Research On Robust Intension Representation Method Of Finger Feature Granules

Recently, multimodal biometric-based recognition has been a hot topic in biometrics community. Due to high user acceptance and convenience of human finger, finger-based personal identification has been widely used in practice. Hence, taking fingerprint(FP), finger-vein(FV) and finger-knuckle-print(FKP) as the ingredients of a finger trait, finger feature representation is helpful for further improving the universality and reliability of fingers in identification. In this aspect, robustly describing finger features is certainly a very important step since finger pose itself is prone to vary during imaging. To effectively extract the reliable multimodal finger features, the following methods are developed in this thesis.1) Gabor-based Local-invariant Gray Features(GLGFs) are extracted. By coding Gabor orientation features, three Gabor-coded modalities are respectively represented using local-invariant gray description(LGD), and these features are then fused to form global feature representations called GLGFs.2) Gabor-Ordinal-based Local-invariant Gray Features(GOLGFs) are extracted. Considering illumination variation in finger imaging, Gabor Ordinal Measure(GOM) is used to obtain feature maps of three modalities. These maps are then granulated for fusion at three levels of information granularity in a bottom-up manner, and each granule is described by LGD. This kind of feature representation with granularity is called GOLGFs.3) Weighted Local-Gabor-based Invariant Gray Features(WLGIGFs) are extracted. Our former work has been testified that Local Gabor Binary Pattern(LGBP) is better than GOM in dealing with illumination and translation invariance of finger-vein trait, using LGBP, three feature maps are then granulated and described by LGD and called LGIGFs. Considering the inner regions of finger with less sensitivity to pose variation, LGIGFs are weighted by a Gaussian modal. The obtained features are called WLGIGFs.The experimental results show that the proposed robust intension representation methods of granules are capable of weakening greatly finger-pose variation, as well as achieving higher accuracy recognition in a large homemade database.