The Effects on Maximum Marginal Analysis in an Online Training Facial Recognition Environment

J. Zhou and J. Fan (PRC)

Keywords

Face Recognition, ART Network, Online training

Abstract

In this paper, a unique maximum marginal analysis in an online training facial recognition system was presented. The motivation of analysis marginal information is that, in the online training facial recognition, most of the trainings on new faces are unsupervised. Those new faces might be trained or clusted as one instance of old face due to the shorter marginal distance between the old faces and new faces. This problem may make the facial classifier into an unstable situation. The maximum marginal analysis was applied to solve this problem and the experimental results show a satisfactory result the identification reaches 91.75% of hit ratio in an unsupervised facial recognition system. Applying the technique presented in this paper to robotics vision system can be one of the potential and promising applications, which can help improving the personal identification performance in an open environment.

Important Links:



Go Back