Facial recognition filter technology
With the successive peaks of artificial AI in China, facial recognition technology has developed rapidly. According to institutional predictions, by 2020, the market size of facial recognition will reach 200 billion yuan, including approximately 70 billion yuan for customs clearance security products and 50 billion yuan for online payments.
Facial recognition technology
Facial recognition technology is an optical technique for biometric recognition of a person's facial digital information features. The principle is to use a scanning device to collect images or video streams containing faces, and extract and locate features of the faces to achieve the goal of identifying different people's identities through comparison. The computation involved in this entire process is enormous, and the quality of the captured images has a crucial impact on recognition efficiency. In this article, we mainly explain the narrowband filter products used in facial recognition technology systems to help developers and users of facial recognition technology systems better understand their functions and usage methods, in order to correctly select the technical specifications of narrowband filters.
Determination of narrowband filter bandwidth for facial recognition
The bandwidth of a narrowband filter should be determined based on the environment and the light source used. It is not necessarily the case that the narrower the filter, the better. The narrower the filter, the higher the price. For example, the bandwidth of an 850nm infrared LED is around 50nm. When choosing a narrowband filter, consideration should be given to the utilization of light energy, and the bandwidth of the narrowband filter should not be set too narrow. For LED light sources, a bandwidth below 15nm is not suitable. Because the bandwidth is too narrow, it blocks a large part of the strong signal light from the LED, and on the other hand, it reduces the effective use angle, resulting in bright edges in the middle of the captured image. After multiple experiments, it was found that controlling the light intensity of the LED at around 70% resulted in good contrast reproduction of the captured image. Therefore, the bandwidth of the narrowband filter can be selected at around 30nm, while for those with slightly higher anti-interference requirements, a bandwidth of 20nm can be chosen. According to the feedback from our company's customers, Ruicheng Optoelectronics produces 850nm narrowband filters, one with a bandwidth of 20nm and the other with a bandwidth of 30nm. There are many people who choose the 20nm bandwidth.
Determination of the center wavelength of narrowband filter for facial recognition
In theory, narrowband filters are best matched with the center wavelength of LEDs. However, the two factors we mentioned earlier are the incident angle effect and the heating problem of the LED itself. These two factors prompt us to consider the variation in the use of narrowband filters when determining the center wavelength of narrowband filters. Therefore, the center wavelength of the narrowband filter should be predetermined to be about 5nm longer than the center wavelength of the LED. In this way, the issue of incident angles between 0 and 10 can be considered. At the same time, the situation where the temperature rise causes the center wavelength of the LED to shift upwards was also considered.
Determination of the cut-off range of narrowband filter for facial recognition
The cutoff range of narrowband filters is mainly determined based on the response range of the receiver itself and the wavelength range of the interference source in the environment where the receiver is located. The general response range is 400-1100nm. In the case of facial recognition, the interference sources are mainly scattered or scattered sunlight and surrounding artificial light sources, which span a wide range of wavelengths, from ultraviolet to near-infrared. Combining these two reasons, it can be determined that the cutoff range of narrowband filters used for facial recognition is 400-1100 nanometers.
Determination of the cut-off depth of narrowband filters for facial recognition
In theory, the lower the transmittance within the cut-off range, the better. However, considering production costs and practical needs, the cut-off depth should be selected within a reasonable range. In facial recognition systems, when the cutoff transmittance of narrowband filters is less than 1%, it can clearly reflect the isolation effect of interference light. The cut-off transmittance of the narrowband filter produced by Ruicheng Optoelectronics for facial recognition is less than 0.5%. For applications with particularly high interference light intensity in the environment, we can also provide narrowband filters with higher cut-off depth to meet customer needs.
Determination of peak transmittance of narrowband filter for facial recognition
It is generally believed that the higher the peak transmittance of narrowband filters, the better, but this is not always the case in facial recognition applications. When the facial recognition device is in a direct sunlight environment, the intensity of the interference light is very high, and the interference light with the same wavelength as the signal light is also very strong. This interference light cannot be eliminated by narrowband filters. In order to improve the anti-interference ability of LED light, it is necessary to further increase the incident intensity of LED light, so that the signal intensity is several times higher than the interference intensity. Increasing the intensity of LED light sources is relatively simple, as long as the number of LEDs is increased. However, when the energy of the LED light source is strong enough to a certain value, coupled with the interference light energy of the same wavelength as the LED, the response of the CCD receiver is easily saturated. Even if the exposure is reduced through software, it will cause severe image distortion. At this point, a narrowband filter is required to filter out the interference light in the cutoff area. Depending on the actual situation, the peak transmission rate of the narrowband filter should be 40% or 60% or other values.
The facial recognition filter produced by Ruicheng Optoelectronics has successfully achieved narrowband filter coating on glass substrates with a thickness of 0.5mm or more using all dielectric hard film coating technology, allowing it to be embedded in cameras. Due to the absence of absorbing colored glass as the substrate, the stability of the filter is good, making it less prone to mold and fog, and suitable for working under strong light (too much sunlight) interference; The transmittance is as high as 92%, the instrument light attenuation coefficient is small, effectively improving the working distance and light intensity, and compensating for the instability of LED infrared lights; The actual cut-off rate is high, and there is no penetration phenomenon under strong infrared light when vertically incident. The cut-off value can reach OD5 or above.
Facial recognition technology
Facial recognition technology is an optical technique for biometric recognition of a person's facial digital information features. The principle is to use a scanning device to collect images or video streams containing faces, and extract and locate features of the faces to achieve the goal of identifying different people's identities through comparison. The computation involved in this entire process is enormous, and the quality of the captured images has a crucial impact on recognition efficiency. In this article, we mainly explain the narrowband filter products used in facial recognition technology systems to help developers and users of facial recognition technology systems better understand their functions and usage methods, in order to correctly select the technical specifications of narrowband filters.
Determination of narrowband filter bandwidth for facial recognition
The bandwidth of a narrowband filter should be determined based on the environment and the light source used. It is not necessarily the case that the narrower the filter, the better. The narrower the filter, the higher the price. For example, the bandwidth of an 850nm infrared LED is around 50nm. When choosing a narrowband filter, consideration should be given to the utilization of light energy, and the bandwidth of the narrowband filter should not be set too narrow. For LED light sources, a bandwidth below 15nm is not suitable. Because the bandwidth is too narrow, it blocks a large part of the strong signal light from the LED, and on the other hand, it reduces the effective use angle, resulting in bright edges in the middle of the captured image. After multiple experiments, it was found that controlling the light intensity of the LED at around 70% resulted in good contrast reproduction of the captured image. Therefore, the bandwidth of the narrowband filter can be selected at around 30nm, while for those with slightly higher anti-interference requirements, a bandwidth of 20nm can be chosen. According to the feedback from our company's customers, Ruicheng Optoelectronics produces 850nm narrowband filters, one with a bandwidth of 20nm and the other with a bandwidth of 30nm. There are many people who choose the 20nm bandwidth.
Determination of the center wavelength of narrowband filter for facial recognition
In theory, narrowband filters are best matched with the center wavelength of LEDs. However, the two factors we mentioned earlier are the incident angle effect and the heating problem of the LED itself. These two factors prompt us to consider the variation in the use of narrowband filters when determining the center wavelength of narrowband filters. Therefore, the center wavelength of the narrowband filter should be predetermined to be about 5nm longer than the center wavelength of the LED. In this way, the issue of incident angles between 0 and 10 can be considered. At the same time, the situation where the temperature rise causes the center wavelength of the LED to shift upwards was also considered.
Determination of the cut-off range of narrowband filter for facial recognition
The cutoff range of narrowband filters is mainly determined based on the response range of the receiver itself and the wavelength range of the interference source in the environment where the receiver is located. The general response range is 400-1100nm. In the case of facial recognition, the interference sources are mainly scattered or scattered sunlight and surrounding artificial light sources, which span a wide range of wavelengths, from ultraviolet to near-infrared. Combining these two reasons, it can be determined that the cutoff range of narrowband filters used for facial recognition is 400-1100 nanometers.
Determination of the cut-off depth of narrowband filters for facial recognition
In theory, the lower the transmittance within the cut-off range, the better. However, considering production costs and practical needs, the cut-off depth should be selected within a reasonable range. In facial recognition systems, when the cutoff transmittance of narrowband filters is less than 1%, it can clearly reflect the isolation effect of interference light. The cut-off transmittance of the narrowband filter produced by Ruicheng Optoelectronics for facial recognition is less than 0.5%. For applications with particularly high interference light intensity in the environment, we can also provide narrowband filters with higher cut-off depth to meet customer needs.
Determination of peak transmittance of narrowband filter for facial recognition
It is generally believed that the higher the peak transmittance of narrowband filters, the better, but this is not always the case in facial recognition applications. When the facial recognition device is in a direct sunlight environment, the intensity of the interference light is very high, and the interference light with the same wavelength as the signal light is also very strong. This interference light cannot be eliminated by narrowband filters. In order to improve the anti-interference ability of LED light, it is necessary to further increase the incident intensity of LED light, so that the signal intensity is several times higher than the interference intensity. Increasing the intensity of LED light sources is relatively simple, as long as the number of LEDs is increased. However, when the energy of the LED light source is strong enough to a certain value, coupled with the interference light energy of the same wavelength as the LED, the response of the CCD receiver is easily saturated. Even if the exposure is reduced through software, it will cause severe image distortion. At this point, a narrowband filter is required to filter out the interference light in the cutoff area. Depending on the actual situation, the peak transmission rate of the narrowband filter should be 40% or 60% or other values.
The facial recognition filter produced by Ruicheng Optoelectronics has successfully achieved narrowband filter coating on glass substrates with a thickness of 0.5mm or more using all dielectric hard film coating technology, allowing it to be embedded in cameras. Due to the absence of absorbing colored glass as the substrate, the stability of the filter is good, making it less prone to mold and fog, and suitable for working under strong light (too much sunlight) interference; The transmittance is as high as 92%, the instrument light attenuation coefficient is small, effectively improving the working distance and light intensity, and compensating for the instability of LED infrared lights; The actual cut-off rate is high, and there is no penetration phenomenon under strong infrared light when vertically incident. The cut-off value can reach OD5 or above.