Biometrics technology has made great progress in recent years, but in order to make biometrics from theoretical research to practical applications, many research institutes need to break through and solve a series of key technologies. From a statistical point of view, human fingerprints, palms, irises and other physiological features are unique. Therefore, these features can be used as the basis for authenticating the identity of the user.

1. Biometric sensor technology

Biometrics can be measured by a certain principle and converted into digital signals that can be processed by computers. This is the main task of biometric sensors and the first step in biometrics. Most of the biometric features are image signals formed by optical sensors such as CCD or CMOS, such as faces, fingerprints, irises, palm prints, hand shapes, veins, and the like. However, iris and vein images require an active infrared source to obtain clear personality characteristics. Since the active light source can overcome the influence of visible light changes on biological characteristics, researchers have recently designed infrared imaging devices in the field of face recognition to overcome the intra-class differences in face patterns with illumination changes, thereby greatly improving people. The accuracy of face recognition.

Biometrics

In order to improve the ease of use, comfort and user acceptance of biometric systems, while ensuring the quality of biometric signals, in addition to being small and sophisticated, and cost-effective, biometric sensor technology has many areas for improvement. For example, 3D fingerprint sensor technology has been recently acquired through contactless methods. The core technologies of biometric sensors include:

1) Intelligent positioning technology

The biometric acquisition device must allow the user and the recognition system to be at appropriate distances and locations to capture qualified biometric signals. The most ideal solution is to let the acquisition device automatically determine the position of the user, and then actively adjust the optical system or directly move the collection device through the mechanical device, so that the requirements of the user can be reduced, and the collection method is more intelligent and user-friendly.

2) Human-machine interface design

The biometric acquisition system should be “people-oriented”, ergonomically designed, and interface between biometrics and acquisition devices. By developing user-based positioning technology, users can quickly find a suitable imaging location under some guidance. For example, in the existing face recognition and iris recognition systems, a mirror is usually installed on the collection device or a gaze point or a well-designed optical system is provided, and the user can find the position suitable for imaging relatively quickly through visual or voice feedback.

3) Optical system design

Mainly the design and processing of the optical lens set. If active light source illumination is required, a filter should be installed on the lens to set the active light source according to the imaging distance.

4) Mechanical control technology

The design of the electronic control unit including automatic zoom, mechanical unit design for program adjustment in accordance with the user's height and distance.

The core technology of biometric sensors also includes sensor circuit design; signal transmission and communication technology; anti-smashing alarm technology and organic integration with other technologies.

2, living body detection technology

In order to prevent a malicious person from forging and stealing other people's biometrics for identity authentication, the biometric identification system must have a living body detection function, that is, whether the biometrics submitted to the system are from a living individual. The biometric biometric discrimination technology utilizes people's physiological characteristics. For example, live fingerprint detection can be based on finger temperature, perspiration, and electrical conductivity. Live face detection can be based on head movement, respiration, red eye effect, etc. The in vivo iris detection can be based on the characteristics of the iris vibration, the motion information of the eyelashes and the eyelids, and the contraction and expansion reaction characteristics of the pupil to the intensity of the visible light source.

In addition, spectroscopy information based on biometric images is also an effective way to perform in vivo detection. For example, printed images form regular paper texture features that can be detected using spectral features. In addition, the living characteristics of biometrics can be detected in the form of human-computer interaction; the use of multi-modal biometrics can also increase the difficulty of forgery.

From the current technical level, the living body detection function has always been a weak link in the biometric identification system. Researchers have used fake fingerprints and human faces to break the existing system, which has caused some users to have a crisis of trust in biometrics. Therefore, the living detection technology will be the biggest bottleneck for biometric systems to enter high-end security applications.

3. Biometric signal quality evaluation technology

In an automatic identification system, biometrics are typically acquired in the form of a continuous video stream or audio stream. Since the effective biometric collection range is always limited, coupled with human motion, posture changes and other factors, most of the biometric signals transmitted to the computer are unqualified. High-quality biometric signals are the basis for feature representation and identification. Low-quality biometric signals may cause false reception or false rejection, reduce system stability and robustness (system robustness), and waste a lot of The computational resources are on invalid biometric signal processing.

Based on the above analysis, we can try to eliminate the impact of low-quality biometric signals on recognition performance in three aspects:

â–  Research high-performance imaging hardware platform

â–  Improve the robustness of the recognition algorithm

â–  Introducing intelligent quality assessment software modules in biometric systems that only allow for the registration or identification of higher quality biometric signals.

It is most practical to design an effective quality evaluation algorithm among these measures. Because the signal quality that the robust identification algorithm can accept is also limited. Although high-performance biometric acquisition devices have been available, they are expensive and cannot solve the underlying problem. Therefore, studying the quality evaluation algorithm of biometrics is of great significance for improving the performance of the recognition system.

The quality evaluation of biometric signals can be seen as a two-class pattern recognition problem—dividing the collected biometrics into qualified and unqualified cases. If the quality of the qualified signal is to be scored, the evaluation index should also be quantified. The problem of quality evaluation of biometric signals is a difficult problem, because the causes of poor quality of characteristic signals vary widely, that is, there are too many types of negative samples, and it is difficult to design a classifier to distinguish all positive and negative samples. . Low-quality biometrics that need to be filtered by quality assessment generally include images with defocus or motion blur, signals with too low a signal to noise ratio, images that are occluded, and the like. The quality evaluation algorithm can generally be designed from both the airspace and the frequency domain.

From the perspective of product practicality, one of the biggest bottlenecks encountered by biometric systems is the quality evaluation of signals. On the one hand, in order to broaden the scope of application of the system and improve the ease of use of the product, it is more user-friendly. For this reason, the researchers hope that the system can operate under the conditions of low biometric quality requirements, but at the same time require the system to be stable. High precision. In order to balance this contradiction, designing a "stable, fast, accurate" quality evaluation algorithm will be the only way.

4. Biosignal localization and segmentation technology

The treated palm prints are clearer. The original signal collected from the biometric acquisition device generally includes not only the biological features themselves, but also background information, such as the original iris image including iris, pupil, sclera, eyelid and eyelashes, which can effectively identify people. The image content is also in the iris area. Therefore, the content of interest must be segmented from the original signal for feature extraction. Positioning and segmentation algorithms are generally based on a priori knowledge of biometrics in terms of image structure and signal distribution. For example, face detection is to find and locate the face area from the image, which has always been a research hotspot in the field of computer vision.

In 2001, Viola and Jones in the United States proposed the use of easy-to-calculate Harr wavelet features to describe face patterns, and AdaBoost to train face detection classifiers, which made breakthroughs in the field of face detection and realized real-time detection of video. Face images, and the accuracy is very high. This method has a great impact on the field of computer vision and biometrics. Nowadays, commercial face recognition systems basically use this face detection method or its variants. Moreover, this method of training weak classifiers through machine learning has also been extended to the detection and recognition of general visual objects. Fingerprint segmentation algorithm is generally based on the difference between the grayscale variance of the image block and the background region. The iris is mainly positioned by the pupil/iris/sclera with large grayscale jumps and rounded edge distribution structure features; The positioning of the lines is generally based on a reference point between the fingers to construct a reference coordinate system.

5. Biometric signal enhancement technology

After the segmented feature regions are obtained, some biometric methods need to enhance the region of interest before feature extraction. The main purposes include denoising and highlighting feature content. For example, face and iris images generally use the histogram equalization method to enhance the contrast of image information; fingerprints generally use the frequency domain method to obtain the frequency and direction features of the ridge line distribution and then perform texture enhancement. For relatively fuzzy biometric signals, consider Enhancement using a super-resolution method or a reverse filtering method.

6. Calibration technique for biometric signals

In order to overcome the translation, scale and rotational transformation between biometric signals acquired at different times, it is necessary to align the two biometrics participating in the alignment. Some biometric calibrations are performed before feature extraction, such as common active shape models and active appearance models for face alignment; some biometric calibration processes are the process of feature matching. The calibration result of the biometric signal has a great influence on the recognition accuracy, so some scholars believe that the most important problem of biometric recognition is the calibration technique.

7. Biometric expression and extraction techniques

For biometrics, whether it is layman or layman, the first question people think of is what features are used to identify the machine? What is the essential feature of the individualized differences in biometric signals? This is the basic, principled problem of biometrics. For this problem, there is a consensus in the field of individual biometrics, such as fingerprint recognition. It is recognized that the detail points (including the tip point and the bifurcation point) are the best way to describe the fingerprint features, so there is a unified basis based on the international The fingerprint feature template exchange standard of the detail point information brings convenience to the compatibility and data exchange of fingerprint identification systems of different manufacturers. However, in other fields of biometrics, such as face, iris, palm print and other fields, researchers continue to explore the best feature expression models. Although there are many kinds of feature expression methods in these fields, some algorithms have achieved good recognition performance, but the fundamental problem of face recognition, iris recognition, palmprint recognition - "What is a face, iris or palm print image? The essential characteristics and their effective expression?" has never been answered with authority and universal approval.

This is because the feature expression methods of each face, iris and palmprint image are based on a certain signal processing method or a certain computer vision or a pattern recognition theory, "public theory is reasonable, woman is reasonable", everyone The intrinsic characterization of these images has not been studied in depth. Nowadays, the trend in the field of biometric expression is to test all kinds of classical or newly proposed image analysis methods in turn, and it is a bit of a feeling of big luck. The root of this phenomenon is that everyone has no guidance on basic theory. Work hard in the direction. Since the various methods are “political”, it is difficult to unify and standardize the data exchange format of the biometric template. For example, data exchange standards for faces, irises, and palm prints can only be based on images, because you can't find a unified, authoritative representation of image features.

Compared with feature-based data exchange standards, image-based switching standards are inferior in many aspects such as occupancy and transmission rate of computing and storage resources. For example, in the ePassport application, the unified format of the biometric data is stored in the non-contact IC chip, and the biometric data needs to be read out from the passport IC through the wireless card reader before the identification, and the feature-based method is more image-based than the image-based method. It's 100 times faster, and the image-based approach requires an additional feature extraction step to get the biometrics in the user's passport. Therefore, whether for research or application, the determination of the essential characteristics of biometric signals is the most important.

By simulating the information encoding rules of these biological nerve cells for external visual stimuli, computer vision researchers have proposed Ordinal Measures to express image content. The Center for Biometrics and Security Technology of the Institute of Automation of Chinese Academy of Sciences proposed the concept of multipole filter by expanding the connotation of the original sequencing measurement features, and established a general framework for the expression of iris image features, which proved the order between iris image regions. The measure feature is equivalent to the order relationship between the reflectances at different positions on the physical surface of the iris, and is an essential feature of the iris image independent of external factors such as illumination and contrast.

In this framework, iris feature extraction can even be simplified into simple addition and subtraction operations, successfully solving the computational complexity problem of iris recognition from PC to embedded platform. Through the sequencing measurement characteristics, the research center also established a general framework for the expression of palmprint image features, and unified the three palmprint recognition methods with the best recognition performance in the field. Aiming at the characteristics of the main line and wrinkle line gray mode on the low-resolution palmprint image, a novel cross-shaped differential filter is proposed to extract the sequencing measurement features in the palmprint image. The experimental results show that the new palmprint recognition method not only has higher recognition accuracy than the mainstream method, but also calculates twice as fast as the best method.

8, biometric matching technology

Feature matching is the calculation of the similarity between the feature vectors of two biometric samples. The graph matching algorithm is also successfully applied in the similarity measure of fingerprint minutiae mode, face mode, and iris plaque mode.

9. Biometric database retrieval and classification technology

With the popularity of biometrics in human daily life, the growth in the number of users will inevitably lead to the expansion of the biometric database. This scale of expansion is not only reflected in the expansion of data storage, but also in the increase in the time it takes to search for a record from the database. For example, in a biometric application of one-to-many hyperscale (such as a city, a country, or an industry), the length of time to complete a recognition will be unbearable. This is an inevitable problem when any mature biometric technology transforms from small-scale applications to large-scale applications.

Although parallel computing technology can be used to reduce the time of each recognition, if there is a biometric rough classification method, hierarchical biometric recognition can be realized: according to the biometric vector, all the templates in the database are divided into several large categories, which are large. When the scale is identified, the large class to which the biometrics belong is first judged, and then the database template of the large class is first compared, so that the time for waiting for the recognition result can be reduced (at least from the expected value). For example, the fingerprint can be divided into several categories according to the number of singular points and the position information, such as an arch shape, a pointed arch shape, a left-handed shape, a right-handed shape, and a spiral shape. In the field of iris recognition research, the fractal database is also used to divide the iris database into four categories. The accuracy of these classification methods is higher than 90% and the results are encouraging. The use of biometric models can also achieve ethnic classification, gender classification, and the like. Therefore, the biometric rough classification and database retrieval technology will be a promising research direction. The focus of the next research is to increase the number of categories and improve the accuracy of classification.

Biometrics

10. Performance evaluation of biometric identification system

To date, any biometric system or method has the potential to go wrong. It is a very complicated problem to give objective and accurate evaluation of the recognition accuracy of the system. It is affected by factors such as the quantity, quality and evaluation index of the test samples, but this is a concern for the application unit and the judicial department. Focus. Therefore, the performance evaluation of biometric methods has become an important direction of biometrics research. For the 1:1 alignment authentication system, there are two cases of error: one is to identify different people's biometrics as the same class, called error reception; the other may be to identify the same person's biometrics as different classes. , called error rejection.

The performance indicators of a biometric method can generally be evaluated from both theoretical and experimental aspects. From the theoretical aspect, we can study the uniqueness of biometrics, that is, accurately model the various parameters that affect false reception and false rejection, and give theoretically the error rate that can be obtained from the nature and mechanism of each biometric identification method. Lower bound. This work is very meaningful and difficult. For example, the judiciary still has a lot of controversy about identifying criminals through fingerprint matching results. Although some researchers claim that there are no two people with the same fingerprint characteristics on the earth, how much is needed in automatic or artificial fingerprint identification systems. Can the similarity fully confirm the homology of the two fingerprints? What is the exact probability of identifying an error? Researchers have conducted in-depth research on this issue, but have not completely solved this problem.

Chaff Cutter

Chaff Cutter,Chaff Cutter Machine,Mini Chaff Cutter,Hand Chaff Cutter

Hunan Furui Mechanical and Electrical Equipment Manufacturing Co., Ltd. , https://www.thresher.nl