Three Multimodal Biometrics Recognition Based On Image Quality

Biometrics can be defined that it can automatically recognize one person based on its physiological and/or behavioral characteristics. The physiological characters contain fingerprint, hand-geometry, palm-print etc; while signature and voice belong to behavioral characteristics. Compared with the traditional personal identification, such as: ID card, key, password, the biometric is hard to be copied, missed, and stolen. Therefore, biometric technology will be used in most different fields in the future; nevertheless, the actual application requires more rigid demands such as recognition rate, acceptability, safety, storage space, identification speed. In order to breakthrough the drawbacks of uni-modal biometric, most researchers pay much attention to the multimodal biometric fusion.Three-modal biometric fusion based on fingerprint, palm-print, and hand geometry is chosen, because they have the intrinsic relation, more relative characters can be obtained to make deep levels information fusion and can be captured in one picture. It can reduce the complexity of fusion designing and system management and has special advantage.Three modal biometric fusion algorithms based on fingerprint, palm-print, and hand geometry is deeply studied. The image quality estimation is innovated to multimodal biometric fusion. Series connected fusion modal and parallel connected fusion modal are proposed based on image quality to raise the accuracy and robustness.The main work of this paper is composed of the following two parts: Information fusion theory, levels and methods are addressed in detail; the fusion level can be classed into four levels: data fusion level, feature fusion level, matching score level, and decision level. As for the fingerprint quality, it can be distinguished in both space domain and frequency domain and then the SVM is employed to classify the fingerprint type, it can be divide to five types, namely, arch, left loop, right loop, whorl and tented. The classified result is reasonable. In the end, three modal biometric databases is built for test.Secondly, the normalizations play a very import role in the. processing of multimodal fusion, it can map varied measure scales in different spaces into a common trust domain. Two normalization methods such as Four segments linear (FSL) and Linear-Tanh-Linear (LTL) are proposed, the scores can separately be mapped into intervals [0, 2] and [0, 1]. Two fusion algorithm user weighting based quality(UWQ) and matcher weighting based quality(MWQ) are put in, at lost, the JFV fusion model, fingerprint types fusion model based in score level, MWQ and MWQ fusion models based in image quality are introduced...