Biometrics and business application prospect analysis

HC Security Network News recently, First Research Institute of Ministry of Public Security announced on the National Cyber Security Awareness Week 2016 activities, developed their own "network trusted identity authentication service platform" will be put into pilot and more.

1. From the online copy of the ID card, biometrics is on the rise.

Recently, the First Institute of the Ministry of Public Security released a message at the 2016 National Cyber ​​Security Publicity Week. The “Network Trusted Identity Authentication Service Platform” developed by the Ministry of Public Security will soon be put into trials in many places. Through this platform, individuals can generate a lifetime ID number of “ID card online copy” on the Internet. In the future, it is not necessary to carry an ID card. When you check in hotels, open online stores and other businesses that require real-name authentication, you can complete it by “brushing your face”. Identity authentication while avoiding the risk of identity being spoofed.

The technical principle of face recognition is face recognition in biometrics. The so-called biometric technology is the combination of computer and optical, acoustic, biosensor and biostatistical principles, using the inherent physiological characteristics of the human body (such as fingerprints, faces, irises, veins, etc.) and behavioral characteristics ( Such as voice, gait, etc. to identify the identity of the individual.

Since the human body features the inherent non-reproducibility of the human body, this biometric key cannot be copied, stolen or forgotten, and biometric identification technology is used for identity identification, which is safe, reliable and accurate. Common passwords, IC cards, stripe codes, magnetic cards or keys have many disadvantages such as loss, forgetting, copying and theft. In addition, biometric technology products are realized by means of modern computer technology, and it is easy to cooperate with computers and security, monitoring, management system integration, and achieve automatic management.

For a long time, insufficient technological innovation, limited application and high price and high cost are the three major factors that restrict the development of biometrics in China. In terms of technology, the original innovation of biometric sensors and core algorithms is insufficient, and the technical standards are not well developed. On the application side, biometrics are basically limited to access control attendance, bank internal control, and public security departments' entry and exit management and criminals. To personal consumption; in terms of price, lack of market competition and immature technology lead to high prices. However, in recent years, this situation has begun to turn, and the three major problems are gradually being resolved.

Technology: The maturity of deep learning algorithms has led to a significant increase in the accuracy of biometrics. Advances in computer technology based on deep learning algorithms provide powerful computational and analytical tools for biometrics. In turn, the huge amount of biometric data also provides a wealth of material for machine training, "big data becomes the fuel of artificial intelligence." Face recognition, Face++ team created the world's highest face recognition accuracy rate, has been in the face detection FDDB evaluation, face key point positioning 300-W evaluation and face recognition LFW evaluation, won three consecutive World number one. In terms of fingerprint identification, domestic manufacturers such as Mindray, Aegis, and Weier Technology already have fingerprint recognition autonomic algorithms and core patents. In addition, third-party fingerprint identification algorithm providers such as Hangzhou Qianyuan and Shenzhen Zhixin have emerged. . In terms of iris recognition, Tiancheng Shengye has launched the world's smallest iris recognition module, which can realize natural iris recognition by using a module to achieve natural reading distance (7-30cm). In terms of vein recognition, the School of Electronic Science and Engineering of National University of Defense Technology has successfully developed a vein identification security system.

Application: At present, the application scenarios of biometrics have been greatly expanded. The bank applies biometric identification in the customer identity verification scenario, covering different risk level scenarios such as weak real-name electronic account opening, settlement account opening and deposit and withdrawal. Mobile payment applications such as third-party payments and mobile banking have begun to use biometrics. The public security department has vigorously introduced technologies such as fingerprints and face recognition in video surveillance and personnel access management in various types of places. Under the policy push, biometrics has also been placed in the social security, education, and medical systems. In recent years, domestic smart phone consumption has increased dramatically, mobile phone users' mobile payment habits have gradually developed, and smart homes have become popular. The growth of personal consumption demand in the three aspects has pushed consumer-grade fingerprint recognition on the mobile phone to begin to break out, and the iris has gradually entered.

Price: In terms of fingerprint identification, the price of fingerprint recognition module has been declining in the past two years. In terms of face recognition, Hanwang’s 500-person face recognition attendance machine cost more than 4,000 yuan when it was first introduced in 2009. It was reduced to about 3,000 yuan when the second generation was launched in 2010. Dropped to around 2,700 yuan.

Benefiting from the gradual solution of the three major problems, the domestic biometrics industry is welcoming unprecedented development opportunities, and its application scope and market size are expected to achieve rapid expansion.

2. Preliminary study on biometric technology

2.1, the minimum system and basic working process of biometric identification

Typically, a biometric minimum system consists of three parts: Sensor, memory, and processor. The sensor is the collection mechanism of the user's biological information; the processor is responsible for information preprocessing, feature extraction, feature training, feature matching and feature recognition; the memory is responsible for feature extraction and storage of training results. The biometric work process includes two stages: user registration and identity authentication. The process mainly includes four steps: biometric collection, preprocessing, feature extraction and pattern recognition:

1) Feature collection

Feature acquisition is the process of using sensors to convert the inherent physiological characteristics of the human body into computer-readable digital information. The biometric sensor mainly uses high-precision optical devices such as scanners and cameras, as well as crystal sensor chips based on capacitance and electric field technology, ultrasonic scanning, and infrared scanning.

2) Pretreatment

Preprocessing is the process of normalizing the digital information read by the sensor and processing the raw data into semi-structured data. Generally, preprocessing methods mainly include information compression, noise reduction, and data normalization.

3) Feature extraction

Feature extraction is the most representative part of extracting biometrics and is the process of converting it into structured data. Feature extraction and expression are the key points and difficulties of the biometric recognition process. For example, iris, fingerprint, face and other images are prone to uneven illumination, plane rotation, partial occlusion and three-dimensional deformation. This brings great obstacles to subsequent feature matching and pattern recognition, which may lead to recognition failure and failure to identify.

4) Pattern recognition

Pattern recognition usually includes feature training, feature matching and feature recognition. It is a process of constructing classifiers through machine learning and data mining to match and identify structured feature vectors. Among them, the feature training is to learn the biometric data set collected by the user registration stage through the machine learning method to generate the biometric classifier model; the feature matching is the biometric database model generated by the biometrics extracted during the identity authentication phase and the user registration phase. Matching is performed, and the similarity between the two is calculated; the feature recognition sets the identification criteria such as the similarity threshold, and accepts or rejects the recognized result.

2.2. Analysis of six types of biometrics

At present, the main mainstream of biometrics are fingerprints, face, iris, vein, voiceprint, and gait recognition. This chapter will briefly introduce their respective technical principles, technical features and current development stages.

Fingerprint recognition

Automatic Fingerprint Identification System (AFIS) is the earliest and most mature biometric technology used in pattern recognition. It integrates sensor technology, biotechnology, electronic technology, digital image processing and pattern recognition.

Figure: Sketch of the fingerprint identification system

In general, the AFIS system consists mainly of the following components:

(1) Fingerprint image input

Fingerprint images are acquired in the AFIS system in two ways, optical scanning acquisition and solid-state sensor acquisition. Optical scanning capture fingerprint images generally use total reflection technology (FTIR). When the finger is placed on the prism, the ridge of the finger contacts the prism and the valley does not contact the prism. The laser illuminates the prism at a certain angle to generate total reflection, and the CCD array receives and acquires the fingerprint image. When the solid sensor captures an image, the voltage of the capacitor is changed by the finger on the surface of the sensor to acquire an image. Compared with optical scanning, solid-state sensors have the characteristics of small size, high integration, digitization, etc., but the acquisition range is small, and optical scanning is rarely limited in the acquisition range.

(2) Fingerprint image enhancement

There are many kinds of feature definitions of fingerprint images. The definition of the most commonly used detail features is the detailed model proposed by the American FBI. It divides the most salient features of the fingerprint image into ridge end points and bifurcation points. Each clear fingerprint typically has 40 to 100 such details. The performance of the detail feature extraction algorithm depends heavily on the quality of the input fingerprint image. In order to ensure the performance of the detail feature algorithm, the fingerprint image enhancement needs to be performed first before extracting the feature. The purpose of fingerprint image enhancement is to improve the sharpness of the ridge information of the recoverable area while deleting the unrecoverable area. In general, image enhancement is performed using digital image processing methods such as smoothing, filtering, binarization, and refinement.

(3) Fingerprint image feature extraction

Matching two fingerprint images largely adopts a method based on comparing dot patterns of two fingerprint images. The points used for matching can be divided into two categories: minutiae points and individual points. The points used to match the fingerprint image are called minutiae points, which are the ridge end points and bifurcation points in the fingerprint image topology. Individual points include center points and triangle points, and the distance between them and the number of ridge lines are generally considered not to change with image transformation, rotation, magnification, and reduction. Therefore, this feature is often used to reduce the search space of the database at the time of matching.

(4) Fingerprint image matching

Fingerprint image matching is to determine whether the feature set (template) of two input fingerprints belongs to the same fingerprint. There are many methods for fingerprint matching algorithms, including image-based matching, ridge pattern matching, point pattern matching, and graph-based matching.

Although the fingerprint identification technology is mature and the cost is relatively low, there are still shortcomings such as being easily forged, having low precision, poor concealment, and being easily affected by the state of the examiner and the environment.

Ultrasonic sensors have also been used to acquire images, which are acquired using ultrasonic reflection ranging. Ultrasonic fingerprint technology can be scanned through a smartphone case made of glass, aluminum, stainless steel, sapphire or plastic. The scan can be protected from dirt that may be present on the finger, such as sweat, hand cream or condensation, providing a more stable and accurate method of authentication. In addition, the ultrasound-based solution uses sound waves directly across the surface of the skin to identify three-dimensional details and unique fingerprint features that are currently unrecognizable by capacitive touchscreen-based fingerprinting techniques, including fingerprint ridges and sweat pores. This produces a fingerprint surface map that is rich in detail and difficult to copy.

Face recognition

Face recognition technology is a kind of biometric recognition technology based on human facial feature information, including face image acquisition, face location, face recognition preprocessing, identity confirmation and identity search. The technology combines knowledge and related technologies in biology, psychology, cognition, pattern recognition, image processing, computer vision, etc. It can be widely used in identity verification, identity authentication, access control, security monitoring, and human-machine exchange. Scenes.

The engineering application of face recognition began in the 1960s. After more than 50 years of development, face recognition technology has made major breakthroughs, and many classic algorithms and face libraries have appeared one after another. At present, the highest correct rate of face recognition system can reach 99.5%, and the correct rate of recognition of human eyes under the same conditions is only 97.52%, and the accuracy of face recognition has been more accurate than the naked eye.

At the same time, with the development of high-performance computers, many automatic identification systems for machines have been introduced at home and abroad, prompting face recognition technology to enter the practical stage. The more successful commercial face recognition systems in foreign countries include FaceIt developed by Identix, AcSysFRS of AcSys, and EPL of EyeMatic.

China's face recognition technology ranks first in the world. At present, the world's highest face recognition accuracy rate is created by our Face++ team. At the same time, due to the special national conditions of China's large population base, the government and industry have more urgent needs for face recognition technology, which promotes the commercialization process of face recognition in China, and the face recognition technology and products have been applied to the government. , military, banking, social security, e-commerce, security and other fields.

Compared with other biometrics, face recognition technology has unique technical advantages in practical aspects, mainly reflected in the following aspects:

1. Non-contact: The collection of face images is different from fingerprints and palm prints. It is necessary to contact the palm-grain collection equipment. The collection of palm prints is not hygienic except for certain wear and tear on the equipment. The device for collecting face images is a camera and does not need to be touched.

2, non-intrusion: the collection of face photos can be automatically taken using the camera, without the need for staff intervention, and does not need to be matched by the collector, just in front of the camera through the normal state.

3. Friendly: A human face is a biological feature that is exposed after a person is born, so its privacy is not as strong as palm prints and irises, so the collection of faces is not as unacceptable as the palm print collection. .

4, intuitive: We judge who is a person, by looking at this person's face is the most intuitive way, unlike the palm print, iris and other related fields experts can be judged.

5, fast: the face collection from the camera surveillance area is very fast, because its non-intervention and non-contact, the face acquisition time is greatly shortened.

6, simple: face acquisition front-end equipment - camera can be seen everywhere, it is not a dedicated device, so easy to operate.

7. Scalability is good: its collection end can completely adopt the camera equipment of the existing video surveillance system. The scalability of the back-end application determines that face recognition can be applied to access control, blacklist monitoring, face photo search, etc. field.

Iris recognition

Each iris contains a unique structure based on features such as crowns, crystals, filaments, etc. No two irises are the same. Iris recognition technology uses the characteristics of iris lifetime invariance and difference to identify identity. The iris recognition system is mainly composed of iris image acquisition, image preprocessing, feature extraction, feature matching, and conclusion.

(1) Acquisition of iris images: The primary task of iris recognition is to obtain high-quality iris images, which is one of the most difficult technologies for iris recognition. Since the iris is distinguished mainly by the difference in texture details, and the texture of the iris is not very clear, especially for yellow people, it is difficult to obtain a clear iris image with a normal CCD camera and under normal lighting conditions. There are currently two representative iris image ingesting devices. One is to adjust the LED light source, and the digital camera captures an eye image with a resolution of 256×256 pixels from a distance of 30 to 50 cm. The other is a small portable opto-mechanical device, the upper end of the device casing is fastened to the subject's eyes, and under the illumination of the infrared emitting tube and the light-emitting diode, the iris image is collected by the CCD camera and input into the computer.

(2) Iris image preprocessing: Preprocessing is a prerequisite for effective feature extraction and feature matching. Its main task is to separate the iris image from the image of the ingested eye. For example, adaptive Wiener filtering is used to remove the noise interference due to reflection, localization of the iris image by histogram equalization, accurate positioning of the inner and outer edges of the iris by Hough transform, and drift elimination by translation and image alignment. , scaling, and rotation to make the iris image structure consistent.

(3) Feature extraction: Iris feature extraction uses complex two-dimensional Gabor wavelet to decompose the iris image.

(4) Feature matching: After obtaining the encoding of the iris image, it is necessary to identify the encoding to identify the authenticity of the identity. There are many algorithms for image matching technology, such as correlation function measurement algorithm, correlation coefficient measurement algorithm, and hierarchical search decision algorithm.

In general, iris recognition has extremely high accuracy. If the threshold is set to be reasonable, the false recognition rate of iris recognition is almost zero, and it is extremely difficult to forge, and has good living body recognition characteristics. However, there are still problems such as expensive equipment, difficult collection, and the technology is not yet fully mature.

Vein recognition

Vein recognition technology is a biometric recognition technology. It uses the hemoglobin in human venous blood to absorb light from near-infrared light. Through vein recognition algorithm, it realizes the identification of human beings, which is safe and reliable, and has high recognition accuracy. The most common sites of vein recognition include the finger vein and the palm vein.

From the technical system, vein recognition mainly includes four stages: image acquisition, pretreatment, feature extraction and matching. Firstly, the blood flowing in human veins can absorb the characteristics of specific wavelengths of light, and use specific wavelengths of light to specific parts. The irradiation is performed to obtain a clear image of the vein at the relevant site, and the acquired image is analyzed and processed to obtain the biological characteristics of the vein at the specific site, and the obtained vein characteristic information is compared with the previously registered vein feature to confirm the identity of the registrant.

In terms of safety, the uniqueness of living recognition and vein characteristics makes vein recognition a significant advantage. Since the venous blood vessels are hidden in the human body and the hemoglobin is stored in the venous blood vessels, only blood circulation can provide hemoglobin activity, which is recognized and thus difficult to be copied and stolen. At the same time, for adults, their left and right hand veins are completely different. The vein pattern will not change with age. Even twins will not have the same vein pattern, and the unique characteristics of veins are very strong.

In terms of accuracy, vein recognition is one of the most recognized biometric methods. The internal sample collected by near-infrared irradiation is less affected by the external environment, and the recognition rate is high. After rigorous testing by the medical profession, the rejection rate of vein recognition technology is less than 0.01%, the false alarm rate is less than 0.0001%, and the login failure rate is 0%. The accuracy ranks among the top of other biometrics.

Although the vein identification has the advantages of safety and accuracy, the current vein identification is limited by its own characteristics, and the product is difficult to be miniaturized. It has special requirements for the collection equipment, the design is relatively complicated, and the manufacturing cost is high.

Taken together, in the more common biometrics, fingerprints, faces, and irises are three of the more successful applications. Among them, fingerprint recognition has been widely used due to its low cost and low threshold for collection. The application, face recognition is gradually popularized with the maturity of image recognition technology; iris recognition accuracy is extremely high, but due to high cost, high collection threshold, application scenarios are limited. Vein recognition has also become more and more widely used with the continuous advancement of technology, and in some cases, the replacement of fingerprint recognition is realized.

Voiceprint recognition

The organs used by people in speaking differ greatly in size and shape, so the voiceprints of any two people are different, and for everyone, from the development of the teens to the fifties, Its voiceprint remains basically unchanged. Voiceprint recognition technology makes use of this feature to input sound into the sonograph, so that the mechanical vibration of different frequencies of the sound becomes a spectrum image, displayed on the screen or recorded on paper. This image is the voiceprint.

The working process of the voiceprint recognition system can generally be divided into two processes: the training process and the recognition process. First, the input raw speech signal needs to be preprocessed, such as sampling, quantization, pre-emphasis, and windowing. In the training process, the voiceprint recognition system should learn the training of the extracted speaker's voice features, establish a voiceprint template or a voice model library, or adapt the existing voiceprint template or voice model library in the system. . In the recognition process, the voiceprint recognition system should perform pattern matching calculation on the characteristic parameters of the input voice according to the existing voiceprint template or voice model library of the system, thereby realizing recognition and judgment and obtaining the recognition result.

Beginning with the rapid development of computer technology in the late 1970s, the research on voiceprint recognition turned to nonlinear processing of various acoustic characteristic parameters and new pattern matching methods. In the aspect of feature parameter extraction technology, nonlinear processing methods such as wavelet feature parameters and linear prediction combinations of different feature parameters have been proposed and widely used. In recent years, the adoption of DSP chip computing technology has enabled the current speech feature parameter extraction technology to reach a relatively mature stage. In the aspect of pattern matching judgment technology: from the beginning of the 1980s, the hidden Markov model, artificial neural network and other technologies have been effectively utilized in voiceprint recognition, and gradually become the mainstream pattern matching method of voiceprint recognition system; After the 1990s, Gaussian hybrid model technology has quickly become an important voiceprint recognition technology due to its simplicity, effectiveness and good robustness. Since the 21st century, the support vector machine technology and the fusion of multiple pattern matching methods have also been deepened. Research and development, and entered the practical stage of commercialization.

Compared with other biometric technologies, voiceprints are simple, accurate, economical and scalable. Compared with other biometric technologies, voiceprints have many advantages such as simplicity, accuracy, economy and scalability. Of course, voiceprint recognition technology is also There are some technical difficulties. For example, there are some technical difficulties in using very short voices, such as: training and recognizing models with very short voices, effectively distinguishing models from training and recognition by realizing the elimination or attenuation of changes. Effectively distinguishing the imitation sounds to truly eliminate or attenuate the effects of changes; eliminate channel differences and background noise.

Gait recognition

Gait refers to the posture of people walking, a kind of biological behavior characteristics that can be perceived at a long distance, and identity authentication with gait characteristics is a research hotspot in the field of computer vision and biometric identification in recent years. Gait recognition is to analyze the image sequence containing human motion, usually including gait detection, and the analysis is to analyze the image sequence containing human motion, usually including gait detection, gait characterization and gait recognition. 3 processes.

Gait detection is the extraction of the human gait contour area from the background image in the image sequence. Research in this area includes background estimation, target detection, and post-morphological processing. Effective segmentation of gait contour regions is very important for post-processing such as feature extraction and target classification. Therefore, gait detection is often regarded as a preprocessing part of gait recognition. Gait characterization is a way to represent the detected gait and the known gait in the database, also known as gait feature extraction. Gait recognition refers to comparing the gait information to be identified with the gait characteristics in the database, and determining the category to which it belongs by a certain judgment basis. There are template matching methods, state space methods, and so on.

The advantages of gait recognition are non-contact, non-invasive, easy to perceive, difficult to hide, and difficult to disguise. Based on these advantages, gait recognition has broad application prospects and economic value in the fields of access control systems, security monitoring, human-computer interaction, and medical diagnosis.

Compared with other biometric technologies (such as fingerprints, faces, irises, etc.), the benefits of data availability and non-contact detection, voiceprint recognition and gait recognition have their own unique advantages, so The field has high application value.

3, commercial-grade applications: deep accumulation, new technology and new scenes drive new markets

3.1, access control: fingerprint application is wide, the face is mature

Fingerprint recognition is the most widely used, low-cost, and easy-to-use biometric technology in access control systems. However, fingerprints are the texture of the epidermis after all, the capacity is subject to changes due to external factors, and it is easy to be forged. In industries with high security requirements, fingerprint identification products will be limited to some extent. The new generation of access control products that embed face recognition products into the access control system are becoming more and more mature. At present, the face recognition access control system is mainly used in places with high security requirements, such as prison access control and bank vault. In addition, in recent years, technologies such as iris and vein recognition have also begun to be applied to access control systems. In 2011, CSIC 710 successfully developed a finger vein identification access control system.

Figure: Market size of domestic access control system (100 million yuan)

In recent years, smart communities and banks, smart city security and other projects have promoted the rapid growth of the access control market. The growth rate has been increasing in the past three years. In 2015, the market scale reached 16 billion yuan. With the increase in the penetration rate of overlay biometrics in the access control market, biometrics will achieve rapid growth.

3.2, attendance: fingerprints in the road, face realization technology breakthrough

Applying biometric technology to the employee's attendance system can help solve the problem of the traditional punching card and the C card attendance method, ensure the attendance data is true, and improve the efficiency and quality of the enterprise staff management.

Among all biometric technologies applied to attendance machines, fingerprints account for the largest share. In addition, with the further application of 3D live fingerprint recognition technology in products, the use of fingerprint film generation, fingerprint skin environment abnormalities lead to pattern recognition difficulties, slow recognition speed, causing several problems of queuing attendance will be alleviated, 3D fingerprint attendance machine will To some extent, the replacement of the traditional optical fingerprint recognition attendance machine is realized. Through the attendance of some unique feature recognition tests on the face, the current technical level has broken through the influence of day and night light, and can achieve rapid recognition under natural conditions. Compared with fingerprint recognition, face recognition completely eliminates the occurrence of card attendance and eliminates the embarrassing situation of fingerprint attendance contact. Iris recognition technology was widely used in the coal industry for attendance at an early stage. At present, the application of iris recognition technology to the building real-name system management system has been promoted nationwide, and the cases in Zhuhai and Guiyang are more prominent. The vein recognition attendance system has also been applied in some enterprises and universities.

3.3. Application in the financial industry: deepening bank customers, mobile payment inciting incremental market

Financial institutions have extremely high requirements for security. Therefore, biometric identification technology is first widely used in the internal control of banks. The bank's confidential information safe, treasury access control, and teller business authorization basically adopt biometric technology. In May 2013, Everbright Bank introduced a fingerprint identification unified identity authentication platform, and replaced fingerprints with fingerprints to log in to the bank's internal business system such as the core business system and the inspection system.

From weak real-name electronic account cards to deposits and withdrawals, biometrics expands to higher-risk bank customer identification scenarios. In December 2015, in order to improve personal banking settlement account services, the People's Bank of China issued the “Notice of the People's Bank of China on Improving Personal Bank Accounts to Strengthen Account Management”. The "Notice" clearly proposes to expand the account opening channel and increase the remote account opening channel. The qualified banks can explore the biometrics technology and other safe and effective technical means as the verification of the identity information of the account opening applicant. Clear government and other safe and effective technical means as an aid to verify the identity of the applicant. Clear policy support further promotes the biometrics technology in the bank customer identity verification scenario into the practical phase.

Figure: China Merchants Bank brush face withdrawal

As early as 2014, Jiangsu Bank launched the “brush face” technology in all physical outlets under the jurisdiction, mainly for assisting account opening, network verification and portrait comparison. After years of development, three types of technologies, including face recognition, vein recognition, and voiceprint recognition, have already landed. Face recognition is currently the most widely used, and is mainly used for identity verification when a weak real-name electronic account (such as a direct bank account) is opened. Weak real-name electronic accounts cannot be paid, transferred, cashed, or exchanged, so the application of biometric technology is less risky. For example, it belongs to the weak real name electronic account direct bank account, its function is limited to the purchase of this and the card issuance and distribution of the balance of wealth management products, funds can only go through the binding bank settlement account. In addition, face recognition technology has also been used to assist inbound and outbound accounts only through the bound bank settlement account. In addition, face recognition technology has also been used to assist the household's nuclear body opening, mobile phone bank account opening, credit card application also uses face recognition technology. With the opening of the merchant's nuclear body, mobile banking and credit card applications have also enabled face recognition technology. With the launch of the face-lifting business in 2015, China Merchants Bank officially entered the identity verification scene when depositing and withdrawing money. Vein recognition has been applied to the deposit and withdrawal business of a number of bank self-service devices, and voiceprint recognition has been first applied in the mobile banking login verification service of CCB.

Mobile payments open a huge incremental market. At present, the most widely used and best-selling technology is fingerprint recognition. Third-party mobile payment tools such as Alipay and WeChat and mobile banking of many banks have launched fingerprint direct payment functions. In September 2015, CITIC Bank cooperated with Huawei to launch fingerprint recognition near-field payment. MateS Huawei Wallet APP is bound to CITIC Bank Credit Card, which can open the same amount of electronic credit card online. Users can finish the mobile phone on UnionPay POS machine. Screen payment. “Out-screen payment” means that when the user closes the account, he only needs to hold the mobile phone close to the POS machine, and then the screen can be cancelled to evoke the payment page, and the fingerprint verification can complete the small payment within 200 yuan; if the consumption amount exceeds 200 yuan, Simply type the credit card password on the POS machine to complete the payment. Recently, Minsheng Bank launched the first iris payment in the industry. Face recognition technology has been proposed for a long time, but it has not yet landed in third-party mobile payment tools. Voiceprint recognition has not yet been put into practical use.

Compared with the authentication of bank customers, the application of biometrics in mobile payment scenarios is still in its infancy. At present, only fingerprint payment has achieved a wide range of landings. According to TrendForce, the global mobile payment market in 2015 was US$450 billion, and it is estimated that it will reach US$620 billion in 2016, with an annual growth rate of 37.8%. In 2015, the number of online mobile payment users in China reached 358 million, with a growth rate of 64.5%, which was 1.8 times the growth rate of users in the overall online payment market. The proportion of online payment for mobile phone users increased from 39.0% to 57.7%. With the further improvement of fingerprint recognition penetration rate and the landing of technologies such as face recognition and iris voiceprint, biometric technology will open a huge incremental market in mobile payment applications such as third-party payment instruments and mobile banking.

In 2009, the application of Chinese business in the biometrics industry segment accounted for 84.9% (calculated by sales). (calculated by sales). If this ratio remains unchanged, according to the forecast of China's 2020 biometric market size of 30 billion yuan, the market size of commercial-grade applications will reach 25.47 billion yuan.

4. Government and public-level applications: initial scale, multi-policy support to stabilize the market

4.1, public security department: intelligent monitoring + personnel access management to help face recognition lead

This year, the public security department achieved a number of projects in the field of bio-identification focused on monitoring and personnel access management. In many technologies, face recognition is the leader.

Face recognition promotes the development of video surveillance to intelligence. With the advancement of the construction of safe cities, many cities in China have deployed a large number of security surveillance cameras. The resulting massive video data brings opportunities for public area security and also poses great challenges. Fully relying on manual monitoring or troubleshooting, can not meet the needs of practical applications, the integration of face recognition technology into video surveillance systems, improve the level of intelligent analysis of video surveillance, is an effective way to meet the challenges. This year, Hefei opened the face recognition “Tianwang”. All the trains and bus stations in Hefei have face recognition devices. All the passing people will be captured by the system to collect facial information, and collected and stored in the library. When the case occurs, Hefei's security video surveillance alarm system can record the suspect's appearance into the library for retrieval and find a matching face.

Face recognition began to be used for personnel access management in many types of places: In July this year, Shenzhen Airport became the first airport in the country to embed the face recognition system into the airport security information system and realize the integrated operation of the two systems. The first domestic airport security screenings No.7 and No.8 were embedded in the system. In the past month, the system found 10 passengers who used the fraudulent documents. The airport security checkpoint recently seized nearly 30% of the total number of such passengers. It has shown strong professionalism and practicality in improving the on-site inspection capability and effectively screening passengers for fraudulent use of documents. It can effectively prevent the illegal elements from changing the boarding pass and then deliberately exchange the boarding pass to board the security check hole of the other aircraft. . In September this year, the Nanjing traffic police used face recognition technology for the first time. Since the face recognition system was launched, the public security checkpoint has inspected 3,200 people and found 32 people with problems, including ID card expiration, inconsistency, and 2 The fugitive and two drug addicts were captured.

2014年，公安部部级的生物识别采购金额为819.2万元，2015年采购金额为899.78万元，而今年截止到现在，采购金额已达1471.08万元。如果地方公安机关的采购额能够实现相似的增速，生物识别将在公安行业将迎来翻倍增长。此外，公安部部级近三年的采购项目基本都是人脸识别项目，对于其他生物识别技术，仅在今年采购了236万元的签证指纹采集仪硬备。可见，人脸识别短期内将主导市场的增长。

4.2、社保部门：生物识别正式入驻社保系统

2015年，人社部提出要紧密跟踪芯片技术的发展方向，积极研究无线支付、基于互联网服务的用卡方式、生物特征识别等新技术的应用，为新一代社会保障卡做好技术储备。由此，生物识别技术正式入驻社保卡。社保生物识别系统的基本结构和工作过程如下：

1)特征采集：参保人在系统中注册，填报本人的身份证、年龄、地址、所属单位等基本信息，设置本人的系统登录密码并采集生物特征信息；系统将所有用户信息和其对应的生物特征集中存储于数据库中。

2)认证：参保人定期使用密码或身份证登录系统，并进行生物特征认证，以确参保人的身份和生存状态；系统则根据用户提供的登录信息提取其生物特征进行匹配验证，认证过程中可以录像保存验证资料。 <

3)社会保险信息系统依据认证获得的身份和生存信息决定如何进行社会保险的发放。

年以来，各级人社部门对生物识别的采购包括了人脸识别、静脉识别和指纹识别三类技术。根据下表数据，我们假设每个市的采购金额为200万元。按国内共计293个地级市计算，市级人社部门的生物识别市场约为5.86亿元。同时，假设省级采购金额为1200万元（甘肃省2016年采购金额为1264万元），按国内共计32个省和自治区计算，省级人社部门的生物识别市场约为3.84亿元。加总之后全国人社部门生物识别市场的空间约为9.7亿元。

表：各级人社部门部分生物识别采购项目

4.3、教育部门：各类考试有望搭建新入口

2016年2月29号，教育部发布了《关于做好2016年普通高校招生工作的通知》，通知中教育部明确提出：要结合各地实际，采用二代身份证现场报名确认、现场采集照片和指纹或静脉等生物特征、及时进行信息比对措施严防替考。通知采集照片和指纹或静脉等生物特征、及时进行信息比对措施严防替考。通知表明，指纹识别、人脸识别和静脉识别三类技术有可能成为未来高考身份核查的强制手段。

早在2012年，河北省教育考试院为了进一步规范考试流程，就率先在国内教育为了进一步规范考试流程，就率先在国内教育考试领域试点引入指纹识别技术。2016年，北京、四川、湖北、广东等多个省份的高考采用人脸识别＋指纹识别双重技术确认考生身份。每个考点都配备身份核验终端，考生进入考场时需要在设备上同完成刷身份证、按捺指纹、人脸拍照，并且与之前注册信息进行联网比对，三者核验均通过才能予以参加高考。而从2015年开始，蒙古自治区在高考身份核查中使用指静脉识别技术。

指纹识别率先打开了高考市场，紧随其后的是人脸。当前和指纹识别率先打开了高考市场，紧随其后的是人脸。当前和识别两类技术已经应用于多省的高考身份核查，指静脉识别两类技术已经应用于多省的高考身份核查，指静脉识别技术还未大范围应用。随着教育部通知的进一步落地，更多省将在高考中使用静脉识别技术。

高考之外，自2015年起，司法考试在各省各考区推行国家司法考试人脸识别系统，考生必须先扫描身份证并现场拍照，通过人脸比对后方可进入考场。2016年，部分地区的英语等级考试也引入了人脸识别技术。高考之外各类考试逐渐引入生物识别技术，将放大生物识别在教育部门的市场空间。

4.4、医疗部门：人脸识别带动智慧医疗新进展

2016年3月，全国首家“人脸识别”医保支付系统在武汉市中心医院上线，通过手机刷脸支付，医保患者偶有不适可以在家通过网络院问诊，电子处方开具后，可以直接完成医保费用的缴纳，实现了医保患者网上就诊，药品快递到家。

人脸识别系统应用于医保支付系统，是智慧医疗的新突破。不过由于刷脸支付涉及医保账户线上使用的资金安全等问题，技术能否在国各大院推广还需涉及医保账户线上使用的资金安全等问题，技术能否在全国各大院推广，还需要看政府层面下一步的支持力度。

5、消费级应用：方兴未艾，需求端打开奠定最大增长动力

5.1、手机端指纹识别：指纹识别硬件价跌量涨，NFC支付形成新推力

指纹识别产业链主要由硬件环节和软件环节构成。硬件环节包括指纹识别芯片、指纹识别模组，软件环节包括指纹识别方案、手机终端和软件应用。在硬指环节，从芯片到模组，还需经历封装、coating、盖板、金属环等环节。

需求扩大和价格成本下降推动指纹模组出货量高速上升。近年来，智能手机和移动支付需求的扩张带动了手机端指纹识别技术需求的增长。同时，指纹识别逐渐成为中高端旗舰机型的标配，且在低端千元市场的普及率剧增。两方面因素叠加，推动指纹模组的需求不断扩张。而随着中国本土厂商全面进入识别市场并加快推出产品，指纹识别模组价格进入下行通道。指纹模组的价格2014年底约为10美元，2015年降至5美元，目前市场均价在3-4美元/片。需求的扩张和价格的下行带来指纹模组出货量快速增长。2016年上半年，指纹模组的出货量呈现逐月增长的趋势，上半年出货总达到1.51亿（不计算苹果手机的指纹模组）。我们保守估计2016年整年的出货总量为3亿，假设均价为3美元，2016年整年指纹模组的销售额将达到9亿美元。

据接近FPC的业内资深人士分析称，今年指纹识别渗透率为20%-25%。如果指纹识别渗透率可以达到50%的水平，指纹芯片、模组市场规模将实现翻倍增长。

NFC支付相继推出，移动成为手机指纹识别关键应用场景。2013年9月，苹果推出首款搭载正面按压式指纹识别（TouchID）的手机iphone5s，支持解锁/支付、与Home键完美融合。自iPhone5s之后，全世界的高端旗舰手机都开始攻关指纹识别技术，指纹识别在消费级市场迎来了爆发式增长。手机端指纹识别的应用场景包括：屏幕解锁、对手机的文件或者应用场景包括：屏幕解锁、对手机的文件或者app进行加密、应用互动(比如长按指纹传感器给预先设定的人打电话，双击传感器就能共享照片，打开播放器的时候在传感器上向右滑动就能切换下一首歌等)以及自家服务(各手机厂商独特的指纹解锁服务，例如华为的应用锁、OPPON3的拍照等方面)。而随着第三方支付工具和手机银行逐渐支持指纹识别，移动支付开始成为手机指纹识别的主要应用场景。另外，三星、华为、中兴、小米等厂商相继与银联达成合作，推出NFC支付。NFC支付可在不联网的状态下完成，与手机硬件的结合更加完善，且由于其信用加密等安全功能集成在硬件上，安全性更高。NFC支付在分割当前移动支付存量市场的同时，有望凭借其优势撬动一部分增量市场，这将对指纹识别芯片、模组和指纹识别软件环节形成拉动作用。

5.2、手机端虹膜识别：正处产业化初级阶段，高端机存在空间

2015年5月，富士通推出了全球首款搭载虹膜识别技术的新款机型ArrowsNXF-04G，用户只要短暂凝视触控屏便可自动解锁，并能够实现购买APP的账号输入或具备登陆网页的密码功能。三星今年的旗舰机型GalaxyNote7支持虹膜扫描，用于解锁的同时，SamsungPay也获得了对于虹膜扫描功能的支持。而即将发布的华为年度旗舰Mate9据称也将内置虹膜识别功能。

虹膜识别算法国外领先，国内进入产业化阶段。虹膜识别技术的核心环节是算法。在该环节，国际领先的厂商有美国Iridian、Iriteck和韩国Jiris等。美国Iridian公司是全球最大的专业虹膜识别技术和产品提供商，它和Irisguard、Securimetrics、LG、松下、OKI、NEC等公司进行合作，以授权方式提供虹膜识别核心算法，支持合作伙伴生产虹膜识别系统。Iritech提供核心软件以及镜头，由其他公司在此基础上开发应用系统，并且负责销售。目前，国内的虹膜识别核心算法处于起步阶段，技术提供商主要是上海交大图像所和中科院自动化所，其虹膜研究成果已经产业化，主要由中科虹霸和聚虹光电这两家公司进行运营。目前，两家公司的产品已经在商业和公共领域获得了应用，未来有望拓展到消费级的手机端。

大量黑色虹膜样本为国内研发提供契机。当前，国内虹膜技术的关键研发方向是针对东方人的黑色虹膜识别。相对于西方人的虹膜，黑色虹膜颜色更深，纹理区域更狭窄，纹理数量更少，而且普遍存在比较严重的眼睑或睫毛遮挡，这些特点导致一般的虹膜识别技术在应用于东方人时识别精度会大大下降，甚至成为虹膜识别技术在东方人群中扩大应用的潜在障碍。中国拥有最大的黑色虹膜样本，这将为国内研究机构和厂商提供契机，在该研发方向上实现赶超。

指纹识别可复制的特性一定程度上削弱了它的安全性。而中高端手机用户对手机数据保密性和支付安全性的要求极高，因而，虹膜识别技术在中高端机型上存在市场空间。

5.3、指纹锁：渗透率尚低，作为智能家居入口优势明显

指纹、智能锁应用。指纹锁是将电子技术、集成电路设计、大量的电子元器件，结合计算机网络技术、内置软件卡、网络报警、锁体的机械设计、生物识别等多种创新的识别技术的综合产品。

目前指纹锁在国内渗透率较低。国内指纹锁市场至今已经过了10多年的漫长培育。从2012年开始，指纹锁产品处于高速增长的阶段，2015年比2014年增长了82%，但目前渗透率仍然较低。以往在中国智能锁主要被用于高端酒店、公寓与别墅当中，在普通家庭用户普及率不到2%，而在欧美等西方发达国家智能锁的市场，而在欧美等西方发达国家智能锁的市场，占有率在50%以上，日本、韩国智能锁更占民用锁70%以上的市场。

近年来，在智能家居系统中，指纹锁有着独特的应用场景与集成空间。伴随智能家居功能延伸与智硬件创业浪潮，指纹锁开始成为重要的应用落地点。然而与之前相比，目前市场情况可能出现了以下变化：目前市场情况可能出现了以下变化：

（1）去中间化销售：在2015年之前，指纹锁产品的零售价基本在三千元以上，部分品牌可以达到四千元以上。随着创新型企业进入指纹锁领域，从2016年开始，很多智能家居品牌也延伸到这一领域，再加上互联网品牌的新运作新模式助推，部分做到了去中间化，在价格上有所下降。另外，大量企业进入指纹锁市场，加速行业普及，对传统企业的产业链体系会造成冲击。随着技术和生产门槛进一步降低，各个生产厂家出现竞争现象，压缩了成本。

（2）整体产品设计能力的提升：包括与门厂和孔位匹配兼容性的提升。以前指纹锁和门厂的设计相互都没有贴合，这就导致指纹锁到门厂去需要经过比较复杂的安装调试，安装调试和销售维护的成本都比较高。现在指纹锁和门厂在技术上形成了联通，指纹锁的安装标准遵循着门厂的标准，各方面的成本都会降低。

（3）指纹识别和嵌入式单片机技术的成熟：包括芯片成本的降低，这会让指纹锁产品进一步稳定，推动行业发展。

（4）智能家居和物联网技术的成熟：智能家居和物联网技术的成熟：让指纹锁融入到了智能家居整体的范畴。智能家居在宣传推广方面可以让消费者快速认识指纹锁，有知就会智能家居在宣传推广方面可以让消费者快速认识指纹锁，有知就会费和普及。

（5）使用者消费习惯改变：

过去住宅等的主要购买者多为70后，没有形成指纹识别的习惯，而目前购房者中开始出现80甚至90后，这一部分消费群体由于iphone等智能手机的使用习惯培养，对于指纹识别已经较为熟悉，更愿意在住宅中接受指纹锁。

从“指纹锁”到“智能锁”：在指纹加密外，近年来新型智能锁还能做到APP发短信、云联网报警，使得智能锁的安全级别大大高于机械锁，手指指纹代替了传统钥匙，产品体验也更便捷。

与其他众多智能家居硬件相比，指纹锁作为入口有得天独厚优势：智能锁的更换周期为5-8年，相比于其它智能家居使用频次高，更周期长，智能化成本较低。同时智锁以个人指纹作为ID可以连接其他智能家居硬件，具备生态链的纵深潜力；

在新的市场环境下，指纹锁、智能锁创业项目迎来了投融资的春天：知名智能锁品牌第吉尔、耶鲁被瑞典锁具巨头亚萨合莱收购，行业资源整合力度加剧。国内项目，小嘀云智能锁于2016年3月获得A轮1.23亿人民币融资，代表着智能家居的突破点正从智能家电向智能安防领域转移的趋势。

目前，智能锁市场潜力巨大。随着中国人口基数扩大、新建楼盘增多、楼价总体水平上升、国人对居住条件的要求提高，目前，国内一线、二线城市房价不断上涨，如果每平米售价10000元以上，那么指纹锁的成本平摊到每平米价格上，总价上涨幅度基本可以忽略。因此，越来越多的房开商采用指纹锁来吸引消费者。同时还有对高科技的热衷、社会不稳定因素带来的安防问题，使指纹锁市场逐渐扩大。根据华经视点对物理门禁控制市场的调查研究表明，超过70%的最终用户和80%的行业受访者认为，在未来3至5年内，他们希望以手机、钥匙牌、标签或凭证卡等方式替代现在的传统门锁。目前我国全国城镇住宅在3.4亿套左右，假设有一半能实现智能锁渗透，每户使用智能生物识别锁1.5个，那智能锁的需求量可以达到2.55亿个。按照每个智能锁2000元的价格计算，总市场规模可以达到5100亿元。

6、广阔市场和多种盈利模式带来巨大利润空间

6.1、从硬件销售到数据变现，盈利模式多样

生物识别盈利模式较为多样化：我们认为盈利模式可以分硬件销售，软件、解决方案服务提供，云服务提供，未来可能的数据变现几种。（1）硬件销售。硬件是生物识别的入口，硬件销售也是生物识别领域最基础的盈利模式。提供优秀的硬件，是生物识别领域公司一种较为传统的盈利途径。

例如虹膜识别，由于提取眼白毛细血管信息的过程容易受到光线的干扰，因此首先需要使用近红外LED来提供照射光线，需要使用四元晶粒机台设备，这一技术难度较大，主要红外线LED厂商包括日本EPITEX、欧洲厂商欧司朗、台湾的晶电和研晶。同时，提取信息需要小型摄像头模组（CCM），它将光学图像转化为电子视频信号的装臵，其后电子信号将被转化为于数码影像设备之显示屏上，以供使用者将其储存为数码影像。

例如指纹识别。每一次的指纹解锁经历三个过程，首先传感器检测到手指按下并判断手指覆盖传感器面积达到验证的标准，其次进行指纹数据验证，最后验证通过点亮屏幕并解锁屏幕，每一个过程都依赖于硬件本身的处理能力，因此硬件在整个指纹识别产业链中占有极高地位。在整个指纹识别产业中，芯片价值占比极高，占整个模组的70%；又分为IC设计、晶圆制造、封测三个环节，在我国A股上市公司中，硕贝德、晶方科技等公司都具有一定的封装和模组技术实力。

（2）软件、解决方案服务提供。优秀的软件技术有助于降低识别错误率，提高

识别效率，是生物识别中的核心环节，特别在人脸识别、声纹识别、步态识别等对于算法要求较高的领域。专注软件和平台开发的公司将最受益于该类领域。

例如人脸识别。目前已有多家公司在人脸识别解决方案领域布局。例如川大智胜，公司三维人脸识别技术采用“三维建库，二维识别”的方式，投入小、精度高，具有明显的技术优势。该系统已经开始试点应用。佳都科重点布局的是云从科技公司的人脸识别技术，目前研发的多项产品进入了试点和应用阶段未来，公司还将进一步加大对人脸识别等图形图像智能分析技术的投入力度，加快产品化和产业化进程，向国际领先的技术水平迈进。除此以外，东方网力、海康等上市公司已经在视频管理、人脸识别领域有较强的技术优势和稳固的市场地位。

云服务模式。除了直接提供解决方案或软件外，部分生物识别公司还应用云模式提供服务。例如旷视科技应用PaaS模式，打造智能视觉云服务，通过向下游客户提供人脸或者图像识别的服务来获得收入。例如，美图秀秀就是旷视科技“人脸识别云服务”的客户，美图秀秀不用自己开发相应的人脸识别模块，只需要接入旷视科技提供的API和离线引擎就可以享受现成的人脸检测、分析和识别等服务。这种模式区别于传统的项目制，具有非常好的盈利前景。

极视角科技是一家计算机视觉及大数据分析公司，是一家PaaS平台提供商。公司通过开放云端平台，为算法开发者提供开发环境、海量数据、架构支持等优质算法和解决方案。

（3）大数据变现。计算机公司除了提供软件工具与服务，还有可能通过与客户的合作，共同实现客户数据的变现，拓展新的商业模式。

例如指纹识别领域。作为只能家居

Intellective Electric Meter

Intellective Electric Meter,Industrial Electric Meter,General Electric Meter,Electro Industries Meters

Yuyao Gongyi Meter Co.,Ltd. , https://www.yycj.com