A Detectors-based Approach for Biometric Identification Systems

Autor/autori: Sorin SOVIANY, Sorin PUSCOCI, Mariana JURIAN

Rezumat: Articolul prezinta o solutie inovativa pentru clasificarea datelor biometrice, solutie bazata pe aplicarea mai multor detectori. Abordarea vizeaza problema preciziei identificarii biometrice, deoarece ratele erorilor la identificare in sistemele biometrice actuale sunt inca mult mai mari decat in cazul aplicatiilor de verificare. Metoda propusa se bazeaza pe o ierarhizare a deciziilor in care primul nivel consta dintr-un numar de detector ruland in paralel pentru recunoasterea anumitor persoane de interes. Al 2-lea nivel al ierarhiei contine o schema de fuziune locala post-clasificare pentru celelalte identitati, in functie de rezultatele primului stadiu de detectie. Metoda este utila atat pentru sisteme biometrice unimodale cat si multimodale, si in particular pentru aplicatii de identificare pe scara larga.

Cuvinte cheie: detector, fuziune locala, ierarhie decizionala, identificare

Abstract: The paper presents a novel approach for biometric data classification, based on multiple detectors. This approach is dealing with the identification accuracy challenges as much as the identification error rates are still more higher than for biometric verification applications. The proposed methods is relying on a decision hierarchy in which the first level is resulting from more parallel running detectors for the focused identities to be recognized. The 2nd hierarchy level contains a local post-classification fusion for the other identities, depending on the 1st stage detection results. The method is useful either for unimodal or for multimodal biometric systems and especially for large-scale identification tasks.

Keywords: detectors, local fusion, decision hierarchy, identification

DOWNLOAD PDF