SYSTEM FOR CAPTURING A COLOR IMAGE AND AN INFRARED IMAGE OF A SCENE

Description

[0001] The invention relates to a method for acquiring a color image and an infrared image of a scene, as well as the acquisition system implementing said method.

[0002] The invention applies, in particular, to technical fields such as security for the recognition of biometric information, including faces, of an individual. Prior art

[0003] To create a facial recognition system with anti-fraud, a camera system that simultaneously acquires both a near-infrared image of the face and an image in the visible spectrum is classically used.

[0004] To reduce the cost and size of such a system, we are familiar with document FR3102324, which describes a device for combining the two images on a single image sensor. This RGB image acquisition device does not include an infrared filter and incorporates a high-pass filter configured to attenuate the blue component of the light in order to recover infrared information on blue photosites. This device offers good resolution of the green component; however, in the lighting environments where such a device is used, there can be very significant differences in intensity between infrared and visible light, making it difficult, if not impossible, to use both infrared and visible images simultaneously due to the limited dynamic range of the image sensors.Furthermore, with only one set of exposure parameter settings such as exposure time and gain, this device cannot correctly reproduce both images. EP 3242249 A1 describes a fingerprint capture device that performs sequential acquisitions at different wavelengths with illumination adjustment based on the acquired image.

[0005] Use cases where the differences in level between infrared and visible light are very significant include: Bright environments without natural ambient light, i.e. including ambient light only from an artificial source, such as LED lighting (Light Emitting Diode), because these environments contain almost no infrared or near-infrared light signals, but only visible light signals; bright environments where natural light is filtered by means of, for example, heat-resistant film placed on windows, because these environments also contain very little infrared light signal; non-bright environments, because in complete darkness no signal is received.

[0006] One solution could involve adding visible light (also called white light) and infrared light to the system when the two cameras are separate, each sensitive to only one light source, either visible or infrared. However, if there is only one camera sensitive to both wavelengths, this solution cannot be applied directly. Furthermore, integrating two cameras creates constraints related to design, cost, and size.

[0007] The present invention aims to overcome at least some of these drawbacks, possibly leading to other advantages. Description of the invention

[0008] The invention relates to a system for acquiring a color image and an infrared image of a scene, said acquisition system comprising: an image acquisition device, said device being sensitive to visible and infrared wavelengths, comprising: a sensor; a lens configured to focus the light received at the input of the acquisition device onto the sensor, a dual-band pass filter being placed in front of the sensor, and in particular between the lens and said sensor; a processing module connected to said image acquisition device configured to separate, at the sensor output, the infrared image and the visible image and to define at least one signal to be controlled and an associated control setpoint, said at least one control setpoint being a control setpoint for an infrared and / or visible signal according to which at least one artificial light source is controlled;at least one artificial light source illuminating said scene, said at least one source comprising an artificial infrared light source emitting a light signal in the infrared and / or an artificial visible light source emitting a light signal in the visible spectrum.

[0009] This acquisition system allows, with a single camera, and therefore an advantage in terms of compactness, to obtain two usable images (infrared and visible), for biometric recognition or visual code reading, in all lighting conditions.

[0010] For example, the dual-bandpass filter could consist of a network of filters.

[0011] Advantageously, said image acquisition device includes a dual-bandwidth filter of which: in the case of a sensor with at least four different photosites, the first bandwidth of the filter is configured to allow visible wavelengths to pass through, in particular those below 650 nm, and in the case of a sensor with three different photosites, the first bandwidth of the filter is configured to attenuate the blue component of the light, in particular with a first bandwidth extending for example between 530 and 650 nm; and the second bandwidth is configured to allow infrared wavelengths to pass through, in particular the second bandwidth extending from 800 to 875 nm; This allows, in particular, thanks to the restricted infrared band, for repeatability in the rendering of infrared images versus external lighting conditions, the spectrum is therefore controlled.

[0012] For example, in these sensors, the four different photosites include in particular the Red, Green, Blue and Infrared photosites and the three different photosites are in particular the Red, Green and Blue photosites, knowing that according to the invention in the case of such a three-photosite sensor, this last photosite is only sensitive to Infrared, due to the application of the filter, which allows good acquisition of faces since the blue component is hardly represented there.

[0013] Advantageously, the second bandwidth extends from 760 nm to 800 nm, allowing for the recovery of sufficient infrared signal even in the presence of a heat-resistant film, as this film cuts off around 800 nm. Indeed, without this extended bandwidth, in the presence of a heat-resistant film, there could be more than 10 klux of visible light and virtually no infrared. Therefore, the system would need to be capable of providing infrared illumination that generates the same signal level as the 10 klux visible light in order to distinguish the infrared.But thanks to the widening of the bandwidth, there is enough infrared to obtain an image, without the need for infrared lighting, the size of which can be reduced so that it is capable of generating a signal level equivalent to a visible illuminance of only 1000 lux, in order to meet the use case corresponding to strong artificial ambient light (not emitted by the system), by LED for example.

[0014] In an advantageous way: the artificial light source emitting in the infrared is configured to emit, within a range of usage distance, in particular from 40 to 100 cm, a signal equivalent to the signal generated by intense ambient visible light, in particular 1000 lux, for example a light-emitting diode at 850 nm, whose spectrum extends from 800 to 875 nm and / or the artificial light source emitting in the visible is configured to illuminate satisfactorily in complete darkness, in particular 50 lux at a usage distance of about 70 cm; This allows for the use of powerful visible and infrared or near-infrared light sources without presenting any risk or discomfort (glare) to the user, as the infrared lighting can be modulated in intensity, notably between 0 and the equivalent of 1000 lux.

[0015] Advantageously, the system includes another imaging device sensitive only in the visible spectrum, synchronized with the acquisition device and whose exposure parameters are controlled independently of those of the acquisition device, which allows stereoscopy and thus the detection of fraud of all types at a limited cost.

[0016] Advantageously, the other device can include a monochrome sensor or a sensor with at least Red, Green and Blue photosites, which makes it possible to obtain stereoscopy at a reduced cost.

[0017] The invention also relates to a method for acquiring a color image and an infrared image of a scene, said acquisition method comprising the following steps: reception of an image from an image acquisition device sensitive to visible and infrared wavelengths; processing of said received image separating the infrared image and the visible image and defining at least one signal to be controlled and an associated control command, said at least one control command being a control command of infrared and / or visible signal according to which at least one artificial source of infrared and / or visible light is controlled; the process offers the same advantages as the system.

[0018] Advantageously, the received image includes metadata, such as gain and exposure time specific to said received image.

[0019] Advantageously, the processing step includes a step of detecting an object in the color and / or infrared image and determining the distance between the detected object and said acquisition device, which makes it possible to estimate the additional signal produced by the artificial lighting of the system on the object.

[0020] Advantageously, the processing step includes a calculation step of at least two metrics on all or part of the image, the first metric characterizing the infrared signal, the second metric characterizing the visible signal, which makes it possible to quantify the two types of signals and then to compare them, in particular between them.

[0021] Advantageously, said part of the image is a region of interest of the image delimiting said detected object, said object being in particular a body part or information, such as a visual code, in particular printed on a physical document, which makes it possible to circumscribe the calculation of metrics to a region of interest in the image, such as a face in the context of, for example, a facial recognition application, or such as information in the case of an application to read physical documents.

[0022] Advantageously, the at least two metrics are calculated as being at least the average over all or said part of the image on each wavelength channel of said characterized signal, which makes it possible not to work with images close to saturation.

[0023] Advantageously, the processing step includes: a step of estimating a component of the first metric characterizing a quantity of infrared signal due to ambient infrared, i.e., not emitted by at least one artificial source of infrared and / or visible light; and; a step of calculating the ambient lighting distribution, in particular in the form of the ratio of the quantity of the component of the first metric characterizing the infrared signal due to ambient infrared to the second metric; and; a step of comparing said distribution to a predetermined target distribution. and in that the control setpoint of the infrared signal is a function of said comparison, and in particular in that the control setpoint of the visible signal is a function of said comparison; which allows a simple and low resource consumption calculation of the light distribution.

[0024] Advantageously, this predetermined target distribution is a function of the ratio of the infrared logic level to the visible logic level, these logic levels being determined in such a way that the sum of these two settings is less than half the maximum signal that can be reproduced by the camera. Indeed, since the visible channels on the camera are also sensitive to infrared, the sum of these two settings is preferably less than half the maximum logic level of the signal. N max which can be captured by the camera, for example N max = 1023 in the case of a camera with a 10-bit output, in order to avoid saturation and maintain sufficient margin.

[0025] Advantageously, the control setpoint for the visible signal is a function of said comparison, in that if the calculated ambient lighting distribution is greater than the target distribution then the controlled signal is the infrared signal on the infrared setpoint and otherwise the controlled signal is the visible signal on the visible setpoint, which makes it possible to reach the setpoint even in the case of less ambient infrared, knowing that there are hardly any opposing use cases, that is to say in which there is a significant excess of infrared in the scene.

[0026] Advantageously, the processing step includes: a step of calculating the exposure parameter of the acquisition device to be applied, such as gain and / or exposure time, in particular as a function of the ratio of at least one control setpoint to said related signal to be controlled, to reach said at least one control setpoint; and / or a step of calculating the control command of the artificial infrared and / or visible light source; This allows the control signal setpoint to be reached using easily controllable parameters.

[0027] Advantageously, the exposure parameters to be applied (in particular gain and exposure time) are determined in such a way that their product is equal to the product of the parameters being applied multiplied by a ratio of the setpoint to be controlled by the controlled signal, which allows for proportional and linear control.

[0028] Advantageously, the calculation step for the control command of the artificial infrared light source of the system is carried out by first calculating the additional infrared signal needed to reach the infrared signal control setpoint, knowing the ambient infrared signal, by subtracting from the infrared signal control setpoint said ratio multiplied by the ambient infrared signal: S IRadditionnel ′ = C IR − Ratio . S IRambiant , then we calculate a duty cycle to apply to the infrared lighting source.

[0029] Advantageously, the control of the artificial infrared light source of the system is achieved by modulation of the duty cycle, which makes it possible to generate a signal equivalent to a visible illuminance between 0 and 1000 lux in the range of distance of use.

[0030] Advantageously, during the calculation step of the control command for the artificial visible light source of the system If the white lighting was off, and the product of the exposure parameters exceeds a predetermined activation threshold, then the white lighting is turned on; if the white lighting was on, and the product of the exposure parameters is less than a predetermined extinction threshold, then the white lighting is turned off.

[0031] Advantageously, the control of the system's artificial visible light source is binary, allowing for simple hysteresis control.

[0032] Advantageously, the acquisition process includes an initialization step performing an iterative scan of configurations until the object is detected, each configuration comprising at least one exposure parameter, such as gain and / or exposure time, and one intensity parameter of at least one artificial light source, the values ​​of said parameters being fixed to predetermined values ​​in each configuration; this makes it possible to detect the object sought regardless of the lighting environment, because the fixed values ​​of the parameters of each configuration correspond to optima in distinct use cases, and as many configurations as identified use cases are tested one after the other iteratively until the object sought is "locked on".

[0033] Advantageously, the predetermined target distribution is defined during the initialization step.

[0034] Advantageously, these configurations are saved as a table in the processing module.

[0035] Advantageously, the acquisition method according to the invention is implemented by the processing module of the device according to the invention.

[0036] The invention also relates to a method for authenticating a body part, characterized in that it comprises: acquire a color image and an infrared image of said body part by the acquisition method according to the invention and in that the detected object is a body part; and authenticate the body part from said color and infrared images; said authentication method has the same advantages as the acquisition method, and enables biometric recognition.

[0037] The invention also relates to a device for authenticating a body part, comprising: an image acquisition system according to the invention; and the processing module being configured to detect a body part from the received image and to perform authentication of said detected body part from said color and infrared images; which has the same advantages as the acquisition system according to the invention and allows biometric authentication, in particular using the visible image to "recognize" the person and the Infrared image to verify the authenticity of the face (in other words to detect possible fraud).

[0038] The invention also relates to a computer program comprising instructions for implementing, by means of a device, the acquisition method according to the invention, when said program is executed by a computing unit of said device, said program having the same advantages as previously mentioned. The computing unit may, in particular, be that of the processing module of the device according to the invention.

[0039] The invention further relates to storage means for storing a computer program comprising instructions for implementing, by a device, the acquisition method according to the invention, when said program is executed by a computing unit of said device, said means having the same advantages as previously mentioned. The computing unit may, in particular, be that of the processing module of the device according to the invention. Brief description of the drawings

[0040] The invention will be better understood and its advantages will become clearer upon reading the following detailed description, given by way of example only and not as a limitation, with reference to the accompanying drawings in which: there figure 1 schematically presents an image acquisition system according to one embodiment of the invention; the figure 2 schematically illustrates an example of the hardware architecture of an image acquisition system processing module according to one embodiment; the figure 3 illustrates a method for acquiring an infrared image and a color image according to one embodiment; the figure 4 schematically presents an image acquisition system according to another embodiment of the invention; the figure 5 presents an authentication device according to one embodiment; and the figure 6 illustrates an authentication process according to one embodiment.

[0041] Identical elements represented in the aforementioned figures are identified by identical numerical references. Detailed presentation

[0042] There figure 1 illustrates an image acquisition system according to an embodiment of the invention.

[0043] The image acquisition system 100 includes an image acquisition device 1 comprising a lens 10, which includes at least one lens element. The lens 10 is configured to focus the light received at the system's input, and more specifically from the acquisition device 1, onto a sensor 14, such as a CCD (Charge Coupled Device) or CMOS (Complementary Metal Oxide Semiconductor) sensor. The sensor 14 is composed of a plurality of photosites. The image acquisition device 1 also includes a dual-band pass filter 12 positioned in front of the sensor 14 such that each photosite of said sensor sees only one color, specifically between the lens 10 and said sensor 14. The dual-band pass filter 12 may, in particular, be bonded to the lens, constitute a coating on the lens, constitute a coating on the lens glass 10, or be positioned in front of the lens.Alternatively, filter 12 could be an array of filters 0, notably placed at different positions in the optical path, for example a high-pass filter upstream of the lens and an array of color filters placed between the lens and the sensor.

[0044] The 14 sensor can have only three different photosites or at least four different photosites.

[0045] In the case of a sensor 14 comprising only three different photosites, the first bandwidth of the dual-band filter 12 is configured to attenuate the blue component of the light, so as to recover infrared information on photosites initially dedicated to blue, with the result that each photosite of the sensor 14 sees only one color: the red channel is thus sensitive to Red and Infrared via a first photosite, the green channel is sensitive to Green and Infrared via a second photosite, and the infrared channel is sensitive to Infrared via a third photosite; the first bandwidth therefore extending, for example, from 530 to 650 nm. The resulting device 10 is then called RVPIR for Red Green Near Infrared.

[0046] In the case of a sensor 14 comprising four different photosites, the first bandwidth of the dual-band filter 12 is configured to allow visible wavelengths to pass through, so that each photosite of the sensor 14 sees only one color, the red channel is thus sensitive to Red and Infrared via a first photosite, the green channel is sensitive to Green and Infrared via a second photosite, the blue channel is sensitive to Blue and Infrared via a third photosite, and the near-infrared PIR channel is sensitive to Infrared via a fourth photosite, the first bandwidth notably covering wavelengths below 650 nm.

[0047] Regardless of the sensor 14, the second bandwidth of the filter 12 is configured to allow infrared wavelengths to pass through, for example the second bandwidth extends from 790 to 875 nm, and preferentially extends from 760 nm-800 nm, so as to recover sufficient infrared signal even in the presence of heat-resistant film, this type of film cutting wavelengths around 800nm, while requiring only a significant oversizing of the infrared illumination.

[0048] The acquisition device does not include any infrared filter (called Infrared Cut in English), so it is sensitive to infrared on all its channels.

[0049] The acquisition device 1 with a sensor 14 composed of only three photosites is obtained in particular from an acquisition device consisting, for example, of a conventional Red Green Blue RGB camera from which the infrared filter has been removed and in which the color filter array is replaced by a filter designed to allow only red, green, and near-infrared (IR) light to pass through. The resulting device 1 is then called RVPIR for Red Green Near Infrared.

[0050] Acquisition device 1 with a sensor 14 composed of at least four photosites is obtained, for example, from a Red Green Blue and Near Infrared (RGBIR) camera combined with a dual-bandpass filter, also called a notch filter, that passes through the entire visible spectrum, i.e., specifically between 400 nm and 650 nm, and through the infrared spectrum, i.e., specifically between 800 nm and 875 nm, while blocking the remainder. The resulting device 1 is then called RGBIR for Red Green Blue Near Infrared. The difference compared to the previous acquisition device 1 with a sensor 14 composed of only three photosites is that it allows blue light (between 400 nm and 530 nm) to pass through instead of blocking it.

[0051] The acquisition device 1 also includes a processing module 16. The processing module 16 receives as input a raw image I B (Raw image) from sensor 14 and generates an infrared image I IRand a color image SCREW A raw image consists of the photonic information obtained after the conversion by the photosites of the incident photons into a digitized electrical signal.

[0052] In the case of a sensor 14 with only three photosites, the processing module 16 is configured to calibrate, by determining calibration parameters, in a known manner, the image acquisition device 1.

[0053] Similarly, in the case of a sensor 14 with at least four photosites, the image device 1 is calibrated in a known way.

[0054] The processing module 16 is also configured to generate an infrared image I IR and a color image SCREW from a raw image I B acquired by sensor 14 and predetermined calibration parameters. The processing module 16 is therefore capable of implementing the image generation process I IR and I SCREWAdvantageously, the processing module 16 generates I images IR and I SCREW which are synchronized.

[0055] The acquisition device 1 may include other elements well known from conventional RGB cameras. These elements are not shown in Fig. 1. The elements of the lens 10 may contain anti-reflective optical coatings to increase the signal-to-noise ratio in the acquired images. The acquisition device 1 generally includes a microlens array positioned above the sensor 14, whose role is to optimize light collection by the photosites. The acquisition device 1 also includes a signal processing module (not shown in Fig. 1) at the output of the sensor 14. Its role is to reconstruct, pixel by pixel, the two missing color information, via a demosaicing algorithm. The paper by Alleyson et al. entitled "Linear demosaicing inspired by the human visual system," published in April 2005 in IEEE Transactions on Image Processing 14 (4), 439-449, is an example of such an algorithm.

[0056] The signal processing module and the processing module 16 can be integrated into a single module or be separate modules.

[0057] Acquisition device 1 advantageously allows for obtaining color images I SCREW having good resolution of the green component, which is particularly important for algorithms operating on skin images, such as facial recognition algorithms. The acquisition device 1 simultaneously allows for the acquisition of infrared images I IR.

[0058] The acquisition system 100 here includes two artificial light sources 20,30 illuminating said scene, the first artificial light source 20 emitting a light signal in the infrared and the second artificial light source 30 emitting a light signal in the visible.

[0059] The said artificial sources 20,30 are positioned so as to illuminate the scene, and in particular the object 150, here a body part, represented in the form of a face.

[0060] The first artificial infrared light source 20 is sized to provide the sensor 14 with a signal equivalent to that generated by very strong visible light, for example, approximately 1000 lux. The infrared illumination can therefore be controlled to modulate its intensity between zero and the equivalent of approximately 1000 lux. This first artificial infrared light source 20 consists, for example, of one or more 850 nm LEDs, whose spectrum extends approximately from 800 to 875 nm.

[0061] The second artificial source 30 of visible light, also called white light source, is sized to allow sufficient signal in complete darkness without generating glare for the user; it generates an illuminance of, for example, 50 to 100 lux at a working distance of approximately 70 cm.

[0062] There Figure 2schematically illustrates an example of the hardware architecture of the processing module 16. The processing module 16 then comprises, connected by a communication bus 160; a processor or CPU (Central Processing Unit) 161; a random access memory (RAM) 162; a read-only memory (ROM) 163; at least one communication interface 165 allowing, for example, the processing module 16 to communicate with the sensor 14 of the acquisition device 1. Optionally, the processing module 16 includes a storage unit 164 such as a hard drive or a storage media reader, such as an SD card reader (Secure Digital).

[0063] The processor 161 is capable of executing instructions loaded into RAM 162 from ROM 163, external memory (not shown), storage media (such as an SD card), or a communication network. When the processing module 16 is powered on, the processor 161 can read instructions from RAM 162 and execute them. These instructions form a computer program that causes the processor 161 to implement the processes described in the following figures.

[0064] Preferably, module 16 has a centralized architecture, but it can also be implemented as a distributed architecture, both in terms of data and components, hardware or software, particularly in the case of dematerialized computing resources.

[0065] The processes described in relation to the following figures can be implemented in software form by executing a set of instructions by a programmable machine, for example a DSP (Digital Signal Processor), a microcontroller or a GPU (Graphics Processing Unit), or be implemented in hardware form by a dedicated machine or component, for example an FPGA (Field-Programmable Gate Array) or an ASIC (Application-Specific Integrated Circuit).

[0066] There figure 3This illustrates a method for acquiring an infrared image and a color image according to an embodiment applied to an acquisition system 100 comprising, in particular, a sensor 14 having only three photosites, as well as an infrared light source 20 and a visible light source 30. It should be noted, however, that in another embodiment the system might not include a visible light source 30, the infrared light source 20 being in this case the only one driven according to the method, for example in the case of an acquisition system whose operating conditions preclude complete darkness.

[0067] The first step represented here of the acquisition process 300 is an initialization step 301 setting the exposure parameters, such as gain and / or exposure time, and intensity parameters of at least one artificial light source of the acquisition system 100, to predetermined values.

[0068] Preferably, the initialization step 301 also includes a sub-step for determining the target logic (digital) levels of infrared and visible signals, as well as a predetermined target distribution of said signals: C IR / C visible , however this substep could be part of another step in process 300, for example comparison step 306. It's visible etc IR These are the target logic levels to which the signal is to be controlled in the visible and infrared images, respectively. These logic levels are determined based on the intended use of the images, particularly the needs of algorithms, especially biometric ones (recognition and fraud detection). Given that the visible channels on the sensor 14 photosites are also sensitive to infrared, the sum of these two inputs must necessarily be lower than the maximum logic level. N maxwhich can be restored by each photosite, for example N max = 1023 in the case of an acquisition device 1 with a 10-bit output. To avoid working with images close to saturation, a sufficient margin must be maintained. For example, these target logic levels are determined in such a way that: C IR + C visible < N max / 2 .

[0069] For example C visible = 120 and C IR =240 (for 10-bit images, signal divided across 1024 channels ranging from logic level 0 to 1023), knowing that these values ​​are sized here relative to low light, i.e., with a gain of 4, configured here as the maximum gain to limit noise in the image, because if there is more light, these targets will not necessarily be reached, but the image quality will be sufficient. This results in a predetermined target distribution of said signals, said target distribution being here C IR / It's visible :the ratio of the infrared logic level to the visible logic level.

[0070] Once the initialization step is completed, the 300 process of acquiring a color image and an infrared image of a scene takes place based on said exposure parameters.

[0071] Step 302 involves receiving an image acquired from the image acquisition device 1, which is sensitive to visible and infrared wavelengths. The image is acquired with the parameters defined in the previous iteration, namely: the exposure time of sensor 14, the gain of sensor 14, the intensity of the infrared illumination source 20, and the intensity of the white illumination source 30. To ensure the correct association between the image and these parameters, they are stored as metadata associated with the acquired image, so that upon image reception, this metadata is included in the image transfer. This association can alternatively be achieved using the solution presented in patent application FR2212301.

[0072] The processing of the received image itself comprises several steps. The processing separates the infrared and visible images and defines at least one signal to be controlled and a corresponding control command. This control command is a control command for the infrared and / or visible signal, based on which at least one artificial source of infrared and / or visible light is controlled. The processing is implemented by the processing module 16 of the acquisition system 100.

[0073] More specifically, the processing step includes a step 303 of detecting an object in the image and determining the distance d between the detected object and the acquisition device 1. Object detection in the image is, for example, performed in a known manner using an artificial intelligence algorithm or a neural network, as described, for example, in the article "You only look once: Unified, real-time object detection" by Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016), Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 779-788). The object 150 being sought is, for example, a face, and the choice of the face of interest is, for example, the closest face. A region of interest (ROI), for example, a rectangle, corresponding to the chosen face, is defined. The image used for the implementation of this step 303 of detection and distance determination is preferably the raw image I B (composite) or the color image I screw , but it could also be the infrared image I IR.

[0074] The distance d from the face to the device is also estimated, for example, based on the object's size in pixels in the image, and knowing the camera's intrinsic characteristics (focal length and pixel size in particular, which are known parameters, or may even be part of the metadata associated with the image during its acquisition). This distance d estimate is approximate but sufficient; a more precise distance d estimation method will be defined later.

[0075] Preferably, if no object is detected in the first received image, a loop, represented by a dashed line, is performed back to the initialization step 301. This step iteratively scans through configurations, acquiring and analyzing images until the desired object is detected in the acquired image. Each configuration includes exposure parameters, such as gain and / or exposure time, and intensity parameters (expressed in the table below as a duty cycle, ranging from 0 to 1) for at least one artificial light source. The values ​​of these parameters are set to predetermined values ​​in each configuration.

[0076] Thus, if the target is a face, the goal is to provide images that allow for face detection as soon as possible, regardless of ambient lighting conditions, particularly backlit situations where simple automatic gain control based on the average of the entire image or a fixed region would be insufficient. An initialization mode is defined according to the target object. In face mode, the acquisition system is configured with a set of parameters optimized for face detection under a given lighting condition. For each successive image, the parameters are modified to adapt to various conditions until a face is detected in one of the images.

[0077] All the configurations can be grouped in a table, for example: [Table 1] Parameter set No. Exposure time Gain Infrared intensity White light intensity 1 0,000625 s 1 0 0 2 0,01 s 1 0 0 3 0,02 s 3 0,1 0

[0078] For example, configuration 1 allows for the detection of a face with very strong illumination, such as in the case of direct sunlight; Configuration 2 allows for face detection in intermediate lighting conditions, such as indirect sunlight or strong ambient artificial lighting (i.e., not emitted by the system's artificial light sources). Configuration 3 allows for face detection in low lighting or complete darkness. In this configuration, the infrared light source (20) is activated at a level sufficient to obtain the necessary signal.

[0079] We chain together image acquisitions using configuration 1, then configuration 2, then configuration 3, and then repeat the process back to configuration 1 until a face is detected. If the object being searched for is, for example, a visual code, then the entire set of configurations will be different, adapted to the visual code mode.

[0080] Then, once the object is detected, process 300 continues with the optimization phase of the processing based on the detected object 150. In some cases, there may be temporarily no object in the capture volume, resulting in a lack of detection during step 303. In this case, system 100 can, for example, freeze its parameters until the next detection, and, for example, after a given time threshold (1 second, for example), loop back to the initialization step with configuration scanning.

[0081] The processing step includes step 304, which calculates at least two metrics on all or part of the image. The first metric characterizes the infrared signal, and the second metric characterizes the visible signal. This part of the image is, in particular, the region of interest (ROI) of the image, delimiting the detected object. The object is, in particular, a body part, such as a face, or information, such as a visual code (e.g., Quick Response Code, Machine Readable Zone), notably printed on paper or displayed on a reader with e-ink technology.

[0082] Each metric can be, at a minimum, the average over the portion of the image corresponding to the ROI, on each wavelength channel. For example, let's take S IR average of the infrared channel, and S visibleaverage of the visible channel. In the case illustrated here of a 1 RVPIR device with a 14 sensor having only three photosites, the visible signal has 2 components, red and green, and we can take for example: S visible = max ( Red S, green S ) . In the case of a 1 RGBPIR device with a 14-photosite sensor, we would take, for example: S visible = max ( Red S, Green S, Blue S ) .

[0083] In the case of a magnification lens, once the detection of object 150 in the first image has been carried out, a second image can be acquired with a magnification so that the entirety of the second image corresponds to the ROI of the first image, this implementation is nevertheless delicate in the case of a moving person because the movement of the object must also be taken into account.

[0084] Next, we proceed to step 305, which estimates the logic level of the signal due to ambient infrared. To do this, we execute a step estimating an S component. IRambiant of the first metric S IR characterizing a quantity of infrared signal due to ambient infrared, that is, not emitted by the sources 20,30 of system 100. Indeed, the first metric S IR The characterization of the infrared signal is based on an estimation of the logic level of the signal due to the presence of infrared, and knowing that the infrared signal in the image is the sum of the signal due to ambient infrared lighting outside the product, and the signal due to the infrared illumination emitted by the system, we obtain: Calculation of the additional signal S Additional IR produced by the infrared light source 20 S IRadditionnel = R IR ⋅ DC IR ⋅ T expo ⋅ Gain d 2 with d being the distance to object 150 previously estimated in step 303, T expo And Gain :the exposure time (in seconds) and the gain (unitless) applied during image acquisition, and R IR the system parameter characterizing the sensitivity of the optical chain (m² / s) and DC IR The current duty cycle of the infrared lighting source (unitless, between 0 and 1); Calculation of the ambient infrared signal S IRambiant that is to say, not emitted by at least one artificial source of infrared and / or visible light (knowing that the visible light source does not emit in the infrared): S IRambiant = S IR − S IRadditionnel .

[0085] Then comes a calculation step (not shown) of ambient lighting distribution, notably in the form of the ratio S IR ambiant S visible of the quantity of the component of the first metric characterizing the infrared signal due to ambient infrared on the second metric.

[0086] We then proceed to step 306, which compares the said distribution of ambient lighting. S IR ambiant S visible to the said predetermined target distribution C IR C visible during the initialization step 301, and the control setpoint for the infrared signal is a function of said comparison, and in particular in that the control setpoint for the visible signal is a function of said comparison.

[0087] Indeed, based on this comparison, we decide whether to try to control the visible signal from source 30 or rather the infrared signal from source 20: If S IR ambiant S visible < C IR C visible So we are in the nominal case where there is less infrared than visible light. In this case, we will control the visible signal to the predetermined target visible logic level, which becomes the visible setpoint; otherwise, there is more infrared than visible light, so we control the infrared signal to the infrared setpoint.

[0088] In what follows, we define S as the signal that we will control, which is equal to S IRambiantor S visible according to the criterion thus defined, and C the instruction to which it will be slaved, that is to say respectively the logic levels C IR or C visible according to the same criterion.

[0089] Sometimes a set point can already be reached, particularly intrinsically depending on the proximity of the place to a window for example.

[0090] Finally, the processing stage includes: a step 307 of calculation of the exposure parameters of the acquisition device 1 to be applied, such as gain and / or exposure time, in particular as a function of the ratio of at least one control setpoint to said related signal to be controlled, in order to reach said at least one control setpoint; a step 308 of calculation of the control command of the artificial infrared light source; a step 309 of calculation of the control command of the artificial white light source, i.e. visible.

[0091] More specifically, in step 307 of the calculation of the new exposure parameters: • S is possibly corrected by a non-linear law (typically a polynomial, for example of degree 3) allowing for the consideration and compensation of any signal saturation; this corrected value is called S'. • The ratio Ratio=C / S' is calculated, which corresponds to the gain that must be applied to the analyzed image to reach the target; • Ratio is restricted between an upper bound ratio_min, set for example at 0.1, and an upper bound ratio_max, set for example at 10; • From this, the new product E = T expo .Gain .Ratio is deduced. • The task then becomes distributing Texpo' and Gain' so as to have T expo '.Gain' = E, following a gain or exposure time optimization strategy, which depends on the desired performance (in terms of motion blur and noise, in particular).Indeed, exposure time and gain act proportionally on the image, resulting in a synergy and their determination is based on a compromise, because lengthening the exposure time can create motion blur (a maximum motion blur can be defined according to the application and the associated algorithm) while increasing the gain generates an increase in noise.▪ For example, the exposure time Texpo is first determined to be as long as possible while remaining less than E, within the limit of a maximum exposure time (e.g., 20 ms or 40 ms), and then Gain' = E / Texpo' is deduced; ▪ It is also advantageous to favor exposure times that are multiples of the ambient lighting period (e.g., 10 ms in Europe and 8.33 ms in the United States) to avoid flicker effects due to non-continuous lighting; ▪ It is also possible to avoid updating these sensor configuration parameters 14 if the variation compared to the previous image is very small (e.g., variation <10%), which allows for a more stable image in terms of brightness. Indeed, since the measurements are noisy, small variations may be due to measurement noise (e.g., in the distance estimation) rather than a true variation in ambient brightness. .

[0092] More specifically, during step 308 of the calculation of the control command for the artificial infrared light source 20, the signal is calculated first S IRadditionnel ′ Additional infrared signal is desired on the next image based on the target infrared setpoint C IR (target infrared logic level) and the ambient infrared signal. S IR ambiant ′ that we anticipate in the next image, knowing the current ambient infrared signal S IR ambient and the ratio Ratio that we will apply to the exposure of device 1: S IR ambiant ′ = Ratio . S IR ambiant , which gives S IR additionnel ′ = C IR − Ratio . S IR ambiant

[0093] Then we calculate the new duty cycle DC IR ′ for infrared lighting according to: DC IR ′ = S IR additionnel ′ . d 2 R IR . T expo ′ . Gain ′

[0094] During step 309 of calculating the control command of the artificial white light source 30, given the desired level of quality in terms of noise on the image for the exploitation of visible images, and knowing that the noise increases with the gain of device 1 (at a given output signal level), it is necessary not to exceed a certain gain of device 1.

[0095] Here, the control of the artificial white light source 30 is binary, on or off, for the sake of simplicity, but its intensity could alternatively be modulated. Thus, If the white light source 30 was off and T expo '.Gain' ≥ the ignition threshold, then the white light source 30 is turned on, with, for example, the ignition threshold Ignition Threshold = 4 x 0.02 (gain of 4 and exposure time of 20 ms); if the white light source 30 was on and T expo '.Gain' <Seuil ex-tinction alors on éteint l'éclairage blanc, le seuil d'extinction Seuil extinction étant choisi de manière à ce que l'allumage de l'éclairage blanc sur l'objet observé ne provoque pas une chute du gain suffisante pour passer à l'itération suivante, c'est-à-dire à l'acquisition d'image suivante, sous le seuil d'extinction, ce qui provoquerait un clignotement et une instabilité du système, par exemple Seuil extinction = 1 x 0,02 (gain de 1 et temps d'exposition de 20 ms), avec un éclairage générant environ 50 lux sur l'objet.

[0096] Once the light is switched on, it is also advantageous not to switch it off until a certain amount of time has passed, in order to avoid discomfort for the user.

[0097] Preferably, the two control commands for the artificial infrared light source 20 and the artificial white light source 30 are defined at each iteration.

[0098] In the above description of the acquisition method 300, only the value of the ambient infrared signal, i.e. not emitted by the artificial infrared light source 20 or by the artificial visible light source 30 of the system 100, is calculated because the modulation of the infrared signal is preferred over the modulation of the visible signal in order to avoid the lighting of the visible light source 30.

[0099] However, we could also calculate the value of the ambient visible signal S not emitted by the sources 20,30 of the system 100 (especially 30 here because the source 20 intrinsically only generates infrared) and calculate the additional visible signal S produced by the lighting source 30 in order to avoid the deviation when the visible light source 30 is switched off and the related reconvergence time.

[0100] This 300 iterative control acquisition process allows for: a. Optimize the exposure time and sensor gain, according to the ambient visible illumination, to reach a target level C visible on the visible image. b. Continuously adjust the intensity of the infrared illumination so that the infrared image reaches a determined target C IR (in addition to the infrared component already present in the ambient illumination). c. Control the activation of the visible light source 30 (here on or off) in the case where it detects a lack of visible light, and control its deactivation when the visible illumination becomes sufficient again.

[0101] There figure 4This schematically illustrates an image acquisition system according to another embodiment of the invention. Indeed, the acquisition system 100 here comprises not only the acquisition device 1 from which a visible image I and an infrared image I are produced, but also another visible image acquisition device 2, this other device being only sensitive to visible light, such as an RGB camera, operating at the same time as the RVPIR image acquisition device 1.

[0102] The exposure parameters of this other device 2: exposure time and gain are then controlled according to an algorithm independent of the RVPIR image acquisition device 1.

[0103] The image acquired by this other device 2 is used in particular to provide video feedback to the user and for biometric recognition.

[0104] Furthermore, by coupling this image with the visible image from the RVPIR image acquisition device 1, it is possible to detect flat forgeries using stereoscopy. A 3D model of the target object is obtained. This 3D model is used to detect flat forgeries, or those whose overall shape does not closely resemble a face. The spacing e between the two devices 1 and 2 is approximately 16 cm, which allows the detection of flat forgeries up to approximately 90 cm.

[0105] It is therefore advantageous to position the infrared illumination source 20 at an angle α greater than 5° relative to the RVPIR image acquisition device 1, as viewed from the face 150. The aim is to avoid generating a "red-eye" effect (also known as "bright pupil"), which would be dependent on the distance to the person and introduce variability into the images, thus reducing the performance of fraud detection. Given the depth of the capture volumes (distance d on the order of approximately 40-90 cm), this implies a distance D of 6.5 cm to 7.5 cm between the RVPIR image acquisition device 1 and the infrared illumination source 20.

[0106] Advantageously, the infrared (IR) channel, or all the channels of the second camera, are processed by artificial intelligence in order to detect 3D mask-type fraud.

[0107] The combination of these processes makes it possible to detect fraud of all types, and to ensure biometric recognition, in all lighting conditions.

[0108] Stereoscopy also allows the distance d to be determined precisely.

[0109] The processing module 16 then communicates in particular via its communication interface 165 with the cameras of the two devices 1,2.

[0110] There figure 5 illustrates an authentication device A according to one embodiment. The authentication device A for a body part includes an image acquisition system such as the acquisition system 1 described with reference to figure 1 and an image analysis module 3.

[0111] The image processing module 3 includes a module 30 for body part recognition, such as a face, and a module 32 for fraud detection. It can optionally include a module 34 for activating access to a location, e.g., a building, a room, etc.

[0112] The body part recognition module 30 is linked to a database 4. This database 4 may be part of the authentication device A or external to it. The database stores images or image descriptors of the body parts of authorized individuals, e.g., individuals authorized to access a building.

[0113] There figure 6 illustrates an authentication process as implemented by authentication device A of the figure 4 .

[0114] The user wishing to be authenticated presents their face in front of the image acquisition system 1.

[0115] During an S30 step, a raw image of a face 150 illuminated by infrared and visible light is obtained by the acquisition system 1.

[0116] During step S32, a color image Ivis and an infrared image IIR are obtained from the optimized configuration parameters of the acquisition system 1. Known demosaicing algorithms can be applied to obtain the missing values ​​for each pixel. Such an algorithm makes it possible to obtain, for each pixel of the color image, an R value and a V value, and for each pixel of the infrared image, an IR value.

[0117] During step S34, a face recognition algorithm is applied. For example, the resulting color image Ivis is processed, e.g., segmented, to extract a face. The extracted face is then compared with face images or their descriptors stored in database 4. If the extracted face is similar, according to a certain metric, to a face in database 4, then the extracted face is recognized; otherwise, the face is unknown. In a variant, the face recognition algorithm uses the color image Ivis and the infrared image IIR. In the case S36 where the face is unknown, access is denied at step S38.

[0118] During step S40, a fraud detection algorithm is applied. For this purpose, the infrared (IR) image is used. In a variant, the infrared (IR) image and the color (I) image are used. A neural network can be used for this purpose.

[0119] In the case S42 where fraud is detected, i.e. that the face is not authentic (e.g. use of a mask), access is denied at step S38.

[0120] Access is validated during an S44 step only if face 150 is recognized and authenticated.

[0121] In a particular embodiment, an image i' is generated by concatenating the images I IR and I vis. To this end, a demosaicing process is applied to the images I IR and I vis, i.e., after correction by calibration parameters. The image I' therefore has three components: R, V, and IR. This image I' is then used in steps S34 and S40 by face recognition and fraud detection algorithms. These algorithms can use convolutional neural networks that have learned their coefficients on images of the type of image I'. For this purpose, resnets can be used. Such networks are notably described in the paper by He et al. entitled "Deep residual learning for image recognition" and published in 2016 in the Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778).

[0122] In another embodiment, the color image I with the two components R and G is colorized in order to recreate the blue component and thus obtain an image I" having 3 components R, G and B. The process described in the document by Zhang et al. entitled "Colorful image colorization" published in ECCV in October 2016 is an example of a colorization process based on convolutional neural networks.

[0123] This solution allows the application of traditional face recognition algorithms that typically use RGB color images. The algorithm described in the paper by Parkhi et al. entitled "Deep face recognition," published in September 2015 in BMVC (Vol. 1, No. 3, p. 6), is an example of such a face recognition algorithm. The image "I" is used by the face recognition algorithm during S34. The fraud detection algorithm can use either only the image "I IR" or both the image "I IR" and the image "I". For this purpose, the algorithm described in the paper by Agarwal et al. entitled "Face presentation attack with latex masks in multispectral videos," published in 2017 in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (pp. 81-89), can be used.

[0124] Colorization is particularly useful for viewing or exporting images.

[0125] Preferably the entire authentication process is implemented by the processing module 16, knowing that this module can be centralized or not.

Claims

1. System (100) for acquiring a color image (Ivis) and an infrared image (IIR) of a scene, said acquisition system comprising - at least one artificial light source illuminating said scene, said at least one source comprising an artificial source (20) of infrared light emitting a light signal in the infrared: - an image acquisition device (1), said device being sensitive to visible and infrared wavelengths, comprising: - a sensor (14); - a lens (10) configured to focus the light received at the input of the acquisition device (1) onto the sensor (14), a double band-pass filter (12) of which: - in the case of a sensor (14) with at least four different photosites, the first passband of the filter (12) is configured to pass visible wavelengths, and in the case of a sensor (14) with three different photosites, the first passband of the filter (12) is configured so as to attenuate the blue component of the light; and - the second passband is configured to pass infrared wavelengths, said double band-pass filter (12) being placed in front of the sensor (14), and notably between the lens (10) and said sensor (14); - a processing module (16) connected to said image acquisition device (1) configured to separate, at the sensor output (14), the infrared image (IIR) and the visible image (Ivis) and to define at least one signal to be servo-controlled and an associated servo control setpoint (C), said at least one servo control setpoint being a servo control setpoint for an infrared signal (CIR) as a function of which at least the artificial source (20) of infrared light is driven, said processing module being configured to: calculate (304) at least two metrics (SIR, Svisible) on all or part of the image, the first metric (SIR) characterizing the infrared signal, the second metric (Svisible) characterizing the visible signal; estimate (305) a component (SIRambient) of the first metric characterizing a quantity of infrared signal due to ambient infrared, that is to say not emitted by the at least one artificial source (20, 30) of light, and; calculate an ambient lighting distribution ( S IRambient S visible ) form of the ratio of quantity of the component of the first metric characterizing the infrared signal due to ambient infrared of non-artificial source over the second metric, and; compare (306) said distribution ( S IRambient S visible ) to a predetermined target distribution ( C IR C visible ), and in that the servo control setpoint (Cvisible, CIR) is a function of said comparison.

2. Acquisition system (100) according to the preceding claim characterized in that said at least one artificial light source comprises an artificial source (30) of visible light emitting a light signal in the visible and in that said at least one servo control setpoint is a servo control setpoint of visible signal (Cvisible) as a function of which the artificial source (30) of visible light is driven.

3. Acquisition system (100) according to any one of the preceding claims characterized in that said image acquisition device (1) comprises a double passband filter (12) of which: - in the case of a sensor (14) with at least four different photosites, the first passband of the filter (12) is configured to pass visible wavelengths lower than 650 nm, and in the case of a sensor (14) with three different photosites, the first passband of the filter (12) extends between 530 to 650 nm; and - the second passband extends from 800 to 875 nm.

4. Acquisition system (100) according to the preceding claim characterized in that the second passband extends from 760 nm-800 nm.

5. Acquisition system (100) according to any one of the preceding claims characterized in that: - the artificial source (20) of light emitting in the infrared is configured to emit, within a usage distance range, of 40 to 100 cm, a signal equivalent to the signal generated by intense external ambient visible light, at 1000 lux, for example a light-emitting diode at 850 nm, whose spectrum extends from 800 to 875 nm.

6. Acquisition system (100) according to any one of claims 2 to 5 wherein the artificial source (30) of light emitting in the visible is configured to illuminate satisfactorily in complete darkness, at 50 lux at a usage distance of approximately 70 cm.

7. Acquisition system (100) according to any one of the preceding claims characterized in that it comprises another shooting device (2) only sensitive in the visible, synchronized with the acquisition device (1) and whose exposure parameters are controlled independently of those of the acquisition device (1).

8. Method (300) for acquiring a color image (Ivis9. and an infrared image (IIR10. of a scene, said acquisition method comprising steps of: • reception (302) of an image (IB) from an image acquisition device (1) sensitive to visible and infrared wavelengths; processing of said received image (IB) separating the infrared image and the visible image and defining at least one signal to be servo-controlled and an associated servo control setpoint, said at least one servo control setpoint (C) being a servo control setpoint of infrared signal (CIR) as a function of which at least one artificial source (20) of infrared light is driven., the processing step comprising: a step (304) of calculating at least two metrics (SIR, Svisible) on all or part of the image, the first metric ( SIR) characterizing the infrared signal, the second metric (Svisible) characterizing the visible signal a step (305) of estimating a component (SIRambient) of the first metric characterizing a quantity of infrared signal due to ambient infrared, that is to say not emitted by the at least one artificial source (20, 30) of light, and; a step of calculating ambient lighting distribution ( S IRambient S visible ) in the form of the ratio of quantity of the component of the first metric characterizing the infrared signal due to ambient infrared of non-artificial source over the second metric, and; a step of comparing (306) said distribution ( S IRambient S visible ) to a predetermined target distribution ( C IR C visible ), and in that the servo control setpoint (CIR) of the infrared signal is a function of said comparison.

9. Acquisition method (300) according to the preceding claim characterized in that said at least one servo control setpoint (C) comprises a servo control setpoint (Cvisible) of the visible signal as a function of which an artificial source (30) of visible light is driven, and in that the servo control setpoint (Cvisible) of the visible signal is a function of the comparison.

10. Acquisition method (300) according to any one of claims 8 to 9 wherein the processing step comprises a step (303) of detecting an object (150) in the color or infrared image and determining distance (d) between the detected object (150) and said acquisition device (1).

11. Acquisition method (300) according to the preceding claim characterized in that said part of the image is a region of interest of the image delimiting said detected object, said object being notably a body part (150) or information, such as a visual code, notably printed on a physical document.

12. Acquisition method (300) according to any one of claims 8 to 11 characterized in that the at least two metrics (SIR, Svisible) are calculated as being at minimum the average on all or said part of the image on each wavelength channel of said characterized signal.

13. Acquisition method (300) according to any one of claims 8 to 12 characterized in that the processing step comprises: - a step (307) of calculating exposure parameter of the acquisition device (1) to be applied, such as gain and / or exposure time, as a function of the ratio of the at least one servo control setpoint (CIR, Cvisible) over said associated signal to be servo-controlled, to reach said at least one servo control setpoint; or - a step of calculating (308, 309) driving command for the artificial source (20, 30) of infrared and / or visible light.

14. Acquisition method (300) according to any one of claims 8 to 13 characterized in that it comprises an initialization step (301) performing an iterative scan of configurations until the detection of the object, each configuration comprising at least one exposure parameter, such as gain or exposure time, and an intensity parameter of the at least one artificial light source, the values of said parameters being fixed at predetermined values in each configuration.

15. A method for authenticating a body part characterized in that it comprises: acquiring a color image (Ivis) and an infrared image (IIR) of said body part by the acquisition method (300) according to one of claims 8 to 14 and in that the detected object is a body part (150); and authenticating the body part from said color (Ivis) and infrared (IIR) images.

16. A device (A) for authenticating a body part, comprising: - an image acquisition system (100) according to one of claims 1 to 7; and - the processing module (16) being configured to detect a body part (150) from the received image and to perform an authentication of said detected body part from said color (Ivis) and infrared (IIR) images.