Other Free Encyclopedias » Online Encyclopedia » Encyclopedia - Featured Articles » Contributed Topics from K-O

Open Systems for Online Face Recognition

learning authors training techniques

Definition: As opposed to the traditional supervised learning approach, online approaches using unsupervised learning techniques have been explored for online face recognition.

Generic person identification is important for novel applications such as video news library and continuous lifelog video. Face recognition technology has seen significant advances, and the FERET evaluation program has created sound testing methodologies . In spite of these advances, face recognition to date, remains a very hard problem. As opposed to the traditional supervised learning approach, online approaches using unsupervised learning techniques have been explored. In authors investigate such an “open” system based on eigenfaces and clustering techniques that allows a humanoid robot to automatically learn the faces of people it interacts with. In , authors describe an algorithm that uses virtual labels created from clustering in the output space (name or gender of a person) to incrementally derive discriminative features in input space (face pictures). Next, in authors present a technique for optimally selecting face exemplars from video with the initial help of an expert for online learning of gender.

Finally, in authors present a method for an online-learning face recognition system for a variety of videos based on Modified Probabilistic Neural Networks (MPNN). This system can automatically detect unknown faces and train new face classifiers such that an “unknown face” can be recognized if it appears again. Figure 1 distinguishes a training phase and classification phase just like any other face recognition system (depicted with a dashed line). The difference with the closed systems is the feedback arrow to the training phase for unknown faces. These unknown persistent faces (assumed to be important) become new sample faces for the online training (dotted ellipse in Figure 1).

Operation Petticoat [next] [back] Oort, Jan Hendrik

User Comments

Your email address will be altered so spam harvesting bots can't read it easily.
Hide my email completely instead?

Cancel or