With the development of modern pattern recognition technology, some specific branches of pattern recognition technology have gradually matured, and many of them have reached the level of application. Of course, most of these applications are still based on limited environmental conditions. This paper introduces a sub-branch of pattern recognition specialty-biometric technology and its application in financial industry, and analyzes and compares them.
2. Biometric technology classification
The so-called biometric technology is: through the close combination of computer with optics, acoustics, biosensors, biostatistics and other high-tech means, using the inherent physiological characteristics of the human body, such as fingerprints, face images, irises and so on to identify personal identity. And behavioral characteristics such as handwriting, voice and gait. Traditional identification methods use identification items (such as keys, certificates, ATM cards, etc. ) and identification knowledge (such as username and password). However, due to foreign objects, once the identification items and identification knowledge that prove their identity are stolen or forgotten, their identity will easily be impersonated or replaced by others.
Biometric technology is safer, more confidential and more convenient than traditional identification methods. Biometric technology has the advantages of not forgetting, good anti-counterfeiting performance, not being forged or stolen, "carrying" with you, and being available at any time and anywhere. Biometric technology can be widely used in government, military, banking, social welfare, e-commerce, security and national defense. For example, a depositor walks into a bank, and he withdraws money directly without a bank card and password. When he withdraws money from the ATM, a camera scans the user's eyes, and then quickly and accurately completes the user identification and handles the business. This is the real shot that happened in a business department of United Bank of Texas, USA. The business department uses the "iris recognition system" in modern biometric technology. At present, the related technologies being studied at home and abroad can be divided into the following seven specific technologies: face recognition, iris recognition, fingerprint recognition, palmprint recognition, vein recognition, voice recognition, handwriting recognition, behavior recognition and video surveillance.
The first six of these seven technologies belong to the category of identity authentication, which is to verify who this person or object is. Another branch of behavior recognition is video surveillance, which is an auxiliary means of identity authentication. But it is found that it has more uses than identity authentication, so it gradually becomes independent.
2 Advantages of fingerprint identification technology:
1, strong practicability: fingerprint samples are easy to obtain, and it is easy to develop identification systems.
2. Reliability can be easily improved: by registering more fingerprints and identifying more fingers, accuracy can be improved.
3, good convenience: the speed of scanning fingerprints is very fast, and the use is very convenient;
4. Wide application: Fingerprint identification technology has occupied most of the market.
5. Fingerprint identification products are cost-effective: the fingerprint acquisition head is more miniaturized and the price is low.
7, the influence of attachments (glasses, beard).
8. Change the camera: The images taken by the same person with different cameras are different.
The main performance indicators of face recognition are: 1, false acceptance rate: this is the probability of mistaking others for designated personnel; 2. False rejection rate: This is the probability of mistaking the designated person for another person.
The thresholds used by computers are different, and so are these two indicators. In general, the error recognition rate increases with the increase of the threshold (relaxation condition), and the rejection rate FRR decreases with the increase of the threshold. Therefore, the error rate (equal error rate; ; ERR) as a performance indicator, this is the adjustment threshold, so that when the two indicators of FAR and FRR are equal, FAR or FRR. Compared with other biometric technologies, face recognition has one of the most obvious advantages, that is, it is easy to posterior. Basically, judging by human eyes can verify whether this person's identity authentication is wrong, while other technologies can't judge by such a simple method, and basically need the cooperation of experts and special equipment.
2.2 Iris Recognition According to clinical medical observation, the iris is located behind the cornea and in front of the lens and has a unique structure, and its color changes with the amount and distribution of pigment, and this unique iris structure has good stability. At present, the mainstream application of iris recognition system is that the system uses black-and-white TV and video combined with camera technology and software to obtain iris digital information, and compares the scanned information with the pre-stored template information for identity recognition.
Advantages of iris recognition technology:
1, biometric collection is more convenient.
2. High accuracy: According to statistics, the error rate of iris recognition is the lowest among all kinds of biometrics so far.
Disadvantages of iris recognition technology:
1, the application popularization scheme is low: iris recognition system has been tested a lot, but it has not been applied on a larger scale.
2. High cost: it is difficult to miniaturize the image acquisition equipment; At the same time, you need expensive cameras.
In the research of iris recognition technology, the research technology of the State Key Laboratory of Pattern Recognition of Institute of Automation, Chinese Academy of Sciences is at the international leading level. They not only have an international leading position in software algorithms, but also can design and manufacture iris recognition equipment by themselves, which is unique in the field of iris recognition all over the world. Even in 2004, they made a hand-held miniature iris acquisition device, which greatly reduced the hardware cost of iris equipment. In 2006, they also made a major breakthrough in the research of long-distance (more than three meters) iris collection equipment. If this research is successful, iris recognition can be carried out at a long distance without interference, and there is no need for such high-precision alignment requirements.
2.3 Fingerprint identification Fingerprint refers to the uneven lines on the front skin at the end of the finger. Medical proof shows that these lines are unique and permanent in patterns, breakpoints and intersections. At present, the mainstream application of fingerprint identification system is that users put a single finger on the prism surface or glass plate and scan it through CCD sensor. The obtained fingerprint image is digitized, processed and analyzed, and finally extracted as acceptable fingerprint digital feature information, and then stored in a memory or card as a reference template. When in use, the information scanned by the fingerprint reader is compared with the template information for identity recognition.
Advantages of fingerprint identification technology:
1, strong practicability: fingerprint samples are easy to obtain, and it is easy to develop identification systems.
2. Reliability can be easily improved: by registering more fingerprints and identifying more fingers, accuracy can be improved.
3, good convenience: the speed of scanning fingerprints is very fast, and the use is very convenient;
4. Wide application: Fingerprint identification technology has occupied most of the market.
5. Fingerprint identification products are cost-effective: the fingerprint acquisition head is more miniaturized and the price is low.
Disadvantages of fingerprint identification technology:
1, poor universality of fingerprints: the fingerprints of individuals or certain groups are difficult to image because of the few fingerprint features, which has a certain impact on the application of this technology.
2. Poor user acceptance: In the past, users used fingerprints because of criminal records, and they had certain psychological obstacles in using them.
2.4 Palmprint recognition Palm geometry is based on the fact that almost everyone's hand shape is different, and this hand shape will not change significantly after a certain age. When a user puts his hand on the hand reader, a three-dimensional image of the hand is captured, and then the shapes and lengths of fingers and knuckles can be measured and compared.
According to the different data used to identify people, hand shape reading technology can be divided into the following three categories: the pattern of hand blood vessels and the geometric analysis of palm and finger. Drawing different features of the hand is very simple and will not produce a large number of data sets. However, even if there are a considerable number of records, palm geometry may not be able to distinguish people, because the characteristics of hands are very similar. Compared with other biometric methods, palm geometry can not obtain the highest degree of accuracy. When the database is increasing, it is necessary to increase the obvious characteristics of hands in quantity in order to clearly identify and compare people with templates.
2.5 vein recognition vein recognition is a new technology that has emerged in recent two years. In this technology, the venous blood flow distribution map near the hand is mainly studied for identity authentication and identification. It generally uses infrared CCD camera to collect vein images. When the recognition system obtains the vein image of the back of the hand through the infrared CCD camera, it will store the feature code of each vein image. Then, the comparison and feedback results between the vein map of the field user and the stored vein map feature code are realized.
Features of vein identification system:
1, the vein recognition system depends on the state of the back of the hand.
2. Even if the back of the hand changes slightly, the recognition of the vein recognition system will be affected.
3. If the user suffers from arthritis or rheumatism, the vein identification system cannot be used.
4. Contact with this system can make you feel comfortable and convenient to the maximum extent. The system is superior to other biometric systems in humidity, sweat, dirt, pen marks and minor injuries.
Because its technology is very close to fingerprint identification, and its application scope and application environment are similar, many fingerprint identification companies promote and apply it as a new alternative update technology.
2.6 Speech Recognition The physiological, psychological and behavioral characteristics of people will be reflected in the speech waveform, and the spectrum of people's voice, including the time variation of curves and the characteristics of driving sound sources, is different. We can extract the characteristics that different people's voices change greatly but the same person's voice changes little for analysis, comparison and recognition. At present, the mainstream application of voice recognition system is to input human voice through microphone, digitize it through digital signal processor and compress it through software, extract voice image information and store it in database, and match the collected voice with the characteristic information in database for identity recognition.
Advantages of speech recognition technology:
1. Speech recognition is a non-contact recognition technology, which is naturally acceptable to users.
2. Speech recognition technology has good convenience, economy and accuracy.
Disadvantages of speech recognition technology:
1, low accuracy: the sound varies widely and it is difficult to match accurately.
2. High technical complexity: the volume, speed and quality of sound will be affected by certain conditions (such as colds), so it is necessary to increase system functions to adapt to this change.
3. High cost: Sound collection equipment (such as high-fidelity microphone) is very expensive.
2.7 Handwritten characters are like people, and people in China pay attention to calligraphy. After choosing their favorite calligraphy style, people integrate their own writing characteristics, ranging from the bookshelf structure of one word to the vertical and horizontal layout of the whole article. Everyone has his own writing habits and format planning; Handwriting has become one of the important means for people to identify themselves. At present, the mainstream application of handwriting (signature) recognition system is: the system uses wired pen and sensitive graphic input board to extract the dynamic process information characteristics of the signature, and determines the true identity of the signer by distinguishing the habitual part and the changing part of the signature.
Advantages of handwriting recognition technology: good acceptability: using handwriting recognition is a recognized identity recognition technology. Easily accepted by the public.
Disadvantages of handwriting recognition technology: 1, high technical complexity: with the change of people's temperament and lifestyle, the signature will also change, so it is necessary to increase system functions to adapt to this change.
2. Higher price: The handwriting board for signature is complex in structure and expensive.
2.8 Behavior recognition Behavior recognition technology is a video analysis system that monitors, classifies, tracks and counts objects. Behavior recognition technology is to analyze and judge according to certain rules, so as to set an alarm for specific behaviors.
Behavior recognition is a video technology based on some patented technologies, as follows: 1. Intelligent video recognition: A series of video image algorithms, which can be used to detect and compensate a series of changes caused by the environment and cameras: camera stability, background recognition, perspective accuracy, adaptive limit, shadow neglect and PTZ camera control.
2. Target segmentation: The engine can accurately segment the target from the background, ignoring the changes of non-target objects, such as the movement of trees and the changes of light. It can also divide the target group into individual targets.
3. Trajectory tracking: When the target object is detected over a certain time limit, the motion, trajectory and velocity functions of the target are established, so as to determine the size, quantity and shape of the object more accurately. The displayed target trajectory is updated in real time, so as to identify the intruder's invasion direction and post-invasion path.
4. Target detection: mainly judge the position, size and shape of the target, and filter non-target objects with high precision.
5. Behavior recognition: apply certain rules to recognize the position, speed and direction of the target; In addition, the number of targets can be judged.
6. Efficient development tools: The multimedia instruction set and advanced CPU with parallel running mode are developed, which gives the highest cost performance in the industry.
2.9 Video Surveillance Video surveillance is traditionally called intelligent video surveillance technology.
The application of intelligent video surveillance technology must be combined with other algorithms and technologies besides basic technologies such as moving target detection. In intelligent video analysis, image segmentation and moving target detection are basic problems. In recent years, many studies have been carried out around these problems, but they are still challenging topics. The core technical problems to be solved are motion blur, light change in background subtraction method, real-time requirements, occlusion and so on.
The early shot segmentation algorithm was carried out in the pixel domain, but this method is very sensitive to the fast motion of pixels, resulting in a large number of false detections. Later developed shot segmentation algorithm based on histogram difference between frames has become a popular method because of its low complexity and good segmentation effect.
Moving target detection and tracking is the basis of advanced applications of automatic or semi-automatic video surveillance, such as event detection, behavior recognition, video image compression coding and semantic indexing. At present, there are time differential method, background differential method and optical flow method for moving target detection.
Background subtraction method is a moving target detection method based on gray level, and the commonly used method is feature-based method. Feature-based detection is to detect moving objects according to the features (points, lines and moments) of images or models (polygons and polyhedrons) composed of features. It is often used when the target is large, the features are easy to extract, or there is a standard target model base for matching. Background subtraction method can extract a very complete target, but it is easily affected by background changes caused by illumination. In recent years, some statistical methods have been introduced to realize background modeling and background removal, which greatly enhances the robustness of background subtraction to noise such as illumination changes and shadows. There are many methods for background modeling according to model features, among which statistical model modeling based on pixel intensity can adapt to gradual illumination, but there are problems for abrupt illumination. As a simple method, Kalman filter has been widely used in target tracking.
Texture analysis based on optical flow field and motion parameter estimation is another commonly used motion region detection algorithm, but the reliability of optical flow is poor due to aperture and occlusion problems. The motion segmentation method based on Bayesian probability statistics can perform motion segmentation and motion estimation at the same time, and the effect is good, but the calculation is complex and large, which is not suitable for real-time processing.
At present, there are two kinds of digital video monitoring systems at home and abroad, one is based on digital video recording equipment, and the other is based on embedded intelligent video monitoring system. Embedded video surveillance system is a special computer system that takes application as the center and adapts to the strict requirements of application system on function, reliability, cost and volume. Digital signal processor (DSP) is a special processor for high-speed real-time processing of digital signals. Its processing speed is 10-50 times faster than the fastest CPU, and it has been widely used in video surveillance systems. Front-end integration, video digitization, monitoring networking and system integration are the development directions of intelligent video surveillance system, and digitalization and networking are the main characteristics of intelligent video surveillance development.
3. Application status of biometric technology
By analyzing the characteristics of biometric technology, we can see that the application objects of biometric technology are almost similar, but the application environment is only slightly different. However, due to the differences between technology and technology development, different technologies have produced different specific application scenarios. Here can be divided into the following aspects: 1, beyond defense authentication: used for access control, channel management, bank withdrawal ATM and other applications; 2. After-the-fact analysis and identification: the way of analysis and verification according to the characteristic information obtained at the scene after the crime, such as the fingerprint and DNA collection and analysis of the crime scene by the police; 3. On-site behavior analysis: According to the behavior of moving objects and the attributes of surrounding objects, the judgment and analysis results of object behavior are obtained, such as road traffic accident analysis, lane change judgment of vehicles running red lights, and alarm in prison restricted areas.
3. 1 bank ATM application analysis In view of the application of biometric technology in the financial industry, Beijing is going to apply face recognition analysis to bank ATM. Let's talk about face recognition technology.
Generally speaking, the process of adopting face recognition technology on ATM can be divided into two application forms: adopting face recognition as a means of identity authentication (belonging to transcendental defense authentication): when a person inserts a card near an ATM device, the camera takes a positive photo of the face during the process of inputting a password, and sends the photo back to the server for comparison and verification, and sends the information back to ATM, and then the ATM system judges the password at the same time to decide whether to carry out subsequent operations. Face as a means of data collection (which belongs to posterior analysis and identification): when a person inserts a card near ATM equipment, the camera takes a front photo of the face during the process of inputting the password, and the photo is sent back to the server for storage. The ATM system performs normal operation after judging the password.
Generally speaking, as a technician who has developed face recognition technology and has some knowledge of related technologies and equipment at home and abroad, we suggest that face recognition is mainly used as a means of data collection when authenticating related financial ATM equipment, which is the second form above. This is determined by the application environment characteristics of ATM equipment. The application environment characteristics of ATM equipment are as follows: 1, with a large number of users. Because the current face recognition technology rarely carries out large-scale comparative experiments on real face data, the current data experiments are all collected in the laboratory environment, the surrounding environment is relatively stable, and the angle and fit of the face are high, so it can not be used as the actual data reference for the existing practical environment application.
Most of them work outdoors and in a few sheltered places. Face recognition technology requires relatively strict light, especially visible light. Light is the difference of light source, such as direction, size, shape, color temperature, distance, light intensity and other factors, which can make the collected face images have many different differences, make the face look like a thousand faces, and thus make the recognition system misrecognize. Outdoor lighting conditions are too complicated, especially the solar spectrum will cover all visible and invisible light, so the research of outdoor face recognition equipment has been in an impossible state.
3.2 The feature analysis of biometric identification technology and the adoption of other identification technologies can only rely entirely on the judgment of the identification equipment itself, and it is impossible to make judgments directly through employees in general banks or financial industries. Here we can sort out some possible problems in related technologies: 1, iris recognition: the most accurate biometric technology, but the required alignment accuracy is high. In addition, when people are sick and take medicine, the iris will change in a certain period of time because of the medicine.
2. Fingerprint identification: Fingers may not be identified by fingerprint identification equipment because of oil stains or their own secretions.
3, palmprint recognition: the palm area is large, although the accuracy is much higher than the fingerprint, and even claims to be close to iris recognition, but there will be similar fingerprint problems. Its equipment is also much larger than fingerprint identification equipment.
4, vein recognition: Even if there is a slight change in the back of the hand, the vein recognition system will be affected. If the user suffers from arthritis or rheumatism, the vein identification system cannot be used.
5, speech recognition: illness, especially throat disease will make the vocal cords change and unrecognizable. At the same time, recording on tape and other recording equipment is also easy to be cheated.
6. Handwriting recognition: It will change because of people's mood and age, and it will also change people's handwriting habits after short-term special training.
7. Behavior recognition: At present, the recognition technology with the lowest recognition accuracy has not reached the level available in the laboratory, let alone commercial use, due to too many influencing factors.
3.3 Intelligent video surveillance technology Above, we analyzed the application classification of biometrics in authentication technology, and here we want to analyze intelligent video surveillance as an auxiliary authentication means.
At present, mature intelligent video surveillance technology can analyze the behavior of objects. At present, this kind of behavior analysis can't analyze the detailed behavior such as fighting, but it can analyze the overall behavior of the object, such as the shape track, color, shape, driving speed, speed change, volume and so on. Through the use of these analysis data, the following functions can be realized: 1, restricted area alarm: used in bank vaults and some areas where personnel need to be restricted. When someone enters the area without normal identity authentication, the system will automatically give an alarm, and save relevant images and video materials as evidence or start relevant capture equipment to capture and analyze the details.
2. Direction control: Monitor the moving direction of the financial management center and all objects around it. During a certain period of time, such as at night or during a certain alert time, no one is allowed to enter or get close to security institutions or related facilities. If someone approaches these security institutions or related facilities, the system can automatically alarm and save relevant images and video data as evidence or start relevant snapshot equipment to capture and analyze details.
3. Quantity calculation: Intelligent video surveillance technology can calculate how many moving objects are in the camera coverage, and can limit the number of people entering the vault and other areas. When the number exceeds the limit, even if the authentication is passed, the access control system cannot start to open the door.
The application of this technology can be used as an auxiliary means of authentication, and the authentication equipment can also be used as an auxiliary means of this technology to realize the security control of the whole area.
3.4 Possible application analysis in financial industry 3.4. 1 Bank ATM and other peripheral independent devices use face recognition technology for face capture and auxiliary authentication, but it is not the only authentication means. Intelligent video surveillance technology is used to obtain behavior and object characteristics that may damage ATM and other equipment.
3.4.2 Face recognition technology is used for face capture and auxiliary authentication in closed spaces such as vaults, but it is not the only authentication method. * * * Authentication can consider other biometric technologies to avoid possible threats such as the loss of keys and RFID cards. Images captured by human faces can be used as a posteriori technology in the future.
Using intelligent video surveillance technology to calculate the number of people near the entrance guard can be set by rules. For example, if there are more than two people within a few meters near the entrance guard, the entrance guard system cannot be opened even after authentication.
3.4.3 Bank business hall can provide special priority service for some VIP customers by analyzing the face image information obtained when waiting for business personnel to enter the business hall and getting the queuing number. At the same time, after comparing the two images of the person who went to the service window to handle business, we can find out whether the business person and the number taker are the same person and whether these two people are the owners of this card, so as to avoid the customer from suffering heavy losses due to the loss of the bank card.
4. Analysis of financial industry application scheme
Here we classify and analyze according to the size of the technical control area. Basically, the application scope of biometric technology can be divided into point-based access control and area-based control.
4. 1 Access control mainly refers to access control, such as the key points that need authentication, such as the vault doors of banks and the passages of important places. Access control is divided into general access control system and important access control system to analyze the system architecture.
4. 1. 1 General access control General access control is the management of garage doors, park gates, general office doors or some less important entrances and exits. In these scenarios, other high-precision authentication technologies related to face recognition (such as iris recognition or RFID card) can be considered for identity authentication. At the same time, intelligent video monitoring technology and DVR equipment can be used to record the behavior of personnel entering the site, providing basic data for auxiliary authentication analysis and subsequent analysis.
4. 1.2 Control of important entrances and exits The important entrances and exits are the entrance and exit management of important core institutions such as archives, treasury, finance office and general manager office. In this scenario, the most perfect authentication analysis system can be adopted: other high-precision authentication technologies related to face recognition (such as iris recognition or RFID card) are used for identity authentication, and intelligent video monitoring technology is used to control the number of visitors and related information, so as to realize the chain management of identity authentication and security alarm and avoid accidents caused by personnel coercion or outsiders intrusion. It can be said that such an access control system is basically impeccable.
4.2 Regional Security Management All access control technologies are part of the whole regional security management, and regional security management is an effective measure to ensure the safe and stable operation of relevant institutions in the financial industry without accidents. Here we will provide a set of regional security management measures that we think are effective.
In view of regional security management, it is suggested that the whole building/community monitoring and access control should be combined to realize the overall intelligent control management system.
We use a variety of biometric technologies to realize this system: 1, monitor fences and abnormal passages through intelligent video monitoring technology, record all abnormal situations and give an alarm in time; 2. For gates and entrances and exits, adopt the two forms mentioned in section 4. 1 for security defense; 3. For all objects entering the area, the camera fusion technology is used to track and record the whole process, marking the moving tracks and behaviors of all objects. When there is an abnormal situation, the source of relevant meaningful objects is immediately queried in reverse, so as to avoid further deterioration of the situation and directly control suspicious objects; 4. Each camera will have a local image and video storage space and a centralized storage space located in the computer room; 5. Intelligent video monitoring technology can greatly reduce the time and space occupied by DVR to store video images, effectively improve the utilization rate of equipment, and make the subsequent analysis more targeted and operable; 6. The weakness of the whole system mainly lies in power consumption. If the circuit is interrupted, the whole system will not work.
The above are the regional security control systems that we think are practical and feasible, and the related technologies adopted are mature and can be put into practice directly.
About the author: Bai Huidong, pen name Qing Run, is an independent software technical consultant and author of the book "Modeling and Realization of the Whole Process of Software Engineering". As a senior software architect/senior project manager of a domestic research institute, I participated in the research and development, management and planning of biometric technology products.