Problems Face ID associated with a lack of training data

Problem with Face ID started during the presentation, when the iPhone is not unlocking the first time, and had to enter a password. When the device went on sale, the technology began to test the journalists, but the main problem faced by the twins and close relatives — they were able to unlock smartphones to each other because of the similarity. Apple this problem, not deny it.

Also the repercussions of the story on unlocking using a 3D mask of the face, which costs $150.

Why relatives can unlock the iPhone

In the industry of computer vision and machine learning are all solved by the quality of data on which to train classifiers. In this case facial recognition. If a particular category of data, for example, when children were able to unlock the phone parents and Vice versa, there are problems, they can be taken into account in the training set and boobookitty model.

Problems with the Apple twins really recognizes and yet does not recommend to use the technology without a password. But this problem is familiar to many solution vendors for face detection. In practice, in the presence of a large number of examples of twin pairs in the training set, this problem can be minimized.

Adult twins in most cases have visible differences: in real life, they suffer from various diseases and are not 100% identical environment. Children under 13 years of Apple also advises not to use the iPhone X, explaining that the facial features are not fully developed.

Ten-year-old boy unlock my mom’s iPhone

To understand the scale of the disaster with the lock bypass, you need to test the technology on a large number of people. Now every single registered a case gets big media coverage, but so far neither of the problems nor their true scale.

If the result of check on millions of users about the percentage recognition is higher than the acceptable values defined by the manufacturer (False Positive Rate no more than ten to the minus sixth power), then it will turn into a real problem.

Then the Face ID technology will improve by review of a representative sample of problem cases when the smartphone is unlocked is not the owner, or when the mobile can not unlock the owner.

When we did the first version VisionLabs LUNA, we also had a similar problem with relatives. The degree of similarity among family members is very high. For example, father and son, mother and daughter often really very similar to each other, but just different age.

Apple is now the same “childhood diseases”, because the product is new and is undergoing massive adaptation, but in the future they this point I will consider and train the neural network to better recognize relatives, twins and other difficult categories.

Can I order the mask to bypass iPhone lock

9 Nov Vietnamese company Bkav, which deals with cybersecurity, wrote in a blog that Face ID, you can cheat by using a mask. Apple during the presentation said that the company has specially cooperated with the producers of masks to teach technology to distinguish a real face from the inanimate.

But if you print in color a 3D copy of a person that really easy to do and to choose a matte finish which will not reflect the radiation of the infrared sensor, the smartphone can be fooled.

Vietnamese experts have created a model, using a special scanner and printed the mask on a 3D printer. On top of it put a realistic silicone nose and the lips and eyes glued printed pictures of these parts of the body of the device owner. The mask cost $150.

Company Bkav lied Face ID with a mask

Obviously, the opportunity bypass Apple not considered. Apple taught the technology on real people and their masks are made by Hollywood artists. But she had not considered the option of creating a “semi-human” mask, as did the Vietnamese investigated. Apple may consider it in the next iOS updates.

Terms of iron, that is built-in smartphone camera and sensors, recognition can have even greater margin of reliability. The company need not release a new iPhone or change the built-in camera technology to work better in these risk categories is sufficient to update the software.

Some wondered why Face ID to unlock the device if the mask was white? The technology recognizes faces using information about the shape and texture of faces, and on the basis of these data complexioned decisions.

Camera True Depth in the visible RGB and in infrared. That is taken into account a set of parameters of 3D shape, texture and complexion. Therefore, in some cases the decision on when triggered at least part of the conditions. For example, it helps to quickly unlock your device in the dark.

What technology Apple uses in Face ID

In 2013, Apple bought Israeli company PrimeSense, which develops hardware for 3D zoning. The technology PrimeSense certainly have a hardware-based True Depth and used in the development of Face ID and other functions of the smartphone.

The company is also known for its introduction of its sensors ASUS xTion and Microsoft Kinect, which uses an RGB-camera and depth sensor to capture the movements of players and incorporate them into the gameplay Xbox.

PrimeSense has developed a chamber, which includes a conventional color camera and IR sensor depth with a structured IR illumination. The camera is designed so that it can work in total darkness when flowers cannot be seen. So Apple implements the recognition in the two spectra: visible given the intensity values of the pixels, and in 3D the image — mask obtained from these IR points.

Apple talk about the new technology at the presentation

Technology gets in the dark depth map, the volume of the face and its details. It is not surprising that not even a color mask can be used for cheating. If match is true 3D copy of the face — mask and standard loaded in the device, there is no reason not to unlock.

There is the opinionthat to set Face ID need in a dark room, but it is not so. If you configure the device in full darkness, you can potentially be not taken into account the number of parameters of the face in the visible spectrum. During testing the device, we realized that the technology works in most cases quickly and efficiently. We have not seen such counterparts, at least among smartphones. When Apple will collect enough statistics, we will fix this in a future firmware update.

Apple uses a complex combination of brightness and wavelength of the IR illumination, which minimizes the negative influence of bright sunlight on the build quality 3D mesh of the face. Used to have a Kinect-like device problem, if you go outside with bright sunlight: the obtained depth map was damaged.

But the IPhone works well even in bright light — we have specifically tested it in various stressful conditions. So this old Apple problem, if not completely solved, reduced to a minimum.

If you look at the videos about the API for Face ID, we see that the model of a person, which you can pull from the device, quite accurate in the dark and in a bright color. According to this, the camera is close to perfection. This means that all the problems in the software, but it can be taught.

Why biometrics is a reliable method

Although it is possible to bypass the lock, facial recognition and other biometric systems are considered the most secure methods of data protection. Because the pin or password you can spy and copy the finger to unlock the Touch ID is more difficult. In turn hack, the facial recognition system is much harder than Touch ID.

Manufacturers of user devices and enterprise solutions are moving EN masse for biometrics at least due to rising security requirements and banal human laziness. But laziness — the engine of progress. Nobody wants to remember passwords and in many companies, security policy passwords are to be changed at least once every three months. So people want the easiest way of authorization, as, for example, to sign using your finger or face.

While the technology of face recognition has its own advantages over the fingerprint scanner. For example, to unlock the iPhone with your face in any weather, and due to wet fingers or a small deformation of the skin may have problems with unlocking, you will have to enter the password. Cold hands in the cold too, do not give unlock a smartphone, get access to their smartphone you can even use part of the face.

We VisionLabs together with its partners conducted pilot projects with different types of biometrics among people who work in the mines, building houses and streets. Many of them due to mechanical work erased papillary patterns on the fingertips. So fingers it is impossible to recognize. Hardly in the world, there is as much hard work to wear off the face.

All the problems Apple is expected to be resolved by the clarification of the training sample and using machine learning. Therefore, the more people will take advantage of the technology, the faster she will learn to distinguish between twins and parents with children.

The use of facial recognition ensures data security. Moreover, four years after buying PrimeSense, Apple has improved the cameras that recognize your face. Therefore, from the iron significant problems, definitely not.

Leave a Reply

Solve : *
22 − 17 =