Biometric facial recognition and liveness detector using AI computer vision
AssigneeElm
InventorsArshad Ali Khan, Syed Adnan Yusuf
Life of this patent
Estimating the term…
Abstract
A method, system, and computer readable medium are described to capture, detect, and recognize faces using machine learning and a single-stage face detector. A method to determine live faces from spoof 2D and spoof 3D images using a liveness score as well as a method to classify faces using machine learning deep convolutional neural networks is also described.
Claims17
A biometric facial recognition method, comprising:
- capturing one or more images of a face by a camera;
- storing the one or more images of the face in a memory;
- extracting one or more key features of each of the one or more images of the face with a single stage face detector module;
- collecting a set of two dimensional feature vectors from the key features;
- collecting a set of three dimensional feature vectors from the key features;
- collecting spontaneous eye blink information from the one or more images of the face;
- detecting liveness of the face based on the collected spontaneous eye blink information,
- when liveness face is detected:
- aligning a geometrical structure and a rotational angle of the one or more images of the face;
- and
- performing facial recognition by matching, with a trained neural network model, the one or more key features of the face with encoded features of a known face stored in the memory;
- and
- wherein the extracting, collecting, aligning and matching are carried out with a processor connected to the memory,
- wherein the trained neural network is trained using low quality spoofed faces in a grey-scale space and high-quality real face images in a color space.
A biometric facial liveness determination system comprising:
- a camera configured to capture one or more images of a face,
- a memory configured to store the one or more images of the face and a trained
- deep neural network model, and
- one or more processors configured to
- extract one or more key features of each of the one or more images of the face with a single stage face detector module;
- collect a set of two dimensional feature vectors from the key features;
- collect a set of three dimensional feature vectors from the key features;
- collect spontaneous eye blink information from the one or more images of the face;
- detect liveness of the face based on the collected spontaneous eye blink information,
- when liveness face is detected:
- align a geometrical structure and a rotational angle of the one or more images of the face;
- and
- perform facial recognition by matching, with a trained neural network model, the one or more key features of the face with encoded features of a known face stored in the memory;
- and
- wherein the trained neural network is trained using low quality spoofed faces in a grey-scale space and high-quality real face images in a color space.
A non-transitory computer readable medium having instructions stored therein that, when executed by one or more processors, cause one or more processors to perform a liveness determination method comprising:
- accepting two-dimensional feature information of a face from a camera image;
- accepting three-dimensional feature information of a face from the camera image;
- accepting eye blink feature information of a face;
- sending the feature information to a trained deep neural network;
- receiving from the trained deep neural network a liveness score associated with the camera image;
- and
- comparing the liveness score to a previously determined minimum accepted liveness score to distinguish the face from a spoof attempt,
- wherein the trained deep neural network is trained using low quality spoofed faces in a grey-scale space and high-quality real face images in a color space.
Dependencies are inferred from the claim text.