometrics. Scientifically, it is called presentation attack detection (PAD). The core function of a PAD technology is to determine whether a biometric feature (e.g. a picture), was captured from a live person. As BioID's algorithm is software-based, this anti-spoofing is independent from special hardware and requires standard cameras only. This way it is ideal for end-user applications where the usage scenario and device is unknown, e.g. digital selfie verification. These unsupervised situations give way to fraud though, and thus need to incorporate sophisticated anti-spoofing, especially for critical processes like deriving trusted credentials about a person’s legal identity.
According to the international PAD standard ISO/IEC 30107, fraudulent biometric traits can for instance be derived from photo printouts, paper masks or video replays presented to a camera. Such attacks are carried out at the sensor level and referred to as replay or presentation attacks, as formerly captured facial traits are replayed to a sensor, e.g. a camera. Attacks at the application level, e.g. using a virtual camera driver to feed a video into the application directly, are not subject of the international PAD standard ISO/IEC 30107. Anyhow, they are becoming a growing challenge, especially due to the enhancing quality and availability of deepfake tools.
State-of-the art ISO/IEC 30107-3 compliant technologies prevent biometric fraud through printed photos, cutouts, prints on cloth, 3D paper masks, videos on displays, video projections and more. Deep fakes presented at the sensor level (e.g. on displays) can be rejected through the same methods, e.g. texture analysis and artificial intelligence. For counter fighting attacks at the application level, namely through the injection of modified camera streams, challenge-response mechanisms can be utilized to reject prerecorded videos/deep fakes.
For both, the agent-based as well as the fully remote identity verification processes, deepfakes (especially injected via virtual cameras) become a growing challenge. Deepfake faces and videos modified by artificial intelligence become more and more realistic and neither the human eye nor biometric systems can detect these with ease.
While ensuring secure applications which prevent virtual cameras and modified video streams as input, further research needs to be pursued for detecting deep fakes in the photo/video material directly.
For more information about BioID you can also subscribe to
LinkedIn: LinkedIn: bioid-gmbh
Facebook: Facebook: bioidgroup
Twitter: Twitter: BioID_GmbH
And our Newsletter: https://landing.mailerlite.com/webfor...
ABOUT BIOID
https://www.bioid.com
BioID is a German biometrics company. The team originated from the German research institute Fraunhofer IIS in 1998. The proprietary software for face liveness detection and facial recognition is used by companies worldwide for biometric authentication. The BioID Web Service also includes PhotoVerify, a software for verifying document ownership biometrically for remote identity proofing.
#biometrics
3 سال پیش
در تاریخ 1400/06/10 منتشر شده
است.
4,576
بـار بازدید شده