HENRIQUE SÉRGIO GUTIERREZ DA COSTA BIOMETRIC IDENTIFICATION WITH 3D FINGERPRINTS ACQUIRED THROUGH OPTICAL COHERENCE TOMOGRAPHY Thesis presented to Programa de Pós-Graduação em Informática at Universidade Federal do Paraná Supervisor: Prof.Luciano Silva to obtain the doctoral degree Co-Supervisor: Prof. Olga Regina Pereira Bellon Curitiba, Pr 2016 2 CONTENTS 1 - INTRODUCTION ..................................................................................................... 11 2 - PERSON IDENTIFICATION USING FINGERPRINTS ......................................... 14 Dermatoglyphics and Ridge Formation ....................................................................... 14 Alterations .................................................................................................................... 14 3 - BIOMETRIC IDENTIFICATION / VERIFICATION SYSTEMS ........................... 16 4 - ACQUISITION OF 3D FINGERPRINTS ................................................................. 18 5 - OPTICAL COHERENCE TOMOGRAPHY ............................................................. 21 OCT Design parameters .............................................................................................. 26 OCT Unique Advantages ............................................................................................. 29 6 – STRUCTURE OF THE FINGER SKIN ................................................................... 31 7 – IMAGE EXTRACTION AND 3D OCT FINGERPRINT DATABASE ................. 34 3D Fingerprint extraction ............................................................................................ 34 OCT 3D Fingerprint database (OCTDB) .................................................................... 36 ROI Extraction to Matching ........................................................................................ 38 KH Maps ...................................................................................................................... 39 8 – 3D FINGERPRINT MATCHING ............................................................................. 41 9 - EXPERIMENTS AND MATCHING RESULTS ...................................................... 44 3D Fingerprint Images Database from Hong Kong Polytechnic POLYUDB ............. 44 KH Map Matching (CORR2D) ................................................................................... 45 Tests with the OCTDB Database ................................................................................. 45 Tests with POLYU database ........................................................................................ 47 Comparison with 2D matching results ........................................................................ 49 Compatibility of 3D Dermal and Epidermal Fingerprints ........................................... 50 Unrolled 2D fingerprints comparison .......................................................................... 52 10 - QUALITY MEASUREMENTS .............................................................................. 55 Quality Measurements for OCTDB ............................................................................. 57 Quality Measurements in POLYUDB database .......................................................... 59 11- ALTERED FINGERPRINTS ................................................................................... 62 Finger Sweating ........................................................................................................... 65 12 - OCT SCANNER PROTOTYPE .............................................................................. 66 Design Specifications .................................................................................................. 66 OCT Implementation ................................................................................................... 67 Full-field OCT ............................................................................................................. 67 Line- Field OCT .......................................................................................................... 70 13 – CONCLUSIONS AND FUTURE DIRECTIONS .................................................. 73 REFERENCES ................................................................................................................ 75 Index of Figures Fig. 1- KH map:(a) 3D plotting, colors correspond to curvature types (light blue-peak, dark blue-saddle ridge, yellow-pit and orange-saddle valley) and (b) 2D representation ............... 13 Fig. 2 – Biometric System Modules .................................................................................... 17 Fig. 3 – OCT and other medical imaging modalities .......................................................... 21 Fig. 4 – Basic OCT implementation using a Michelson interferometer ............................. 22 Fig. 5 – Skin images acquired by OCT: (a) B-scan of the skin(a), the white arrows in white detail of some A-scans;(b) a 3D image (volume) .................................................................... 23 Fig. 6 – Full-Field (TD) typical configuration. From: (DUBOIS, 2001) ............................ 24 Fig. 7 – En-Face image of the skin ..................................................................................... 24 Fig. 8 – Block diagrams of FD-OCT configurations: (a) Spectral Domain OCT; (b) Swept Source OCT. From (WOJTKOWSKI, 2010). ......................................................................... 25 Fig. 9 – Line-field schematics and line scanning on the sample. From ( YASUNO, ENDO, MAKITA ,AOKI, ITOH,YATAGAI, 2006) ........................................................................... 26 Fig. 10 – Skin layers. From (CUMMINS; MIDLO, 1976) ................................................ 32 Fig. 11 – B-Scan of the finger skin showing the stratum corneum and the dermis-epidermis interface.................................................................................................................................... 32 Fig. 12 – Detail of the ridge shapes (rendered 3D OCT image of the finger) .................... 32 Fig. 13 – A fingerprint acquired by a touch-based scanner (the yellow circles show an ending and a bifurcation). .................................................................................................................... 33 Fig. 14 – 3D fingerprints: (a)3D epidermis fingerprint; (b) 3D dermis fingerprints; (c) 3D registered epidermis and dermis fingerprints;(d) a region around the minutiae (a line ending) known as minutia cloud (corresponding to the rectangle in figure (a)). .................................. 34 Fig. 15 - Detection of external fingerprint: (a) Original (top) and detected external layer (bottom); (b) detection directions (white arrows). ................................................................... 35 Fig. 16 – Epidermis-dermis detection: (top) A-scan represented by the white arrow; (bottom) the intensity graph of the A-scan ............................................................................................. 35 Fig. 17 – Low quality images: (a) Motion artifacts (appear as longitudinal lines in the picture); (b) Holes and low-density regions............................................................................. 37 Fig. 18 – Rendered images from the OCT 3D Fingerprint database (OCTDB) ................. 37 Fig. 19 – 2D fingerprints scanned from the same volunteers (samples) ............................. 37 Fig. 20 – Minutiae clouds (rendered images): (a) Bifurcation; (b) Ridge ending............... 38 Fig. 21 - KH map: (a) 3D plotting. Colors correspond to different curvature types (light blue-peak, dark blue-saddle ridge, yellow-pit and orange-saddle valley) and (b) 2D representation ........................................................................................................................... 39 Fig. 22 – KH maps of: (a) line ending; (b) bifurcation ....................................................... 40 Fig. 23 – KH maps extracted from: (a) core; (b) delta ........................................................ 40 Fig. 24 – Matching method overview ................................................................................. 41 Fig. 25 - Examples KH-maps of the entire fingerprint ...................................................... 42 Fig. 26 – LGP calculation, from original intensity values to the final code (8 bits and decimal).................................................................................................................................... 43 Fig.27 – Black and white representation of entire fingerprint KH maps: (a) bad quality image with ridges unclear and noise; (b) acceptable quality image ................................................... 44 Fig. 28 – Fingers x amount of minutiae for OCTDB .......................................................... 46 Fig. 29 – FAXxFRR and CMC curves for matching (OCTDB): (a) single minutia; (b) multiple minutiae ..................................................................................................................... 47 Fig.30 - Fingers x amount of minutiae for POLYUDB ...................................................... 48 Fig. 31 – FARxFRR and CMC for finger matching with multiple minutiae (POLYUDB) 49 Fig. 32 – FARxFRR and CMC curves for 2D matching .................................................... 50 ............................................................................................................................................. 51 Fig. 33 – Curvature maps of the external (left) and internal (right) fingerprints ................ 51 Fig. 34 – FARxFRR (top) and CMC (bottom) for epidermis KH maps ............................. 51 Fig. 35 – 3D Fingerprint aligned with the Z axis: (a) perspective view; (b) top view ........ 52 Fig. 36 – Mapping of clouds points at height z to: (a) linear representation ;(b) the resulting unrolled fingerprint .................................................................................................................. 53 Fig. 37 - Examples of unrolled fingerprints (top) and their correspondent regular fingerprints(bottom), scanned with a regular fingerprint scanner. (a) left ring, (b) left middle, (c) left ring and (d) left pinky. ................................................................................................. 54 Fig. 38 – Matching scores distribution (CORR2D). Match (blue), No-match (red). The arrow indicates the threshold to eliminate the uncertain no-matches ...................................... 55 Fig. 39 – Quality problems. (a) holes; (b) noise. ................................................................ 55 Fig. 40 – Quality indicators for OCTDB: (1) 1/ρ1; (b) 1/ρ3 ............................................... 57 Fig. 41 – Quality indicator for OCTDB: dens .................................................................... 58 Fig. 42 – Normalized (OCTDB) scores (Gaussian) to: (a) ρ1;(b) ρ3. In blue the “match” scores and in red the “no-match” scores. ................................................................................. 58 Fig. 43 - Normalized (OCTDB) scores (linear) for dens quality indicator ........................ 59 Fig. 44 – CORR2D score distribution to POLUYDB ......................................................... 60 Fig. 45 – Normalized scores for 1/ρ1 (top) and 1/ρ3 (bottom): (a) linear normalization; (b) Gaussian normalization ............................................................................................................ 61 Fig. 46- Linearly normalized scores: (a) ρ1/ ρ1’ and (b) ρ2/ ρ2’ ........................................ 61 Fig. 47 – Skin alterations: (a) finger skin replaced by skin from the sole; (2) fingers that were bitten; (c) fingers burnt by acid; (d) stitched fingers; Figure from (YOON; FENG; JAIN, 2012) ........................................................................................................................................ 62 Fig. 48 – A Z-shaped fingerprint cut. From (YOON; ZHAO; JAIN, 2012). By observing the ridge flow it can be noticed that parts of the skin have been transplanted other positions of the same finger. .............................................................................................................................. 63 Fig. 49 – Scar, clearly visible in (a) rendered OCT image;(b) external 3D fingerprint and (c) internal 3D fingerprint strongly affected. ................................................................................ 64 Fig. 50 – Finger skin damaged by abrasion or scratches. (a) OCT image of ridges flattened by abrasion; (b) scratched skin image; (c) OCT image of scratched ridges ............................ 64 Fig.51 – Sweating fingers: (a) OCT rendered image; (b) 3D external fingerprint;(c) 3D internal fingerprint ................................................................................................................... 65 Fig.52 – Full-field OCT implementation schematic ........................................................... 68 Fig. 53 – FF-OCT prototype .............................................................................................. 69 Fig. 54 – A rat ear (B-Scan) imaged with the FF-OCT system .......................................... 69 Fig. 55 – FF-OCT images of the finger skin: 3D rendered image of some ridges (top) and a B-scan showing the stratum corneum (bottom) ....................................................................... 70 Fig. 56 – Skin images acquired by a Line-field OCT device ( YASUNO, ENDO, MAKITA, AOKI, ITOH,YATAGAI, 2006) ............................................................................................. 71 Fig. 57 – Prototype being implemented in IMAGO Research Group laboratory (Universidade Federal do Paraná) ............................................................................................ 71 Index of Tables Table I – Comparison of biometrics technologies based on the perceptions of experts 21 Table II – Sensitivity of FD-OCT, SD-OCT and SS-OCT 32 Table III – File name convention used for the extracted minutiae clouds 47 TABLE IV - Results for OCTDB and POLYUDB 53 TABLE V - Results for OCTDB: 3D and 2D 55 Table VI - UNROLLED FINGERPRINT COMPATIBILITY 59 Table VII – Matching scores ranges (OCTDB) 61 Table VIII – Matching scores ranges (POLYUDB) 64 Table of Acronyms ERR Equal Error Rate CCD Charge Coupled Device CMC Cumulative Matching Characteristic CL Cylindrical Lens COMPDC Compatibility Database CORR2D 2D Normalized Cross-Correlation EURODAC European Asylum Seeker register DLP Digital Light Processing FAR False Acceptance Rate FD-OCT Fourier-domain Optical Coherence Tomography FF-OCT Full-field Optical Coherence Tomography FRR False Rejection Rate LED Light Emitting Diode LGP Local Gradient Patterns NIST National Institute of Standards and Technology OCT Optical Coherence Tomography OCTDB Optical Coherence Tomography 3D Fingerprint Database OFDI Optical frequency Domain Interferometry POLYUDB The Hong Kong Polytechnic University 3D Fingerprint Images Database PPI Pixels Per Inch RMS Root Mean Square ROC Receiver Operating Characteristic ROI Region of Interest SD-OCT Spectral Domain Optical Coherence Tomography SLD Superluminescent Diode SNR Signal-to-noise Ratio SS-OCT Swept-source Optical Coherence Tomography TD-OCT Time-domain Optical Coherence Tomography