ASICLAND (CEO Jongmin Lee) announced on the 8th that it has registered a patent for an "Edge AI-based facial recognition device."
The Edge AI-based facial recognition device uses a 3D camera that generates a depth map image corresponding to the user's face and a 2D camera that generates a 2D face image of the user. This device recognizes the user's face using these two types of cameras.
Facial recognition technology detects face images present in still images or videos captured by the camera. It performs preprocessing such as face alignment, brightness correction, and image normalization on the detected face images, then extracts features from the preprocessed face images to identify the faces. Existing technologies have issues as they rely solely on verifying the three-dimensionality of the user's face through distance sensors or depth maps, failing to improve the basic accuracy and reliability required for facial recognition.
Regarding this, Jongmin Lee, CEO of ASICLAND, stated, "The device performs facial recognition in two different ways: a first neural network model trained to calculate the similarity between the depth map image corresponding to the user's face and a pre-registered depth map image, and a second neural network model trained to calculate the similarity between the 2D face image corresponding to the user's face and a pre-registered 2D face image. By complementarily considering the results, it derives the final facial recognition result, preventing spoofing attacks using photos or videos, and improving the required accuracy and reliability of facial recognition."
The company has also registered a patent for a "data management device and method for on-chip systems." This technology is expected to contribute to reducing power consumption in on-chip systems, which is applicable to various devices that use limited power, such as mobile terminals, sensor nodes, drones, and robots.
ASICLAND (CEO Jongmin Lee) announced on the 8th that it has registered a patent for an "Edge AI-based facial recognition device."
The Edge AI-based facial recognition device uses a 3D camera that generates a depth map image corresponding to the user's face and a 2D camera that generates a 2D face image of the user. This device recognizes the user's face using these two types of cameras.
Facial recognition technology detects face images present in still images or videos captured by the camera. It performs preprocessing such as face alignment, brightness correction, and image normalization on the detected face images, then extracts features from the preprocessed face images to identify the faces. Existing technologies have issues as they rely solely on verifying the three-dimensionality of the user's face through distance sensors or depth maps, failing to improve the basic accuracy and reliability required for facial recognition.
Regarding this, Jongmin Lee, CEO of ASICLAND, stated, "The device performs facial recognition in two different ways: a first neural network model trained to calculate the similarity between the depth map image corresponding to the user's face and a pre-registered depth map image, and a second neural network model trained to calculate the similarity between the 2D face image corresponding to the user's face and a pre-registered 2D face image. By complementarily considering the results, it derives the final facial recognition result, preventing spoofing attacks using photos or videos, and improving the required accuracy and reliability of facial recognition."
The company has also registered a patent for a "data management device and method for on-chip systems." This technology is expected to contribute to reducing power consumption in on-chip systems, which is applicable to various devices that use limited power, such as mobile terminals, sensor nodes, drones, and robots.