Optical Comparator for Iris Recognition

Transcription

1 "Can you pick the Kiwi from among the Aussies?" Optical Comparator for Iris Recognition Timothy D. Wilkinson Introduction What is a comparator? The joint transform correlator (JTC) Liquid crystal over silicon (LCOS) Test case: Optical head tracker Optical iris recognition 1

2 What is a Correlator? Comparator? A correlator allows us to compare an unknown image with a known reference object. If they match, then the output will be a peak intensity. Input Image Reference Image Optical Output What is a Comparator? The position of the peak locates the object Input Image Reference Image Optical Output 2

3 What is a Comparator? Different objects do not correlate (no output peak) Input Image Reference Image Optical Output What is a Comparator? Objects can be partially hidden (occlusion) Input Image Reference Image Optical Output 3

4 What is a Comparator? Multiple objects can be detected/differentiated Input Image Reference Image Optical Output Applications: Industrial Inspection Compare frame by frame a sequence of images. Rogue objects are indicated by no correlation. We don t need to know what the objects are. Input Output Frame n Frame n-1 Frame n+1 Frame n Frame n+2 Frame n+1 Frame n+3 Frame n+2 4

5 Applications: Motion tracking Frame sequences can be searched for object motion. Search frame is segmented into small pieces. Each piece is then compared with the next frame. Correlation motion directly translates to segment motion. Segments Current Frame Previous Frame Correlation #1 Correlation #3 Optical Correlation Correlation can be done using coherent optics (lasers). A correlation uses 2 Fourier transforms. ( ) ( ) = ( ) [ [ ] [ ( )]] sxy, rxy, F F sxy, F rxy, T T T A lens performs a Fourier transform at the speed of light! 5

6 The Joint Transform Correlator (JTC) Light Display device FT lens Camera Computer The Joint Transform Correlator (JTC) Pass 1 (reference) Input plane FT (lens) (input image) Joint Power Spectrum 6

7 The Joint Transform Correlator (JTC) Pass 1 Binary threshold the JPS a d P c b P > a + b + c + d 4 The Joint Transform Correlator (JTC) Pass 2 Binary JPS FT (lens) 1.0 Normalised Camera Intensity Correlation peaks Output plane 0 Position (CCD pixels) 7

8 The Joint Transform Correlator (JTC) Light Display device FT lens Camera Computer Liquid Crystal Over Silicon (LCOS) The power of silicon VLSI electronics with the flexibility of liquid crystals. 8

9 Liquid Crystal Over Silicon Aluminium Edge Contact Glass Cover ITO Pattern Glue Seal Silicon Chip QVGA SXGA, HDTV Mirrors Home grown LCOS devices The fast bitplane 320x240 pixels 34um pitch High speed (44kHz) The stretched fast bitplane 640x512 pixels 17um pitch High tilt (44º) 9

10 Commercial LCOS devices Displaytech, QVGA 320x240 pixels ($22) CRL-Opto, SXGA 1280x1024 pixels Optical Comparator based on LCOS Input fibre from laser diode Compact design No alignment problems Backplane SLM λ/2 Polarisor Control/Processor Fast Detector Fast LCOS, 10kHz Potentially lightweight Future plastic mould opto-mechanics 10

11 Optical Comparator based on LCOS LCOS lens Laser Camera Optical Comparator based on LCOS 11

12 Test case: Head tracker Eurofighter pilot has a directional helmet mounted display Need to know which direction the pilot's head is pointing Eurofighter Cockpit Janes Ours Test case: Head tracker Before the head moves Camera view Compare view #1 with view #1 12

13 Test case: Head tracker After the head moves Camera view Compare view #1 with view #2 Test case: Head tracker Compare Compare 13

14 Test case: Head tracker Compare Test case: Head tracker 14

15 Correlation for Iris Recognition Current hot topic for anti-terrorism and security. The iris is like a fingerprint each one is unique. Difficult to fake/duplicate Good images for use with correlation. Correlation for Iris Recognition Reflection from the cornea Pupil Sclera Iris 15

16 Pupillary zone Collarette Correlation for Iris Recognition Structural features Radial contraction folds of Schwalbe Structural folds of Schwalbe Circular contraction folds Crypts of Fuch Ciliary zone Correlation for Iris Recognition Sectoral heterochromia 16

17 Correlation for Iris Recognition Direct correlation with database. Reduced resolution to 512x512. Grayscale 98% success, RGB colour 100% Correlation for Iris Recognition RED GREEN RGB emphasises features Increased accuracy of detection 3 x as many correlations BLUE 17

18 Correlation for Iris Recognition Right Left Correlation did not predict left right pairs Correlation for Iris Recognition Images used were very high resolution (2400x2000). Image can be broken into 512x512 sections. Each section is then cross correlated with the others Correlation matrix

19 Initial Conclusions: Correlation matrices combine to strengthen detection. Structural defects can be emphasised and searched. Possible right/left relationship indications. More thorough study needed. Rotation of each section to remove radial symmetry. Pupil dilation size is an important issue. The lure of the start-up company CMOS Sensor Optical system Megapixel Resolution Capable of over 500 corr/s Successfully used in HOVMON field trials (Prototype limited by generic drive electronics) Laser Input 15cm SLM 19

20 The lure of the start-up company Many Thanks The assistance of Dr Nick New 20

21 A Smart Camera Non-linear processing can be done on a CMOS backplane a d P b P > a + b + c + d 4 Average c a d P b P = a - c + b - d Robert's Cross c A Smart CMOS Camera 128 x 128 detectors processor 128 x 3 Programmable control Output Programmable nonlinear processing: Average, Robert, Low pass Processing disabled on second pass 21

22 A Smart CMOS Camera 22

23 A Smart Camera: Pass 1 Grab spectrum Process (Robert's X) Send processed spectrum back to the SLM 23