Video-Based Face Recognition
- PDF / 470,033 Bytes
- 5 Pages / 504.467 x 719.855 pts Page_size
- 21 Downloads / 204 Views
Video-Based Face Recognition Jingxiao Zheng1 and Rama Chellappa2 1 University of Maryland, College Park, MD, USA 2 Johns Hopkins University, Baltimore, MD, USA
Related Concepts Face Association Face Detection Face Recognition Face Tracking Video Analysis
Definition Video-based face recognition is a type of face recognition where the test data are videos rather than still images. Similar to traditional face recognition, approaches for video-based face recognition attempt to identify a person in a video (identification) or decide whether two subjects in two different videos have same identity (verification).
© Springer Nature Switzerland AG 2020 K. Ikeuchi (ed.), Computer Vision, https://doi.org/10.1007/978-3-030-03243-2_816-1
Background In computer vision and biometrics, video-based face recognition has received significant amount of attention in recent years. It has a wide range of applications including visual surveillance, access control, video content analysis, etc. Large amounts of video data are becoming available everyday since millions of cameras have been installed in buildings, streets, and airports around the world and people are using billions of handheld devices that are capable of capturing videos. An example of video-based face identification from the IARPA Janus Surveillance Video Benchmark (IJB-S) dataset [1] is shown in Fig. 1. In this example, gallery consists of pre-enrolled still images, and a test video is captured from a surveillance camera. Every face in the video should be detected and matched to a subject in the gallery or to an unseen class for the open-set scenario. Notice that the low-quality faces in the video frames make this problem very challenging. A video-based face recognition pipeline usually consists of components including face detection, face alignment, feature extraction, face association, and set/sequence-based face matching. The first three frame-wise components
2
Video-Based Face Recognition Pre-Enrolled Gallery G1
G3
G2
?
Frame 2
Frame 1 Test Video
Video-Based Face Recognition, Fig. 1 An example of video-based face identification problem consisting of a pre-enrolled gallery of three subjects and two frames from the test video
are almost identical to still image-based face recognition. Their differences are mainly in the last two components: (1) Since a video contains lots of frames with much more information than a single image, faces are usually associated before matching, where faces with the same identity across different video frames are associated into sets/sequences. (2) Face matching is also performed in set/sequence, instead of using just a single image. The overall algorithm includes the following steps: 1. Given a video, faces are first detected from video frames and aligned using the estimated fiducial points. 2. Face features are extracted from the detected faces using feature extractors for face recognition. 3. Face sets/sequences with unique identities (ideally) are constructed by face tracking/association approaches. 4. Face sets/seq
Data Loading...
