Study On 3D Finger Vein Verification Technique

The finger vein modality plays an important role in biometrics due to its unique advantages.However,the usual approach to vein imaging and information acquisition that is adopted in current vein verification systems employs a monocular camera to acquire a single-view 2D vein image on one side of the finger under near infrared light,which causes two problems: limited vein pattern information can be acquired for verification,and there is a clear difference among samples of the same subject captured from different positions in contact-free mode.Both problems have adverse effects on the system's performance.In general,existing systems are more sensitive to positional variations of the finger,particularly for variations caused by rotation around by finger axis.Subject to the limitations of image acquisition and imaging modes,the problem of texture information changes and losses caused by finger pose variation still remains to be done,though there are a few works considered to address this problem.To solve the major problems explained above,we study a brand new finger vein verification scheme based on 3D reconstruction and convolution neural network technique.The key contributions of this paper can be summarized as follows:1.we proposed a scheme that uses three cameras to capture three finger vein images of the same finger in all range of angle,and develope the corresponding software and hardware system.Three cameras can cover all range of finger and acquire the entire vein information under skin.Since the off-the-shelf algorithms based on multi-view 3D reconstruction technique need to extract the feature points in the vein region,but it is difficult to obtain enough effective feature pionts on the finger vein image.Moreover,other classical 3D reconstruction algorithms cannot meet the requirements of finger vein verification system for high efficiency,small size and low cost.Under the premise of considering all factors,we proposed a new 3D finger vein verification system structures.2.We proposed a 3D finger vein reconstruction algorithm based on elliptic model.Since the perpendicular to the axial section of the finger approximates an ellipse,and the vein information in image is only corresponding to the vein on the finger epidermis,so we minimize the reconstruction error by constructing the 3D finger model throught concatenating the ellipse in different positions and map the 2D texture onto the 3D model to construct the 3D finger vein image.3.we proposed a 3D finger vein feature extraction and matching algorithm.In order to solve the multi-pose problem,we first rotate and translate the reconstructed 3D finger vein image,making the finger axis coincides with the Z axis,and the midpoint of the line segment is located at the origin of the system coordinate system.Then,we rotate the normalized 3D finger vein image around the finger axis to generate a flattened texture map and geometric map.Finally,we construct a lightweight convolution neural network to learn the texture feature and we fuse the texture and geometric feature in an effective way to obtain a best matching result.From our experiments,the lightweight network we proposed can obtain promising result in the 2D finger vein verification task compared to other recent algorithms,and the result that using the 2D finger vein texture map obtained by flattening 3D finger vein images improve 4.16% of EER when compared to using singel 2D finger vein image based on the multi-pose dataset.In addition,it can also get a result close to that using single 2D finger vein image when we only use the depth map.Moreover,when fuse the texture and geometric feature,the EER is 2.13% which improve 4.4% compared to traditional 2D finger vein verification system.Our proposed 3D finger vein verification system obtains more biometric information and is more robust to finger pose variation,and it shows good potential in real application.