Research On Finger Multimode Identification Technology

Compared with single mode,biological multimodal recognition technology has higher recognition performance and security,and has been widely applied in various fields of life.Human fingers contain fingerprint,finger vein and other biological information,which can be used for multimodal fusion recognition.At present,most of the research on fingerprint and finger vein fusion recognition technology is based on independent image.In practical application,it is difficult to realize because of the high requirements of acquisition equipment.Because near infrared finger image contains fingerprint information of visible light reflection imaging besides finger vein of infrared imaging,how to separate fingerprint and finger vein information by single near infrared finger image and use it in finger multi-modal fusion recognition has high research value.In this thesis,based on a single near-infrared finger image,the algorithm of image enhancement and separation,the algorithm of quantization layer and feature layer fusion recognition are studied.Finally the finger multimodal recognition is realized,and the concrete contents are as follows:Firstly,this thesis proposes a near-infrared image enhancement algorithm based on histogram equalization with limited contrast to separate fingerprint and finger vein in a single near-infrared finger image.In the finger image of near-infrared imaging,because the contrast difference of fingerprint Valley ridge is not obvious,the fingerprint information is difficult to be fully utilized,and the traditional fingerprint preprocessing algorithm is not suitable for the fingerprint of this imaging mode.By analyzing the gray and width of fingerprint and digital vein pixels in the image,adjusting the clipping coefficient and the size of image blocks,enhancing the fingerprint information in the image,and combining the block enhancement model to separate the fingerprint and digital vein information in a single image,the corresponding separated image is obtained,which also provides the image basis for the multi-modal and multi-level fusion recognition of fingers.Secondly,this thesis proposes a fingerprint vein quantization layer fusion recognition method based on thin line distance order statistics,which can effectively solve the problem of poor performance of image feature point recognition method of near-infrared special imaging.Due to the difference of near-infrared light penetration and visible light reflection in different parts of fingers,the fingerprint and vein imaging are not stable,and the pixel gray level in the background area changes greatly.The location of feature points extracted from the same kind of fingers is obviously different,and false feature points are easy to appear.Compared with feature points,thin linestructure is more stable.Therefore,the fine line distance order statistics is used to obtain the matching quantization values of fingerprint and finger vein,and the matching quantization values are weighted and fused based on the best weight theory.Theoretical analysis and experimental results show that the proposed algorithm has better performance than the traditional feature point algorithm and single modal recognition algorithm.Thirdly,this thesis proposes an adaptive radius LBP feature layer fusion recognition method based on the combination of equivalent patterns,which can effectively improve the description ability and recognition performance of texture features.Because the width and thickness of the fingerprint vein striations separated from a single near-infrared image are quite different,the traditional LBP feature is fixed window radius,which has some limitations in dealing with this kind of texture with different sizes,so it is necessary to adjust the LBP window radius to adapt to the fingerprint and finger vein texture information of the special near-infrared image.In this thesis,an adaptive radius LBP Operator Based on equivalent mode is proposed to improve the anti rotation and anti noise performance,and enhance the description ability of different sizes of texture.Firstly,LBP feature histogram is fused,and SVM multi class classifier is used to train the fused feature vector.Finally,the self collected database is used to verify that the feature layer fusion has better recognition performance than the quantization layer for low-quality fingers,and the open database experiment simulation shows the effectiveness of the proposed fusion recognition algorithm.