Font Size: a A A

Research On Quality Assessment And Recognition Of The Finger Vein Image

Posted on:2020-08-12Degree:MasterType:Thesis
Country:ChinaCandidate:Y ChenFull Text:PDF
GTID:2428330572970983Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
In recent years,finger vein recognition,as one of the emerging biometric identification technologies,has attracted more and more researchers' attention.Some low-quality images that may be generated during the acquisition will seriously affect subsequent feature extraction and matching,so it is necessary to perform accurate quality assessment after collecting finger vein images to eliminate low-quality images.There are some drawbacks such as that manually design features is difficult and vein detection procedures is complex in traditional finger vein quality assessment methods.A few researchers have used convolutional neural networks(CNN)for finger vein quality assessment,but these methods based CNN have shortcomings such as inaccurate image labeling,inadequate features express ability based on single vein pattern.In addition,feature extraction is a key step in the process of finger vein recognition.The traditional recognition system relies on the hand-crafted characteristics,but these features are susceptible to environmental illumination and finger posture changes in practical applications.Convolutional neural network(CNN)has strong feature expression ability,and currently,CNN-based finger vein recognition techniques have achieved good results.However,the techniques usually adopt complex network structure or step-by-step process,making they cannot be applied to a hardware platform with limited computing power,small memory and realize an end-to-end identification process.This article has carried out in-depth research on the above issues,the main work and contributions are as follows:(1)A Light-CNN based finger vein image quality assessment method is proposed.Firstly,the finger vein image is automatically labeled based on the traditional finger vein image quality evaluation method.Secondly,the finger vein image is divided into image blocks to expand the training set,and the image block instead of the whole image is used as the input of CNN.Then,the Light-CNN and its variants are trained.In the final test,the average quality score of multiple image blocks corresponding to an image is the final quality score of the image.This method has a good trade-off between computational complexity and performance and can distinguish between high and low quality finger vein images to a certain extent.(2)A cascade-optimized CNN based finger vein image quality assessment method is proposed.The method learns the vein quality information of the finger vein binary image and the gray image hierarchically: Firstly,the finger vein binary image is used as the input of the network,and the pre-training model is obtained after learning.Then,the finger vein grayscale image is used as the input of the pre-training model to fine-tune this model.Finally,an optimized model is obtained.The fused quality feature learned by the method has showed better performance than the existing manual features and the features obtained from the single vein form,and can effectively distinguish between high and low quality finger vein images.(3)An end-to-end finger vein recognition method based on SqueezeNet is proposed.First,one difference image with the differential operation of the two images for authentic matching or imposter matching and one 2C image(the image' channel is two)by regarding this image pair as a two-channel image are separately obtained,meanwhile,the 3C image(the image's channel is three)is further obtained with the channel connection of the difference image and 2C image.Then,the SqueezeNet(this network has been pre-trained on ImageNet)that receives 3C image as the input is fine-tuned and the best optimization mode is determined.Finally,a cascading optimization framework is designed to integrate the difference images and 3C image.This method not only achieves high recognition accuracy but also streamlined model and end-to-end identification process.
Keywords/Search Tags:finger vein quality assessment, finger vein recognition, convolutional neural network, feature fusion, SqueezeNet
PDF Full Text Request
Related items