Embedded Fingerprint Authentication System

							                    Embedded Fingerprint
                    Authentication System


                        Christian Narváez

                            AXSionics
       School of Engineering and Information Technology HTI
                     christian.narvaez@axsionics.ch


Christian Narváez              12.10.2004                 1
     Outline
                    Biometrics introduction
                    Fingerprint recognition basics
                    Fingerprint feature extraction
                    Minutiae matching
                    Conclusions



Christian Narváez                12.10.2004          2
      Biometrics




Christian Narváez   12.10.2004   3
     Goal & Constraints
         Goal
                    Implementation of a complete, fast, robust and
                    reliable Fingerprint Recognition System


         Constraints
                    Platform: ARM7TDMI
                    Execution time < 2s
                    Sensor: Fujitsu MBF310


Christian Narváez                   12.10.2004                       4
     Fingerprint Recognition Basics
                                 Ridge bifurcation




                                          Ridge ending




     Core


                      Delta




Christian Narváez   12.10.2004                       5
     Fingerprint Flow
               Image Sensing    ➔   Choice of the sensor



           Feature Extraction   ➔   Extract distinctive characteristics of
                                    the fingerprint

                    Matching    ➔   The fingerprint is accepted or rejected




Christian Narváez                    12.10.2004                               6
     Feature Extraction           Generate Image Maps




                                    Binarize Image




                                   Detect Minutiae




                                 Remove False Minutiae




                                 Count Neighbor Ridges




                                   Minutiae Quality




Christian Narváez   12.10.2004                           7
     Feature Extraction – Generate
     Image Maps
                    Direction map
                    Low contrast map
                    Low flow & high curvature maps



                    Binarize



Christian Narváez   12.10.2004                  8
     Feature Extraction – Detect
     Minutiae




Christian Narváez   12.10.2004     9
     Feature Extraction – Remove False
     Minutiae
                            Too wide or to narrow


                            Side minutiae


                            Ridge interruptions


                            Hooks           ...etc.


Christian Narváez   12.10.2004                        10
     Feature Extraction – Additional
     information
                Count neighbor ridges
                        Number of ridges between every minutia
                        and it's 5 nearest neighbors




                Minutiae quality
                        Based on image maps and the mean and
                        standard deviation of the surrounding
                        pixels


Christian Narváez                 12.10.2004                     11
     Matching
                                        Rotation

                                        Translation

                                        Non-linear
                                        deformation

             Template   Query
                                        Feature extraction
                                        errors


Christian Narváez          12.10.2004                 12
     Matching – Local Structure
                                   Find a set of few minutiae that
                                   are invariant to:
                                        Rotation
                                        Translation
                                        Distortion
                                   Relations between minutiae:
                                        Distance i – k
                                        Angle i – k
                                        Number of ridges between i and k
                                        etc...
                    ⇒ Register query minutiae set
Christian Narváez               12.10.2004                         13
     Matching – Global Structure
                Compare the absolute position and
                orientation of the query and the template
                minutiae sets
                     Criteria for each pair of matched minutiae:
                        Are we close enough?
                        Are we the same type?
                        Are we in the in the same direction?
                        Are we reliable enough?



Christian Narváez                 12.10.2004                       14
     Matching – Results
                                              Nm: number of matched
                                              minutiae
                                              NQ: number of minutiae
                                              in the query
                                                        N   m
                                                  s =
                                                        N   Q


                    Threshold too high:
                          Higher False Rejection Rate (FRR)
                    Threshold too low:
                          Higher False Acceptance Rate (FAR)
Christian Narváez                12.10.2004                      15
     Conclusions
                System implemented and working
                First results
                     On small fingerprint set (100 images)
                Important improvements can be done
                     Quality of minutiae
                     Local matching algorithm



Christian Narváez                12.10.2004                  16
                                 Thank
                                 you...
Christian Narváez   12.10.2004            17