Study Of Finger Vein Image Recognition Technology

The finger vein recognition system performs biometric recognition by capturing an image of the finger vein pattern.Different from traditional biometrics such as fingerprints and faces,the finger vein features located inside the human body are unique,permanent,safe,and living,and the acquisition process is user-friendly,which has gained widespread attention in recent years.The feature extraction based on the minutiae of the vein pattern is relatively simple,and can also better represent the topology of the vein network.However,due to the low quality of the finger vein images and the limitations of the vein segmentation algorithm,the accuracy of the existing minutiae-based methods is generally unsatisfactory.In addition,the traditional finger vein recognition algorithm relies too much on vein segmentation and skeleton algorithm,which has numerous parameters and is computationally complex,so it can not meet the real-time requirements.In order to overcome these limitations,the finger vein image recognition technology is studied in this paper.The main work is as follows:First,the extraction algorithm of the finger vein pattern is improved.Based on the Contrast Limited Adaptive Histogram Equalization(CLAHE)algorithm,this paper combines the Gabor filter method to enhance the finger vein image and improve the sensitivity to the vein edge.In addition,for the enhanced vein image,the paper extracts the vein pattern by multi-scale matched filtering method,and obtains the vein skeleton image after the processing such as refinement,denoising,and deburring to maximize the topology.Secondly,an algorithm for feature extraction and matching based on minutiae and vein structure is proposed.This paper adds detection of inflection points while extracting endpoints and branchpoints,which enriches the types of minutiae points.Then,the curve segment connected with each minutiae point is tracked,and a structural feature coding method is proposed,which encodes the structure information of the minutiae point and the curve segment.This paper also proposes a matching method for the structural feature codes,which matches codes by calculating the similarity matrix.The finger vein recognition system based on minutiae and structural feature is successfully constructed.Thirdly,an algorithm based on deep learning for feature extraction and matching is proposed.In order to get closer to the actual recognition situation,this paper proposes a lightweight two-channel network for open-set recognition,and chooses Cosine Loss as the loss function of the network to obtain the finger vein features with high discriminative performance.Finally,a deep vein-based finger vein recognition system is successfully built.Experimental results show that the finger vein recognition method based on the minutiae and structure feature and the finger vein recognition method based on deep learning proposed in this thesis are effective,and the accuracy of finger vein recognition is improved.