Cutting-Edge Fingerprint Indexing and Matching Innovations at i-VisionGroup
Explore groundbreaking advancements in fingerprint indexing and matching techniques developed by i-VisionGroup at Tsinghua University. Discover how ConvNets are utilized to generate fixed-length feature vectors for efficient indexing, along with a robust approach for aligning fingerprints into a unified coordinate system. Witness superior performance compared to state-of-the-art algorithms on benchmark databases. Dive into the development of a unified deep network, FingerNet, for minutiae extraction and more.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
i-VisionGroup@Tsinghua 2017-10-20 i-VisionGroup@Tsinghua
Kai Cao and Anil K. Jain. Fingerprint Indexing and Matching: An Integrated Approach Yao Tang Fei Gao Jufu Feng* Yuhang Liu. FingerNet: An Unified Deep Network for Fingerprint Minutiae Extraction Fingernet 2 i-VisionGroup@Tsinghua
Fingerprint Indexing and Matching: An Integrated Approach Contribution Designed a fingerprint indexing algorithm by leveraging a large longi- tudinal fingerprint database to train a ConvNet to generate a fixed-length feature vector is efficient for indexing. Developed a robust approach for aligning fingerprints into a unified coordinate system. Demonstrated superior indexing performance on two different benchmark databases over state-of-the-art algorithms. 3 i-VisionGroup@Tsinghua
4 i-VisionGroup@Tsinghua
Alignment Use SOM to train dictionary Contrast-enhanced image I* gradient of I* ROI TRAINING: ( ) is manually marked/reference point (x, y) is detected automatically 32 32 OF blocks train SOM (6X6) With the 2,000 aligned OF 5 i-VisionGroup@Tsinghua
SOM Initialize W Iteration Input X and Y=W[X] Find argmin d(X,Y)=C WC*[X] WC*[X]+n(x-w)n (0,1) and n decay C* is neighbor domain of C Final W is the centroids of OF 6 i-VisionGroup@Tsinghua
Alignment Use SOM to train dictionary COMPUTING: is the only param to compute point-based Hough transform I rotated by opt and translated to ensure reference point 7 i-VisionGroup@Tsinghua
ConvNet Training Training Set: 3,741 subjects with more than 12 impressions from MSP longitudinal fingerprint database. which results in around 440K fingerprints from 37,410 different fingers Net: inception-V3 Augmentation: 512X512 down sample to 333X333,randomly cropped to 299X299 Output the last fully connected layer with 2048 d 8 i-VisionGroup@Tsinghua
Online Indexing Simply finds the top-k largest values in a set. 9 i-VisionGroup@Tsinghua
FingerNet: An Unified Deep Network for Fingerprint Minutiae Extraction Background: In latent fingerprints, ridge patterns are drowned in background noises. Handcrafted features are difficult to adapt to complex background variance. Our basic idea is to combine domain knowledge and deep learning representation ability. Contributions: 1.A new way to guide the deep network s structure design and weight initialization to combine domain knowledge and the representation ability of deep learning, while preserving end- to-end differentiability. 2. A novel network for fingerprints called FingerNet is proposed. Typical fingerprint representations including orientation field, segmentation, enhancement and minutiae can be acquired from the unified network. 10 i-VisionGroup@Tsinghua
3.Reliable minutiae have been extracted on both rolled/slap and latent fingerprints automatically without any fine tuning. 4.One way to generate weak labels to latent fingerprints from the matched rolled/slap fingerprints, which helps to achieve modular training. Traditional Methods to Equivalent ConvNets Expand to FingerNet Label, Loss and Training 11 i-VisionGroup@Tsinghua
Traditional Methods to Equivalent ConvNets Normalization== pixel-wise activation layer Orientation Estimation== shallow ConvNet with 3 handcrafted kernels (Gxy,Gxx,Gyy) + merge layers + complex activation layers 12 i-VisionGroup@Tsinghua
Segmentation == a shallow ConvNet Enhancement Convolution operations are conducted on local fingerprint block == selective convolution method. 13 i-VisionGroup@Tsinghua
Gabor Filter Grouped Phases Orientation Choose Enhanced Map Extraction ==ConvNet with one convolution layer and one maxout layer 14 i-VisionGroup@Tsinghua
15 i-VisionGroup@Tsinghua
Expand to FingerNet Normalization pixel-wise normalization Orientation Estimation 3 atrous convolutional layers with different sample rates (ASPP) averaging Segmentation share the entire multi-scale feature maps with orientation estimation part a multi layer perception upsampling Enhancement as plain network 16 i-VisionGroup@Tsinghua
Minutiae Extraction 3 convpoolingbranching map 17 i-VisionGroup@Tsinghua
Label, Loss and Training Weak orientation labels: generated from matched rolled/slap fingerprints. Use minutiae to align. Weak segmentation labels:convex hulls. strong orientation labels: unoriented minutiae directions. Ground truth labels: manually mark 18 i-VisionGroup@Tsinghua
Experiments training data: 8000 pairs of matched rolled fingerprints and latent fingerprints. test experiments are conducted on NIST SD27 and FVC 2004 database set A. 19 i-VisionGroup@Tsinghua
CMC 20 i-VisionGroup@Tsinghua
