Exemplar-based face recognition from video.pdf

Exemplar-based Face Recognition from Video ? Volker Krueger and Shaohua Zhou University of Maryland, Center for Automation Research A.V. Williams Building College Park, MD 20742 USA vok@cfar.umd.edu Abstract. A new exemplar-based probabilistic approach for face recog- nition in video sequences is presented. The approach has two stages: First, Exemplars, which are selected representatives from the raw video, are automatically extracted from gallery videos. The exemplars are used to summarize the gallery video information. In the second part, exem- plars are then used as centers for probabilistic mixture distributions for the tracking and recognition process. Probabilistic methods are attrac- tive in this context as they allow a systematic handling of uncertainty and an elegant way for fusing temporal information. Contrary to some previous video-based approaches, our approach is not limited to a certain image representation. It rather enhances known ones, such as the PCA, with temporal fusion and uncertainty handling. Ex- periments demonstrate the effectiveness of each of the two stages. We tested this approach on more than 100 training and testing sequences, with 25 different individuals. Keywords: Surveillance, Video-based Face Recognition, Exemplar-based Learning 1 Introduction Face recognition has been a major research topic in recent years. Among the most successful approaches are [19; 14; 20]. The techniques have been thoroughly evaluated in the FERET-Protocol [15] and produce acceptable recognition rates in ideal conditions. However, if ideal conditions are not met, e.g., in case of out-of-plane rotation or variations in facial expressions, recognition rates drop drastically. The major reason is, that all the recognition approaches use the still- to-still technique: gallery and probe sets contain still face images (mug-shots), and recognition rates are high only if geometrical and photometrical conditions of the test images in the probe set match those in the g