Fingerprint Recognition - WKUpeople.wku.edu/qi.li/teaching/595_BA/ba21_fingerprint.pdfFingerprint Recognition Fingerprint Biometrics • The uniqueness of a fingerprint can be determined

1

Fingerprint Recognition

Fingerprint Biometrics

• The uniqueness of a fingerprint can be determined by the pattern of ridges and furrows as well as the minutia points.

• Minutia points are local ridge characteristics that occur at either a ridge bifurcation or a ridge ending.

2

Two Recognition Strategies

• Minutia based • Correlation based

Minutia Types• Ridge ending

– the point at which a ridge terminates

• Bifurcation– the point at which a single ridge

splits into two ridges • Short ridge (or dot)

– ridges which are significantly shorter than the average ridge length on the fingerprint

3

Recognition Steps

• Local Ridge Orientation• Segmentation• Singularity and Core Detection• Enhancement• Binarization/Thinning• Feature Extraction/Minutia Detection• Minutia matching

4

Local Ridge Orientation• An average of the angle formed by ridges within

a local window intersecting with the horizontal axis

Calculating Ridge Orientation

• The local ridge orientation is the hypothetical edge orthogonal to the direction of the calculated gradient phase angle for that region– Gradient phase angle

• Iy = image after Sobel-Y Convolution• Ix = image after Sobel-X Convolution• phaseimage = arctan( (Iy(i,j) / Ix(i,j) )

5

Segmentation

• Segmentation separates the fingerprint from the background.

• Calculate directional variance in shades of gray– Noise would not have any directionality

and gets assigned to the background– Ridges would have variance orthogonal to

the local ridge orientation and would get assigned to the foreground

• A quality index is also derived from the variance that alerts us to which regions are good, medium, or poor.

6

Singularity and Core Detection

• Optional step depending on implementation• Sum the orientation differences between

indexes of the orientation image angles in an 8 neighborhood surrounding a pixel.

• 360 degrees is a whorl has been detected• 180 degrees is a loop• -180 degrees is a delta

7

Singularity and Core Detection

Enhancement• Linear Contrast

• Gabor Filtering

G(x, y : θ, f) = e−1/2[(x2θ/σ

2x)+(y

2θ/σ

2y)] ∗ cos(2πfxθ),

xθ = xsin(θ) + ycos(θ)yθ = xcos(θ)− ysin(θ)

f is ridge frequencyθ is ridge orientation

8

A Fingerprint Image

Filtered Images

Gabor Filter Mean Filter Median Filter

9

Binarization

• Adaptive Thresholding works well• It accounts for the differences in

contrast over the different regions of the print.

Thinning

• The structuring element is placed at every possible pixel position. If they match then set center to 0.

• Apply the left then the right, then rotate each structure element 90 degrees 3 times.

10

Illustration of Thinning

Thinned Fingerprint Image

Based on Gabor Image Based on Median filteredImage

11

Pruning

• Apply these structure elements to prune spurs

Minutia Detection

• Crossing Number Method

Examples of 8 neighbor sets. cn (p)=2,cn (p)=3 and cn (p)=1 representing a nonMinutia region, a bifurcation and a ridge ending

12

Detecting False Minutia

• Once initial Minutia have been detected they need to have the false Minutia filtered out.

• There are less points to check so each one can be examined closer under a wider scope

• Consider– the length of associated ridges– the minutia angle– the number of facing Minutia in a neighborhood

13

Minutia Matching

Mapping Sets of Minutia

• Minutia matching can be thought of a mapping algorithm that maps a set I to a set M trying to maximize the matching of points in the two sets.

• Two points from a different set match if they are with physical and angular tolerance of each other

• This approach embeds alignment in the matching phase