Near infrared palm image acquisition and two-finger valley point-based image extraction for palm vascular pattern detection

Human palm vascular pattern is one of biometric modality that can be used for authentication purpose. It is concealed under the skin and unseen through human visual in visible light spectrum. To enable visibility of palm vascular pattern, additional illumination from near infrared (NIR) light is needed. With NIR-sensitive imaging device, the palm vascular pattern can be recorded. Even so, palm vascular pattern does not directly seen in the recorded image. As datasets available for research communities originated from multispectral palm print images that contain information other than vascular pattern, supplementary image processing is needed to reveal the vascular pattern in the image captured. Given variations imposed by human hand and specifications of imaging components, the enhancement processing in detecting palm vascular pattern differs accordingly. This thesis explores one of the options available in developing a NIR-sensitive imaging setup that can capture only palm vascular pattern. The setup was constructed using Raspberry Pi single board computer (SBC) to enable portability of the device. Experiments were conducted to observe different imaging setup and related components combinations that can help imaging the palm vascular pattern. Based on assessments of image contrast (Michelson contrast, standard deviation and RMS contrast) executed on acquired images through the experiments, an imaging configuration was finalized to acquire a selfdeveloped dataset. Additional two palm image datasets were used in observing the related enhancement processing that can visually detect palm vascular pattern from a NIR illuminated palm image. The palm vascular detection processing was also executed on the self-developed dataset constructed earlier for validation. Based on the processing, a framework in extracting two fingers’ valley points to identify region-of-interest (ROI) was proposed; based on the nature of the acquisition process either it is guided or unguided acquisition. The ROI extracted was assessed by mean squared error (MSE) and structural similarity (SSIM) index to check the ROI stability, every time it is extracted from different palm samples. A vascular image enhancement processing comprises of several enhancement techniques were recommended based on their ability in enhancing palm vascular pattern visually. Assessment of the enhanced vascular pattern was done by biometric recognition process; measured in its correct recognition rate (CRR). The biometric recognition process was done by extraction of vascular line features by Local Binary Pattern (LBP), and classification by K-nearest neighbour (KNN) algorithm using cross-validation technique. The average CRR achieved were 13.8%, 38.7% and 64.2%; for the CASIA, PolyU and self-developed datasets respectively.Although the average CRR were quite low for an accurate biometric recognition system; it indicates that the palm image dataset developed in this thesis has distinctive ability such that it can be used as biometric data. This is because, the unguided image acquisition device in this thesis had been catered to capture only palm vascular pattern for recognition purpose compared to other datasets that contain additional information other than palm vein pattern. In summary, vascular pattern can be detected visually from the palm image acquired by the NIR palm image acquisition device developed in this research.

Text FK 2019 127 - ir.pdf Download (1MB)

Actions (login required)

View Item