3D ultrasound palm vein recognition through the centroid method for biometric purposes

Background, Motivation and Objective A few years ago, some of the authors proposed an ultrasound technique for achieving 3D palm vein patterns for biometric recognition purposes. That technique was developed with a commercial imaging system using doppler modality. Unfortunately, even if results were promising, the acquisition time was too long for practical applications. More recently, they proposed an effective recognition system that employs an ultrasound research scanner (ULA-OP), a CNC pantograph and a high frequency (12 MHz) commercial probe. The system is able to acquire 3D image of the human palm (38x25x15 mm3) in about 4 seconds. Successively, they developed a procedure to extract palmprint features and validated it with verification and identification experiments on a consistent ad hoc established database. In this work, a new method for extracting palm vein pattern is proposed. The procedure uses as input the same 3D images already exploited for extracting palmprint features, as matter of fact upgrading the recognition system to a multimodal system. Statement of Contribution/Methods The proposed method is based on the individuation of areas in the B-mode image of lower brightness than a predefined threshold. Then, it measures a set of properties for the individuated dark areas in the binarized image and selects the vectors that specify the centers of mass of these dark regions (centroids). The same technique in applied to the successive B-mode images. Vein patterns are finally defined whenever centroids are found in the surrounding of the same coordinates for a certain number of consecutive B-mode images. Results/Discussion The proposed procedure has been tested on several samples. The figure below shows main steps for a single sample and a single vein pattern. As can be seen in the 8 bit gray scale B-mode (a) and C-mode images (b), the high frequency probe guarantees a sufficient resolution that allows to clearly distinguish main vein patterns of the human hand. Figure (c) shows the result of the binarization for three successive B-mode images; the threshold value was set to 64 by evaluating the experimental results. Finally, figure (d) shows the sequence of centroids (cyan color) individuated by the procedure for the selected vein pattern. Results of a validation carried out through verification experiments on a subset of the database are finally presented and discussed.