Research On Finger Multi-modal Feature Recognition Algorithm

In recent decades,with the development of information technology,traditional identity-authentication faced with enormous challenges.Nowadays,there are two bottleneck problem: low recognition rate for large scale population,and low ability of anti-counterfeiting.Biological recognition technology can effectively solve these problems.Finger is a vital organ of the human body which can help us to feel the world,it has a lot of identity information,such as fingerprint,inner knuckle print,finger vein,and so on.In recent years,finger identity-authentication gradually got the attention of academia.Compared to the single-mode biological recognition system,multi-modal biological recognition has fused variety of biological information,it has better security features and higher recognition rate.Firstly,this thesis introduces the basic theoretical knowledge of imaging acquisition from finger vein and knuckle print,image preprocessing,feature extraction and recognition,and fusion recognition.Then this thesis also introduces the procedure of finger multi-modal feature recognition.In order to improve the recognition accuracy of the recognition system,this thesis has respectively researched on the image preprocessing,feature extraction and recognition,and fusion recognition.The main work and results of this thesis can be summarized as follows:(1)Research on the image preprocessing of finger vein and knuckle print: This part mainly includes three aspects,they are extracting the Region of Interest from finger vein images,image normalization and Image de-noising.In allusion to traditional filter easy to lose detail information and blur the image,this thesis proposed a method based on wavelet filter and mean filter.This method not only removed noise while preserving details of the finger vein and knuckle print,but also would not make the image become blur.(2)Research on feature extraction and recognition of finger vein: In order to extract and recognize the feature of finger vein,this thesis proposes a method for finger vein recognition based on scattering convolution network algorithm.First of all,we extract the region of interest from the original finger vein images,then process the image preprocessing.Secondly,multi-layer scattering convolution network is used to extracting the scattering energy distribution feature matrix.Finally,process sample training and matching recognition by using Support Vector Machine classifier.Besides,this paper also used a feature extraction method based on subspace for comparison,and compared the two methods on the low frequency sub-band images.(3)Research on finger multi-modal fusion recognition: From the three levels of finger vein and knuckle print fusion recognition,this thesis has research on the fusion recognition algorithm.In the data layer,using a method based on wavelet transform to fusion recognition;in the feature layer,using a method based on scattering convolution network to fusion recognition;in the decision layer,using a method based on weight match matching algorithms to fusion recognition.Finally,three different fusion strategies are compared and analyzed.