English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Face recognition on still images has been extensively studied. Given sufficient training data (many gallery stills of each person) and/or high resolution images, the 90% recognition barrier can be exceeded, even for hundreds of different people to be recognized (Phillips et al., 2006). Face recognition on video streams has only recently begun to receive attention (Weng et al., 2000; Li et al., 2001; Gorodnichy, 2003; Lee et al., 2003; Liu and Chen, 2003; Raytchev and Murase, 2003; Aggarval et al., 2004; Xie et al., 2004; Stergiou et al., 2006). Video-to-video face recognition refers to the problem of training and testing face recognition systems using video streams. Usually these video streams are near-field, where the person to be recognized occupies most of the frame. They are also constrained in the sense that the person looks mainly at the camera. Typical such video streams originate from video-calls and news narration, where a person’s head and upper torso is visible. A much more interesting application domain is that of the far-field unconstrained video streams. In such streams the people are far from the camera, which is typically mounted on a room corner near the ceiling. VGA-resolution cameras in such a setup can easily lead to quite small faces – down to less than ten pixels between the eyes (Stiefelhagen et al., 2007), contrasted to over two hundred pixels in many of the latest face recognition evaluations (Phillips et al., 2006). Also, the people go about their business, almost never facing the camera directly. As a result, faces undergo large pose, expression and lighting variations. Part of the problem is alleviated by the use of multiple cameras; getting approximately frontal faces is more probable with four cameras at the corners of a room than with a single one. The problem is further alleviated by the fact that the goal is not to derive a person’s identity from a single frame, but rather from some video duration. Faces to be recognized are collected from a number of frames; the person identity is then established based on that collection of faces. Far-field unconstrained video-to-video face recognition needs to address the following challenges: • Detection, tracking and segmentation of the faces from the video streams, both for system training and recognition. • Selection of the most suitable faces to train the system and to base the recognition upon. • The face recognition algorithm needs to cope with very small faces, with unconstrained pose, expression and illumination, and also with inaccurate face framing. • Fusion of the individual decisions on faces, to provide the identity of the person given some time interval.

Far-Field, Multi-Camera, Video-to-Video Face Recognition