Method, computer readable medium and vehicle for detecting spoofing in facial recognition

Abstract

Disclosed are methods for detecting spoofing in facial recognition, including: selecting an illumination pattern comprising at least three illumination intensities; controlling a light emitting device to emit light pulses in accordance with the illumination pattern; capturing respective images of an eye region of a face with a camera during each emitted light pulse; determining whether a glint is present in the eye region of each captured image; measuring a glint intensity of each determined glint; extracting a respective digital feature for each captured image, the digital feature representing a curve of all measured glint intensities at the glint intensity of the respective image; and outputting a signal that the face is spoofed if one or more extracted digital features do not correspond to a reference digital feature. Further disclosed are computer readable media storing instructions to perform the method and vehicles comprising a processor capable of performing the method.

Description

Technical Field

[0001] This invention relates to a method for detecting forgery in facial recognition, a computer-readable medium storing instructions for performing the method, and a vehicle comprising at least a processor configured to perform the method. Specifically, the invention relates to a method for detecting forgery in facial recognition using light pulses of varying intensities, and further to a computer-readable medium storing instructions for performing the method, and a vehicle comprising a camera, a light source, and a processor configured to perform the method. Background Technology

[0002] Facial recognition is a popular tool used to authenticate users and allow authenticated users to activate certain functions or use components, devices, etc. Facial recognition is based on using traditional hardware (such as CCD or CMOS sensors) to capture images or videos and extract facial features (such as the biometric properties of the captured face). If the facial features match the stored facial features of a pre-authenticated user, the user is allowed access to functions, components, devices, etc.

[0003] This authentication system is vulnerable to spoofing attacks, such as presenting a printed photograph or a video played on a monitor to an image sensor or camera, or having someone else wear a mask that mimics the identity of an authenticated user. Printed photographs and video displays sometimes allow spoofing attacks to succeed due to the high-quality printers and high-resolution displays used for forgery, and because only two-dimensional image or video data is available for facial recognition.

[0004] Therefore, by employing 3D sensors to obtain dense depth maps to acquire spatially distributed point patterns and by using triangulation to obtain sparse 3D depth information, the detection of various attack types (such as attacks using planar printed objects or flat panel displays) has been solved. However, this 3D technology is costly and cannot be installed in all devices. Summary of the Invention

[0005] The object of this invention is to provide a cost-effective yet efficient method for detecting forgery in facial recognition, as well as computer-readable instructions and a vehicle for performing this method.

[0006] According to a first aspect of the invention, a method for detecting forgery in facial recognition includes: selecting an illumination mode comprising at least three illumination intensities; controlling a light-emitting device to emit light pulses according to the illumination mode; and capturing corresponding images of the eye region of the face with a camera during each emitted light pulse. The illumination mode defines the number of light pulses emitted by the light-emitting device and the illumination intensity of each light pulse. Multiple illumination modes may be available, from which one is selected. The number of light pulses (equal to or greater than three) may be different or the same for each illumination mode.

[0007] Capturing the corresponding images may include capturing a number of images corresponding to the number of light pulses. Depending on the lighting pattern, each image is captured simultaneously with a light pulse, or at least over a period of time including one light pulse emitted by the light-emitting device. The number of captured images may be equal to or greater than the number of light pulses. The camera may be adjusted manually or automatically to frame the face or the area around the eyes of the face. Alternatively, the camera may be equipped with a wide-angle lens and / or guided to the specific position where the face will be captured by the camera. Moreover, the camera may be a digital camera that captures digital images, such as a camera including a CCD or CMOS sensor.

[0008] The method also includes: determining whether a flash is present in the eye region of each captured image; and measuring the flash intensity of each determined flash. A flash is a corneal reflection, i.e., the reflection of a light source on the outer surface of the cornea. In other words, the method includes: determining whether a reflection of a light-emitting device can be detected in the eye of the face during a light pulse. The closer the light-emitting device is to the eye, the more pronounced the flash, which helps in identifying a flash. For example, a flash is a bright reflection in the eye with a small circular shape. Measuring the flash intensity may also include measuring the absence of a flash, i.e., an intensity with (almost) zero or below a minimum threshold. Depending on the illumination pattern, such a "flash" may correspond to a light pulse with (almost) zero intensity.

[0009] The method further includes: extracting corresponding digital features for each captured image, the digital features representing a curve of all measured flash intensities at the corresponding image's flash intensity; and outputting a signal indicating facial forgery if one or more extracted digital features do not correspond to reference digital features. The curve of all measured flash intensities depends on the illumination intensity of all light pulses emitted according to the illumination pattern. For example, specific values ​​(Y-axis) for each measured intensity can be calculated, and the calculated values ​​can be mapped according to the time between measurements (i.e., between light pulses (X-axis)). The resulting graph forms a curve from which digital features cannot be extracted at each flash intensity (at each opportune position along the X-axis corresponding to a light pulse and / or a captured image).

[0010] This provides a strong anti-spoofing measure because the digital signature depends not only on the light intensity of a specific light pulse, but also on the measured flash intensity of temporally adjacent flashes / images. Since the latter further depends on the chosen illumination pattern, forgery of the digital signature can be prevented.

[0011] As an example only, if a printout of a face image including a flash is held in front of a camera, the flash intensity, depending on the illumination intensity (i.e., the intensity of light reflected by the printout (e.g., paper or photograph) at the flash location), will not change or may only change slightly. However, a "real" flash will vary more significantly, making the digital characteristics of a particular image or light pulse obtained from the curve different for the printout compared to a real flash, and thus detectable as a forgery. Similarly, even if a high-resolution display showing an image or video of a face is held in front of the camera, the displayed "flash" will likely result in digital characteristics different from what was expected.

[0012] In one implementation variation, extracting digital features may include determining the intensity level of the flashes in each captured image, or determining a normalized intensity level of the flashes in each captured image, and calculating the gradient of a curve defined by the determined intensity level or the determined normalized intensity level, or calculating the curvature of a curve defined by the determined intensity level or the determined normalized intensity level. For example, the absolute intensity level of the flashes in each captured image may be determined such that any extraction of digital features is based on a value or signal that can be directly derived from the camera (sensor). Alternatively, a normalized intensity level may be used as the basis for calculation. Normalizing the intensity level may be based on the maximum intensity of all flashes in all captured images or on the minimum intensity of all flashes in all captured images. Normalized intensity levels further improve anti-spoofing because the resulting curve or graph depends on the maximum or minimum intensity of the emitted light pulse, i.e., on the selected illumination mode, making the resulting curve more difficult to predict and therefore the resulting digital features more difficult to predict.

[0013] These intensity values ​​(absolute or normalized values) are then used to form a curve or graph, from which gradients or curvatures are calculated at the corresponding locations along each intensity value. The gradient or curvature reflects not only the flash intensity of a particular image, corresponding to a specific light pulse, but also depends on the flash intensity, and therefore on the illumination intensity of light pulses emitted before and after the particular image. This gradient and curvature are virtually unpredictable, thus preventing forgery.

[0014] In another implementation variation, the method may further include comparing digital features with reference digital features, wherein the reference digital features are associated with a selected lighting pattern. The selected lighting pattern may be pre-stored along with one or more digital features corresponding to individual light pulses defined by the lighting pattern. For example, using the same lighting pattern, multiple reference digital features for a specific selected lighting pattern can be calculated using multiple faces and / or one or more faces at different distances from the camera. This data base can be used when comparing digital features during a "live" session to detect forgeries.

[0015] In another implementation variation, the comparison can be performed by a machine learning process trained with multiple reference digit features and / or multiple lighting patterns. The machine learning process is capable of classifying multiple reference digit features captured and extracted from one lighting pattern and multiple lighting patterns. Therefore, the machine learning process can easily compare “real-world” digit features with one or more reference digit features. The machine learning process may not even be aware of the chosen lighting pattern, but when well-trained (i.e., based on a large number of lighting patterns and facial training), it can classify “real-world” digit features as valid or fraudulent.

[0016] In another implementation variation, the comparison may include comparing the difference between the extracted digital features and reference digital features with a threshold, and determining that the extracted digital features do not correspond to the reference digital features if the difference is greater than the threshold. In other words, the extracted digital features, represented by a specific value or vector, can be subtracted from the reference digital feature values, vectors, etc., and the resulting difference can then be easily compared with a threshold. For example, if the extracted digital features deviate from the reference digital features by less than 5% or less than 2%, the captured face can be authenticated as a valid user. If the difference is greater than the threshold (e.g., greater than 2% or greater than 5%), a signal indicating that the face has been forged is output.

[0017] In another implementation variation, the illumination pattern can define arbitrary levels of illumination intensity for each light pulse. In other words, the illumination intensity of each light pulse can differ from, be the same as, be zero, and / or correspond to the maximum light intensity of the pulses. While conventional anti-spoofing measures might use two varying light intensities one after another in time, arbitrary illumination intensity levels for each light pulse are much harder to forge. As an example, the illumination pattern can define a sequence of intensity levels such as 100%, 80%, 60%...0% or 0%, 20%, 40%...100% or 100%, 60%, 40%, 0%, 20%, 80% or 40%, 0%, 100%, 20%, 80%, 60%, etc. (where 100% is the expected or possible maximum intensity level of the emitting device).

[0018] It should be understood that the number of strength levels in the above example can also vary. For example, the number of strength levels could be three, five, ten, twenty, etc. The number of strength levels can be pre-selected, for example, based on the security level used for authentication and / or the environment (e.g., how much time is available for user authentication).

[0019] Therefore, the selection of the illumination pattern can include choosing a predetermined illumination pattern with this illumination intensity. This selection can be random to prevent flash forgery by predicting the next illumination pattern from an unauthorized user. To further enhance the security level of the method, the selection can alternatively or additionally include dynamically generating the illumination pattern. This can include selecting a predetermined pattern and randomly modifying it, for example, changing the order of intensities or changing one or more intensity levels. It should be understood that the method can also be based solely on randomly generated illumination patterns (including multiple intensities and / or the intensity levels of individual light pulses), i.e., without any predetermined pattern. Therefore, forgery is further prevented because pattern prediction is (virtually) impossible.

[0020] In another implementation variation, the method may further include illuminating the face using an additional light-emitting device that emits diffuse light to avoid generating flashes in the eye region. This additional light-emitting device can be used to ensure that the face is generally visible. To avoid a second flash, the additional light-emitting device can be mounted at a greater distance than the (first) light-emitting device emitting the light pulse, or it can be equipped with a diffuser. In any case, the additional light-emitting device may, for example, help determine the normalized intensity level of the flash. For example, the flash intensity can be correlated with the intensity level of another facial region or the average intensity level of the face or a portion thereof.

[0021] The other light-emitting device can provide constant illumination to the face throughout the entire duration of the emitted light pulse, depending on the illumination pattern. Therefore, if the illumination intensity level decreases, the flash becomes darker relative to the background, and vice versa. This makes flash detection easier because other facial areas are only slightly affected by the light pulse, as the passive illumination remains constant. These characteristics can be used, for example, by means of image processing (e.g., using differential imaging) to detect flash. If illuminated by other light-emitting devices, artificial flashes (e.g., a print of a face with flashes in the eyes) remain (almost) unchanged relative to the level of pulsed illumination.

[0022] Alternatively or otherwise, sunlight can provide diffused light, making the other light-emitting device usable only at night or when there is insufficient sunlight.

[0023] In another implementation variation, the method may further include estimating the gaze direction of the eye in at least one eye region of the captured image. For example, conventional eye tracker techniques may be employed, which utilize reflections in the eye, such as flashes of light.

[0024] Furthermore, if one or more estimated gaze directions do not correspond to the expected gaze direction, outputting a signal indicating facial forgery may include outputting that signal. For example, the method may also include requesting the user to look in a certain direction, such as looking at a display associated with a light-emitting device and / or a camera. Therefore, additional security measures can be implemented when detecting forgery in facial recognition.

[0025] In another implementation variation, the light-emitting device can emit infrared light. Infrared light can be used even at night, allowing the method to be performed independently of daytime or the presence of sunlight. Furthermore, the actual anti-spoofing measures are undetectable to the user, making it unknown when a printed object or display must be placed in front of the camera. Additionally, there is no glare that could distract the user (e.g., a driver of a nighttime vehicle) from light pulses.

[0026] According to a second aspect of the invention for better understanding, a computer-readable medium is configured to store executable instructions that, when executed by a processor, cause the processor to perform a method according to the first aspect or one or more associated implementation variations. For example, the computer-readable medium may be a volatile or non-volatile memory, such as a CD, DVD, USB stick, RAM, ROM, etc.

[0027] According to a third aspect of the invention for better understanding, a vehicle includes: a camera; at least one light-emitting device; a storage device storing a plurality of lighting modes including at least three lighting intensities; and a processor configured to perform a method according to the first aspect or one or more associated implementation variations. Therefore, facial recognition can be employed in a vehicle for any authentication purpose. The processor may be a dedicated processor or may be part of a vehicle computer system, such as an engine control unit (ECU).

[0028] In one implementation variation, the vehicle may also include a security system configured to deactivate vehicle components when the processor outputs a signal indicating that the face has been forged. The vehicle components could be, for example, an engine control unit, preventing the engine from starting when the processor outputs a signal. This allows for a keyless security system for using the vehicle. Alternatively, the vehicle components could be an infotainment system, preventing, for example, activation by a young driver. Other use cases involve authorizing payment processes (e.g., for parking or refueling) or accessing personal data via the infotainment system based on facial recognition (e.g., entering the nearest destination in a navigation application). In the case of forged facial data, such user requests should be denied.

[0029] In one implementation variant, the processor is also configured to execute the method whenever the vehicle begins its journey.

[0030] In another implementation variation, the vehicle may further include a driver assistance system comprising a camera configured to observe the driver of the vehicle. Some driver assistance systems observe the driver, for example, whether the driver is becoming fatigued. Therefore, the camera is pointed at the driver, which can be used for the anti-spoofing measures of the present invention, eliminating the need for additional cameras and / or light-emitting devices.

[0031] In another implementation variation, the vehicle may also include a display, wherein the processor is configured to use the display's illumination as an additional light-emitting device. Since displays are typically equipped with backlight units that provide diffused light, or since displays typically emit diffused light, such displays can be used to provide passive illumination for the face when displaying information. This passive illumination aids in the identification and extraction of flashes, as explained above regarding the first aspect.

[0032] In addition, the display can be used to provide guidance information to users, such as annotations in a specific direction, so that the direction of eye gaze can be used as an additional measure for detecting forgery in facial recognition.

[0033] According to another aspect, an electronic device includes a processor configured to perform a method according to the first aspect or one or more associated implementation variations. Such an electronic device can be any handheld device, such as a mobile phone, tablet computer, laptop computer, PC, etc., or a device installed in a machine.

[0034] This invention is not limited to the aspects and variations in the form and order described. Specifically, the description of aspects and variations should not be construed as a specific limiting grouping of features. It should be understood that this invention also covers combinations of aspects and variations not explicitly described. Therefore, various variations or optional features can be combined with any other aspect, variation, optional feature, or even combination thereof. Attached Figure Description

[0035] Preferred embodiments of the invention will now be explained in more detail with reference to the schematic accompanying drawings, in which:

[0036] Figure 1 An image describing the eye area of ​​the face;

[0037] Figure 2 A schematic diagram illustrating exemplary flash intensity is shown;

[0038] Figure 3 A flowchart illustrating a method for detecting forgery in facial recognition is shown schematically.

[0039] Figure 4 and Figure 5The detailed steps of a method for detecting forgery in facial recognition are illustrated schematically.

[0040] Figure 6 The vehicle is illustrated schematically; and

[0041] Figure 7 An electronic device is illustrated schematically. Detailed Implementation

[0042] Figure 1 An image depicting only a partial view of the eye region 15 of a human face 10 is described. Based on conventional image processing, the eye region 15 can be located in a larger image. Specifically, due to the specific form of the eyes 17 on the face, the eyes 17 can be identified, and then the image region 15 including the eyes 17 can be selected from the larger image.

[0043] The face is illuminated by at least one light-emitting device, wherein one light-emitting device forms a reflection 20 in the eye 17, which can be viewed as a bright circular spot 20 near the pupil of the eye 17. This reflection 20 is also referred to as a glint 20. Such a glint 20 is mainly present if a point light source emits light toward the cornea of ​​the eye 17 and forms a corresponding reflection on the cornea. The more diffuse the light illuminating the face 10, the less present or less visible the glint 20 will be. The present invention relates to generating multiple glints 20, for example, by emitting light pulses toward the eye 17, as will be discussed regarding Figures 3 to 5 A more detailed overview.

[0044] Figure 3 A flowchart illustrating a method for detecting forgery in facial recognition is shown schematically. The method begins by selecting an illumination pattern comprising at least three illumination intensities (step 110). Then, in step 120, a light source (e.g., light-emitting device 220) is controlled according to the illumination pattern. Figure 6 )) Emit light pulses. Optionally, in step 125, for example, by using another light-emitting device 225 ( Figure 6 The face 10 is further illuminated using diffused light. Although the light source controlled in step 120 emits light that causes a flash 20 in the user's eyes 17, the diffused light illuminates only the entire face 10 without causing a separate flash.

[0045] In step 130, a camera 210 is used during each emitted light pulse. Figure 6The corresponding image of the eye region 15 of the face 10 is captured. Therefore, for each light pulse, at least one image is captured. Then, in step 140, it is determined whether the flash 20 exists in the eye region 15 of each captured image. In an optimal scenario, each light pulse generates the flash 20, except for specific light pulses with very low or zero illumination intensity. The determination of the flash 20 can be achieved through image processing, because... Figure 1 The flash 20 shown can typically be detected as a bright spot within the eye 17.

[0046] Subsequently, the flash intensity of each determined flash 20 can be measured in step 150. (This is just an example.) Figure 2 Two curves or graphs are shown, the lower one representing the flash intensity measured in step 150 with respect to six light pulses (step 120) in six associated captured images (step 130). In the exemplary illumination mode, the illumination intensity of subsequent light pulses decreases by 20% from one light pulse to the next (see X-axis value). The measured flash intensity (Y-axis value) decreases accordingly.

[0047] Return to reference Figure 3 The illustrated method extracts corresponding digital features for each captured image in step 160. The digital features represent a curve of all measured flash intensities at the corresponding image's flash intensity. In other words, referencing... Figure 2 The digital feature of the third captured image (see value "60" on the X-axis) represents the curves adjacent to each other on either side of the flash intensity value (see value "60" on the Y-axis). In this particular case, the curve is concave or turns to the left (when viewed along the X-axis). In the case of the next captured image (see value "40" on the X-axis), the digital feature represents a convex or rightward curve (when viewed along the X-axis).

[0048] If one or more extracted digital features for all images (i.e., all light pulses and measured flash intensities) do not correspond to reference digital features, then in step 180, a signal indicating that face 10 is a forgery is output. Figure 3 For example, a printed object or display device can be held in front of the camera 210, wherein the printed object or display at least shows the eye area 15 having the eye 17 and even including the flash 20. However, reference Figure 2 During each light pulse, the printed material (paper / photograph) or display will reflect the light from the light-emitting device in almost the same way. Due to the lack of a “real” flash 20, the measurement of “flash intensity” in step 150 will actually measure the intensity (or brightness) of the printed or display material at the location of the captured flash.

[0049] Figure 2An exemplary second graph, namely the one above, is shown, which identifies the measured light intensity during six light pulses at the location of the fake flash. (As from...) Figure 2 The derived "flash intensity" decreases only slightly for the last two measurements, where the illumination intensity is very low depending on the illumination pattern. However, the measured "flash intensity" remains very high, for example, due to illumination of the print or display by another light-emitting device or sunlight. Since the digital features extracted from each of the second and subsequent measurement points differ from the reference digital features (i.e., the expected digital features), facial deception can be detected, and in step 180 ( Figure 3 The output of face 10 in the image is a fake signal.

[0050] Figure 4 and Figure 5 The detailed method steps for detecting forgery in facial recognition are illustrated schematically. Specifically, in order to extract digital features in step 160, the intensity level of each corresponding flash in each captured image can be determined in step 162. This intensity level can be obtained directly from the camera (sensor) output or after performing normal image processing on the sensor signals / data.

[0051] Alternatively or additionally, a normalized intensity level of flash 20 can be determined for each captured image in step 164. This normalization can be implemented in various ways, such as relative to the maximum illumination of the face 10 or eye region 15 during a maximum illumination pulse, or the corresponding minimum illumination, or relative to the average illumination of the face or eye region 15 or any other portion of the image when no light pulse is emitted. For example, in Figure 2 In the method, a normalized flash intensity is applied relative to the background (e.g., the entire eye region 15). Since the skin in the eye region 15 is quite bright, the simulated measured "flash" intensity is at or near 100%. The true flash intensity of the curve decreases as the illumination intensity of the light pulse decreases, because the background eye region 15 remains bright throughout the method.

[0052] Return to reference Figure 4 Regardless of the determination in step 162 or step 164, digital feature extraction 160 can be achieved by calculating the gradient of the curve defined by the determined intensity level or normalized intensity level in step 166. Alternatively or additionally, in step 168, the curvature of the curve defined by the determined intensity level or normalized intensity level is calculated. (As from...) Figure 2 This can be derived from the data, that is, for each light pulse and the captured image, at each measurement point, in... Figure 2 The gradient of the curve or curvature between the two figures shown in the figure is significantly different.

[0053] While traditional anti-spoofing measures compare the difference between the measured light intensity and the expected light intensity with a threshold, this comparison will lead to, for example, for Figure 2 The first two measurement points show false positives. On the other hand, according to the present invention, even for the first measurement point, the gradient and curvature are different, making it more reliable and also more efficient in detecting forgeries in facial recognition.

[0054] Go to Figure 3 and Figure 5 In step 170, the digital features extracted for each measurement point are compared with reference digital features. The reference digital features can be associated with the selected illumination mode. In other words, when an illumination mode is selected in step 110, associated reference digital features for each measurement point (i.e., each light pulse) can be derived (e.g., from a memory that also stores one or more illumination modes). Therefore, this reference digital feature is the expected digital feature.

[0055] The comparison of digital features may further include: in step 172, comparing the difference between the extracted digital features and the reference digital features with a threshold. Then, if the difference is greater than the threshold, it is determined that the extracted digital features do not correspond to the reference digital features, and the signal is output in step 180.

[0056] Figure 6 and Figure 7 The vehicle 1 and electronic equipment 200 are schematically illustrated, wherein the electronic equipment 200 may be part of the vehicle 1. A driver 5 sits in the vehicle 1, for example, in a driver's seat. The vehicle includes a camera 210 capable of capturing images of at least the eye region 15 of the driver 5, and also includes at least one light-emitting device 220 and a processor 250. The processor 250 is configured to perform... Figures 3 to 5 The method is illustrated and explained herein. Electronic device 200 also includes storage device 240, which stores multiple lighting modes including at least three lighting intensities, from which one can be selected in step 110.

[0057] The processor 250 can control the light-emitting device 220 to emit light pulses according to a selected illumination mode, while the camera 210 captures an image of the driver 5's eye region 15. The light-emitting device 220 can be an infrared (IR) light-emitting device, such as an IRLED, to generate a flash 20. Of course, instead of IR light or other than IR light, the light-emitting device 220 can emit visible light. The processor 250 can also perform method steps 140 to 180 to determine whether the face 10 captured by the camera 210 is fake.

[0058] The vehicle 1 may also include a safety system 260 configured to deactivate vehicle component 270 when the processor 250 outputs a signal that face 10 is spoofed. For example, if a spoofed face 10 is detected, the processor 250 may prevent the engine from starting.

[0059] Furthermore, the display 225 can be installed in the vehicle 1, whereby it can serve as an additional light-emitting device to provide diffused light for illuminating the face 10 of the driver 5. On the other hand, the display 225 can be used to present instructions to the driver 5 on how to use the anti-spoofing system, i.e., the functions of the processor 250 when performing method steps 110 to 180.

[0060] In a specific example, processor 250 can be configured to display instructions on display 225 to driver 5 to look in a specific direction. In another method step 190 ( Figure 3 In the image, processor 250 can estimate the gaze direction of eye 17 from at least one captured image, for example, based on one or more determined flashes 20. If the estimated gaze direction corresponds to the expected (indicated) gaze direction, additional security measures can be obtained to determine whether face 10 is fake.

[0061] The vehicle 1 may also be equipped with a driver assistance system 265, which includes a camera 210 and is configured to observe the driver 5 of the vehicle 1. Such a driver assistance system 265 can be configured to control the driver 5's gaze or estimate the driver 5's fatigue by processing images captured from the driver 5, particularly from the driver 5's face 10. The camera 210 of the driver assistance system 265, as well as the optional display 225 and / or any processing device (e.g., processor 250) of the driver assistance system 265, can be used in the disclosed method, making it unnecessary to install redundant equipment in the vehicle 1.

[0062] Figure 7 Electronic device 200 is illustrated schematically. Although electronic device 200 is illustrated as including storage device 240, processor 250, safety system 260, and driver assistance system 265, this configuration is to be understood as a configuration suitable for a vehicle. Electronic device 200 may also include camera 210, light-emitting device 220, and / or display 225. Safety system 260 and / or driver assistance system 265 may be omitted in different configurations of electronic device 200. For example, electronic device 200 may be a mobile phone, tablet computer, laptop computer, PC, etc.

[0063] The above description of the accompanying drawings should be understood as providing only exemplary embodiments of the invention, and should not limit the invention to these specific embodiments.

Claims

1. A method for detecting forgery in facial recognition, the method comprising: Select a lighting mode that includes at least three lighting intensities; The light-emitting device is controlled to emit light pulses according to the lighting pattern; During each emitted light pulse, a camera is used to capture corresponding images of the eye area of ​​the face; Determine whether a flash is present in the eye region of each captured image; Measure the flash intensity of each determined flash; Extract corresponding digital features for each captured image, the digital features representing a curve of all measured flash intensities at the flash intensity of the corresponding image; and If one or more of the extracted digital features do not correspond to the reference digital features, a signal indicating that the face is fake is output. Extracting the digital features includes: Determine the intensity level of the flash in each of at least three captured images, or determine the normalized intensity level of the flash in each of at least three captured images; and Calculate the gradient of a curve defined by at least three determined intensity levels or at least three determined normalized intensity levels, or calculate the curvature of a curve defined by at least three determined intensity levels or at least three determined normalized intensity levels.

2. The method according to claim 1, characterized in that, Also includes: The digital feature is compared with the reference digital feature, wherein the reference digital feature is associated with the selected lighting mode.

3. The method according to claim 2, characterized in that, The comparison is performed by a machine learning process trained using multiple reference digital features and / or multiple lighting patterns.

4. The method according to claim 2, characterized in that, The comparison includes: comparing the difference between the extracted digital features and the reference digital features with a threshold, and if the difference is greater than the threshold, determining that the extracted digital features do not correspond to the reference digital features.

5. The method according to claim 1, characterized in that, The lighting mode defines the lighting intensity at any level for each light pulse.

6. The method according to claim 1, characterized in that, Also includes: The face is illuminated using other light-emitting devices that emit diffuse light to avoid generating flashes in the eye area.

7. The method according to claim 1, characterized in that, Also includes: Estimate the gaze direction of the eye in at least one of the captured images in the eye region. The step of outputting a signal that the face is fake includes: outputting the signal if one or more of the estimated gaze directions do not correspond to the expected gaze direction.

8. The method according to claim 1, characterized in that, The light-emitting device emits infrared light.

9. A computer-readable medium configured to store executable instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 8.

10. A vehicle, comprising: camera; At least one light-emitting device; A storage device that stores multiple lighting modes with at least three lighting intensities; and A processor configured to perform the method according to any one of claims 1 to 8.

11. The vehicle according to claim 10, characterized in that, Also includes: A security system configured to deactivate vehicle components if the processor outputs a signal that the face is fake.

12. The vehicle according to claim 10 or 11, characterized in that, Also includes: A driver assistance system that includes the camera and is configured to observe the driver of the vehicle.

13. The vehicle according to claim 10 or 11, characterized in that, Also includes: monitor, The processor is configured to use the illumination of the display as another light-emitting device.

Images