Towards pose-independent face recognition

The area in which face recognition technology arguably has the most potential is in policing; but perpetrators of crimes tend to be camera-shy, so full-face imagery is rarely available. Hence, police forces are interested in pose-independent schemes for face recognition. Indeed, the approach outlined herein, though very much work in progress, is being developed in association with Essex Police. Some police forces routinely capture video sequences of suspects but currently archive only a small number of frames from them. However, by processing the video sequence from a moderately cooperative subject, one can construct an image that "wraps around" the sides of the head and hence can be used for pose-independent face recognition. This paper is organized as follows. The basis of the authors' scheme is outlined. The authors discuss how a suitable "texture map" may be generated from a video sequence. This is followed by a description of a simple scheme for searching for faces in a database of texture maps. Results from this scheme are presented. Concluding remarks and a discussion of directions for further work are given.