Derawi Biometrics

Fingerprint recognition is the most matured approach among all the biometric techniques ever discovered. With its success of use in different applications, it is today used among many access controls as each individual have an immutable, unique fingerprints.

To understand how fingerprint recognition works, the knowledge of the hand skin needs to be analyzed. The hand skin or the finger skin consists of the so called friction ridges with pores. The ridges are already created in the ninth week of an individual’s fetal development live \cite{biohis3}, and remains the same all life along, only growing up to adult size, but if severe injuries occur the skin may not reconstruct the same as before. Researchers have found out that identical twins have fingerprints that are quite different and that in the forensic community it is believed that no two people have the same fingerprint \cite{biohis7}.

Acquisition Devices

There have been developed many new acquiring technologies over the last decades taking over the old ink technology. The old technology was based on sensing ridges on an individual’s finger with ink, where newer technologies uses a scanner placing the surface of the finger onto this device. Such technologies are referred to as live-scan and based on four techniques:

• Frustrated total internal reflection (FTR) and optical methods is a first live scan technology. In Figure 1 you see that the reflected signal is acquired by a camera from the underside of a prism when a finger touches the top of the prism. The typical image acquisition surface of 1 inch by 1 inch is converted to 500 dots per inch (DPI) using either CCD or CMOS camera. This kind of technology has been out in market for a couple of time now and was in the start mainly used for medical studies \cite{art16}. Some issues about these reflection technologies are that the light is a function of the skin characters. For instance, if the finger is either wet or dry, the finger can be difficult to process and therefore gives bad results. But these kinds of problems can be handled and fixed by using the so-called extent by FTIR imaging using ultrasound instead of visible light, but the resulting system is bulkier.

• CMOS Capacitance: The ridges and valleys create different charge accumulations, when a finger hits a CMOS chip grid. This charge is converted to an intensity value of a pixel using various competing techniques such as AC, DC and RF. The typical image acquisition surface of 0.5 inch by 0.5 inch is converted to 500 dots per inch (DPI). The resultant images also have a propensity to be affected by the skin dryness and wetness.
• Ultrasound Sensing: The thermal sensor is developed by using pyro-electric material, which measures temperature changes due to the ridge-valley structure as the finger is swiped over the scanner and produces an image. In this case the skin is a better thermal conductor than air and thus contact with the ridges causes a noticeable temperature drop on a heated surface. This technology is claimed to overcome the dryness and wetness of the skin issues of optical scanners. But the resulting images are not affluent in gray value images. The thermal sensor is becoming more popular today, because they are small and of low cost. Swipe sensors based on optical and CMOS technology are also available as commercial products.

Recognition Approaches

A fingerprint authentication system gives some sort of a similarity degree or sort of “distance” (dissimilarity) degree between two fingerprint images. Furthermore it should also give a report if these two measures are accurate and reliable against each other. For example, the two impression shown in the Figure 2 should be large or equivalently, whereas the distance between the images should be small.

Thus, the similarity between two impressions of the same finger should be invariant to (1) translation, (2) rotation, (3) the pressure applied and (4) elastic distortion between the impressions due to the elasticity of the finger skin.

The fingerprint comparison has been studied by many researchers over several decades. Two known classes of recognition techniques can be distinguished:

• Image Techniques: This class includes both optical and numerical image correlation techniques. Several image transform techniques have also been explored. These comparison techniques will become important when the area of the finger that is sensed is small (for example as with CMOS sensors). This technique is also known as the correlation-based technique and is demonstrated in the next section.
• Feature Techniques: This class of techniques extracts features (e.g. minutia points) and develops different machine representations of a fingerprint from these landmarks. This technique is the most widely used approach to fingerprint recognition, which is being described in more details later. This technique is also known as the minutia-based technique and is demonstrated in the next section.

Furthermore there exists a third recognition technique for fingerprints that combine the two above mentioned approaches:

• Hybrid Techniques: This class of comparison technique combines the image and feature or uses neural networks in interesting ways to improve accuracy.

Minutia

The uniqueness of a fingerprint can be determined by the pattern of ridges, also known as the minutiae points. Minutiae points are local ridge characteristics that occur at either a ridge bifurcation or a ridge ending, as illustrated in Figure 3.

• Endings: The points at which a ridge stops.
• Bifurcations: The point at which one ridge divides into two.
• Dots: Very small ridges.
• Islands: Ridges slightly longer than dots, occupying a middle space between two temporarily divergent ridges.
• Ponds or lakes: Empty spaces between two temporarily divergent ridges.
• Spurs: A notch protruding from a ridge
• Bridges: Small ridges joining two longer adjacent ridges
• Crossovers: Two ridges which cross each other
• Core: The inner point, normally in the middle of the print, around which swirls, loops, or arches center. It is frequently characterized by a ridge ending and several acutely curved ridges.
• Deltas: The points, normally at the lower left and right hand of the fingerprint, around which a triangular series of ridges center.

The techniques for fingerprint recognition can be placed into two categories:

1. Minutiae-based
2. Correlation-based

In the minutiae-based approach, a template is created consisting of all minutiae points.  Different types of minutiae’s that are stored into the template are coordinates, angles and qualities of the points. This technique, however, may in some cases give difficulties extracting the minutia’s if the fingerprint image is of a very low quality. Figure 4 summarizes the pre-processing of how to extract the minutia’s from a fingerprint image:

• Fingerprint Input: The Authentication Fingerprint is captured. The user will be advised to accurately position his finger on the device. The quality is measured and only good images will be accepted. Once an image is accepted, its relevant information will be extracted and stored into a database.
• Color Reduction: The extraction of information starts with the color reduction to black and white. Several filters reduce the noise in the relevant image areas without destroying required information.
• Mask filter: A mask is computed to be able to exclude noisy, un-sharp or blank areas from the recognition process. This mask is also used to determine the overall image quality.
• Thinning: Ridge width is reduced to one pixel based on the image from step “Color Reduction”.
• Feature Extraction: Minutiae locations are extracted. False minutiae are rejected by the quality mask.  Various features are extracted from each minutia. A template is now ready to be created.

The correlation-based approach is a measurement of image similarity and compensates variations of brightness, contrast, ridge thickness, etc. The method is able to overcome some of the difficulties of the minutiae-based technique.  However, it has some of its own shortcomings. For recognition it requires good image quality because the recognition is performed directly with gray-level fingerprint images. Figure 5 summarizes the pre-processing, which is based in \cite{art22}. The main steps are: normalization, low frequency noise filtering, and orientation field estimation and frequency estimation with their respective coherences, Gabor filtering and finally equalization.