Research On Key Technology Of Finger Vein Recognition

Compared with the traditional identity authentication technology that uses specific objects or knowledge,the biometric technology that uses the unique physiological characteristics or behavior of the human body for identity recognition has attracted widespread attention.Among the many biological characteristics of the human body,the use of near-infrared light with a specific wavelength to irradiate the finger to extract the finger vein features of the irregular blood vessel structure existing in the finger is a biological feature that has received extensive attention in recent years.Finger vein features are used in various fields due to their in vivo identification,internal features,uniqueness and stability.However,due to low contrast,uneven illumination and light scattering from the skin,finger veins still have many problems in practical applications.Need to be resolved.In terms of recognition,finger movement and more noise in the collected images will affect the results;in terms of security,finger vein features are unique,once they are stolen,the individual vein features will no longer be safe;in terms of application,recognition Performance often brings a large hardware overhead,which is not conducive to the promotion of finger veins.Therefore,how to realize fast,accurate and safe finger vein recognition has always been a challenging subject.To promote the development of finger vein features,this work focus on the recognition,safety and application of finger vein features.The main research contents are as follows:(1)Research on finger vein recognition technology.The finger vein recognition algorithm based on convolutional neural network shows advanced performance due to its powerful feature extraction ability,but the existing methods lack the common problems in translation,rotation,etc.Discussion of severe deformation problems.In order to solve the above problems,this study aims at the problem of finger deformation in the acquisition process of finger vein images,and plans to use deformable convolution to learn the finger deformation information;and then correct the finger deformation according to the finger deformation information.In the stage of finger vein image feature extraction and recognition,for the problem that the extracted feature information is less distinguishable,a squeeze and excitation module is introduced to improve the accuracy of the finger vein recognition algorithm.(2)Relying on the powerful feature extraction capabilities of deep learning algorithms,the performance of finger vein recognition algorithms has been continuously improved,but such methods often require the construction of larger network scales and hardware devices.In fact,in the application scenarios of finger vein recognition,some scenarios are not suitable for deep learning methods with large network scale due to space and cost constraints.In order to solve the contradiction between finger vein recognition performance and hardware resources in resource-limited scenarios,a metalearning-based finger vein recognition scheme is proposed to convert the finger vein recognition problem into a meta-learning problem.Design an adaptive finger vein recognition algorithm.(3)Due to low contrast,uneven illumination and light scattering of the skin,among different vein images collected by the same finger,there are usually some low-quality images with poor image quality.The false and missing information contained in such low-quality images will cause performance errors of the recognition algorithm.Therefore,it is necessary to consider the quality assessment of finger vein images and the screening of low-quality images to reduce the impact of low-quality images on the performance of the recognition algorithm.Existing finger vein image quality assessment work often only considers the difference properties of some low-quality images,and has poor generality.To this end,statistical knowledge is used to analyze the matching of different samples between fingers to automatically mark possible low-quality images;at the same time,a lightweight convolutional neural network is used to mine the hidden differences between low-quality images.properties for fast,accurate,and versatile finger vein image quality assessment.(4)In terms of the security of finger vein templates,because of the uniqueness of finger vein features,losing the vein feature template in one application scenario will lead to the loss or misappropriation of the template in all other application scenarios that use the same vein feature.risks,leading to serious privacy and security concerns.There are few researches on template protection of the existing finger vein recognition system,which usually sacrifices part of the recognition performance while protecting the vein template.Aiming at the balance between template security and recognition performance of the finger vein recognition system,a finger vein image encryption scheme is proposed to be developed by using the RSA algorithm;and the ciphertext is destroyed twice by grayscale normalization to further ensure the security of the encrypted image.Fast and accurate recognition is achieved under the premise of protecting the safety of finger vein templates.