Skip to main content
Version: 3.12.0 (latest)

Face SDK Components

Base components#

Face Detector#

Face Detector is a basic Face SDK component, which allows to detect faces on input images or videos (when used with VEE, see below). The result of data processing is a special internal object representing a user's face, which is a normalized (normally rotated in the frame plane and cropped) face image standardized for further biometric processing. Due to the built-in tracking mechanism, a face is tracked and treated as the same object even after a person had left and entered the frame. For streaming video data, it's recommended to use the VideoEngine component instead of FaceDetector.
Licensing: by Data Channel (photo or video).
See the detailed info in Tracking Faces and Capturer Class Reference.

Encoder#

Encoder is a basic Face SDK component, which allows to create the Face biometric template. Encoder uses the normalized face image obtained from FaceDetector as input data. The result of processing by Encoder is a biometric template Face (numerical array of identification features), which is stored in a biometric database and/or used for biometric operations. Internal composition and accordingly the size of the Face template depend on the selected method of biometric processing.
Licensing: by Data Channel (photo or video).
See the detailed info in Face Identification.

MatcherDB#

MatcherDB is a basic component of Face SDK, which implements matching of biometric templates: 1:1 (verification), 1:N (identification) or their combination. MatcherDB uses the Face biometric template calculated by the Encoder component as input data. For verification (1:1), it's sufficient to use MatcherDB(1). Identification (1:N) is performed in the array of biometric templates. The MatcherDB size determines the search index size (permitted search boundaries) but doesn't limit the size of the biometric database.
Licensing: by the number of biometric face templates in search index of a biometric database.
See the detailed info in Face Identification.

3D Liveness Detector#

3D Liveness Detector is an optional Face SDK component, which allows to estimate the liveness of the subject. Liveness is estimated by means of detecting 3D surface of a real face and can detect spoofing attacks by means of presenting a photo or a video image of a face. As input data, 3D Liveness Detector uses the normalized image obtained from FaceDetector and a depth map received from a 3D (RGBD) sensor.
Licensing: by Data Channel (photo or video).
See the detailed info in Liveness.

Gender-Age Detector and Emotions Detector#

Gender-Age Detector and Emotions Detector are optional Face SDK components, which allow to estimate gender and age of a person and get a rough estimate of the prevailing emotional state of the face at a given time. As input data, Gender-Age Detector and Emotions Detector use the normalized face images obtained from the FaceDetector module. The age is detected within the range of +/- 5 years. The emotional states are Happy, Surprise, Neutral, Angry.
Licensing: by Data Channel (photo or video).
See the detailed info in Age & Gender and Emotions.

Macro Components#

Video Engine Standard (VES)#

Video Engine Standard (VES) is a macro component (component bundle) of Face SDK, which provides basic processing of video stream in 1 channel (for example, real-time video data from 1 camera). VES processes video in cycles. Each cycle includes:

  • detection and tracking of faces in a frame
  • selecting the best shot and encoding the selected face (computing the Face biometric template)
  • estimation of age, gender, and emotions (separately licensed component)

VES receives the sequence of video frames as input data. It means that a video stream should be decompressed. VES by default includes the following components: Face Detector, Encoder, one load balancer (to manage the sequence of input video frames). For complex tasks of real-time video processing in high-load systems, the internal composition of VES can be extended: you can add additional components to improve performance. As compared to VideoEngine Extended, VES does not support face matching feature.
Licensing: by Data Channel (photo or video).
See the detailed info in Video Stream Processing.

Video Engine Extended (VEE)#

Video Engine Extended (VEE) is a macro component (component bundle) of Face SDK, which provides basic video stream processing in 1 channel (for example, real-time video stream from 1 camera) and makes the data sequence for further search (in the watchlist or in the database). VEE processes video in cycles. Each cycle includes:

  • detection and tracking of faces in a frame
  • selecting the best shot and encoding the selected face (computing the Face biometric template)
  • comparing the template with the database (separately licensed component)
  • estimation of age, gender, and emotions (separately licensed component)

VEE receives the sequence of video frames as input data. It means that a video stream should be decompressed. VEE by default includes the following components: Face Detector, Encoder, and MatcherDB(N) where N is the size of the search index, which is typically equal to the number of faces (biometric templates) in the database. VEE also includes two internal load balancers: the first one is designed to manage the sequence of input video frames, the second one is designed to manage the sequence of biometric database search queries based on the face template. For complex tasks of real-time data stream processing in high-load systems, the internal composition of VEE can be extended: you can add additional components to improve performance.
Licensing: by Data Channel (photo or video) and size of the search index of MatcherDB.
See the detailed info in Video Stream Processing.

Last updated on