engleski

Novel pattern recognition-based methods for re- identification in biometric context

The object reidentification problem is to authenticate an object across multiple disjoint fields of view. Towards that end, once the object has been detected and initialized at one location, one seeks to match it with a feasible set of candidates detected at other locations and over time. Note that detection and seeding for process initialization do not presume known identity and that one allows for reidentification of objects and/or faces whose identity might remain unknown. Standard pattern recognition methods can be exploited, including Bayesian analysis, in order to cope with the dynamic nature of the problem. In the context of biometrics, advanced video (data stream) surveillance applications (including sensor networks) for airport and subway surveillance require the ability to reidentify an individual. Face- and gait- based recognition are especially promising biometrics for reidentification, since they can operate at a distance and do not require detailed and/or high resolution images of the subject and/or its biometric traits. Two core aspects this special issue addresses are: (1) registration to establish correspondences between parts of a pair of images ; and (2) invariant template representations suitable to compare the corresponding parts. The invariance requirement comes from the inherent variability in the data capture process with respect to both sensors and subjects, e.g., PIE (pose, illumination, and expression). The registration requirement facilitates linking corresponding parts into meaningful assemblies that can be then authenticated.

reidentification; biometrics

nije evidentirano