MINU-EXTRACTNET: Automatic Latent Fingerprint Feature Extraction System Using Deep Convolutional Neural Network

Abstract

The performance of any fingerprint identification system depends on the accuracy of the feature extraction stage. Many latent fingerprint matching methods failed to produce good results because of the presence of background noise. To improve the performance, we propose a simple Deep Convolutional Neural Network (DCNN) model called ‘‘Minu-ExtractNet’’. Firstly, latent fingerprint pre-processing is implemented using a Convolutional Neural Network (CNN) model called ‘‘Pre-ProcessNet’’. This model enhances the quality of the latent and produces the orientation information along with different segmentation masks. Secondly, pre-processed information is then used to extract the minutiae feature points using another CNN model called ‘‘ExtractNet’’. This feature extractor model performs the image quality assessment to determine the threshold value to filter out spurious minutiae points. A dynamic thresholding algorithm is developed to achieve this goal. Experiments are carried out on both plain (FVC2004) and latent (NIST SD27) public fingerprint datasets. The results show significant improvement in precision, recall and Fl score values under different settings in comparison with the state-of-the-art CNN methods.