Full recordShow full item record
AbstractFace recognition is one of the few biometric methods that possess both accuracy and intrusiveness. For this reason it has drawn attention of many researchers and numerous algorithms have been proposed. Various fields such as network security, surveillance benefits from the face recognition because it provides more efficient coding scheme. Since the face recognition is a real world problem and there are cases when not all the input data is not known beforehand. In this project the focus is on the online learning strategy. We implemented online nonparametric discriminant analysis methodology for long-term face recognition problem. The advantage of using NDA over LDA is explained briefly. Besides reviewing the online version of NDA, we propose an optimized version based on 'affective forgetting'. In order to guarantee real-time response, the online learning strategy has been extended with a pruning mechanism which gets rid of the oldest samples. Experimental results on the FRIENDS dataset demonstrated that the performance of classification is not affected by replacing the former samples with new ones.