Face Recognition

1
Face Recognition

• Joshua I. Cohen

2
Overview

• Cutting Edge
• Constraints
• - Speed
• - Accuracy

3
Data Acquisition

• Digital Camera

4
Data Acquisition Contd
5
Data Acquisition Contd

• Video Camera w/ IR Filtering

6
Algorithm

1. Pre-Processing
2. Convert .bmp to .pgm
3. Isolate Largest Contour
4. Profile Curve
5. Curve Matching
6. Parse Data to Extract Costs

7
I. Pre-Processing

• Starting Images

8
I. Pre-Processing Contd

• Convert To Gray Scale

9
I. Pre-Processing Contd

• Binary Threshold

10
I. Pre-Processing Contd

• Centroid

11
I. Pre-Processing Contd

• Filter Noise Using Connected Regions

12
I. Pre-Processing Contd

• Fill Region Up Until Contour

13
I. Pre-Processing Contd

• Remove Hair-Line Noise

14
I. Pre-Processing Contd

• Crop Image Area

15
I. Pre-Processing Contd

• Normalize
• Fill Border To Black
• Save to .bmp File

16
II. Convert .bmp to .pgm

• SUN utility
• convert-to-pgm f1.bmp f2.pgm

.bmp ? .pgm
17
III. Isolate Largest Contour

• SGI utility
• computeBoundaryLength f1.pgm smooth 1

CONTOUROPEN/CLOSEnx1 y1 x2 y2 xn yn
.con file
18
IV. Profile Curve
19
V. Curve Matching

• OpenCurveMatch f1.con f2.con 10 1
• Intrinsic Properties length and curvature
• Optimal Correspondence
• Alignment Curve
• Invariant rotations, translations, viewpoint
  variation, articulation, occlusions

20
VI. Parse Data to Extract Costs

• f1-points-f2-points-match.txt
• cost-summary.txt

f1 f1 0 f1 f2 403.668 f1 f3 468.081 f1 f4
454.315 f2 f2 0
21
Preliminary Results

• Amir vs Everyone

22
Preliminary Results Contd

• Brian vs Ming / Vinesh

23
Preliminary Results Contd

• Tom vs Vinesh

24
Preliminary Results Summary

• 50-60 success rate
• margin of error very small
• - 450 or less is a match
• - greater than 450 not a match

25
Remaining Issues

• Angle

Eye
26
Remaining Issues Contd

• Angle Contd

Normal
Forehead / Eye
27
Remaining Issues Contd

• Hair-Line

• Normalizing ? Distortion

28
Possible Solutions

• Use More Criteria (local extrema)
• - contour of nose
• - contour of chin
• Better Method to Filter Hair-Line
• Curve Match Segments of Face Profile

29
References

• Ross Cutler, Face Recognition Using Infrared
  Images and Eigenfaces, April 1996.
• Thomas B. Sebastian, Philip N. Klein, Benjamin B.
  Kimia, Alignment-based Recognition of Shape
  Outlines, 2001.
• Zdravko Liposcak and Sven Loncaric, A
  Scale-Space Approach to Face Recognition from
  Profiles, 1999.