Recently, face recognition technology for repeatedly in the course of fugitives caught in "meritorious service", once again take the "red." Since the 1960s, open face recognition research, development to which progress today? The competition in the industry, is artificial intelligence investment bubble caused by impetuous, or size of the market will continue plunging?
What is face recognition technology?
In the 1960s, applied research engineering recognition official opening. The main advantage of the method of initial geometry of the face, carried out through the identification and analysis of the relationship between the topology of the face organ feature points. This method is simple and intuitive, but once face pose, facial expression changes, the accuracy of serious decline.
Today, solutions are mostly based on face recognition technology multi-source initiative to near-infrared images. It can overcome the effect of changes in the light, have achieved excellent recognition performance, overall system performance in terms of accuracy, stability and speed over a three-dimensional face recognition.
Draw focus!
1. Face recognition technology works:
One is to establish a high-volume database containing facial image; the second is to get the current target face image to be identified through a variety of ways; the third is the target face image database in both face images to compare and screening.
2. Face recognition technology process:
(1) image acquisition and pre-processing
Capture of a facial image is generally divided into two ways, real-time collection of bulk import someone face image and face images; the purpose of preprocessing is in face image detection systems on the basis of facial images to make further the characteristics of the face image processing to facilitate the extraction.
(2) Face Detection
Face detection methods can be divided into three categories, which are based on detection of skin color model, based on edge detection feature, based on statistical theory.
(3) facial feature extraction
Current mainstream face recognition system can support the use of the features of a human face is usually divided into visual features, facial features and other image pixel count, and the characteristics of the face image is extracted for some of the specific features of the human face to extraction.
(4) identification of a living body and a face recognition
The actual face recognition systems typically need to increase the in vivo identification of links, because one of the common problems biometrics is to distinguish whether the signal from the real organism, including the requirement that the signal man turned around, blinking, opening and opening a word and so on.
Technology development
The most widely used on the market three programs, including structured light 3D program, the program and the ToF binocular stereo 3D imaging scheme.
1.3D structured light techniques
3D structured light techniques (Structured Light) is to obtain a three-dimensional structure of the object to be photographed by optical means, then the information obtained deeper applications. This program will be very promising in the future, may expand a wider space, such as 5G-related communications fields, AR as well as with other 3D modeling can play a critical role it unique.
For example, iphone x / xr / xs / xsmax, oppo Find x, mate20pro the 3D results are based on other technical principles of light.
3D structured light has its limitations, as compared to a conventional camera, the light from the 3D structure of the work to be shorter, the longest distance is generally about 1 m, the data given are 60cm far OPPO accuracy can be controlled in the range of ± 1mm .
2.ToF
ToF (Time Of Flight, the ToF technique), simply refer to emit light after a treatment, will be reflected back later hits an object, the capture time back and forth, since the wavelength of light and the modulated light is known, it is possible to quickly and accurately calculate the distance to the object.
ToF popularization and development of technology just around the corner, vivo has been to seize the initiative, and analysts have predicted Apple's iPhone in 2019 there may be a significant innovation including using ToF technology.
3. The binocular stereo imaging
Binocular stereo imaging (Stereo System) using dual cameras photographing the object, then the object distance is calculated by a triangular principle. Huawei has been listed nova3 use is the binocular 3D face recognition program.
The simplest principle of binocular scheme, the cost is relatively low.
Technology development
Combined with three-dimensional information: the two-dimensional and three-dimensional information integration features more robust multi-feature fusion
Feature fusion: a single feature difficult to deal with complex lighting and attitude change
Large-scale human face than: people-oriented than face massive data search
Depth study: full depth of the powerful neural network learning under conditions of large data