Low Quality Fingerprint Recognition Based On Deep Convolutional Neural Network

Fingerprint is regarded as a widely accepted biometric feature because of its uniqueness,persistence and acceptability.Through the study of fingerprint identification,the demand for access control can be met,the efficiency of government services can be improved,and the detection and prevention of various diseases can be greatly facilitated.Due to the different methods of fingerprint extraction,there is a lot of noise in the foreground area and the background area of the fingerprint,and the accuracy is not high.Therefore,it is difficult to identify the fingerprint with these low quality fingerprints.In recent years,with the rapid development of deep learning,convolutional neural network has been widely used in various fields such as industry,medical treatment,finance,environment,military,etc.,and has achieved quite good results in image recognition and classification.Therefore,the method of convolutional neural network is proposed to study the low quality fingerprint.Specific work includes:Firstly,NIST DB10 and FVC 2004,which are poor in fingerprint quality,are selected as the fingerprint database for text research,which greatly increase the difficulty of fingerprint identification,and the training set and test set of this paper are 3200 and 800 respectively.In this paper,the selected fingerprint database is preprocessed by a number of methods,including histogram equalization,fingerprint image segmentation,fingerprint directional field estimation,fingerprint image enhancement based on STFT,fingerprint image thinning,singularity detection and ROI extraction,which can improve the quality of fingerprint images.Secondly,using the Inception V3 network to compare and analyze the results before and after preprocessing.Subsequently,a deeper,wider and more complex Inception V3 network was used to further improve the accuracy of fingerprint identification.Finally,all the experimental results are compared and analyzed in detail.When the fingerprint data set is not preprocessed,the fingerprint identification accuracy is 75% using the Inception V3 network;after a number of preprocessing operations,the same method is used to obtain the recognition rate of 86.25% on the data set.To achieve better identification results,the more complex Inception-ResNet-V2 network was used and the fingerprint identification accuracy reached 93.75%.Therefore,it can be seen that the study in this paper effectively improves the robustness of fingerprint image and the recognition accuracy of low-quality fingerprint.