Book Description
A palm vein biometrics system is essentially a pattern recognition system that operates by acquiring an image of the palm veins, extracting a feature set from the image, and comparing this feature set against the template saved in a database. Unlike other biometric technologies such as fingerprints or face recognition, the palm vein scanner works by capturing the images of the vein patterns that are beneath the skin of the palm. Thus, palm vein based biometrics are more secure than fingerprints and palm prints. Moreover, the palm vein scanner captures the images of vein patterns in a contactless manner, which makes it more sterile and hygienic to use. However, the palm prints are also available in the Near Infrared (NIR) illumination, around 760nm wavelength. In some multispectral palmprint databases, the palmprint and palm vein images are both available in the same image. In this thesis, the features of both palm vein and palmprint are used for recognition. The process of palm vein recognition can be divided into several stages: image acquisition, pre-processing, feature extraction, matching and decision making. In order to build a reliable and accurate system, the unchangeable features of the palm prints/veins must be efficiently extracted from the original image. The difficulty of this problem, combined with the development of hyperspectral imaging techniques, has motivated the research presented in this thesis. Line or linear prints/veins detection have played an important part in palm prints/vein recognition. These techniques have been reviewed and explored. However, the vulnerability to the change of palm position and ambient illumination, together with low accuracy, encouraged researchers to find more stable algorithms. Fourier transform, wavelet transform and other frequency domain transforms are more robust and stable in palm prints/veins recognition. Due to the sparsity of the palm image, which is usually composed of some prints and veins, it's convenient to only transform the linear features to the frequency domain. Thus, in this thesis, the Curvelet Transform is introduced to extract the curve-like features from the palm print/vein images for accurate and sparse representation. The palm image is decomposed to several scales of coefficients, while in each scale the coefficients represent different features of the palm image. This technique can reduce the storage of the palm image to several hundred bytes and improve the accuracy as well. In order to increase the recognition accuracy, a combination of several biometrics features should be considered as well. The Curvelet Transform is good at extracting the curve-like features for accurate and sparse representation while the Gabor Filter can preserve local orientations. A combining scheme is proposed to utilise both of the two recognition methods at the same time with single near-infrared palm image. This combining scheme improves the recognition accuracy a lot, compared with techniques with solely Curvelet Transform or Gabor Filter. The experimental results demonstrate the effectiveness of the proposed method. As we can get information more easily and more accurately, the problem of information redundancy emerges, especially in hyperspectral imaging. As the number of palm images in the database increases, more effective methods of categorizing palm veins and fast matching algorithms should be developed. In hyperspectral palmprint recognition systems, it's not convenient to use exhaustive search for the optimal band selection and combination. A bands selection scheme is developed at the pre-processing stage.