A real-time face recognition method on video sequence using KFDA and NIB2DPCA

An approach to real time face recognition from video sequences is present in this paper. Motion region is used to segment foreground firstly and skin colour used to detect human face from the video sequence. The detected face images are reconstructed by a feature space defined by NIB2DPCA (Non-Iterative Bilateral-projection-based 2D Principal Component Analysis). KFDA (Kernel Fisher Discriminant Analysis) is used to extract the non-linear features of the reconstructed images. Project a probe image onto the feature spaces of KFDA and classify the face by comparing its position in the feature spaces with that of known individuals to recognize face. Experiments on different face videos indicate that developed approach is more robust by comparing to other methods.