Biometric face recognition systems are widely used in many different types of applications. At present, a smart TV with face recognition system is a typical example of such application. Face recognition in smart TV is used for viewer authentication and based on this, personalized services or different recommendations can be provided. Face recognition systems should works in real time and should be able to recognize one or more identities. The most of this systems include also graphical user interface for automatic training process (Fig. 2.2).
Usually the 2D face recognition task requires processing of the input from a camera. The main face recognition process consists of following sub processes like:
- image acquisition - reads an image from the camera, converts it to the system format and pass it to the system process
- face localization - localizes the faces in the image and associate found coordinates with the image. Depending on the camera which is used the localization algorithm is implemented.
- training process – clustering algorithms are used , e.g. K-means
- pre-processing of localized faces includes histogram equalization
- normalization – e.g. resizing
- feature extraction - extracts features from pre-processed faces, LBP can be used
- classification of faces - use methods like Support Vector Machines or K-Nearest neighbor distance matching
- face tracking - usually only frontal faces in the image are tracked because the vast majority of face recognition methods is reliable only with use of frontal face images. Once the face has been recognized, it is tracked, what significantly saves computational resources and can follow the subject even after changes in pose [3]. So the information about recognized user is send as output from the system.
Fig. 2.2 – Example of training GUI of face recognition system