Dalton

Swipe sensor optimized fingerprint matching algorithm

  • Extract traditional minutia instead of characteristic of fingerprint image
  • Translate data of spatial domain into frequency domain (extract frequency)
  • Fast matching-speed to comparison of code level

Fingerprint image

Convert into frequency domain

Encode and save

Scalability

  • Can be used on any mobile sensor
  • Unlimited size of fingerprint image
   

Fermion

Ultra small sensor optimized fingerprint matching algorithm

  • Strong performance of learning and authentication from ultra small area (such as 4mm x 4mm, 5mm x 3mm)
  • Recognize characteristic of all fingerprint ridges
  • Design to put up with distortion of stretched fingerprint ridge

Minutia

Only one traditional minutia found

in small area, traditional minutiae

(ridge end, bifurcation etc ..) are hardly can be found

Fermion

Features extracted from CS algorithm

Fermion extracts features from whole ridge shape and track down pixel intensity change not wanting to find any minutia

Accuracy

  • High security guaranteed based on FAR as 0.001%
  • Statistical 1:1 match FRR as 2.4% but learning FRR can be below 1% according to enrollment and learning policy
Version FRR FAR
Reject Genuine Ratio Accept Impostor Ratio
Fermion v3.0.8 86 3573 2.406941% 15 2006847 0.000747%

Tested with 11,000 fingerprint images from 100 people

Invariant

  • Rotation invariant : no limitation to matching angle (360 degree match)
  • Contrast invariant
  • Strong against noise

Open Source License

Name Version Purpose Download URL
nanoflann 1.1.8 k-d tree https://github.com/jlblancoc/nanoflann/archive/v1.1.8.tar.gz
openssl 1.0.2 AES encryption http://www.openssl.org/source/openssl-1.0.2.tar.gz
uuid Unknown To identify finger index http://code.dyne.org/hdsync/plain/src/uuid/
   

Muon

Ultra fast and lightweight fingerprint matching algorithm

  • Fast verification and learning of fingerprint with lower CPU
  • Change of pixels information and every shape of ridge as its features
  • Rotation and contrast invariant being strong against noise

Lightweight

Template size for 1 finger : about 4 kB

Basically it follows UI policy such as enrollment count and whether it permits learning capability or not but usually the template size can be grown up to 4 kB

Speed

Extracting feature time : about 350 ms

Verification time : about 100 ms

With 300 dmips on STM32F42XX board 240MHz

Also it can verify significantly fast on high dmips devices