Fingerprint Recognition Algorithms and Systems

• Lecture 5 is about fingerprint recognition algorithms and systems.
• The lecture covers introduction, fingerprint features, fingerprint analysis and representation, local ridge orientation, gradient-based approaches, enhancement, contextual filtering, minutiae detection, binarization-based methods and fingerprint matching.

Fingerprint Analysis and Representation: Introduction

• The skin on the palms and soles exhibits a flow-like pattern of ridges and valleys.
• This is different from the skin on other parts of the body.
• Papillary ridges on the finger, called friction ridges, help the hand grasp objects better.
• They increase friction and improve tactile sensing of surface textures
• The pattern of friction ridges on each finger is considered unique and immutable for biometric recognition.
• Fingerprints can differentiate even identical twins.

Features of Fingerprints

• Fingerprint details are characterized at three levels, from coarse to fine.
• Coarse level features can be derived from finer levels under ideal conditions.

Level 1 Features

• At the global level, ridge line flow delineates a pattern in fingerprints.
• Singular points, like loop and delta, act as control points around which ridge lines are "wrapped".
• Singular points and coarse ridge line shape help in fingerprint classification and indexing.
• Singular points are not distinctive enough for accurate matching.
• External shape, orientation image, and frequency image are global-level detectable features.

Level 2 Features

• At the local level, around 150 different local ridge attributes called minute details exist.
• Minute details are not evenly distributed.
• Impression conditions and quality heavily affect them, making them rarely observable.
• Ridge endings and bifurcations are the two most prominent ridge characteristics.
• A ridge ending is where a ridge ends abruptly.
• A ridge bifurcation is where a ridge forks or diverges into branch ridges.
• Minutiae are generally stable and robust.

Level 3 Features

• At the very-fine level, intra-ridge details, including the ridge width, shape, curvature and edge contours, are detectable.
• Other permanent details include dots and incipient ridges.
• Finger sweat pores locations and shapes are considered highly distinctive.
• Extracting very-fine details, including pores, is only feasible in high-resolution (e.g., 1,000 dpi) good quality fingerprint images.
• This representation isn't practical for non-forensic applications.

Fingerprint Images

• A fingerprint is a reproduction of the exterior appearance of the fingertip epidermis.
• Interleaved ridges and valleys form its most evident structural characteristic.
• Ridges (ridge lines are also called ridges!) are dark, and valleys are bright

Approaches of Fingerprint Analysis and Representation: Local Ridge Orientation

• Local ridge orientation at a pixel [x, y] is the angle θxy that the fingerprint ridges form with the horizontal axis.
• Fingerprint ridges aren't directed; θxy is an unoriented direction in within 0 to 180°.
• A fingerprint orientation (or directional) image is a matrix D.
• Its elements encode the local orientation of the fingerprint ridges.
• Each element θij, which corresponds to the node [i,j] of a square-meshed grid, is located over the pixel [x,y].
• It denotes the average orientation of the fingerprint ridges in a neighborhood of [x,y].

Gradient-Based Approaches

• Computing gradients in the fingerprint image is the simplest approach to extract local ridge orientation.
• The gradient at point [x, y] of I is a two-dimensional vector [∇x(x, y), ∇y(x, y)].
• ∇x and ∇y components are the derivatives of I at [x, y] by the x and y directions.
• Sobel and Prewitt operators provide achieve a good result with a simple approach .

Enhancement and Contextual Filtering

• Contextual filters are most used for fingerprint image enhancement.
• Conventional image filtering uses a single filter for convolution throughout the image.
• Contextual filtering changes the filter characteristics based on the local context.
• Typically, a set of pre-computed filters is used.
• One filter is selected for each image region.

Defining the Context

• Context in fingerprint enhancement is defined by the local ridge orientation and frequency.
• A local orientation and frequency varies slowly across the fingerprint area.
• It defines the sinusoidal-shaped wave of ridges and valleys.
• Tuned to the local ridge frequency and orientation.
• An appropriate filter can efficiently remove noise, preserve ridge and valley structure.
• Many contextual filters have been proposed.
• Intended filter behavior aims to provide a low-pass (averaging) effect along the ridge direction.
• It links small gaps and filling impurities due to pores or noise.
• The other aim is to perform a bandpass (differentiating) effect in the ridges' orthogonal direction.
• It increases discrimination between ridges and valleys while separating parallel linked ridges.

Gabor Filter-Bank

• Wan, Hong, and Jain (1998) suggested an effective method founded on Gabor filters.
• Gabor filters have both frequency-selective and orientation-selective attributes.
• Gabor filters also have optimal joint resolution in spatial and frequency domains.
• As shown in Figure 4.6, A Gabor filter is defined by a sinusoidal plane wave.

Gabor Filter-Bank Based Feature Extraction

• The even symmetric two-dimensional Gabor filter form defined as formula (4-1).
• In the formula, θ is the filter's orientation.
• [xv, yq] are after a clockwise rotation of the Cartesian axes by an angle of (90°-0).
• In the above expressions, f is the frequency of a sinusoidal plane wave.
• σx and σy are the Gaussian envelope's standard deviations along the x- and y-axes.
• The four parameters (Θ, f, σx, σy ) for the Gabor filters must be specified to apply the filters to an image .
• Local ridge frequency completely determines the filter's frequency.
• Local ridge orientation determines the filter's orientation.
• Selection of the values of σx and σy has involves a tradeoff.
• Higher values make filters more robust.
• Larger values are more likely to generate spurious ridges and valleys.
• Smaller values are less likely to introduce spurious ridges and valleys.
• They are less effective in removing the noise.
• Increasing σx and σy reduces the bandwidth of the filter.

Empirical Data

• Hong, Wan, and Jain (1998) set σx = σy = 4
• Figure 4.12 shows an example of thd filter set for no = 8 and nf = 3.
• Figure 4.7 reveals the application of Gabor-based contextual filtering on images.
• It shows images of both medium and poor quality.

Minutiae Detection

• Binary images go through a thinning stage.
• The stage lets ridge line thickness drops to one pixel.
• This produces a skeleton image.
• A simple image scan allows easy pixel detection that correspond to minutiae.
• This method uses a global threshold to set pixel to 0, the other one to equal 1.

Fingerprint Matching

• The fingerprint identification issues (search for an input fingerprint in a database of N fingerprints)
• It can be executed as N one-to-one comparisons (verifications) between fingerprint pairs sequentially.
• Fingerprint classification and indexing techniques speed up the retrieval in fingerprint identification problems.