Fingerprint Analysis and Representation

1
Fingerprint Analysis and Representation

• Handbook of Fingerprint Recognition
• Chapter III Sections 1-6

Adaptive Flow Orientation based Feature
Extraction in Fingerprint Images
N.K. Ratha, S. Chen, A.K. Jain, Pattern
Recognition, vol. 28, no. 11, pp. 1657-1672,
1995.
Presentation by Tamer Uz
2
Fingerprint Analysis and Representation

• Handbook of Fingerprint Recognition
• Chapter III Sections 1-6

3
Outline

• Introduction
• Estimation of Local Orientation
• Estimation of Local Ridge Frequency
• Segmentation
• Singularity and Core Detection

4
Introduction

• Fingerprint
• Interleaved ridges and valleys
• Ridge width 100µm-300 µm
• Ridge-valley cycle 500 µm

5
Introduction

• A Global Look
• Singularities In the global level the
  fingerprint pattern shows some distinct shapes
• Loop ( )
• Delta (?)
• Whorl (O)Two facing loop

6
Introduction

• A Global Look
• Core
• A reference point for the alignment.
• The northmost loop type singularity.
• According to Henry(1900), it is the northmost
  point of the innermost ridgeline.
• Not all fingerprints have a core (Arch type
  fingerprints)

7
Introduction

• A Global Look
• Singular regions are commonly used for
  fingerprint classification

8
Introduction

• Local Look
• Minutia Small details. Discontinuties in the
  ridges. (Sir Francis Galton)

9
Introduction

• Local Look
• Ridge ending / ridge bifurcation duality

10
Introduction

• Local Look
• Sweat Pores
• High resolution images (1000 dpi)
• Size 60-250 µm
• Highly distinctive
• Not practical (High resolution, good quality
  images)

11
Estimation of Local Ridge Orientation

• Quantized map
• Average orientation around indices i,j
• Unoriented directions
• Weighted (rij)

12
Estimation of Local Ridge Orientation

• Simple Approach
• Gradient with Sobel or Prewitt operators
• Tij is orthogonal to the direction of the gradient

• Drawbacks
• Non-linear and discontinuous around 90
• A single estimate is sensitive to noise
• Circularity of angles Averaging is not possible
• Averaging is not well defined.

13
Estimation of Local Ridge Orientation

• Averaging Gradient Estimates
• (Kass, Witkin 1987)
• dij rij.cos2?ij, rijsin2 ?ij

14
Estimation of Local Ridge Orientation

• Reliability (rij)

• calculated according to variance or least sq.
  residue
• Like detecting outliers and assigning low weights
  to them.

15
Estimation of Local Ridge Orientation

• Effect of averaging

16
Estimation of Local Ridge Frequency
17
Estimation of Local Ridge Frequency

• Simple Algorithm
• 32x16 oriented window centered at xi, yi
• The x-signature of the grey levels is obtained
• fij is the inverse of the average distance
• To handle noise interpolation and/or low pass
  filtering is applied.

18
Estimation of Local Ridge Frequency

• Other Algorithms
• Mix-spectrum technique (Jiang, 2000)
• Energy of 2nd and 3rd harmonics in the spectrum
  (Fourier) domain is imposed on the fundamental
  frequency.
• Variation function technique (Maio Maltoni 1998a)

19
Estimation of Local Ridge Frequency

• Example on Variation Function Tech.

20
Segmentation

• Segmentation Methods
• Orientation histogram in neighborhood.
• Variance orthogonal to the ridge direction
• Average magnitude of gradient in blocks
• Threholding the variance of Gabor Filter
  (Band-pass) responces.
• Classifying pixels as forground or background
  using gradient coherence, intensity mean and
  intensity vaience as features

21
Segmentation

• Example Segmentation

22
Singularity and Core Detection

• Singularity Detection Methods
• Poincare method
• Methods based on local characteristics of the
  orientation image
• Partitioning based methods

23
Singularity and Core Detection

• Poincare Method

24
Singularity and Core Detection

• Poincare Method

25
Singularity and Core Detection

• Poincare Method

26
Singularity and Core Detection

• Poincare Method
• If we know the type of the fingerprint
  beforehand, false singularities can be eliminated
  by iteratively smoothing the image with the help
  of the following observation
• Arch fingerprints do not contain singularities
• Left loop, right loop and tented arch
  fingerprints contain one loop and one delta
• Whorl fingerprints contain two loops and two
  deltas

27
Singularity and Core Detection

• Methods based on local features
• Orientation histograms at local level
• Irregularity

28
Singularity and Core Detection

• Partitioning based methods

29
Singularity and Core Detection

• Core Detection
• Core North most loop type singularity
• It is generally used for fingerprint registration
• It needs to be found for the arches from scratch
• Has to be validated for the others

30
Singularity and Core Detection

• Core Detection
• Popular Algorithm (Wegstein 1982)
• Orientation image is searched row by row
• The sextet best fits a certain criteria is found
  and the core is interpolated
• Accurate
• Complicated and heuristic

31
Singularity and Core Detection

• Core Detection
• Other idea
• Voting based line intersection

32
Adaptive Flow Orientation based Feature
Extraction in Fingerprint Images

• N.K. Ratha, S. Chen, A.K. Jain, Pattern
  Recognition, vol. 28, no. 11, pp. 1657-1672,
  1995.

33
Outline

• Introduction
• Related Work
• Proposed Algorithm
• Experimental Results
• Conclusion

34
Introduction

• This paper proposes a feature extraction method
  from fingerprint images.
• Extracted features are minutiae (x,y,T)
• Method Extracting orientation field followed by
  segmentation and analysis of the ridges

35
Introduction

• General Stages of the Feature Extraction Process
• Preprocessing
• Direction Computation
• Binarization
• Thinning
• Postprocessing

36
Related Work
37
Proposed Algorithm
38
Proposed Algorithm

• 1)Preprocessing and Segmentation
• Goal To obtain binary segmented ridge images.
• Steps
• Computation of orientation field
• Foreground/background separation
• Ridge segmentation
• Directional smoothing of the ridges

39
Proposed Algorithm

• 1.1 Computation of the Orientation Field
• An orientation is calculated for each 16x16 block
• Steps
• Compute the gradient of the smoothed block.
  Gx(i,j) and Gy(i,j) using 3x3 Sobel Masks
• Obtain the dominant direction in the block using
  the following equation
• Quantize the angles into 16 directions.

40
Proposed Algorithm

• 1.1 Computation of the Orientation Field

41
Proposed Algorithm

• 1.2 Foreground/Background Segmentation
• Variance of grey levels in the direction
  orthogonal to the orientation field in each block
  is calculated.
• Assumption fingerprint area will exhibit high
  variance, where as the background and noisy
  regions will exhibit low variance.
• Variance can also be used as the quality
  parameter of the regions.
• High variance (high contrast) good quality
• Low variance (low contrast) poor quality

42
Proposed Algorithm

• 1.2 Foreground/Background Segmentation

43
Proposed Algorithm

• 1.3 Ridge Segmentation
• Orientation field is used in each (16x16) window
• Waveform is traces in the direction orthogonal to
  the orientation
• Peak and the 2 neighbouring pixels are retained
• The retained pixels are assigned with the 1 and
  the rest are assigned with 0.

44
Proposed Algorithm

• 1.3 Ridge Segmentation

45
Proposed Algorithm

• 1.3 Ridge Segmentation

46
Proposed Algorithm

• 1.4 Directional Smoothing
• A 3x7 mask (containing all 1s) is placed along
  the orientation
• The mask enables to count the number of 1s in
  the mask area.
• If the 1s are more than 25 percent of the mask
  area than the ridge point is retained.

47
Proposed Algorithm

• 2) Minutiae Extraction
• We are a few steps away from extracting the
  minutiae.

• First ridge map is skeletonized.
• Ridge boundary aberrations result
• In hairy growths.

• It is smoothed by using morphological binary
  open operator

48
Proposed Algorithm

• 2) Minutiae Extraction
• Morphological binary open operator

http//documents.wolfram.com/applications/digitali
mage/UsersGuide/Morphology/ImageProcessing6.3.html
49
Proposed Algorithm

• 2) Minutiae Extraction

50
Proposed Algorithm

• 2) Minutiae Extraction

51
Proposed Algorithm

• 3) Post Processing
• Ridge breaks (insufficient ink or moist)
• Ridge cross-connections (over-ink, over-moist)
• Boundaries

52
Experimental Results

• Summary of the procedures

53
Experimental Results

• Summary of the procedures

54
Experimental Results

• Performance Evaluation
• Detected minutiae is compared with the ground
  truth (extracted by human experts)

L Number of 16x16 windows in the input image Pi
Number of minutiae paired in the ith window Qi
Quality factor of the ith window (good4,
medium2, poor1) Di Number of deleted minutiae
in the ith window Ii Number of inserted minutiae
in the ith window Mi Number of ground truth
minutiae in the ith window
55
Experimental Results

• Performance Evaluation
• Base Line Distribution
• Generate same number of random minutiae in the
  foreground of (512x512) image
• Calculate the GI.

56
Experimental Results

• Performance Evaluation

57
Conclusion

• Robust feature extraction based on ridge flow
  orientations
• Novel segmentation method
• An adaptive enhancement of the thinned image
• Quantitative performance evaluation
• The execution time must be substantially reduced