| Hi-index | 0.00 |
A novel, template-based method for face recognition ispresented. The goals of the proposed method are to integratemultiple observations for improved robustness and toprovide auxiliary confidence data for subsequent use in anautomated video surveillance system. The proposed frameworkconsists of a parallel system of classifiers, referred toas observers, where each observer is trained on one faceregion. The observer outputs are combined to yield the finalrecognition result. Three of the four confounding factors -expression, illumination, and decoration - are specificallyaddressed in this paper. The extension of the proposed approachto address the fourth confounding factor - pose -is straightforward and well supported in previous work. Afurther contribution of the proposed approach is the computationof a revealing confidence measure. This confidencemeasure will aid the subsequent application of the proposedmethod to video surveillance scenarios. Results are reportedfor a database comprising 676 images of 160 subjectsunder a variety of challenging circumstances. Theseresults indicate significant performance improvements overprevious methods and demonstrate the usefulness of theconfidence data.