Hand Vein Recognition Based On3D Point Clouds Matching

In recent years, hand vein recognition has drawn more and more attentions because itssecurity. The hand vein pattern can only be extracted when the object is alive so it’s hard toforge. Existing methods usually recognize different hand veins according to2D infraredimages, in which the hand vein patterns can be extracted after image denoising andenhancement. However, the hand vein patterns extracted from2D images are seriouslydistorted with the posture of the hand varies so the false rejection rate is high. Besides, thevein feature extracted from2D infrared images is usually not plentiful enough to supportlarge database recognition. Aiming at these problems, a hand vein recognition methodbased-on3D point clouds matching is proposed in this thesis. With the3D point cloudsreconstructed by the stereovision system, an improved kernel correlation algorithm is usedto measure the match score of different hand veins. This method might provide a solution tolarge database hand vein recognition and could achieve a low false rejection rate with ahigh dimensional feature. The major work of this thesis are as following:1. A hand vein stereo vision reconstruction system has been designed and calibratedbased on precise calculation.2. An image processing algorithm for infrared hand vein images has been proposed,including image denoising, enhancement and hand vein pattern extracting using adaptivethreshold.3. Reconstruction of hand vein3D point cloud has been accomplished. Extracted handvein pattern has been used as the index of pixel position, while SAD algorithm been used tomatch the two camera images. The depth information of hand vein can be gain with theoptical parallax by triangulation.4. A improved kernel correlation algorithm for point cloud matching has beenproposed, in which the decision of authentication is made by the average KC value of thetwo point cloud.5. A database which includes50persons’ hand vein3D point cloud has been built. Theregister experiments have been conducted for the testing of recognition rate and falserejection rate, which is98%and0%respectively. The experimental results show that themethod proposed in this thesis is effective and potential.