Feature


Facial recognition: the security option of choice Security concerns have been raised over the recently announced health ’passports’


which depend on barcodes and QR codes for identification purposes. Facial recogni- tion has been put forward as a alternative – UC News reports.


A facial recognition system is a tech- nology capable of identifying or verify- ing a person from a digital image or a video frame from a video source. There are multiple methods in which facial recognition systems work, but in gen- eral, they work by comparing selected facial features from given image with faces within a database.


Facial recognition software has countless applications in consumer markets, as well as the security and surveillance industries. In fact, facial recognition technology is already be- ing used to improve security protocols and payment procedures in China, and it is possible that the rest of the world will follow suit.


There are two main tasks that fa- cial recognition models perform. The first is verification, which is the task of comparing a new input face to a known identity. A good example of this is the unlocking of smartphones with


to a database of multiple face identi- ties. This task is often used for secu- rity and surveillance systems. A good example is facial recognition in law enforcement. On the INTERPOL web- site, there is a forensics section which explains how they use facial recogni- tion to identify persons of interest at airports and border crossings.


How it works


Facial recognition works by feeding an input image to an algorithm. The algorithm creates a facial embedding for the input image. The algorithm compares the input image’s facial em- bedding to the embeddings of known faces in the database.


Every solution has a different method of loading and coding the comparison database often adjusting or adding el- ements to the raw data. Most systems employ triplet loss for building the al-


embeddings of person A are more sim- ilar to each other than they are to the embedding of person B. Subsequent- ly, this teaches the algorithm to use facial measurements that allow it to accurately classify images of the same person as being similar to each oth- er. This process is then repeated hun- dreds of thousands or even millions of times. Finally, the network should then be able to produce accurate fa- cial embeddings for faces it has never seen before.


Before the algorithm can compare faces, the solution converts the face images into data that the algorithm can understand. The system calcu- lates measurements based on facial features and landmarks. These facial landmarks are also known as facial keypoints. Depending on the neural network, these landmarks may or may not be used in creating the embed- ding. Sometimes the landmarks are just used to crop the face image and remove noise in the background of the image.


Deterministic Face Embedding


China already has widespread mobile payment as a primary method of making pur- chases. In some places, cash isn’t accepted — only mobile payment.


facial identification. When setting up the system, the phone will register a face as the phone’s owner. Therefore, the only task when unlocking is to compare new input faces to your regis- tered face on the device.


The second is recognition, which is the task of comparing an input face


gorithm. Triplet loss works by feeding the algorithm three images. Two of these images are of person A and the other image is of person B. The algorithm creates a facial embed- ding of each image and then compares them. After the comparison, the net- work will be adjusted slightly so that


An alternative methodology is called Deterministic Face Embedding. This works by creating face embeddings that render a face image into numeri- cal data. That data is then represented as a vector in a latent semantic space. The closer the embeddings are to each other in the latent space, the more likely they are of the same person. It’s worth noting that the accuracy of solutions which use deterministic face embeddings depends on the clarity of the input images. These models are often tested under constrained set- tings. In practice, input images (from surveillance video for example) are of- ten taken in unconstrained or uncon- trolled settings. The image quality may be low or partially obscured image. In such cases, approaches that use de- terministic face embeddings might suffer in performance.


Probabilistic Face Embedding is a relatively new technique which rep-


17


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30