Fingerprint Recognition

1
Fingerprint Recognition

• Professor Ostrovsky
• Andrew Ackerman

2
The Idea

• Including Region information in minutiae matching
• Reduces amount of matches need
• Can help better identify matches

3
The Process

• Preprocessing image
• Getting image into proper input
• 24-bit Bitmap of ridge detail properly enclosed
• Thinning Image
• Zhang-Suen Thinning algorithm
• A Fast Parallel Algorithm for Thinning Digital
  Patterns
• Edge Enhancement
• Fill in gaps in edges due to input quality or
  thinning process
• Code works to extent
• Region Coloring
• Colors in regions
• Spurious Region Removal
• Remove regions erroneously created in thinning
  process

4
Thinning (Zhang-Suen)

• A(P) Number of 01 patterns in order set P2 P9
• B(P) Number of non-zero neighbors
• Algorithm
• Do until picture is stable
• First Subiteration Delete P from pattern if
• A) 2 B(P) 6
• B) A(P) 1
• C) P2 P4 P6
• D) P4 P6 P8
• Second Subiteration Delete P from pattern if
• A and B same as above
• C) P2 P4 P8
• D) P2 P6 P8

5
Thinning Example

• Spurious Regions

Edge Gaps
6
Scanning Image

• Scan image
• Move a window down the image
• Offset window horizontally
• Offset window vertically
• Offset window horizontally
• and vertically

7
Edge Enhancement

• Scan the Image
• For each window
• Find all the endpoints in the window
• For each endpoint pair
• Look at line that goes through the points
• If line is strong
• Draw line between points

8
Region Coloring

• Color in Regions
• Does in one scan of image

9
Spurious Regions Removal

• Scan the Image
• For each window
• If a region is encased in the window and is small
  enough
• Calculate how often the regions around it border
  it
• Remove the border between the region and the
  region that least borders it

10
Testing

• Tested 8 different fingerprints of same finger
• Choose different minutiae pairs
• Calculated amount of edges crossed between the
  minutiae
• Calculated amount of regions crossed between
  minutiae pair
• Compared Results

11
Input Fingerprints
12
Output of Program
13
Minutiae Pair 1

• Image 1 13 Regions, 20 Edges
• Image 2 9 Regions, 16 Edges
• Image 3 11 Regions, 19 Edges
• Image 4 11 Regions, 17 Edges
• Image 5 12 Regions, 19 Edges
• Image 6 10 Regions, 16 Edges
• Image 7 11 Regions, 15 Edges
• Image 8 13 Regions, 18 Edges

14
Minutiae Pair 2

• Image 1 7 Regions, 11 Edges
• Image 2 5 Regions, 13 Edges
• Image 3 6 Regions, 11 Edges
• Image 4 7 Regions, 12 Edges
• Image 5 7 Regions, 11 Edges
• Image 6 6 Regions, 10 Edges
• Image 7 4 Regions, 8 Edges
• Image 8 7 Regions, 12 Edges

15
Statistics

• Minutiae Pair 1
• Max Region Difference 4
• Average Region Difference 1.64
• Max Edge Difference 5
• Average Edge Difference 2.21
• Minutiae Pair 2
• Max Region Difference 3
• Average Region Difference 1.25
• Max Edge Difference 5
• Average Edge Difference 1.7

16
Future Work

• Improve Preprocessing Code
• Improve on edge enhancement via algorithms that
  use edge orientation
• Matching Algorithm
• Using processed fingerprints, come up with
  algorithm that does fingerprint matching
• Use idea of regions and lines between minutiae
• Fingerprint of Fingerprint
• Run X amount of binary tests on a fingerprint and
  return a vector of size X. This binary vector
  would identify the finger print
• Matching prints would have similar vectors
  (hamming distance)
• Tests could include information about minutiae as
  well as regions