Fingerprint Analysis (part 1) Pavel Mrázek
Dec 18, 2015
What is fingerprint• Ridges, valleys
• Singular points– Core– Delta
• Orientation field
• Ridge frequency
Small scale: Minutia• 150 types in theory• 7 used by human experts• 2 types for the machine:
– Ending– Bifurcation
SensingTraditional (off line): rolled ink impression
+ paper scan
• Plus: big area
• Minuses: – Inconvenient– Distortion– Too much/little ink
SensingOptical sensors
• Good: large area possible, good image quality, contactless scanning available
• Bad: size
Orientation fieldOrientation field (or ridge flow)
estimation:
• Crucial step before
image enhancement
• Various methods:– Gradient-based– Gabor filters– FFT
Orientation estimation• Gradient direction
– local characteristics
– same ridge orientation, opposite gradients
– more global view needed
• Classical solution: Structure tensor(second moment matrix, interest operator)– start from a 2x2 matrix
(positive semidefinite)
– safe to average information
Orientation estimationStructure tensor• Local:• Larger scale: average componentwise
(Gaussian window, linear/nonlinear smoothing)
• 2 nonnegative eigenvalues– both small: backgroung / low contrast– one big, one small: regular ridge area– both big: multiple orientations
(core, delta, scar)
Orientation estimationStructure tensor
• system of 2 orthogonal eigenvectors
• shows dominant direction
References• Maltoni et al.: Handbook of Fingerprint Recognition. Springer
2003.• Maltoni. A tutorial on fingerprint recognition. In LNCS 3161,
Springer 2005.• Hong, Wan, Jain. Fingerprint image enhancement: algorithm
and performance evaluation. IEEE PAMI 1998.• Zhou, Gu. A model-based method for the computation of
fingerprints’ orientation field. IEEE TIP 2004.• Weickert. Coherence enhancing shock filters. DAGM 2003.
• Contact: mrazekp -at- cmp.felk.cvut.cz