Deepfake and digital attacks
Counterfeit visual data that deceives camera-based verification systems.
Facing fraudsters
The person behind the camera could be a fake
They could be manipulating your systems to appear as someone else, or they might not be a live person at all.
Physical presentation attacks
Visual data manipulated with high-quality screens or printed images to trick camera-based verification systems.
Evasion attacks
Individuals alter their appearance to make it harder for facial recognition systems to verify their biometric features.
0X
Surge in deepfake fraud attempts over the last 3 years
0%
Of businesses claim identity fraud losses have increased in recent years*
Common attacks
Thirteen attacks, three families
Deepfake and digital attacks
- Face swapping Replacing the face in a target image with a face from another source.
- Face morph Digitally blending facial features from two people to create a new synthetic face.
- 2D Synthetic assets AI-generated images of people who don’t exist – all facial features are entirely synthetically generated, unlike face morphs that blend real people’s facial features to generate a new face.
- Face reenactment Digitally replicating real facial movements and expressions to make manipulated videos appear authentic.
- Video replays Footage is replayed on high-resolution screens in front of the camera to simulate a live individual, making it difficult to distinguish between real and live people.
Physical presentation attacks
- 2D masks Flat, printed masks are used to mimic someone's face and deceive verification systems, but lack the depth and detail of a real human face.
- 3D masks Lifelike, three-dimensional masks made of materials like silicone are crafted to closely resemble a real person, making them harder to detect.
- Paper printouts Printed photos of a face are presented to the camera, attempting to pass off a static image as a live individual.
- Cardboards Printed images mounted on cardboard provide a sturdier, more rigid appearance in an effort to fool systems.
- Video replays Footage is replayed on high-resolution screens in front of the camera to simulate a live individual, making it difficult to distinguish between real and live people.
Evasion attacks
- Exaggerated expressions Overly dramatic facial movements, like extreme smiles or wide eyes, are used to distort the face and confuse recognition systems
- Occluding objects Items like hats, glasses, or scarves are used to block parts of the face, preventing systems from capturing a clear image.
- Heavy makeup Complex makeup techniques are applied to change the appearance of key facial features, making it harder for systems to recognize the individual.
How it works
Incode’s advanced liveness technology
Within Incode's liveness, we employ multi-modal intelligence to maximize accuracy of liveness detection without slowing down the end-user experience.
Technical benefits
Why choose Incode’s liveness detection?
Discover how Incode Deepsight’s advanced liveness technology tackles fraud without impacting your user experience.
Full ownership of our ML models and tech stack
Unlike other providers who rely on third-party vendors, we build and own our entire technology stack. Our AI/ML models, powered by deep learning and designed for identity verification, allow us to train on the latest document and biometric fraud vectors. This full control enables us to tailor our models to the unique needs of our clients, ensuring superior performance and flexibility.
Rich, well-organized data to train models
Using advanced neural networks, including both standard Convolutional Neural Networks (CNNs) and cutting-edge Large Vision Models (LVM) and transformers, we train our models to achieve state-of-the-art results across various tasks. Over nearly a decade, we’ve curated large, statistically representative datasets, so our models deliver balanced performance across variables like age, skin tone, and gender. Our in-house Fraud Lab has compiled over 1 million unique presentation attacks, from basic printouts to advanced 3D masks. We also generate synthetic data such as face swaps and synthetic faces, enhancing the robustness of our models.
Internal testing for flawless detection
Our internal testing environment is designed to be more challenging than real-world spoof attempts. By testing our models against complex attacks, we ensure perfect detection rates in production. This way, we can prevent known fraud networks, repeat verification attempts, and other fraudulent behaviors before they impact your business.
Leveraging multiple input modalities
Incode advanced liveness incorporates detection across various modalities, such as depth and motion, and multiple frames, multiplying its accuracy.
Fraud Lab: real-time adaptation to emerging threats
Our liveness detection AI models are continuously refined by Incode’s Fraud Lab, where our team trains our proprietary AI technology in response to the latest emerging fraud techniques. Unlike competitors who rely on third party providers, we can immediately train our in-house models to analyze input for emerging fraud types, ensuring these attacks are neutralized before they impact your business.
Verified proof
Global banks, fintechs, and marketplaces trust Incode.
0%
false positives and false negatives in independent testing by the Georgia Department of Driver’s Services (iBeta Level 2 protocols, 2024).
0
false positives and false negatives · Georgia DDS, iBeta Level 2 protocols, 2024
0%
false positive and false negative rate · Michigan State MSU-MFSD evaluation, 2023
1M+
unique presentation attacks compiled by Incode's in-house Fraud Lab
What types of deepfake attacks does Incode's liveness detection prevent?
Incode's liveness detection prevents face swapping, face morphing, 2D synthetic assets, face reenactment, and video replays through advanced digital spoof detection algorithms.
What physical presentation attacks can Incode detect?
Incode detects 2D masks, 3D masks, paper printouts, cardboards, and video replays through ML models designed to identify physical spoofing attempts.
How does Incode handle evasion attacks?
Incode's algorithms check for evasion attempts including exaggerated expressions, occluding objects like hats and glasses, and heavy makeup that aims to confuse recognition systems.
What is Incode's false positive rate for liveness detection?
Incode achieved 0% false positives (false acceptance of fraudulent users) and 0% false negatives (false rejections of genuine users) in independent testing by the Georgia Department of Driver's Services according to iBeta level 2 protocols in 2024.
What's next