In our first post explaining the buzzwords associated with facial recognition, we explored ePassports automated gates and facial recognition rates amongst other terms. Now it’s time for the remaining five industry terms common when talking about facial recognition.
#6 True Positive, True Negative,False Positive and False Negative Rates
It sounds messy and complicated, but fear not.
Accuracy in verification and authentication is defined in terms of true positive and true negative, false positive and false negative rates.
- True positive: ability to recognize a legitimate user
- True negative: ability to recognize an illegitimate user
- False positive: Recognizing an illegitimate user as legitimate
- False negative: Recognize a legitimate user as illegitimate.
A facial identification system will capture a face and determine the person’s identity by examining a biometric pattern calculated from the person’s biometric features and compared to a set of templates.
The system will assign the pattern to the person with the most similar biometric template. To prevent patterns of persons not known by the system from being correctly identified, the similarity has to exceed a certain level. If this level is not reached, the pattern is rejected.
#7 False Acceptance and False Rejection Rates
These levels or thresholds are rates and they are called False Acceptance Rate (FAR) or False Rejection Rate (FRR). They apply for verification as well as for identification.
- FAR is the instance of a biometric security system incorrectly accepting or identifying an unauthorized user.
- FRR is the instance of a security system failing to verify or identify an authorized person.
FAR and FRR always work together. If you choose a threshold that is too low, a very small amount of total facial patterns will be rejected. Of course, this also means that impostor patterns could be falsely accepted.
This means that the value of the thresholds is very important, as these will determine the proportion of the client patterns that will be falsely rejected.
The bottom line is that the thresholds of facial recognition systems have to be adjusted depending on the application.
In a system used for passenger clearance, a false rejection may cause bottlenecks when it stops travelers from going through a gate or an area to another when they SHOULD be recognized and authorized.
On the other hand, a false acceptance could be terrible if it permitted access to an unauthorized or fraudulent person. This explains why facial recognition analysts and vendors are very demanding in terms of FAR when it comes to accuracy:
- A 1:100 FAR would be considered as low security, as you’d have 1 person out of 100 able to board a plane or go through the immigration process.
Unimaginable!
Now, this rate could also be considered as good or high depending on the service and context. If applied to unlocking a mobile device for example, the fraudulent user would have to get hands on the phone first, then attempt to unlock it.
- For sensitive uses such as boarding or immigration control, Frontex, the European Border and Coast Guard Agency, recommends a 1:1000 FAR in their Best Practice Technical Guidelines for Automated Border Control (ABC) Systems. This is considered as the minimum required security level for ABC systems.
Frontex was created in October 2016, to help implement integrated border management at the EU level, oversee an effective functioning of border control at the external borders and provide increased operational and technical assistance to EU member states.
But what factors affect the accuracy of a facial recognition system?
#8 Reference Image
Context is important here. In the case of typical border control or access control, the system will compare a live capture of one traveler’s face with a face registered somewhere. We are talking about a one to one comparison in most cases when compared to a travel document, or it can also be a one to many comparison when comparing against a biometric data base.
So what’s the difference with CCTV video surveillance ?
For video surveillance purposes, the system will have to search though a crowd, and then compare one face to many that are stored in the database.
There are several factors that may impact accuracy:
- The quality of the image stored that serves as a reference for comparison, as well as the quality of the live image captured. So the position of the camera used for capture is important, as it will impact image resolution, focus, angle etc. Nevertheless, it is not the only element to be taken into account to ensure best accuracy.
Lighting has a big impact on image quality as well, not to mention elements that cannot be controlled easily such as a person’s movements and accessories such as glasses, caps etc. The higher the quality of the images, the lower potential error rate you will have.
- The number of reference images from different angles: if you want to do recognition from different angles rather than frontal recognition only, you will need to train the system with these type of images, i.e. by having at least one image for each pose per person.
- The total size of the database to compare with, or in other words the number of records to which the live captured image will be compared to. The bigger the base, the more matching possibilities and potential false positive responses there will be, and the more time it will take to find matches. So for a fast process, you will need to seek for ways to reduce the database you are matching against.
#9 Artificial Intelligence
AI has been around for a long time now, as it was founded as an academic discipline in 1956. Artificial Intelligence has already become an essential part of the technology industry, helping to solve complex problems in knowledge, perception, planning, learning and moving objects for example. Simply put, AI is about allowing machines to act like humans, in a much faster and precise manner, while continuing to learn. To be able to learn automatically, a machine will identify patterns in streams of inputs, by identifying and categorizing items, and then taking action. AI uses algorithms to do this.
Today, AI is used behind the scenes in many applications and allows efficiency improvements and cost reduction beyond mere automation. Recently, AI has made tremendous steps forward using deep learning, and deep learning is extremely well adapted to image recognition and facial matching.
#10 Deep Learning
Deep learning is a subset of machine learning, but radically innovative in the sense it relies on deep neural networks to learn. Computation power combined with more layers of neurons enables learning of very complex mathematical problems, with millions of parameters.
As a result, one can achieve a result that will be much more accurate, much faster than any human could. The capability to handle gigantic sets of data allows the system to improve and become more and more accurate.
Said simply, the network of millions of artificial neurons will automatically “learn” from huge data sets received in input, and map them into mathematical functions to deliver an intelligent output, instantly.
Applied to facial recognition, AI and deep learning algorithms allow analysis in seconds and instant verification of the authenticity of an identity and perform biometric facial comparison to ensure:
- that an individual submitting documentation is indeed the person pictured on the ID
- that a person is indeed recorded into a database and can be cleared for entrance or exit
AI is also able to build in mechanisms to reduce or filter through a set of records, to gain in accuracy and speed of response.
Get ready for tomorrow, when your face will be your token to travel.
Facial Recognition will change border control and air travel as we know it
In busy places such as airports or border control points, speed is a must to avoid travelers having to queue for hours. Of course, this speed should certainly not compromise security, and this is exactly why facial recognition technology, combined with security by design to protect the data and AI to ensure best accuracy will play an essential role in biometric recognition systems.
There is still quite some work to be done from a legislative perspective, around the use and storage of biometric data. Nevertheless, the technology is there, and the travelers are hungry for stress less and seamless journeys.
So be ready to walk through control points just using your face to prove you are who you say you are!
So that’s it, our guide to the buzzwords surrounding facial recognition technology. We hope you found it useful, and if you have any questions or comments on any of the terms explained above, get in touch in the comments below or @Gemalto.
The post Facial Recognition 10 Buzzwords Demystified (Part 2) appeared first on Cybersecurity Insiders.
August 20, 2018 at 09:09PM
0 comments:
Post a Comment