University of Twente Student Theses
Fingerprint recognition based on spectral minutiae representation and deep learning
Yu, S. (2020) Fingerprint recognition based on spectral minutiae representation and deep learning.
PDF
3MB |
Abstract: | This paper proposes to apply spectral minutiae representation and deep learning for fingerprint recognition. The fingerprint is one important biometric feature, and its recognition typically incorporates four steps: image acquisition, processing, feature extraction, and comparison. The powerful functionality of deep learning in imaging processing makes it plausible to recognize the fingerprint patterns. Conventionally, deep learning has mainly been used to extract the minutiae or the feature vectors from raw fingerprint images. There has been no hybrid use of the two. In this paper, we propose to use the spectral minutiae representation and the convolutional neural network (CNN) in combination to advance direct matching of spectral minutiae representation in fingerprint recognition. In the proposed approach, a minutia set is represented by a spectrum with a fixed size, specifically, this spectral minutia representation converts a minutiae set into a 128×256 sized magnitude spectrum. This spectrum serves as the input to CNN, while the output of CNN is a 128-dimensional feature vector. The fingerprint recognition is then completed by feature vector comparison. In this paper, the CNN with 19 layers is used and the whole network is trained by triplet loss. This proposed approach makes the fingerprint recognition using CNN more efficient, as no complicated pre-processing is needed compared to process endowing raw images to CNN. The performance of the proposed approach is compared to direct matching of complex spectral minutiae representation. |
Item Type: | Essay (Master) |
Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
Subject: | 54 computer science |
Programme: | Electrical Engineering MSc (60353) |
Link to this item: | https://purl.utwente.nl/essays/85579 |
Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Repository Staff Only: item control page