Finishing of facial imaging technology

The three-dimensional face recognition is through 3D camera stereoscopic imaging, can identify the space in the field of view of each point of the coordinate information, so that the computer to get 3D of space data and can restore the complete three-dimensional world, and realize a variety of intelligent three-dimensional positioning. Simply said that the machine gets more information, the accuracy of analysis and judgment has been greatly improved, the face recognition function can distinguish between flat image/video/Makeup/skin mask/twins and other states, suitable for the financial sector and smart phones, such as the security level of high-demand application scenarios.

1, three kinds of mainstream 3D imaging technology

(1) Structured light: The structured light structured the specific light information onto the surface of the object and is captured by the camera. According to the change of the light signal caused by the object, the information such as the position and depth of the object is computed, and the whole three dimensional space is restored.

(2) TOF (time of Flight, flight times): Through the use of a proprietary sensor, capturing near infrared light from the launch to the receiving time of flight, judging the distance of the object.

(3) Binocular ranging (Stereo System): The use of two cameras to shoot objects, and then the triangle principle to calculate the distance of objects.

Which of the three technologies is better suited for smartphone applications?

As the above figure shows, only binocular vision in three technologies is not suitable for dim environments, which means that our smartphones are not able to unlock the face recognition at night and are excluded first.

The following is a look at TOF technology and structured light technology.

The TOF technology has the advantages of faster response time, better anti-illumination performance, high accuracy of depth information and far-distance recognition, but it also has the disadvantage of low resolution, high cost, high power consumption and too large module.

And the structure of light technology advantage is low light performance, higher resolution, cost, moderate power consumption, the main disadvantage is susceptible to sunlight, identification distance is short, the corresponding time slightly slow shortcomings.

However, for the face recognition function on smartphones, structured light technology should be more advantageous than tof technology. Because face recognition through the front 3D system of the smartphone is very close to the recognition of this scenario itself, there is no question of the need to support further recognition distances. In addition, the structural light compared to the TOF technology, short-distance precision is higher, but also more suitable for use in the front of the mobile phone camera. And its resolution, the corresponding time is enough to deal with the needs of mobile phone face recognition (the TOF Project Tango phone is a rear 3D system, and its role is not primarily for facial recognition).

In addition, in view of the depth graph produced by the two technologies, the tof depth graph has many problems, such as noise generated by multiple reflections, fine edge-over, and lag in time domain filtering. However, the depth map of structural light is only slightly lower than the definition of boundary line. Finally, because it is used in consumer mobile devices such as smartphones, cost and power consumption are also factors to be considered.

Therefore, in general, if the mobile phone front 3D facial recognition system, structural light technology compared to the TOF technology more advantages.

At present, the international giants Apple, Microsoft, Facebook/oculus, Intel, Google and so on have already targeted 3D imaging face recognition technology, in recent years acquired a dozen companies in this field and the momentum is not reduced. Unfortunately, the above-mentioned companies are no exception to their own products to build the core technology threshold for internal ecological services, at least not currently committed to become a professional supplier of depth sensors and technical services.

Looking back, let's take a look at the 3D face recognition module used by iphone x:

From the above picture, we can see that in the "Liu Qi" part of the iphone x, there are many components, namely: Infrared camera, Flood illuminator (floodlight original), distance sensor, ambient light sensor, speaker, microphone, front camera, Projection spot (structural light) emitter.

It is understood that when the IPhone x faces recognition, it will call the structure light emitter, infrared camera, front camera and distance sensor. Among them, the structural light emitter casts more than 30,000 invisible light points to the naked eye, covering the entire face, and the infrared camera is responsible for collecting these light points for accurate and detailed facial images with depth information. Because of the invisible infrared light, facial recognition can be performed even in the dark. The front camera is responsible for acquiring 2D images and then co-synthesizing 3D stereoscopic images. The distance between the structure light emitter and the face is limited, so a distance sensor is required to remind the user to adjust the iphone x to the optimal distance for 3D sensing.

As for the flood illuminator seems to be Apple's own words, there are insiders said that this seems to be "multispectral scanning", if so, then you can measure and analyze the facial epidermis spots, pores, wrinkles and skin texture, so that the face recognition can play a high anti-false effect. There are insiders said that the industry said that this is the actual infrared light emitter, then whether it means that the iphone x structure light +tof Two kinds of 3D facial recognition technology. This has yet to be examined.

To enhance the recognition speed and accuracy of face ID, Apple also integrates an independent neural network engine in the A11 processor to memorize facial data and quickly identify faces. Apple said that the mobile phone verification is very fast, basically you can see instantly unlock, more than fingerprint recognition faster.

According to Apple, face ID has a failure rate of 1/1000000. But awkwardly, when Apple executive Craig came on stage to demonstrate the face recognition feature of the iphone x, it was the first time that it had encountered a failure to identify faces ID, failed again for the second time, and succeeded for the third time. People in the industry said that this is mainly because "the conference scene Light is dark, and the presenter side has a relatively strong large screen light, which indicates that Apple's face ID is not enough for the scene prediction".

This article is organized from the following network resources:

http://www.sohu.com/a/191481400_99950678

Http://baijiahao.baidu.com/s?id=1578500066293318023&wfr=spider&for=pc