Especially thanks to rapid technological advances in computer science and electronics. Measured from a market share perspective, facial recognition is currently the second most widely used biometric method in the world after fingerprints. Every day, more and more manufacturers are integrating facial recognition into their products in cyber security courses. For example, Apple uses Face ID technology and banks are implementing eKYC solutions for the onboarding process. Contrary to the main goals of facial recognition research to improve the performance of validation and identification tasks, the security gaps in facial recognition systems are much smaller in the past and have only been studied in the last few years. Some attention has been paid to the detection of different types of attacks, which consist of determining whether the biometrics are from a living person or are counterfeit.

Two types of attacks on facial recognition systems: presentations and indirect attacks.

Presentation Attack

Presentation attacks are performed at the sensor level without accessing the inside of the system. Presentation attacks threat is purely related to biometric vulnerabilities. In these attacks, the intruder uses certain artefacts. They usually use artificial things (face photos, masks, synthetic fingerprints, printed iris images, etc.) or try to imitate real user aspects (walking, signatures, etc.). Unauthorized access to the biometric system. Because "biometrics is not a secret," attackers are aware that a large amount of biometric data showing people's faces, eyes, voices, and behaviours is open to the public and use these sources to bypass faces. I will try the method. A recognition system using the following example. The attacker uses the user's photo to impersonate. They use the video if the user is disguised. Alternatively, hackers can create and use  3D models of the attacked face, such as surreal masks. We use anti-spoofing technology to prevent these attacks. Indirect attack Indirect attacks (2-7) can be performed against matching databases, communication channels, etc. This type of attack requires the attacker to access the inside of the system. Indirect attacks can be prevented by methods related to "traditional" cybersecurity rather than biometrics, so we won't cover them in this post. Attacking Methods Without implementation presentation attack detection measures, most of the state of heart facial biometric systems are vulnerable to simple attacks. Typically, face recognition systems can be spoofed by presenting to the camera a photograph, a video, or a 3D mask of a targeted person/ infect users or using makeup or plastic surgery. However, using photographs and videos are the most common type of attack due to the high exposition of the face and the low cost of high-resolution digital cameras.

Photo Attacks

a photo attack consists of displaying a photograph of the attacked identity to the sensor of the face recognition system.

Video Attack

An attacker could play a legitimate user's video on any device that plays the video and present it to the sensor/camera for cyber security certifications.

3D Mask Attack

In this type of attack, the attacker creates a 3D reconstruction of the face and presents it to the sensor/camera.

Other Attack

Makeup, surgery Spoofing prevention technology Most facial recognition systems can be easily attacked by spoofing methods. Therefore, in order to develop a secure facial recognition system in a real-world scenario, anti-spoofing technology should be a top priority from the initial planning of the system.

Anti-Spoofing Technology

Facial recognition systems try to distinguish between real users and cannot determine if the biometric sample presented to the sensor is genuine or fake. You can do that in four ways: sensor The available sensors are used to detect signal patterns that are characteristic of living organisms in cyber crime courses online. Dedicated hardware It has dedicated hardware for detecting evidence of life, such as a 3D camera, but it is not always available. Challenge/response method Use a challenge-response method that can detect presentation attacks by asking the user to interact with the system in a specific way. smile Expression of sadness and happiness Head movement algorithm use of the following detection algorithms that are inherently robust against attacks.

Glossy Feature Projection: First, the mirror feature space is characterized according to the actual image, based on which the actual data and spoofing data projection is learned. The SVM model is then trained according to the actual projection, and the 3D mask projection and the printed photo projection are used as an anti-spoofing model to identify the ID change.

Deep Feature Fusion: By carefully studying the meaning of facial image colour feature information for human face recognition, deep convolutional neural networks ResNet and SENet have constructed a deep feature fusion network structure and related face spoofing prevention data. Effectively train.

Image Quality Rating: This method is based on a combination of image quality measurements. This solution compares the original image with the processed image. Deep learning: this method is based on a multi-input architecture that combines a pretrained CNN model and the local binary patterns descriptor.

How to Implement?

We can build a presentation attack detection system (PAD) using anti spoofing techniques and integrate it with the facial recognition system. With this approach, the anti spoofing system makes its decision first, and only if the samples are determined to come from a living person, then they are processed by the face recognition system.