Biometric Face Recognition System

Ing. Stanislav Kovář – skovar@fai.utb.cz,
Ing. Jan Valouch, Ph.D. - valouch@fai.utb.cz.

Tomas Bata University in ZlinFaculty of Applied Informatics Nad Stráněmi 4511, 76005 ZlínCzech Republic

Abstrakt příspěvku v angličtině.

The article contains a description of the functions of biometric system for face scanning, which also the Broadway 3D from Artec Group uses. Determination of basic facial characteristics that are necessary for the correct identification of users of the system. The results of practical measurements of the device in certain situations and their evaluates are presented in the article.

Introduction.

These days biometric systems are well known and their use is very common. Especially for objects, where the great emphasis on safety. Biometric systems, which are discussed in this article, work with physiological biometrics. It is derived from unique biometric characteristics of each human. As is the shape and distribution of papillary lines on the inner side of the hand, the quantity of stored pigment in the iris or quantity of photoreceptors in the retina, and so on. Some of these characteristic elements are changing throughtout of life, what it is a major disadvantage of biometric systems. Therefore, it is necessary to perform in defined time intervals refresh of the reference biometric sample. These elements also include the face, which is used to identify people by Broadway 3D. [4]

The face is made up of three characteristic external structures (eyes, mouth, nose), which help to shape the overall image of the face. These parts are variable in ethnic groups, which reduces the probability of occurrence of two different persons with identical characteristics. For clear identification of a concrete person is necessary to know the overall shape of the face, whereon are located further morphological characters. Cheeks and chins are the other features determining the resemblance face, which have importance for 3D face readers. In practice exist two approaches of scanning individual users - with user interaction and without user interaction. In the first case, as like the other biometric systems, the device requires a certain cooperation by user, ie. user has to stop before the reader and look at the designated place or postpone clothing which prevent a clear identification (hats, glasses, etc..). In the second case, the user doesn't need cooperate with a biometric device and identification is done during walking, respectively during the passing around the reader. During the detection of a user also environmental influences can cause some problems, eg. lighting, inappropriate background, many people in the image, etc. Basic property of the face is its variability, when except the above examples, visage, facial expression and aging are the major factors which most affect the shape of the face. Therefore, Biometric systems must be designed that they would be able to correctly identify the user even with these changes. [1],[2], [6]

Biometric systems for identification face are divided by the sensed characteristics:

•  identification by 2D image,
•  identification by 3D image,
•  identification by thermal image.

Also Broadway belongs among biometric systems for recognizing people by 3D images. 3D scanning is the safest and most effective method of identification. Apparatus for generating 3D facial images eliminates the shortcomings of 2D images, where during the transformation of the face into 2D are lost some essential information. Currently there is no known way, how could an attacker trick the 3D readers, unlike 2D face readers (photography). However, unlike the 2D scanners are 3D more expensive and especially image processing is more complex, due to the more information. The main difference between 2D and 3D face scanning is in scanning device that by the 3D scanning works as 2.5D scanners. 2.5D scanner creates a classic 2D image, which in each point saves information about the depth of the image and it allows you to present spatial data. In such a case it isn't a 3D image, because it is not possible to present the points which have the same coordinates and different depth. To create 3D images using a light source (IR, visible) through which scanning is performed at a certain angle and creates a 2.5D image. To ensure the formation of a coherent image, it is necessary to repeat the scan at different angles and from different locations, thereby preventing the formation of holes in the image. All obtained 2.5D images are in the final stages tag together into the final 3D image. However, the creation of a 3D image through multiple sites is considered to be impractical, costly (need more sensing devices), so system in the recognition face has to make do with informations gathered in one place. [1]

Gathered information about the face can be visualized by system a number of ways, the most often by three basic ways. First, as a cluster of points (cloud of points), expressing the coordinates of the individual points of the face, without any mutual bonds. It is not user friendly, even though it is very simple. The second method is called Polygonal network, it forms the polygonal 3D model grid on the surface of face, it has usually triangular shape. The third way of representation of the face is called depth map, which is basically 2.5D scan, wherein the intensity of points is equal to the distance in the space. All these ways visualize information about the faces in a 3D model, which is ready for the next part of the proces called recognition.

Methods

Deployment of biometric reader Broadway 3D is expected in the case of protection of objects which need a high degree of security. Broadway uses the same rules as all biometric devices. biometric samples face (template) of the user by equipment are obtained during the registration, then they are used for the identification or verification. Registration process needs requires the presence of the operator (e.g. Administrator), identification and verification processes not. Broadway autonomously evaluates whether the user is authorized or not. The result is displayed on the connected display or indicated by LEDs on the top of the recognition device. [3]

Broadway 3D is characterized by a high degree of reliability, especially it eliminates the shortcomings of the human factor. Security staff may do mistakes during controlling of a large number of people, but Broadway can check up to 60 people per minute. It is impossible to cheat it with using a mask or with partial covering of the face (hats, sunglasses, etc.). For verification, in case of loss or theft of an ID isn’t any danger of intrusion. Broadway 3D can serve as an attendance systém, because the information about face can‘t be faked and information about attendence can’t be misused, too. Scanning and saving face in a database, unlike other biometric systems, does not bring the risk of misusing of sensitive data, because the face information are publicly accessible (social networks, identity cards, etc.) against fingerprints, etc.

Face reader Broadway 3D is the first device in the world that is able to visually identify a person as easily as people can identify each other, and identification is a fraction of a second long. One glance into the device while walking is needed for Broadway to recognize the identity of a one user from tens of thousands of pre-registered individuals. Face Detection, unlike scanning other biometrics characters, does not require direct physical contact or the exact location in front of the identification device. Broadway 3D identifies persons regardless of their age or height. 3D face reader Broadway is equipped with a 3D camera system which records and saves the unique three-dimensional shape of the face. Unlike the human eye the device has the ability to distinguish subtle differences in geometry accurate to within fractions of a millimeter, which it allows to recognise even identical twins. Technology of 3D scanning allows fast and accurate acquisition of the surface of the scanned object using the methods of structured light. Light is projected through a pattern (in this case through the grid) to the surface of the object in the parallax, which is the angle between the two lines leading from different points in space relatives to the observed point (the place where the lines intersect). In other words, the camera scans the surface of the object from a different place than light, thus it leads to the measurement of two parameters (distance and angle). Advantage of 3D face recognition is the ability to measure the angle and distance. This is different from the 2D face readers, which measure only the distances in the structure of the face. The grid form a grid of coordinates on the surface of the scanned object, which is used for precise calculation of each point. Texture of the camera is synchronized with the 3D sensor, which has a wide angle of view. The sensor simultaneously captures the shape and structure of the object‘s surface. The camera is the best aspect of the whole device. It is a camera of new generation, which captures images at a wide angle up to 41°, ie. it is able to capture the face of people which have a half of the adult’s adult (children, users using wheelchair, etc.). Frame rate of the camera is 15 frames per second, even in stroboscopic mode, eg. when the user is running around the reader. The reconstruction algorithms and competent hardware allow an accurate recognition of objects. [1], [5]

Results.

Freeware application Turnstile Enrollment Application (TEA) was used for the testing of the biometric device Broadway 3D which is offered by Artec Group for all the products of Broadway series. TEA allows you to view the percentage of compliance of the biometric sample with a template during the identification and verification. Identification is divided by gender because of the different characteristics of men and women. The graph shows average values ​​of all users those are achieved at different thresholds. The results of the identification are showed by orange color in the graph and the results of FRR are showed by yellow color, it means erroneous denial of authorized users by biometric reader Broadway 3D. The threshold set by the system for a specific situation is showed by blue color. Results of False Rejection Rate were determined on the basis of individual identification of concrete users. Values ​​are approximated, vary according to the number of tested objects.

The identification with using the thredholds of around 60% and 70% runs without any problems, as the results show. With using a higher thresholds the value of FRR is increased, the correct identification was achieved only in exceptional cases. Wrong rejection of authorized user with using the threshold value lower than 70% was mainly due to the poor quality and especially the long time intervals between the processes of registration and identification (approx Monthly). For the ladies even the repeated the attempts of registration fweren’t able to create enough quality templates. Quality value was around the limit of 900, to see Table 1.

Table 1: the range of values the quality of registration

Another part of the testing is focused on evaluating the quality of identification when the heas is covered by common blankets (as a glasses, hats, etc.) or with making grimaces (smile, etc.). Also the results were evaluated after physical exertion, or during the user's movement toward or during movement around the reader. Testing is divided by gender again. Threshold of identification was set to 70%. This value was chosen in order to ensure safety.

Conclusion

Biometric systems have taken an important position in the security field. Reason of it is their reliability, speed, security and clarity during the identification and verification of people. The using of biometric readers for recognizing of human faces increases comfort and reduces the time, which is required for recognition. In the future we can expect further development of biometric systems and their use in access control systems.

From the testing of biometric readers Broadway 3D that the reader can easily identifies the user when the threshold has value of 60%. The value of 70% can be considered as reliable especially with using of high-quality biometric templates. Due to the fact that in practice there is no distinction between the genders, it is necessary to put emphasis on the registration of women to prevent during the operation erroneous rejection of authorized persons. Broadway also identifies more than one user, which are moving in the scene.

During the testing, it was found that Broadway during verification and identification retroact primarily to change the muscles around the mouth. This information has been obtained on the basis of changes these muscles in the smile, or yawn. On the other hand covering the eyes with spectacles or sunglasses been more or less smoothly. For spectacles were problems during identification significantly greater than in sunglasses with a reflective coating. Reflective coating on sunglasses evocate in the resulting model in the place eyes a white squares corresponding to content of glasses. Headgear (hats) do not represent any problem for the biometric reader and users have been reliably identified. When the women used peruke during the identification, the device incorrectly verified them, especially when simulating wigs with long hair. This wig basically extended the original length of the hair, which caused trouble for the reader. Changing hair color isn’t any problem for Broadway. More interesting is the fact that the men were much less problems. Although the testing subjects have had predominantly short hair and they were more visible in the face (no covering of the face with falling hair, etc.), there weren’t any difficulties with log hair wigs. In other words, the quality of the template was so good that the identification was successful in the most cases, despite to the reduction in the area of scanned face. This proves that the most important minutiae are in the area of the mouth. The average probability of false rejection of authorized person were increased by nearly 6%, when the wigs were used.

Biometric reader Broadway 3D is an excellent choice for access systems in buildings with big frequency of people. But it is not completely reliable in all situations guaranteed by the manufacturer. Especially it is problematic in the identification during walking. The device comes with the SDK, which, however, requires extensive knowledge of the programming language C++. Fortunately there is an alternative application TEA, that can be used at least as alternative for the SDK. Finally, the biometric reader Broadway 3D can be considered as a quality facility where the pros outweigh the cons.

References

[1] RAK, Roman – MATYÁŠ, Václav – ŘÍHA, Zdeněk. Biometrie a identita člověka ve forenzních a komerčních aplikacích. 1st ed. Praha: Grada Publishing, a.s., 2008, 631 p., 32 p. colour illustr. appx. ISBN 978-80-247-2365-5.
[2] DRAHANSKÝ, Martin – ORSÁG, Filip. Biometrie. 1st ed. [Brno: M. Drahanský], 2011, 294 p. ISBN 978-80-254-8979-6.
[3] JANEČEK, Tomáš. Biometrika [online] [cit. 2014-01-21]. Retrieved from: http://www.nula.wz.cz/biometrika/.
[4] MARIEB, Elaine N – MALLATT, Jon. Anatomie lidského těla. 1st ed. Brno: CP Books, 2005, 863 p. ISBN 80-251-0066-9.
[5] LI, Haizhou – LI, Liyuan – TOH, Kar-Ann. Advanced topics in biometrics. New Jersey: World Scientific, c2012, xv, 500 p. ISBN 978-981-4287-84-5.
[6] BLAŽEK, Vladimír – TRNKA, Radek. Lidský obličej: Vnímání tváře z pohledu kognitivních, behaviorálních a sociálních věd. 1st ed. Praha: Karolinum, 2009. ISBN 978-80-246-1556-1.