2016 Artificial Intelligence Hunan Forum and Zixing Artificial Intelligence Institute inauguration ceremony, where many top experts from home and abroad gave us presentations. Researcher Sun Zhenan, researcher at the State Key Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, and director of Tianjin Zhongke Institute of Intelligent Recognition Industry Technology, gave a detailed lecture on the topic of iris recognition.

This year is the 60th anniversary of artificial intelligence, like iris, face, fingerprint recognition is more practical artificial intelligence, but also the development of the use of a few more successful direction. Nowadays, smartphones are basically unlocked with fingerprint recognition, and recently Samsung released a new phone that will have iris unlock, so now many people don't know much about iris recognition, so I'll give you a science.

What is an iris?

The iris is a circular film located between the black pupil and the white sclera on the surface of the human eye, which exhibits a wealth of visual features such as spots, stripes, filaments, coronas, and crypts in infrared light.

Why use iris recognition

Nowadays there are mainly face recognition and fingerprint recognition to identify people, in fact iris has a great advantage.

1. Uniqueness: the formation of the iris is largely determined by random factors in the embryonic developmental environment, and the randomly distributed detail features in the iris texture determine the uniqueness of the iris pattern. The iris has a unique texture structure, and the degree of freedom of the detail features above the iris is of the order of hundreds, thus uniquely identifying the individual identity. 2. Stability: the iris texture is stable and formed eight months after birth and remains unchanged for life. 3. Non-intrusive: collection is non-contact, with the possibility of remote access. 4. Security: relatively fingerprint and face features are not easy to steal. Most of the iris needs to be identified in infrared, because the iris is between the black pupil and the white pupil, and most Chinese iris pigments are below near infrared to show detailed features that can identify a person very accurately.

A brief history of the development of iris recognition

In 1985, two other ophthalmologists applied for the first patent on iris recognition in the United States. They thought that iris recognition could be achieved by computer, but they did not know how to achieve it.

Then in the 90's there were some iris recognition products developed one after another, later iris recognition is proved to be a very safe and reliable identification technology, in various fields has been widely used, including customs, airport, national identity card currently there are three countries as an identity card iris identification, like India everyone has a number assigned to collect iris, now there are 1 billion collected iris recognition characteristics, in India application is very successful, Mexico, Indonesia, now their national identity card also added iris characteristics, the Chinese Academy of Sciences Institute of Automation is 1999 began to study iris products, after we invented the mobile phone terminal iris application, probably iris is such a development history.

Applications of iris recognition

At present, the iris recognition application field is very wide, no matter in physical space or in cyberspace, as long as the personnel identification can be used, including like security clearance can be used iris, or security counter-terrorism is also used iris, some terrorists can make up or even can be disguised, or use the way of scalding to replace the fingerprint, but anti-terrorism requires accurate identification, iris is a very important way. Our technology is also currently available in foreign banks where self-service ATMs can withdraw money with iris IDs.

In addition, Samsung's newly released mobile phone is to use iris recognition, Alipay will also iris recognition, they are also clear that the future is a very important direction for the use of iris, because fingerprints can be forged, it is likely that through the forged fingerprints can steal the funds inside Alipay, including touching the fingerprints of the teacups and glasses can be stolen in because of security attacks. In public security and judicial prison management including coal miners' attendance are used, and in access control, iris also has a broad application prospect.

The main processes of iris recognition

The first step is Acquisition of iris images , the capture of iris images is that ordinary cameras cannot capture a clear iris texture because the physical size of the human iris is relatively small and requires some near-infrared light to work with. The second step is Processing of iris images , we need to partition it out effectively and then normalize it. The third step is Iris image feature comparison , which is where the iris is more Key Step . In 1998 we developed the theory and methods of iris computing.

Iris image acquisition.

The first step is to give you an introduction to the process of iris image acquisition, which is a process with Challenging questions， One is the small physical size, resulting in imaging system depth of field is relatively small, and then the imaging process of the user's eye and the camera in an optical axis, effective focusing is not very convenient, that is, the traditional technology is facing a big challenge . for exampleOther factors such as reflections can interfere when you wear your eyes， consequently Iris image acquisition is a relatively large bottleneck for iris recognition.

The principle of iris imaging is mainly to use a near-infrared filter to image after passing through the imaging system, including some visual feedback, the distance between the person and the camera . This iris system is a sophisticated automated system, this is the iris image captured in different situations, our oriental iris texture is using near infrared 800+ nm light source, the iris image is the clearest.

today's Iris recognition systems are primarily proximity, for example, Japan OKI was first handheld, iris equipment used on laptops, including this portable iris imaging system, including also iris imaging devices. Currently there are also companies developing long range, say between 1m and 3m to capture the iris texture of a person, which is more convenient and is a more important trend in security clearance.

There are four main leaps in the acquisition of iris images We started to do it in 1998, and in 1999, we achieved from scratch, when we bought the equipment from the United States, they did not sell us, later through our independent research, in 2001 there was a great progress, we achieved monocular to binocular acquisition, in 2006 and 2008 there were new equipment made, later we researched an image can be photographed first and then focus on the light field camera.

Now we are developing iris recognition technology for mobile terminals, this is our recent development of iris recognition for mobile phones. This is iris recognition registration, which is very convenient as it takes only two to three seconds to achieve iris registration. The first iris security phone was released earlier this year and is being used in the country's security sector, including the application of our own iris devices, and we have also built our own iris project database, which currently has 16,000 research teams applying to use it, including a variety of databases, and we have made the field of iris research significantly more active through the sharing of iris images.

Iris image processing

After acquiring the iris image, we have to pre-process the iris image, which is to take some reflections on top of the iris image, including determining the initial position, which is iris detection, and then locating the inner and outer circle of the iris, sometimes the iris is not exactly circular, including the detection of eyelashes and eyes, we also propose new methods to be able to automatically model the distribution of the iris according to the law of the distribution of the iris, and then combine the texture, the edge of the iris, the boundary of the upper and lower eyelids, and then use some models to filter out these eyelashes and light spots.

Iris image feature comparison

In this case, how to encode iris features, traditional iris recognition methods are more complicated, we propose a relatively new and very simple and practical, very fast calculation Sequencing measurement characteristics , is to count the absolute values between this image regions, just to calculate which region is brighter and which region is darker relative to each other.

These codes are consistent with a binomial distribution, such that it can be determined that the distribution of match scores obeys a binomial distribution if different users are iris matched. Hamming distance match scores will be very high. And the same person is overwhelmingly able to match the fixed-order measurements even if there is some noise, in which case it is easy to identify and very fast to compute. Later we present the linear planning aspect Solving the problem of fixed-order measurement feature selection in iris and palmprint images The best current recognition results are achieved in the mainstream iris and palmprint image databases.

Opportunities and challenges of iris recognition

After completing iris technology identification, we have applications for prisoner management, airport security, family planning, terrorism prevention, and foreign ATMs through technology transfer conversion, and will also release iris security phones this year. Now the iris, although after years of development, can achieve very good recognition effect, but not necessarily completely solve the problem of iris recognition, because in complex scenes, especially in the conditions of long distance, multiple targets including the amount of communication, how to accurately identify the person's identity still needs a lot of research.

This includes the user being in different postures and speeds, different lighting of the environment, etc., all of which are future directions for this system. In particular, network conditions present new opportunities for iris recognition everywhere, such as the realization of iris acquisition and recognition from three to five meters or even ten meters away, including the challenge of how to do it on the Internet and mobile Internet on mobile phones. We think there are Three challenges.

One is the challenge of how easy it is to Second, the challenge of robustness Third, the challenge of security

While the iris is inherently much more secure than fingerprints with faces, it is possible that someone else could use this iris literature or glass eyeballs, or have hidden glasses to access the iris recognition system. therefore The iris is going from near to far, from passive to active imaging, and from static to dynamic imaging. In the future, regardless of the situation, we can capture the iris of multiple people at the same time, including from a distance to a close distance, and even people can present a 3D face, including the acquisition of imaging and the cooperation of the user. I think there is still a need for innovation of change in iris in biometric carry out, because the objects and distances of iris acquisition are ever-changing, so it is difficult to guarantee high quality iris imaging every time with a fixed pattern to cope with changing scenarios.

Limitations of existing iris recognition technology

So we think a lot of everything that's wrong right now. Sensing and knowing are one-way processes The two are not considered, that is, the sense and knowledge are isolated from each other, because we can dynamically configure the sensing framework in the imaging process, because the previous imaging devices are fixed, including the focal length, aperture are dead.

Ideas for the development of iris recognition

In the future we want to make it live, that is, dynamically configurable that can be adjusted to the environment itself, called A dynamically configurable framework for biosensing. Driven by top-down cognition, the optimal parameters are selected to acquire the best image, driven by the task. In the recognition process, it is also bottom-up data-driven, and then a better recognition model can be obtained using deep neural network methods driven by big data, because previous recognition models are manually adjusted filter parameters, which are difficult to cope with complex and variable iris imaging, so the recognition algorithm should be adaptive.

We are geared towards the needs of iris recognition. Using computational imaging active perception mode , a dynamically configurable visual sensing model is built to capture iris imaging information in a multidimensional optical field. The imaging process is all driven by cognitive tasks, all multidimensional, and all multidimensional signal processing for computation in the signal processing stage, for example, there is computation in the imaging process and parameters for imaging in the recognition process to drive it. In this way we introduce the cognitive requirements of the recognition task in the imaging process for targeted iris information detection and establish iris texture imaging for identity recognition.

Computational Light Field Imaging Techniques

We have researched and developed computational light field imaging that Basic ideas of light field imaging It is possible to record the direction of light and the intensity of light at the same time during the imaging process, so that a three-dimensional perception can be obtained, and it is possible to image first and then focus, and it is possible to adopt a computational imaging mode, and it is possible to calculate the refocusing of targets in different positions, so that the iris of different people can be obtained, so that a single image can be taken of more than one person's iris.

light field camera

We've also developed it. light field camera， And present the information in three dimensions through a computational model， And it enables simultaneous iris imaging of multiple targets， Because the original iris imaging uses imaging first and then focusing， And then during the iris imaging process， You can use In vivo testing . for example Use a piece of white paper to match the area around the eye、 Focusing area enables automatic recognition of real or fake irises， This is very convenient.。

Data driven

Like the father of artificial intelligenceMinsky He believes diversity of data to describe intelligent models is more effective， Because the nature of the iris is one of diversity and complexity， In this way we propose Data driven， It is the use of data neural network models to define such an approach using iris recognition big data . for example Using manually set rules to do iris image segmentation， Now simply using machine automation to learn big data models will improve19%， This allows for iris recognition concept elements and data definition， Including this noisy iris image segmentation achieves very good results。

And we put this method， Also it is possible to classify this people， If it is a single neural network with only98% the accuracy of the， But with the iris network you can reach99% the accuracy of the， We apply deep neural networks to iris images， Because iris images can come from the Internet or from mobile phone device vendors， This way the imaging is all differentiated， We've worked on many ways to overcome these differentiations before， We use the common space approach to express the model， Later on we used the Data driven， in other words Deep learning to do iris image recognition is also a very high accuracy improvement over traditional methods.

Security issues facing iris recognition systems

lastly A question on some of the security issues that iris systems will face in the future， Because the iris is an information security system， Every link is potentially vulnerable to security attacks， For example, a sensor can falsify data to carry out an attack， There are security attacks including feature extraction and physical sign matching， The focus is on addressing Two security analyses . for example How to go about the determination and recognition of fake iris textures on the front end and how to protect the iris signature template on the back end to protect the privacy of the user.

Front End Fake Iris Texture Recognition and Judgment Methods - Live Recognition

To identify forged iris textures, we use In vivo testing The method. The main method currently used in the in vivo detection method is the texture classification method, which is the iris captured by the real user, which is very natural and rounded, and then the iris texture of the prosthesis, the imaging will form some rougher texture, because we use the texture classification method to make the difference.

Our in vivo testing methods are also used in Ethnic classification and coarse classification of large-scale databases We've had good results with all of them, and that's what we use A lot of people's iris big data goes to automatically implement hierarchical visual primitive dictionaries The same model can not only solve the problem of iris live detection but also realize the classification of human race.

Back-end iris feature protection - iris key system

For example, traditional iris recognition, iris is not related to genes, later our research iris and genes still have a relationship, Asian and European iris is very different, we can use automatic classification method to achieve 99% accuracy, you give me an iris image I will know whether this is Asian or European, we also proposed the iris key system, we combined it with information security, we proposed the iris key method, while protecting the iris template at the same time can also protect the key, we also proposed the iris key system. Iris Key is a security system that organically combines iris recognition and cryptography to secure both iris features and keys, while the basic idea uses a fuzzy commitment approach.

The basic principle of the 512-byte iris feature code and the key error correction code to generate the key code for the operation, and then the result of the operation is the result of the security, and this result is difficult to attack out, unless you know my key, but as long as the person can be the result of the iris comparison, even if it can not be 100% on the comparison, as long as the error error error within a certain range can use the error correction code to solve the key, so that the iris feature can be achieved, is revocable.

For example, the iris template can take a new template to protect the encryption, in this case can protect the security of the iris template, in the actual process, the iris different time to collect different imaging can not be 100% to match on, may also be some errors, and then through the error correction coding to overcome, in addition to enhance the degree of intrusive key, we adopted the imaging features and extraction module, in the error correction coding to take several?

lastly

Our team is system innovation from the source of iris information acquisition, and has opened the complete chain of iris recognition from scientific research to industrialization, practicing acquisition device, data resource, recognition algorithm, integration system, and application a technical route, although we have made some progress, but iris recognition we think there are still a lot of problems to be solved from controlled scenes to complex scenes, and need to solve the problems of convenience, robustness, and security, and we will continue to do a better iris system.

question session

Question. In the course of what you just said you mentioned the issue of recognition accuracy, I asked how big the accuracy is the result obtained on the database respectively?

Sun Zhennan: The result of the experiment is in several hundred people, there are Han, Mongolian, Tibetan do the classification, this database has a description, Han has more than 400 people, more than 400 men, more than 200 women, Tibetan has 178 men, 124 women, Mongolian people are less, there are 58 men, 72 women, the amount of data has more than 20,000 pictures.

Question. Other than these tests, what is the maximum tolerance inside the other databases? It's keeping it under 90% accuracy.

Sun Zhennan: We have also measured, we have measured thousands of people basically can reach more than 99%, for example, the simple definition of two categories, for example, Oriental and Western people is also very high accuracy rate, because the Eastern and Western people texture distribution and characteristics are still different.

Question. You were talking about the iris recognition group, what kind of results will be done in the future? In addition to the few iris recognition phones mentioned earlier.

Sun Zhenan: We just recently applied for a national key research and development program, cloud computing and big data special project inside a project, we will achieve ten centimeters away from multimodal iris face gait image recognition, is in the anti-terrorism stability they have urgent needs, because there are many extremists, his face plastic surgery, the name also changed, fingerprints all erased, can not confirm the identity of these people, the next person we will public security field to do this system, basically to key personnel to collect iris, basically after leaving the country can not be changed iris, unless the eyes are blind, China's counter-terrorism and in how to identify this person's identity at a distance in the monitoring environment, this iris system or there are many areas to study.

Question. I would like to ask, do color tee and contact lenses have any effect on iris recognition system?

Sun Zhennan: If wearing colored teeel or contacts will have an impact, the texture is not the person's texture, is superimposed on the above texture, we must accurately determine, for example, in foreign banks our system can automatically identify, you were imaging is wearing colored or contacts, we this system can automatically identify, must ask the person to remove this colored or contacts.

Recommended>>
1、How mybatis prevents SQL injection
2、Industry Intelligent logistics accelerates the layout of the future will reshape the spatial pattern of the city
3、Why do you remember so long
4、python crawler cat eye movie top100
5、Why is it that humanity need not fear the robot revolution

已发送

发送中