Product
Key facial geometric features, such as the angle of the bridge of the nose and the area of the cheeks, are crucial to improving recognition accuracy. One of the innovations in facial recognition technology is the thermal recognition mode. At present, the development trend of facial recognition technology is mainly divided into several aspects, such as commercial systems, mainstream software and algorithms. We believe that the integrated application of terminal equipment requires a complete solution, and large manufacturers have obvious advantages in this regard. The components of facial recognition mainly include software parts, such as databases and algorithms; hardware parts, such as camera modules, integrated devices, sensors, chips, ICs, hardware interface circuits, LCD screens and memories; and system service providers that integrate software and hardware. Among them, the algorithm is the core link of the industrial chain and the part with the highest technical barriers. From the perspective of domestic companies that design facial recognition, they mainly provide camera algorithms, etc., and the value of hardware may be underestimated.
Main commercial systems, mainstream software and mainstream algorithms for facial recognition
Infrared LED narrowband filtering technology is expected to become a core element. Traditional facial recognition technology is mainly based on visible light images, but this method has insurmountable defects. Near-infrared facial recognition system can completely solve the problem of ambient light influence. When the ambient light changes, the recognition effect of traditional recognition technology will drop sharply and cannot meet the requirements of the actual system. For example, the "yin-yang face" phenomenon that occurs when taking pictures in a side-lit environment may lead to incorrect recognition.
Solutions to the illumination problem include three-dimensional image face recognition and thermal imaging face recognition. However, these two technologies are not mature enough and the recognition effect is not ideal. The core technology and system of face recognition based on near-infrared images can capture near-infrared face images that are not affected by changes in ambient light under different lighting conditions. Combined with advanced algorithms, high recognition rates can be achieved.
Near-infrared face images not affected by ambient light
Near-infrared face recognition includes two parts: active near-infrared face imaging equipment and corresponding illumination-independent face recognition algorithms. Using active near-infrared light sources with an intensity higher than that of ambient light, combined with optical filters of the corresponding band, an environment-independent face image can be obtained. The face image will only change monotonically with the distance between the person and the camera.
Using specific feature extraction methods on this image, such as local binary pattern (LBP) features, can further eliminate the monotonous changes of the image and obtain feature expressions that are completely independent of illumination. In recent years, near-infrared face recognition has been widely used in real life, such as "Shenzhen-Hong Kong Biometric Passport Self-service Customs Clearance System", "Macau-Zhuhai Biometric Passport Self-service Customs Clearance System", "Beijing Airport T3 Terminal Self-service Customs Clearance System", etc., all of which have achieved good results.
Active near-infrared face imaging equipment can provide high-quality face images that are not affected by ambient light for face recognition. The so-called high quality includes: appropriate image brightness, uniformity, appropriate contrast, and no overexposure. Active near-infrared face imaging equipment generally includes the following units:
Active near-infrared light source with intensity higher than ambient light in the corresponding band, usually high-power 850nm and 940nm infrared LEDs;
Camera capable of receiving near-infrared light, usually a CCD image sensor. CCD has the advantages of small size, light weight, low distortion, low power consumption, low voltage drive, strong impact resistance, vibration resistance, and electromagnetic interference resistance, so it is widely used in various image acquisition systems. The CCD in the face recognition system is basically based on silicon substrate, and its spectral response range is 400nm~1100nm, which is also the spectral range that the narrowband filter should consider;
The narrowband filter is placed outside the camera lens, allowing near-infrared light to pass through while filtering ambient light. It is mainly used to isolate interference light, transmit signal light, fully highlight useful information, reduce interference information, and lay the foundation for subsequent image processing and recognition.
Technically, infrared LED narrowband filters are expected to become a core factor. At present, some solutions use infrared glass that isolates visible light and transmits infrared light as a filter. However, ordinary infrared glass only isolates visible light and ultraviolet light, and does not isolate the part of the interference light in the infrared band. Therefore, in order to obtain a good anti-interference effect, a narrowband filter must be used.
Comparison of ordinary absorption color filter and narrowband filter curves
Spectral distribution of 850nm LED
The selection of narrowband filters needs to consider multiple optical indicators, including bandwidth, center wavelength, cutoff wavelength, cutoff depth, peak transmittance, product thickness, etc. Judging from the recent development of cameras and AR, domestic optical companies have become the main suppliers, which reflects that the domestic optical strength is sufficient to meet the needs of consumer electronics and special displays. It is expected that major international customers will continue to use domestic suppliers for narrow-band filter modules.