Palm Vein Recognition Based on Adaptive Region-of-Interest Segmentation and Modified Deep Learning Model.

Region of Interest (ROI) is the basis of palm vein recognition. The ROI segmentation methods based on key points location can provide an accurate ROI. However, these methods require the manual labeling of auxiliary points in advance and can't perform well in palm vein images that don't contain the whole fingers. To solve that, this paper proposes an adaptive palm vein recognition scheme:(1) The center of the ROI was located by the centroid of the binary palm vein image and calibrated by image erosion. The size of the ROI was determined by the maximized inscribed circle. The tangent point between the inscribed circle and hand contour was used as an auxiliary to calibrate ROI in angle. (2) A deep learning model that can receive input of any size is designed to extract vein features. Feature extraction can be implemented without scaling images to avoid image distortion. The research results and experiments show that this recognition scheme is superior to most traditional methods.