Fingerprint Spoof Detection using Multispectral Imaging Robert K. Rowe, Ph.D. Chief Technology Officer

Transcription

1 Fingerprint Spoof Detection using Multispectral Imaging Robert K. Rowe, Ph.D. Chief Technology Officer September 21, 4

2 Program Goal and Topics Goal Develop multispectral imaging as a means to augment a conventional optical fingerprint reader to enhance the spoof detection capability of the overall system Topics Technology introduction Optical simulations Test bed results Miniaturization and integration activities 2

3 A Multi-Spectral Imager for Fingers Customized multispectral imager with the following distinctions from a conventional TIR-based optical fingerprint sensor: Optical geometry laid out to avoid critical-angle phenomena System configured to interrogate the skin, not the interface between the skin and sensor Use crossed linear polarizers Preferential imaging of scattered (subsurface) light Use multiple illumination wavelengths See different depths and structures in the tissue Linear Polarizer Polarizer (Crossed) Digital Camera Platen Illumination 3

4 Relevant Fingertip Physiology 10 3 Hemoglobin at Blood Concentrations De-Oxy Oxy Absorbance [mm-1] X Wave len g th [n m] Optical absorption due to blood (Note: semi logarithm scale) From S. Sangiogi et al., Microvascularization of the human digit as studied by corrosion casting, J. Anat. 204, (4) The structure of capillaries immediately below the surface mirrors external epidermal ridges and contains other distinct features (e.g. arterioles) Blood has distinct and strong absorbance bands in the visible region An image of these capillaries represents a subsurface fingerprint The non-blood portions of skin also have distinct spectral characteristics 4

5 Multispectral Simulations

6 Skin Model Used in TracePro, a nonsequential optical raytracing package This quadrant is the only one with surface features Dermis Epidermis Blood Vessels This quadrant has middepth blood vessels This quadrant has deep, large blood structure This octant is not contacting the platen 6

7 Results of Multispectral Simulations Pse udocolor [R:2 G:5 B:8] nm 450nm nm Pseudo-color image using 450, 600 and 750nm 550nm 600nm 650nm 700nm 750nm 800nm Key Findings: Different wavelengths provide different information Surface features are not necessary Contact is not necessary 7

8 Multispectral Testbed

9 Multispectral Testbed Designed and built bench-top system for technology assessment Able to collect multiple illumination wavelengths -700nm, 20nm HW Sequential imaging Image resolution is ~0 DPI ~ DPI as processed Light Source Filter Controller Sampling Platen Tunable Filter to Illuminator and Polarizer Polarizer & Lens Camera 9

10 TIR and Multispectral Finger Images 1st Three Principal Components Mapped to RGB Conventional TIR Extremely Dry Skin Multispectral (475,, 560, 576, 625 nm) 10

11 Extreme Sampling Conditions TIR Air gap TIR Pooled water 11

12 Spoof Detection for an Optical Fingerprint Sensor

13 TIR Fingerprint Sensor Incorporating a Multispectral Spoof Detector 13 TIR FP Illumination Absorbance (mm -1 ) s o [mm -1 ] TIR FP Imaging Multispectral Illumination and Imaging H e m o g l o b in A b s o rb a n c e De o xy W a v e le n g th [n m ] O x y Blue Green Red Wavelength (nm) Reduce the multispectral technology to a miniature, rugged solid-state subsystem configured to fit in a TIR fingerprint sensor One straight-forward implementation is to use a commercial color imaging module (R-G-B) and matching tricolor LEDs Fortuitously, RGB wavelengths span an interesting region of the optical spectrum of blood The relationship of the intensities of red, green and blue light are representative of blood and other spectrally active features in the skin Work is currently underway to test the characteristics of the RGB concept for spoof detection

14 Static Spectral Spoof Detection Baseline: Testing Against 2 nd Placement of a Finger Finger-1, Filtered Pseudo-Color Image Fing e r-1: S e lf MDis t (ROI m e an=0.9) Self S-Res (ROI mean=0.8) Finge r-2: Te s t MDis t (ROI m e an=0.9) Test S-Res (ROI mean=0.9) Finger-2, Filtered Pseudo-Color Image Outlier Metrics MDist = Mahalanobis Distance, S-Res = Spectral Residuals 14

15 Static Spectral Spoof Detection Test Against Gummy Bear Fingers Finger-1, Filtered Pseudo-Color Image Fing e r-1: S e lf MDis t (ROI m e an=0.9) Self S-Res (ROI mean=0.8) Gel-2, Filtered Pseudo-Color Image Ge l-2: Te s t MDis t (ROI m e an=2.3) Te s t S-Re s (ROI m e an=2.6) Outlier Metrics MDist = Mahalanobis Distance, S-Res = Spectral Residuals 15

16 Static Spectral Spoof Detection Test Against Playdough rob8-1, Filtered Pseudo-Color Image ro b8-1: S e lf MDis t (ROI m e an=0.9) Self S-Res (ROI mean=0.8) Playdough: Test MDist (ROI mean=3.6) Test S-Res (ROI mean=10.0) Playdough, Filtered Pseudo-Color Image Outlier Metrics MDist = Mahalanobis Distance, S-Res = Spectral Residuals 16

17 A Color Matched Spoof Commissioned construction of an ultra-realistic prosthetic finger, including fine detail of fingerprints Multilayer silicone structure Casting made on a real and available finger Color of prosthetic customized to match that of the real finger 17

18 Static Spectral Spoof Detection Test Against Color-Matched Prosthetic Real8-1, Filtered Pseudo-Color Image Real8-1: Self MDist (ROI mean=0.9) Self S-Res (ROI mean=0.9) Pros 8-1: Te s t MDis t (ROI m e an=1.3) Te s t S-Re s (ROI m e an=6.5) Pros8-1, Filtered Pseudo-Color Image Outlier Metrics MDist = Mahalanobis Distance, S-Res = Spectral Residuals 18

19 Spoof Detection Chromatic Texture Comparison Fing e r-2:fing e r-1, Po w e r S pe c trum Ratio Finger-2 Color Planes nm 540nm 625nm B G R Real Prosthetic Pros8-1 Color Planes Power Ratio S patial Fre que nc y [c y c /im ag e ] 10 1 Pros8-1:Real8-1, Power Spectrum Ratio 475nm 540nm 625nm Significantly more power in prosthetic spatial frequencies than the real finger: especially for blue illumination at higher spatial frequencies 19 B G R Power Ratio Spatial Frequency [cyc/image] Note: semi-logarithmic scale

20 Multispectral Spoof Detection Current Efforts TIR FP Illumination TIR FP Imaging Multispectral Illumination and Imaging Collaborating with Cross Match to integrate a commercial camera module and a custom LED illumination board into a standard fingerprint reader as a proof-of-concept prototype Each subsystem separately controllable via a USB interface to a host PC Eventual commercial unit will be tightly integrated 20

21 Summary and Acknowledgements Multispectral images of skin provide significant information about the authenticity of the sample A multispectral imager can be built from readily available components and configured to operate in conjunction with a conventional optical fingerprint reader Lumidigm and Cross Match have teamed to integrate the multispectral spoof sensor into commercial products The multispectral work has been funded in part by the DoD Rob Rowe contact info: RKRowe@Lumidigm.com (cell) 21