| Abstract | Techniques developed for recognition of objects in photographs often fail when applied to recognition of the same objects in video. A critical example of such a situation is seen in face recognition, where many technologies are already intensively used for passport verification and where there is no technology that can reliably identify a person from a surveillance video. The reason for this is that video provides images of much lower quality and resolution than that of photographs. Besides, objects in video are normally captured in unconstrained environments, often under poor lighting, in motion and at a distance. This makes memorization of an object from a single video frame unreliable and recognition based on a single video frame very difficult if even possible. This paper introduces a neuro-associative approach to recognition which can both learn and identify an object from low-resolution low-quality video sequences. This approach is derived from a mathematical model of biological visual memory, in which correlation-based projection learning is used to memorize a face from a video sequence and attractor-based association is performed to recognize a face over several video frames. The approach is demonstrated using a video-based facial database and real-time video annotation of TV shows. |
|---|
