Biometrics Concepts
Introduction
Historically, the term biometrics is related to the Biological Sciences and refers to the application of statistical methods to a wide range of measurable characteristics in biology. Today, it is more commonly used in Information Technology, consisting of electronic identification of human beings based on their physical and behavioral characteristics.
Fingerprints
Fingerprints are considered one of the most reliable characteristics for human recognition due to their individuality and persistence. A fingerprint consists of patterns of ridges and valleys on the surface of the finger and its formation occurs in the early months of fetal development. Its main disadvantage is its intrusiveness, since people need to explicitly cooperate to provide their fingerprints to a biometric system. Moreover, fingerprint-based authentication has traditionally been associated with criminal procedures. The state of the art of fingerprint-based authentication methods demonstrates adequate accuracy and reliability.
Minutiae
A fingerprint is characterized by the pattern of ridge (dark lines) and valley (light lines) spacing. Generally, ridges and valleys run in parallel and sometimes end or bifurcate. At a global level, fingerprints may present regions with peculiar curvature patterns, called singularities. At a local level, there is another important feature called minutia that can be found in fingerprint patterns. Minutia derives from the Latin “small details” (minutiae), and this refers to a discontinuity behavior of the ridges, such as endings, bifurcations, trifurcations or other features like pores (small holes within the ridges), islands (two closed bifurcations), dots (small ridges), etc. Most systems use only bifurcations and endings. To properly compare (match) fingerprints, the fingerprint features, such as minutiae and singularity points, must be extracted. It is also possible to extract other global information such as ridge orientation and frequency of the fingerprint ridge regions. The images below show a fingerprint captured by a fingerprint scanner and the corresponding extracted features: ridges (thin black lines) and minutiae (blue arrows for endings and red arrows for bifurcations).
The process of interpreting these characteristics from digitized images is called extraction. This involves advanced signal processing algorithms and takes considerably more time than comparing the features of two fingerprints. Therefore, biometric systems try to perform this extraction only once and store the extracted features in a dataset called template fingerprint template.
Fingerprint Templates
The extracted features from fingerprints (minutiae, ridges, singularities) are stored in a template file. These can be stored in a wide variety of standard and proprietary formats. Standardized formats generally include only a minimal set of information, useful for exchanging fingerprint templates between different biometric systems.
Proprietary fingerprint template formats include pre-computed properties of the minutiae set that are specific to each ABIS system vendor, allowing faster and more accurate matches.
ABIS is an acronym for Automated Biometric Identification System
ISO
ISO (International Organization for Standardization) and IEC (International Electrotechnical Commission) published in 2015 the ISO/IEC 19794 family of standards, which is commonly applied to biometric data formats, standardizing common contents, meanings and representations of biometric data formats and biometric modalities, in addition to specifying which requirements address the complexities of applying biometrics to a wide variety of person recognition applications, whether those applications operate in an open systems environment or consist of a single closed system.
The ISO/IEC 19794-2 standard specifies data formats for representation of fingerprints using minutiae notation, defining three model formats for exchange and storage of fingerprint minutiae data. It defines a record-based format, a normal format and a compact format for smart card use. It also defines an extended format that can handle additional ridge count and core and delta locations.
The solutions of the Griaule biometric suite are compliant with major international standards and certifications related to biometrics. Among them, the highlights are:
ISO/IEC 19784 – Framework for biometric application programming interfaces (BioAPI).
ISO/IEC 19785 – Framework for biometric information interchange (CBEFF).
ISO/IEC 19794 – Biometric data formats for facial images.
This adherence ensures that Griaule's products meet international requirements for interoperability, quality and security in the processing and storage of biometric data.
ANSI
Fingerprint images have been widely used by agencies and police around the world for criminal identification. The introduction of the Automated Fingerprint Identification System (AFIS) by these agencies created a demand for standard formats for interoperability and information exchange between geographically distributed AFIS subsystems. Automatic Fingerprint Identification SystemsThese agencies created a demand for standard formats for interoperability and information exchange between geographically dispersed AFIS subsystems.
The ANSI 378/2004 standard specifies data formats for representation of fingerprints using the fundamental notion of minutiae. The data format is generic, and can be applied and used in a wide variety of application areas where automatic fingerprint recognition is involved. No specific application or characteristic requirements are addressed in this standard. The standard contains definitions of relevant terms, a description of how minutiae should be defined, a data format to represent fingerprint data and conformance data.
The biometric suite of the Griaule was designed to operate in compliance with the standards established by the American National Standards Institute (ANSI) and by the National Institute of Standards and Technology (NIST). Among them, the highlights are:
ANSI/INCITS 398 – Defines requirements for exchange of facial image data.
ANSI/INCITS 385 – Establishes the data interchange format for iris image data.
ANSI/NIST-ITL (including its updates) – Regulates the structure for exchanging biometric data in civil, criminal and forensic contexts.
ANSI/INCITS 381 – Defines the data interchange format for fingerprint image data.
ANSI/INCITS 378 – Establishes the standard for representation and interchange of fingerprint minutiae templates.
Compliance with these standards ensures that Griaule's solutions are prepared for integration into complex environments, offering compatibility with already consolidated systems and ensuring compliance with practices adopted by government agencies and international institutions
Template Consolidation
Two captures of the same finger are unlikely to be identical, showing small variations in the positioning of minutiae and in quality. Some minutiae can also appear or disappear due to small changes in finger positioning on the sensor in each capture.
To increase the quality of the model, a rolled capture can be performed, but this type of capture is not supported by all scanning equipment, and can present its own difficulties due to the rolling motion that the finger needs to perform. In these cases, several captures can be taken and combined to identify fingerprint minutiae producing a higher quality model.
The process of obtaining multiple captures and combining the extracted models into one is called template consolidation. This method provides several advantages: It is able to increase the overall quality of the minutiae; It can discard false minutiae incorrectly detected due to capture quality problems; It is able to combine minutiae found by subsequent captures, generating richer models.
Fingerprint Scanning
Capture Methods
There are essentially 4 different capture methods for fingerprint images:
Paper/ink fingerprint scanning
Generally required when importing legacy records into a digital biometric system. Prints made on paper are scanned with a flatbed scanner, with a resolution ranging from 500 to 1000 dpi.
GBS Cardscan Web is the component of the Griaule Biometric Suite designed for scanning ink fingerprint images and registering them in the Griaule Biometric Database Server.
Flat digital capture
The user's finger is pressed against a fingerprint digitizer, which produces a digital image of the area of the finger that is in contact with the device. Most digitizers produce 500 dpi images. Some high-quality devices provide images at 1000 dpi. While flat captures are faster, the captured area of the finger is limited, since it is not possible to position the full extent of the finger on the sensor.
Rolled digital capture
The user rolls the finger over the fingerprint digitizer, which produces a digital image with the area of the finger exposed on the sensor. The typical resolution is also 500 dpi and high-quality devices provide images at 1000 dpi. Rolled captures require smooth movements by the individual, which takes longer and may require more than one attempt if the individual rolls the finger too fast or abruptly. Rolled fingerprints contain more spatial distortions than flat fingerprints, due to skin deformation during the capture process, but capture a larger fingerprint area, and this allows biometric systems to compare fingerprints with greater accuracy.
Contactless live captures
This is an experimental fingerprint digitizer that attempts to capture the fingerprint image with a modified photographic camera, without requiring the subject to touch the sensor. While less intrusive, these digitizers do not guarantee an exact resolution (it varies according to the distance of the finger from the sensor), introduce lighting artifacts and can be susceptible to spoofing attacks (spoofing), such as presenting a photo of a finger to the sensor instead of a real finger.
Resolution
Resolution is the density of pixels per area, usually measured as dots per inch (dpi, in English dots per inch). The higher the sensor resolution, the richer in detail the captured image is, favoring better quality templates.
For fingerprint images, the resolution commonly required by security authorities is 500 dpi. This means that a device with a 1x1 inch sensor will produce an image of approximately 250,000 pixels (500x500 pixels).
Note that image dimension and image resolution are distinct measures: A device with a 2x2 inch sensor and a resolution of 500 dpi will produce images of dimension 1000x1000, but the resolution will still be 500 dpi. Both the sensor area and the resolution are important to ensure the quality of the extracted fingerprint features.
Image Quality
If the fingerprint image is noisy, blurred, or presents compression artifacts, the template extractor may not be able to correctly recognize the fingerprint features. For example: two ridges may end up connected due to a blur in the image, preventing detection of an ending minutia.
To mitigate dust particles deposited on the sensor, some readers provide built-in noise reduction technology. Nevertheless, to increase image quality it is always important to keep the device capture surface clean and follow good fingerprint capture practices.
To standardize the assessment of fingerprint image quality, NIST (National Institute of Standards and Technology) published NFIQ in 2015. NFIQ is a standardized measure of fingerprint image quality, designed to predict the performance of minutiae-based fingerprint matching systems. NFIQ classifies fingerprint quality into 5 classes: NFIQ 1 (highest possible quality) to NFIQ 5 (lowest possible quality).
Image Encoding
In many biometric systems it is desirable to store images for long-term purposes, such as re-extracting templates with more modern extractors, or for forensic and deduplication purposes.
Due to the large amount of images that an ABIS usually stores, and the quality required by template extractors to perform an extraction efficiently, traditional lossy compression algorithms are not suitable for encoding fingerprint images, since compression produces artifacts that can hinder template extraction. At the opposite side, lossless image compression algorithms generate very large files, increasing storage requirements.
For lossless compression the most common format for fingerprints is PNG (Portable Network Graphics).
For the specific needs of an ABIS the Wavelet Scalar Quantization format was developed, described below.
Wavelet Scalar Quantization (WSQ):
The Wavelet Scalar Quantization format was developed by the FBI specifically to compress fingerprint images, reducing the compressed file size without loss of detail, especially when compared with established algorithms such as JPEG/JFIF.
The WSQ format is based on wavelet theory, and was designed to compress grayscale fingerprint images at 500 DPI. Images with higher resolutions are better compressed with the JPEG2000 format.
The compression algorithms WSQ, PNG and JPEG2000 used by Griaule are certified by the FBI, ensuring interoperability, quality and reliability in civil, criminal and forensic applications.
Fingerprint Matching
The goal of template extraction is to perform the comparison (or matching) of fingerprints. Matching is the process of comparing two templates that produces a quantitative measure of similarity between them. In a functional biometric system, genuine pairs genuine (templates extracted from captures of the same finger) will produce high similarity scores, while impostor pairs impostor (templates extracted from captures of different fingers) will produce low similarity scores.
Since two templates are hardly identical, even if obtained from sequential captures of the same finger, the corresponding fingerprints must take into account changes in minutiae positioning and minutiae that do not exist in both models.
Scoring and Threshold
A matching algorithm has two steps. In the first, an Evaluation Algorithm assigns a similarity score for the comparison. The second step decides whether the pair is genuine or impostor using the decision threshold. If the score is higher than the threshold, the algorithm decides it is a genuine comparison; otherwise, it is classified as an impostor pair. The decision threshold is a configurable parameter of any biometric system.
Increasing the threshold will increase the occurrence of false negative errors (genuine pairs identified as impostors) and decrease false positive errors (impostor pairs identified as genuine); decreasing the threshold will reduce false negative errors and increase false positive errors.
When choosing the decision threshold for an application the user must take into account the acceptability of false negative and false positive errors, and the size of the template database.
Facial Recognition
Facial recognition is one of the friendliest and least intrusive methods among biometric modalities. Data capture is simple and capture devices are common, inexpensive digital cameras.
For a biometric system, facial recognition involves three tasks:
Detect and delimit the faces present in an image.
Extract a feature vector from a face, and encode it to store as a template.
Matching: Compare two face templates and compute a similarity score.
Modern facial recognition algorithms are based on deep neural networks, an artificial intelligence technique where the algorithm chooses the best discriminative features between faces based on a large training dataset. Given that features can change as training datasets are continuously updated, there is no standard format for facial templates. Biometric systems generally keep the original image stored so that improved templates can be extracted as the dataset improves over time.
Iris Recognition
Iris patterns are formed during fetal development and continue to develop during the first two years of life. Iris biometrics benefit from the fact that iris patterns are unique to each individual, and do not change after initial development. Unlike finger fingerprints, which are susceptible to changes from burns, cuts, abrasions and chemical reactions, iris patterns are very unlikely to be modified, intentionally or not. Iris capture requires cooperation of the individual, who needs to keep the eyes open and still while the capture is performed. Iris capture is performed with specialized scanners using near-infrared light (NIR) and sensors sensitive to near-infrared light. The image below is a typical example of an iris captured by a biometric sensor.
Although iris capture sensors are becoming cheaper as they are integrated into personal electronic devices such as smartphones, they are still more expensive than normal cameras used for facial recognition.
For a biometric system, iris recognition involves 3 tasks:
Detect and delimit the pupil, the iris and detect occlusion from eyelids or eyelashes.
Extract spectral features from the iris regions and encode them to store in a template.
Matching: Compare two iris templates and calculate the similarity score.
Error Rates
An important aspect of a biometric system is its accuracy. From the user's point of view, an accuracy error occurs when the system fails to recognize the identity of a registered person or when the system incorrectly recognizes the identity of an intruder.
There are basically three sources that increase the error rate in a system:
Human Factor
Capturing biometric information always requires human participation at different levels. For fingerprint capture, fingers must be in contact with the sensor surface; in iris image capture, the eyes must be looking at a fixed point of the digital camera. Even for facial captures, being probably the least invasive biometric capture, a good capture requires some restrictions on face positioning. As a consequence, human behavior at the time of biometric capture is a critical factor for processing stages and recognition accuracy. An inadequate capture causes poor extraction of biometric features or even prevents extraction. A critical point for evaluating biometric systems is to train people in good biometric capture practices.
Types of Error
False Positive errors occur when a biometric system classifies an impostor pair as genuine.
False Negative errors occur when a biometric system classifies a genuine pair as impostor.
Last updated
Was this helpful?