DE102016009619A1 - Method for detecting the spatial extent of a camera object as part of a living recognition for devices for recording person-specific data - Google Patents

Abstract

For a "personal data collection device", it is proposed to prevent or at least complicate fraud with flat images, such as printed images or video, by using the lack of spatial extent of printed images or videos as an exclusion criterion. For this purpose, the principle of "spurious" stereoscopy is used. Checking whether the camera object has a spatial extension is carried out in the manner described below: 1. The camera (6) is positioned at the optimum height for taking the facial image. 2. A first image (13) of the person (1) is taken with the camera (6). 3. The position of the camera (6) is changed in the vertical direction by a set amount. There are proposed 6 cm as a measure of the change in height (16) of the camera (6). This corresponds approximately to the distance between the eyes of a person (1). 4. With the camera (6), a second image (14) of the person (1) is recorded. 5. A correspondence analysis in the vertical direction is performed, comparing the first image (13) with the second image (14). The correspondence analysis determines disparities (20) and stores them in a result matrix. Each position of the matrix is associated with a pixel of the first image (13) or second image (14). 5. From the result matrix, the decision is made as to whether the object in front of the camera (6) from which the images were taken has a spatial extent. For this purpose, the result matrix is evaluated using mathematical methods.

Description

Technical area

The invention relates to a method for detecting the spatial extent of a camera object as part of a living recognition for devices for recording person-specific data.

Known state of the art

This invention relates, for example, to the "Personal Data Collection Device" described in the Laid-Open Publication DE 10 2013 203 433 A1 is described. The device described there serves for the automatic acquisition of person-specific data, in particular also the facial image of the person ( 1 ). It contains a camera ( 6 ), a lighting device ( 7 ), one in front of the camera ( 6 ) arranged semitransparent mirror ( 8th ) and a control unit ( 2 ) for performing image acquisition. It can also be a drive unit ( 10 ) with which the camera ( 6 ) can be positioned in the vertical direction. With the drive unit ( 10 ) the camera ( 6 ) is positioned so that it is at an optimum height for taking a picture of the face.

Significant statements on the state of the art of living detection when taking pictures of facial images with cameras can write "Overcoming Scenarios for Biometric Systems" by Stefan Hein and Markus Mahrla, 18 October 2004 , are taken. Here are some passages cited.

"Living detection is understood as the verification of whether the biometric features were generated by a living organism rather than an artificial counterfeit."

"For example, the increasing prevalence of digital cameras, some of which are becoming ever smaller (eg a mobile phone camera) makes it child's play to take several photos unnoticed by a person and misuse them for the purposes of verification by means of facial recognition." "Current methods of detecting a person's life are certain Properties of the skin, as well as movements, for example, in the face and eye to use. Typical skin properties include color, electrical conductivity, reflection behavior and heat radiation. The electrical conductivity check integrates well with fingerprint or hand geometry scanners; Skin reflection and color are also easily implemented in facial recognition systems. The temperature differences of the skin can be analyzed by simple infrared scanning. The fact that we can never put our biometric feature to complete rest makes it possible for live recognition to register minute movements that suggest a living organism. Thus, movements of the lips, eyelids and head can be helpful in distinguishing living persons and facial dummies. In the case of iris recognition, it is known that the pupil widens due to the influence of brightness. This effect and other features such as the curvature of the eyes are helpful indicators to differentiate artificial eyes from real ones. "

To assess the effectiveness of a living detection, the Internet site "Http://www.secupedia.info/wiki/Lenderenderkennung" (28.07.2016) The following statements are to be taken: "For the evaluation of the system overcoming security with the different methods for counterfeit defense not only the technical possibility for overcoming the respective biometric procedure is to be considered, but also the technically and temporally necessary expenditure, which a statement about the probability of a Overcoming possible. In addition, a safety analysis of the respective overall system is always required. Basically, life detection increases the overcoming security of a biometric access system. "

There are stereo cameras known that 2 cameras ( 6 ) whose optical axes ( 15 ) run parallel and which are arranged approximately at a distance from each other, which corresponds to that of the human eye pair. With these two cameras ( 6 ) a picture is taken simultaneously. Both images together contain the information about the spatial extent of the camera object.

Even before the invention of the stereo cameras, the so-called "fake" stereoscopy was known in which with a camera ( 6 ) 2 pictures taken from different positions with approximately parallel viewing angles.

Also known is the creation of a 3D image by projection of recognizable patterns, such as colored lines, in the face of the person to be imaged ( 1 ). This method is also called strip light scanning. This allows very accurate three-dimensional information to be determined.

It is state of the art that systems using only one camera ( 6 ) Capture images, no proof of spatial extent perform.

The problem underlying the invention

The inclusion of personal data, including facial images, is often automated today. This also applies in particular to the "device for collecting person-specific data", which is disclosed in the published patent application DE 10 2013 203 433 A1 is described. Since no persons are involved in the use of such devices ( 1 ) is more necessary for the operation of the device for image acquisition, this process is also monitored by anyone. Fraud attempts can therefore also be carried out without the assistance of a supervising person ( 1 ) is noticed. Such fraud attempts include, for example, the provision of a flat image, such as a photograph of a person printed on paper ( 1 ) or a flat screen displaying a video of a person ( 1 ) is played in front of the camera ( 6 ).

The invention

It is proposed to prevent, or at least to complicate, fraudulent attempts at printed images or video clips by using the lack of spatial extent of printed images or videos as an exclusion criterion. For this purpose, the principle of "spurious" stereoscopy is used.

• 1. Die Kamera (6) wird in der für die Aufnahme des Gesichtsbildes optimalen Höhe positioniert. Dieser Vorgang wird gemäß Anspruch 9 der Offenlegungsschrift DE 10 2013 203 433 A1 durchgeführt.
• 2. Mit der Kamera (6) wird ein erstes Bild (13) von der Person (1) auf genommen.
• 3. Die Position der Kamera (6) wird in vertikaler Richtung um einen festgesetzten Betrag verändert. Es werden 6 cm als Maß für die Höhenänderung (16) der Kamera (6) vorgeschlagen. Das entspricht etwa dem Augenabstand einer Person (1).
• 4. Mit der Kamera (6) wird ein zweites Bild (14) von der Person (1) aufgenommen.
• 5. Es wird eine Korrespondenzanalyse in vertikaler Richtung durchgeführt, die das erste Bild (13) mit dem zweiten Bild (14) vergleicht. Durch die Korrespondenzanalyse werden Disparitäten (20) ermittelt und in einer Ergebnismatrix speichert. Dabei ist jede Position der Matrix einem Pixel des ersten Bildes (13) oder zweiten Bildes (14) zugeordnet.
• 5. Aus der Ergebnismatrix wird die Entscheidung abgeleitet, ob das vor der Kamera (6) befindliche Objekt, von dem die Bilder aufgenommen wurden, eine räumliche Ausdehnung hat. Dazu wird die Ergebnismatrix mit mathematischen Methoden bewertet.

For a "personal data collection device" described in the Laid-Open Publication DE 10 2013 203 433 A1 is described, which automatically takes pictures of people ( 1 ), whereby the position of the camera ( 6 ) can be shifted in the vertical direction, it is proposed to make the check whether the camera object has a spatial extent in the manner described below:

• 1. The camera ( 6 ) is positioned at the optimum height for taking the facial image. This process is according to claim 9 of the published patent application DE 10 2013 203 433 A1 carried out.
• 2. With the camera ( 6 ), a first picture ( 13 ) of the person ( 1 ).
• 3. The position of the camera ( 6 ) is changed in the vertical direction by a set amount. It will be 6 cm as a measure of the height change ( 16 ) the camera ( 6 ) proposed. This corresponds approximately to the eye relief of a person ( 1 ).
• 4. With the camera ( 6 ), a second picture ( 14 ) of the person ( 1 ).
• 5. A correspondence analysis is carried out in the vertical direction, showing the first image ( 13 ) with the second image ( 14 ) compares. The correspondence analysis analyzes disparities ( 20 ) and stored in a result matrix. Each position of the matrix is a pixel of the first image ( 13 ) or second image ( 14 ).
• 5. From the results matrix, the decision is made whether this is in front of the camera ( 6 ), from which the images were taken, has a spatial extent. For this purpose, the result matrix is evaluated using mathematical methods.

• – Die Bildebenen sind identisch, sie entsprechen also lediglich zwei unterschiedlichen Ausschnitten derselben Ebene.
• – Die Zeilen der beiden Stereobilder liegen parallel zur Verbindungsgeraden zwischen den optischen Zentren. Diese Verbindungsgerade wird auch Basislinie genannt.
• – Die Bildkoordinaten der beiden Kameras (6) werden so definiert, dass zueinander kollineare Bildzeilen der Stereobilder die gleiche Zeilenkoordinate besitzen.

By the application of the method according to the invention, the device for detecting the facial image has the properties of the preferred standard stereo geometry. The standard stereo geometry is usually used on two cameras ( 6 ) applied, which have a horizontal distance from each other. It has the following properties:

• - The image planes are identical, so they only correspond to two different sections of the same plane.
• - The lines of the two stereo images are parallel to the connecting line between the optical centers. This connecting line is also called the baseline.
• - The image coordinates of the two cameras ( 6 ) are defined such that mutually collinear image lines of the stereo images have the same line coordinate.

The application of standard stereo geometry is advantageous for obtaining depth information because both correspondence analysis and triangulation are simplified.

In the usual in practice stereo method for obtaining depth information, the cameras ( 6 ) are arranged at a horizontal distance from each other. Accordingly, the correspondence analysis is carried out in the horizontal direction.

• – Die beiden Bildebenen sind ebenfalls identisch, sie entsprechen also lediglich zwei unterschiedlichen Ausschnitten derselben Ebene.
• – Anstatt der Zeilen liegen die Spalten der beiden Stereobilder parallel zur Basislinie.
• – Die Bildkoordinaten der beiden Kameras werden so definiert, dass zueinander kollineare Bildspalten (17) der Stereobilder die gleiche Spaltenkoordinate besitzen.

In the application according to the invention, however, the camera position is changed in the vertical direction. The stereo geometry has to be changed accordingly and now has the following properties:

• - The two image planes are also identical, so they only correspond to two different sections of the same plane.
• - Instead of the rows, the columns of the two stereo images are parallel to the baseline.
• The image coordinates of the two cameras are defined in such a way that collinear image columns ( 17 ) of the stereo images have the same column coordinate.

According to the x-axis and y-axis are reversed in the triangulation.

As a correspondence analysis, for example, the block matching method can be used. Through the correspondence analysis, points of the first image ( 13 ) as corresponding points ( 19 ) In the second picture ( 14 ) found. For each corresponding point ( 19 ), the disparity ( 20 ) is determined as the pixel difference in the vertical direction. A living person ( 1 ) always has a three-dimensional extent. Pictured areas of the person ( 1 ), which are at different distances from the camera ( 6 ), for example the eyes and the tip of the nose, cause points with different disparity in the correspondence analysis ( 20 ). In camera images taken from a plane image, all corresponding points ( 19 ) about the same disparity ( 20 ), where the absolute value of the disparity ( 20 ) depends mainly on the change of the camera position in the vertical direction.

The results of a performed block-matching procedure essentially depend on the size of the block used ( 18 ). As the block size increases, the block matching method finds fewer and fewer corresponding points ( 19 ), if it is a camera object with spatial extent. For camera objects that have no spatial extent, this effect occurs to a much lesser extent. To derive from the result matrix the decision whether in front of the camera ( 6 ) is a three-dimensional body or not, the number of corresponding points ( 19 ) are counted. If the number of corresponding points ( 19 ) exceeds a threshold, the camera ( 6 ) Pictures of a flat illustration and not of a person ( 1 ). The live detection then failed. If an evaluation is done with this variant, then the block matching procedure with large blocks ( 18 ) be performed. A meaningful measure of the block size can be determined empirically, for example.

For another variant of the evaluation, the block matching method has to be set so as to reduce the disparities ( 20 ) corresponding points ( 19 ) is optimally determined over a wide range. The result matrix then becomes the density function of all disparities ( 20 ). The density function uses a density function of disparities ( 20 ) previously determined with a natural person. If both functions correlate, then the live recognition is passed.

As a further technical embodiment, a feature-based correspondence analysis can be used. In the first picture ( 13 ) and in the second picture ( 14 ) each prominent points extracted by a suitable algorithm of image processing. For example, the SIFT algorithm or the SURF algorithm can be used for the determination of prominent points. Both the SIFT algorithm and the SURF algorithm determine a descriptor for each distinct landmark. This descriptor is in each case a data record in which the characterizing properties of the respective prominent point are stored. In a further step, corresponding prominent points ( 19 ) and the associated disparities ( 20 ) in the vertical direction from the two images. For this purpose, the descriptors of the prominent points from the first image ( 13 ) with the descriptors of the prominent points from the second picture ( 14 ) compared. For the comparison of two descriptors a distance function is introduced. For the SIFT algorithm and the SURF algorithm, the vector distance or the ratio distance can be used. The concrete mathematical foundations are described in the literature. Striking points from the first picture ( 13 ) and the second image ( 14 ) are corresponding if the comparison of their descriptors gives the greatest similarity. From these corresponding points then the disparities ( 20 ) and stored in a result matrix. The evaluation also takes place here in that the density matrix of all disparities ( 20 ) is determined. The density function is compared to a density function that was previously done with a natural person ( 1 ) was determined. If both functions correlate, then the live recognition is passed.

In a further technical embodiment, the person ( 1 ) successively taken multiple images, the illumination is in each case from a different direction. The respective different direction of the illumination causes the formation of characteristic shadows due to the three-dimensional extent of the person to be photographed ( 1 ). The characteristic shadows can be recognized by image processing algorithms in the recorded images. If the recognition of the characteristic shadows can be successfully performed, the live recognition is passed.

In a further technical embodiment, a three-dimensional model of the face of the person is 1 ) generated. For this purpose, at defined distances to the camera ( 6 ) Projectors arranged a light pattern on the face of the person to be photographed ( 6 ) project. The three-dimensional model of the face is used to determine the distribution of the coordinates to the optical axis ( 15 ) the camera ( 6 ) are parallel. Furthermore, the correlation coefficient is calculated from the distribution of these coordinates and a distribution as a reference of several persons ( 1 ) was determined. If the Correlation coefficient exceeds a threshold value, the live detection is passed.

In a further technical embodiment, the three-dimensional model of the person's face ( 1 ) with a three-dimensional reference model for similarity. For this algorithms are described in the literature.

It is suggested to put a device after DE 10 2013 203 433 A1 around a display ( 9 ) to extend that directly above the camera ( 6 ) in the semitransparent mirror ( 8th ) is integrated. The display of content in this display ( 9 ) the attention of the person ( 1 ) to attract. This will cause eye movement and head posture of the person ( 1 ) are favorably influenced for image acquisition. The same effect can be achieved with a ring-shaped illumination around the camera ( 6 ) is arranged.

embodiment

By way of example, a technical embodiment of the invention will be explained, which is applied to a "device for collecting personal data", as in the published patent application DE 10 2013 203 433 A1 is described. The published patent DE 10 2013 203 433 A1 For example, from the paragraph [0086], the description of a process of acquiring the personal data may be taken. It describes how, after switching on the "device for recording person-specific data", the program for controlling all elements by the control unit ( 2 ) is started automatically as the device is driven into a ground state and how the camera ( 6 ) is positioned at a height of 1700 mm. Furthermore, it describes how the recording of personal data takes place while the camera ( 6 ) is positioned so that the eye position of the person depicted in the camera image ( 1 ) reaches a predetermined value in the vertical direction. Paragraph [0093] then describes how the camera images are retracted to capture them as personal data. According to published patent DE 10 2013 203 433 A1 At this point, the capture of camera images is completed, with only in a single, optimal camera position images were fed. As a result, there are at first 3 camera images which, with one and the same optimum camera position, are illuminated by the illuminated person ( 1 ) and of which the person ( 1 ) can select one.

According to this invention is in the published patent application DE 10 2013 203 433 A1 described sequence supplemented by the examination of the spatial extent of the camera object. For this purpose, first of the program one of the 3 camera images of the illuminated person ( 1 ) as the first camera image ( 13 ). After that, the camera ( 6 ) by the drive unit ( 10 ) 6 cm upwards. Then the lighting device ( 7 ) and an image as the second camera image ( 14 ) moved in. The program now leads to the first image ( 13 ) and with the second image ( 14 ) performs a block matching procedure and stores the determined disparities ( 20 ) in a result matrix. The block size for the block matching method is chosen so large that only in flat camera objects disparities ( 20 ) can be determined. A concrete value for the block size that satisfies this property was determined prior to commissioning by experiments.

Then the program determines the number of disparities ( 20 ), which could be determined by the block matching method. If the number of disparities determined by the program ( 20 ) exceeds a threshold, then it has been recognized that the images were taken from a non-spatial object and the device outputs the sentence "The captured images are not allowed for capture." by voice output through the speakers.

In a further technical embodiment, images are taken in regular sequence, while the camera ( 6 ) is positioned so that the eye position of the person depicted in the camera image ( 1 ) reaches a predetermined value in the vertical direction. If the intended camera position is reached, then the last picture taken is the first camera picture ( 13 ). The image taken as person-specific data is the second camera image ( 14 ). With the pictures thus taken, the live recognition is then carried out according to the invention.

LIST OF REFERENCE NUMBERS

1

person

2

control unit

3

Storage

4

processor

5

interface

6

camera

7

lighting device

8th

Semitransparent mirror

9

display

10

drive unit

11

Facility

12

data line

13

First picture

14

Second picture

15

Optical axis

16

altitude change

17

Collinear image column

18

block

19

Corresponding point

20

disparity

21

Corresponding distinctive point

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 102013203433 A1 [0002, 0011, 0013, 0013, 0026, 0027, 0027, 0027, 0028]

Cited non-patent literature

• "Overcoming Scenarios for Biometric Systems" by Stefan Hein and Markus Mahrla, October 18, 2004 [0003]
• "Http://www.secupedia.info/wiki/Learning_recognition" (28/07/2016) [0006]

Claims (4)

Method for detecting the spatial extent of a camera object as part of a living recognition for devices for recording person-specific data, including the face image of a person ( 1 ), which checks whether the facial image of a real person ( 1 ) or from a flat image with the following characteristics: 1. The device comprises a camera ( 6 ) for recording the facial image as person-specific data. 2. The device comprises a drive unit ( 10 ), with which the position of the camera ( 6 ) can be adjusted in the vertical direction. 3. The device comprises a lighting device ( 7 ) used for the front lighting of the person ( 1 ) is trained. 4. The device comprises a semi-transparent mirror ( 8th ), between the person ( 1 ) and the camera ( 6 ) and whose mirror surface is perpendicular to the optical axis ( 15 ) the camera ( 6 ) is aligned to receive the facial image. 5. The device comprises a control unit ( 2 ), which has a memory ( 3 ) and a processor ( 4 ) and on which a program operates, which carries out the following procedure: 6.1. With the camera ( 1 ) is a first facial image of the person ( 1 ) as person-specific data. 6.2. The position of the camera ( 6 ) is changed in the vertical direction. 6.3. With the camera ( 6 ) is a second facial image of the person ( 1 ). 6.4. A correspondence analysis in the vertical direction is carried out, which analyzes the disparities ( 20 ) corresponding points ( 19 ) from the first face image ( 13 ) and the second face image ( 14 ) and stored in a result matrix. 6.5. The first face image is allowed as person-specific data if, by analyzing the result matrix with mathematical methods, it is determined by the camera ( 6 ) person ( 1 ) has a three-dimensional extent or the first face image is discarded as person-specific data if it is determined by mathematical means that the camera ( 6 ) person ( 1 ) has a two-dimensional extent. Method for detecting the spatial extent of a camera object as part of a living recognition for devices for recording person-specific data, including the face image of a person ( 1 ), which checks whether the facial image of a real person ( 1 ) or from a flat image, according to claim 1, wherein as a correspondence analysis, a block matching method is carried out in the vertical direction, for which a block size is set, with the disparities ( 20 ) can only be found if the camera object has a two-dimensional extent, wherein the first face image ( 13 ) is discarded as person-specific data when the number of determined disparities exceeds a threshold. Method for detecting the spatial extent of a camera object as part of a living recognition for devices for recording person-specific data, including the face image of a person ( 1 ), which checks whether the facial image of a real person ( 1 ) or from a flat image, according to claim 1, wherein as a correspondence analysis, a block matching method is performed in the vertical direction, which is configured so that it optimally disparities both in images of camera objects with two-dimensional and three-dimensional extent ( 2 ), where the first face image ( 13 ) is permitted as person-specific data if the frequency distribution of the disparities ( 20 ) is correlated with a frequency distribution that coincides with natural persons ( 1 ) was determined. Method for detecting the spatial extent of a camera object as part of a living recognition for devices for recording person-specific data, including the face image of a person ( 1 ), which checks whether the facial image of a real person ( 1 ) or from a flat image, according to claim 1, wherein a feature-based correspondence analysis is performed in the vertical direction, wherein the first face image is permitted as person-specific data, if the frequency distribution of the determined disparities ( 20 ) correlates with a frequency distribution associated with natural persons ( 1 ) was determined.

Images