Fingerprint Minutiae Distribution Statistics Based On Automatic Detection Algorithm

In 1999,the Daubert case in Philadelphia sparked a global debate on the science of fingerprint evidence.Since then,the current fingerprint identification system has been challenged and questioned.The probabilistic identification technique based on the statistical rule of fingerprint minutiae has become a new difficulty and hotspot in fingerprint testing.In this paper,six types of fingerprint minutiae data sets were established from the perspective of fingerprint identification probability assessment.Based on constructing the fingerprint minutiae detection model of YOLOv5s＿FI,the occurrence frequency of six types of fingerprint minutiae in 619297 fingerprint images,including ridge endings,bifurcations,lakes,independent ridges,spurs,crossovers,were counted on each finger position and pattern.Quantifying the value of distinguishing fingerprint minutiae.Specific research work includes the following four parts:Firstly,six types of fingerprint minutiae data sets had been established.Firstly,high-quality fingerprint images with clear and complete lines were selected from the police’s actual combat fingerprint automatic identification system.Then the fingerprint images were preprocessed;finally,label data and data enhancement were made,and the training and test sets were divided.The data set contained 5120 fingerprint images,including 192416 minutiae labels in six categories: bifurcations,spurs,crossovers,points,independent ridges,and lakes,which provided data support for the training and optimization of the subsequent feature detection algorithm.Secondly,an automatic fingerprint detection algorithm has been proposed based on the YOLOv5 network.Because of the small size,dense density and large number of fingerprint minutiae,the YOLOv5 algorithm was improved by deleting 32-fold the under-sampling detection layer and adding 4-fold the under-sampling micro feature fusion layer.FPN,PAN and SPP structures were used to increase receptive fields.In addition,the SE channel attention mechanism module was added.The experimental result showed that compared with the original model,the mAP0.5 value of the improved YOLOv5s＿FI model increased to 97.22%.The weight was reduced by three quarters during the detection speed unchanged.Thirdly,a fingerprint pattern classification algorithm had been constructed based on the ResNet network.In order to understand the distribution of fingerprint features of different pattern patterns,it was necessary to classify a large number of fingerprint images according to pattern patterns(Arch,Left loop,Right loop,Whorl)and detect them in batches.Firstly,Res2NeXt50was selected as the backbone network,and packet convolution and stack convolution were added to optimize the recognition details and global features at a fine-grained level.In addition,replace the activation function PReLU;add the CBAM attention mechanism;fusing the FPN pyramid structure increases the receptive field.The experimental result showed that the accuracy of the optimized fingerprint pattern classification model Res2NeXt＿FI improved to 98.7% in the selfestablished database and 97.2% in the public data set NIST-DB4.Fourthly,the distribution rules of fingerprint minutiae in large-scale fingerprint images were analyzed.The preprocessed 619297 fingerprint images were classified according to ten finger positions and four patterns.They were introduced into the fingerprint minutiae recognition model YOLOv5s＿FI to identify the number and location of the six types of minutiae.The results show that the frequency ranges(unit%)of six types of minutiae per finger are ridge endings [68.49,70.81],bifurcations [26.37,27.26],independent ridges [1.533,1.626],spurs [1.129,1.198],lakes[0.4588,0.4963],crossovers [0.3034,0.3256].The results showed differences in the distribution frequency and quantity of the six fingerprint minutiae in the ten-finger positions and the four patterns.Through the above four aspects of work,this paper calculated the number and frequency of six types of fingerprint minutiae,distinguished the identification value of each type of fingerprint minutiae,and provided essential statistical support for the probabilistic model of fingerprint identification.