A testing device and method for a structured light face recognition system

Abstract

The application discloses a kind of structural light face recognition system's testing device and method.The testing device includes the face recognition module to be measured, illumination projection module and dynamic simulation light field generation module;Wherein, illumination projection module emits optical beam to dynamic simulation light field generation module, the latter generates modulated light field after receiving, is projected to the photosensitive surface of the module to be measured by the transmission path of this optical module;The modulated light field is handled to the module to be measured, generates face recognition confidence score and reconstructs three-dimensional depth map, and is fed back to dynamic simulation light field generation module.The application is by constructing "digital simulation-reverse optical path calculation-physical light field reconstruction-closed loop optimization", solves the prior art only can generate 2D projection image, cannot restore high-fidelity 3D mask's three-dimensional optical characteristics such as light scattering and refraction, and it is difficult to comprehensively evaluate the problem of structural light face recognition system anti 3D attack robustness and three-dimensional reconstruction algorithm stability.

Description

Technical Field

[0001] This invention relates to a testing device for a structured light face recognition system, as well as a corresponding testing method, belonging to the field of image recognition technology. Background Technology

[0002] Structured light facial recognition systems, as a core biometric solution relying on 3D structured light technology, have deeply penetrated multiple key areas such as security access control, financial payments, government services, and smart terminal unlocking, thanks to their high-precision 3D reconstruction capabilities and efficient identity authentication. They have become a crucial technological support for ensuring public safety and personal information security. Its core working principle involves projecting a pre-set coded optical pattern (such as dot matrix, stripes, or grids) onto the face using a structured light projector. This pattern, deformed by the facial surface, is captured by an infrared imaging camera. A built-in 3D reconstruction algorithm then calculates depth information to construct a precise 3D facial model, which is finally compared with a target template in a database to complete identity verification. With the widespread application of this technology in high-security scenarios, the need to test its anti-spoofing capabilities and the robustness of its 3D reconstruction algorithm is increasingly urgent. The ability to accurately simulate real-world attack scenarios and comprehensively verify the system's anti-attack capabilities directly relates to the security and reliability of terminal applications.

[0003] Currently, the mainstream security testing method in the industry is 3D mask / head model attack testing. This method assesses the system's discrimination capabilities by simulating real attack behavior. In practice, a high-precision 3D scan of the target face is first performed to obtain complete 3D data including facial contours, facial features, and skin texture. The model is then optimized using professional modeling software. Subsequently, through multiple complex processes such as 3D printing of the initial prototype, silicone casting, skin texture replication, and color calibration, a highly realistic 3D mask is finally produced. While this method can recreate realistic attack environments to some extent, it suffers from several insurmountable limitations: First, the production cost is exorbitant. The material cost, equipment wear and tear, and labor cost of a single high-fidelity 3D mask can reach tens of thousands of yuan, placing a significant economic burden on enterprises or research institutions that require batch testing and multi-scenario verification. Second, the preparation cycle is lengthy. From 3D scanning to final product delivery, the entire process takes 1-2 weeks or even longer, severely lagging behind the R&D iteration pace of structured light face recognition systems. Third, its versatility is extremely poor. The facial features, size, and texture of the physical mask are fixed values, making it impossible to flexibly adjust parameters to simulate attack scenarios under different facial contours, attack angles, and lighting intensities. Furthermore, a single mask can only be adapted to a single test target. Fourth, the scenario coverage is limited. It cannot simulate complex scenarios such as facial micro-expression changes and dynamic attacks, nor can it achieve continuous parameterized attack intensity adjustment. Consequently, the test results can only reflect the system performance in specific scenarios and cannot comprehensively assess the system's robustness in diverse and complex attack environments. This significantly reduces the reference value of the test results and makes it difficult to meet the industry's demand for large-scale, high-precision, and comprehensive testing of structured light face recognition systems.

[0004] Chinese invention patent CN113963425A discloses a testing method for a face liveness detection system. This method involves projecting a laser pattern of a preset frequency onto a face under test using a 3D structured light camera, acquiring first face point cloud data of the face illuminated by the laser pattern, and then capturing a face image using a 2D camera. The method also acquires texture information of the corresponding pixels in the face image from the first face point cloud data, combines the first face point cloud data with the texture information to obtain second face point cloud data, and converts the second face point cloud data into a projected image according to preset calibration parameters. When the projection device of the 3D structured light camera is turned off, the face liveness detection system is tested using the projected image. However, this testing method cannot reproduce the three-dimensional optical characteristics such as light scattering and refraction of a high-fidelity 3D mask, and suffers from limited scene coverage, high testing costs, and low efficiency, making it difficult to comprehensively evaluate the robustness of the structured light face recognition system against 3D attacks and the stability of the 3D reconstruction algorithm. Summary of the Invention

[0005] The primary technical problem to be solved by this invention is to provide a testing device for a structured light face recognition system.

[0006] Another technical problem to be solved by the present invention is to provide a testing method for a structured light face recognition system.

[0007] To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution: According to a first aspect of the present invention, a testing apparatus for a structured light face recognition system is provided, comprising a face recognition module under test, an illumination projection module, and a dynamic simulated light field generation module; The illumination projection module emits an optical beam to the dynamic simulated light field generation module; the dynamic simulated light field generation module receives the optical beam and generates a modulated light field, which is then projected onto the photosensitive surface of the face recognition module under test through the optical transmission path of the illumination projection module; the face recognition module under test processes the received modulated light field to generate a face recognition confidence score and a reconstructed 3D depth map, and feeds the face recognition confidence score and the reconstructed 3D depth map back to the dynamic simulated light field generation module.

[0008] Preferably, the face recognition module to be tested includes: a structured light projector, an infrared imaging camera, and a target face 3D reconstruction module; The structured light projector is used to emit invisible light of a specific wavelength and project a preset coded optical pattern onto the test target area; The infrared imaging camera is installed at a preset angle to the structured light projector to capture the optical pattern modulated by the test light field. The target face 3D reconstruction module analyzes the pattern captured by the infrared imaging camera, calculates the depth information and constructs the corresponding 3D model. Then, it compares the model with the target face templates stored in the database, outputs the face recognition confidence score, and completes the identity authentication simulation process.

[0009] Preferably, the projection angle, pattern type, and light intensity parameters of the invisible light are configured according to a preset simulated attack scenario to generate a test light field corresponding to the optical effects of different three-dimensional masks.

[0010] Preferably, the lighting projection module includes: a light source, a first polarizer, a homogenizing lens, a projection lens, a second polarizer, and a polarizing beam splitter. The light source is used to emit infrared beams; The first polarizer is used to modulate the polarization state of the light beam emitted by the light source, converting natural light into S-polarized light and ensuring the polarization consistency of the light beam. The homogenizing lens is used to homogenize the S-polarized light that has passed through the first polarizer, eliminate the uneven light intensity of the beam, form a uniform parallel beam, and avoid affecting the test accuracy due to differences in light intensity. The polarizing beam splitter is used to separate and control the transmission of S-polarized light and P-polarized light. When an S-polarized parallel beam is incident, the polarizing beam splitter reflects it to the direction of the second polarizer. When P-polarized light, which is subsequently modulated, is incident, the polarizing beam splitter allows it to pass through, thus achieving orderly switching of the optical path. The second polarizer is used to further polarize the reflected S-polarized light, and at the same time receive the reflected beam and convert it into P-polarized light to ensure that the beam polarization state meets the subsequent transmission requirements. The projection lens is used to project the optically processed and modulated light beam onto the photosensitive surface of the structured light face recognition system under test.

[0011] Preferably, the light source is an infrared laser in the same wavelength band as the structured light face recognition system.

[0012] Preferably, the dynamic simulated light field generation module includes a control and calculation module and a spatial light modulation device; wherein... The control and calculation module is used to generate holograms of corresponding test scenarios, simulate three-dimensional light field information under different facial features and different attack scenarios, and complete the simulation of light transmission path, three-dimensional data extraction, reverse light path calculation and encoding pattern optimization. It also receives data from the face recognition module under test, performs error analysis and algorithm parameter adjustment, and realizes dynamic optimization of the test process. The spatial light modulation device adjusts the phase and amplitude parameters of the incident light wave in real time according to the instructions of the control and calculation module, modulates the light beam, and generates an coded response pattern. After being transmitted through the optical system of the illumination projection module, it forms a three-dimensional light field in space that is consistent with the optical effect of a real 3D mask.

[0013] Preferably, the spatial light modulation device is any one of a reflective liquid crystal spatial light modulator, a transmissive spatial light modulator, or a digital micromirror array.

[0014] According to a second aspect of the present invention, a testing method for a structured light face recognition system is provided, comprising the following steps: S1: Construct a 3D face mesh model in simulation software, add optical parameters, and obtain the structured light pattern of the structured light face recognition system to be tested; S2: From the perspective of the infrared imaging camera in the structured light face recognition system under test, render the infrared image captured by the infrared imaging camera when the virtual face is illuminated by structured light. S3: Extract three-dimensional data, and recover depth information based on the projection pattern and the infrared image captured by the infrared imaging camera using triangulation. S4: Calculate a set of coded patterns such that the difference between the image formed on the sensor of the infrared imaging camera after the coded pattern is modulated by the spatial light modulation device and the infrared image captured by the infrared imaging camera in step S2 is minimized. S5: The encoded pattern is loaded onto the spatial light modulation device, and the modulated light wave is generated by the parallel light wave illumination after beam expansion and collimation, and transmitted to the sensor of the infrared imaging camera. S6: Obtain the 3D depth image reconstructed by the structured light face recognition system under test, perform error calculation with the 3D face mesh model constructed in step S1, adjust the preset algorithm parameters, and generate a coding pattern with higher fidelity. S7: Combine subjective visual comparison results and objective indicators to form the test evaluation results, and complete the three-dimensional mask simulation test of the structured light face recognition system under test.

[0015] Preferably, in step S3, the formula for triangulation is: Where z is the depth of a point; f is the focal length of the infrared imaging camera; B is the baseline distance between the projector and the infrared imaging camera; and d is the deformation of the corresponding point in the pattern calculated using a feature matching algorithm.

[0016] Compared with existing technologies, this invention has the following technical advantages: First, by using a spatial light modulation device as a programmable virtual face and combining reverse optical path calculation and ray tracing technology, it can accurately reproduce the complex optical effects of light scattering, refraction, and phase modulation of a real three-dimensional mask at the physical light field level, overcoming the fundamental limitation of existing technologies that can only generate two-dimensional projection images and cannot simulate three-dimensional optical characteristics. Second, through parameterized digital modeling and dynamic light field control, this invention can flexibly simulate diverse attack scenarios with different facial contours, attack angles, light intensities, and skin materials without the need to create physical masks, significantly reducing testing costs and cycles, and improving testing efficiency and scenario coverage. Third, by feeding back the reconstructed three-dimensional depth map of the system under test to the light field generation stage, a closed-loop optimization mechanism based on error analysis is constructed, which can iteratively improve the fidelity of the simulated light field and ensure the accuracy and repeatability of the test results. Finally, this invention provides a high-fidelity, high-efficiency, and highly controllable comprehensive solution for evaluating the anti-spoofing capabilities of structured light face recognition systems and verifying three-dimensional reconstruction algorithms, which is of great significance for improving the reliability and robustness of biometric technology in high-security scenarios. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of a test device for a structured light face recognition system in the first embodiment of the present invention; Figure 2 This is a flowchart of a testing method for a structured light face recognition system in the second embodiment of the present invention. Detailed Implementation

[0018] The technical content of the present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0019] First Embodiment like Figure 1 As shown, the first embodiment of the present invention provides a testing device for a structured light face recognition system, comprising: a face recognition module under test, an illumination projection module, and a dynamic simulated light field generation module. The illumination projection module provides a stable optical environment that meets testing requirements for the dynamic simulated light field generation module, and its output specific optical beam illuminates the dynamic simulated light field generation module. The dynamic simulated light field generation module generates a modulated light field with complex three-dimensional light field characteristics, which is precisely projected onto the photosensitive surface of the face recognition module under test through the optical transmission path of the illumination projection module. The face recognition module under test processes the received modulated light field to generate a face recognition confidence score and a reconstructed three-dimensional depth map, and feeds the face recognition confidence score and the reconstructed three-dimensional depth map back to the dynamic simulated light field generation module, providing a basis for subsequent encoding optimization and realizing closed-loop control of the testing process.

[0020] In one embodiment of the present invention, the face recognition module to be tested includes: a structured light projector, an infrared imaging camera, and a target face 3D reconstruction module.

[0021] The structured light projector, acting as the light signal emitting unit, is responsible for emitting invisible light of a specific wavelength (such as infrared light) and accurately projecting a preset coded optical pattern (such as a dot matrix, stripes, or grid) onto the test target area. Its projection angle, pattern type, light intensity, and other parameters are consistent with the actual application scenario, ensuring the authenticity of the test. The infrared imaging camera, acting as the signal receiving unit, is installed at a preset angle to the structured light projector. It can clearly capture the optical pattern modulated by the test light field, possessing high resolution and high sensitivity characteristics. It can accurately record the deformation information of the pattern, providing raw data support for subsequent depth calculation. The target face 3D reconstruction module, as the core processing unit, incorporates advanced depth calculation algorithms and face recognition algorithms. First, it analyzes the pattern captured by the infrared imaging camera through algorithms, calculates depth information, and constructs a corresponding 3D model. Then, it compares the model with the target face template stored in the database, outputs key evaluation indicators such as face recognition confidence score, and completes the identity authentication simulation process. Its algorithm performance directly determines the system's recognition accuracy and anti-counterfeiting capability, making it a key evaluation object of this test device.

[0022] In one embodiment of the present invention, the illumination projection module includes: a light source, a first polarizer, a homogenizing lens, a projection lens, a second polarizer, and a polarizing beam splitter (PBS).

[0023] The light source uses an infrared laser in the same wavelength band as the structured light face recognition system under test, capable of emitting a stable infrared beam. This provides an optical foundation that meets wavelength requirements for the entire testing process. Its output light intensity can be fine-tuned according to testing needs, ensuring the stability and adaptability of the optical signal. The first polarizer modulates the polarization state of the beam emitted by the light source, converting natural light into S-polarized light, ensuring the polarization consistency of the beam and laying the foundation for subsequent optical control. The homogenizing lens homogenizes the S-polarized light after passing through the first polarizer, eliminating uneven beam intensity and forming a uniform parallel beam, avoiding the impact of intensity differences on testing accuracy. The polarizing beam splitter, as the core device for optical path switching, separates and controls the transmission of S-polarized and P-polarized light. When the S-polarized parallel beam is incident, the PBS reflects it to the direction of the second polarizer, while the P-polarized light, after subsequent modulation, is allowed to pass through by the PBS, achieving orderly switching of the optical path. The second polarizer is used to further polarize the reflected S-polarized light, and simultaneously receives the beam reflected by the spatial light modulator and converts it into P-polarized light, ensuring that the beam polarization state meets the requirements for subsequent transmission. The projection lens is responsible for accurately projecting the beam, which has undergone a series of optical processing and modulation, onto the photosensitive surface of the structured light face recognition system under test. Its focal length and projection angle can be adjusted according to the test distance and range to ensure the accuracy of the light field projection.

[0024] In one embodiment of the present invention, the dynamic simulated light field generation module includes a control and calculation module and a spatial light modulation device.

[0025] The control and computation module integrates multiple functions such as simulation, data processing, and command control. It can generate holograms of corresponding test scenarios using professional 3D simulation software, accurately simulating three-dimensional light field information under different facial features and attack scenarios. Simultaneously, this module possesses powerful computing capabilities, capable of performing complex calculations such as light transmission path simulation, three-dimensional data extraction, reverse optical path calculation, and coded pattern optimization based on computational holography principles, triangulation, and other algorithms. It can also send precise control commands to spatial light modulation devices based on the calculation results, triggering the display of corresponding response patterns. Furthermore, it can receive data from the face recognition module under test, performing error analysis and algorithm parameter adjustments to achieve dynamic optimization of the testing process.

[0026] Spatial light modulators, acting as "programmable virtual faces," utilize dynamic optical control devices such as reflective liquid crystal spatial light modulators (SLMs), transmissive SLMs, or digital micromirror devices (DMDs). They can precisely modulate the beam by adjusting the phase, amplitude, and other physical parameters of the incident light wave in real time according to instructions from the control and calculation module. The coded response pattern generated by the display control and calculation module is transmitted through the optical system of the illumination projection module, forming a complex three-dimensional light field in space that highly replicates the optical effects of a real 3D mask. Replacing traditional static physical attack props, this device offers advantages such as high flexibility, repeatability, and adjustable parameters, meeting the needs of various testing scenarios.

[0027] Second Embodiment like Figure 2 As shown, the second embodiment of the present invention provides a testing method for a structured light face recognition system, including: S1: Construct a 3D face mesh model in simulation software, add optical parameters, and obtain the structured light pattern of the structured light face recognition system to be tested.

[0028] S2: Render the infrared image captured by the camera when the virtual face is illuminated by structured light, from the perspective of the infrared imaging camera of the system under test.

[0029] S3: Extract 3D data, based on the projection pattern and the pattern captured by the infrared camera, and use triangulation to recover the depth information.

[0030] S4: Calculate a set of coded patterns such that after the pattern is modulated by a spatial light modulator, the difference between the image formed on the infrared imaging camera sensor and the infrared image captured by the camera is minimized.

[0031] S5: The coded pattern is loaded onto the spatial light modulation device, and the modulated light wave is generated by the parallel light wave illumination after beam expansion and collimation, and transmitted to the infrared camera sensor of the system under test.

[0032] S6: Obtain the reconstructed 3D depth image of the system under test, calculate the error with the input model, adjust the algorithm parameters, and generate a coded pattern with higher fidelity.

[0033] S7: The subjective visual comparison results and objective indicators together constitute the test evaluation results, and complete the three-dimensional mask simulation test of the system under test.

[0034] In one embodiment of the present invention, the formula for triangulation in step S3 is: Where z is the depth of a point; f is the focal length of the camera; B is the baseline distance between the projector and the camera; and d is the deformation of the corresponding point in the pattern calculated using a feature matching algorithm, i.e., the positional deviation between the feature point in the projected pattern and its corresponding point in the camera image. Then, the actual two-dimensional coordinates in the x and y directions are obtained based on the camera settings, finally yielding the target's three-dimensional light field information.

[0035] The above method will be explained in detail below: In step S1, firstly, a high-precision 3D modeling software (such as Maya, 3ds Max, etc.) is selected to import a preset high-precision 3D face mesh model. This model must contain complete facial contours, facial features, and other details to ensure the realism of the simulation. Then, in the modeling software, the skin material parameters of the model are set according to the optical properties of real human skin, including refractive index, reflectivity, and scattering coefficient, so that the simulated light field can exhibit an optical effect close to that of real skin when interacting with the light beam. Based on the preset simulated attack scenarios (such as different face contours, different attack angles, and different light intensities), the projection angle, pattern type, and light intensity parameters of the structured light projector are configured accordingly to ensure that the generated test light field can simulate the optical effect of the 3D mask in the corresponding scenario. Next, a virtual model of the structured light projector is built in the modeling software. The actual parameters of the projector in the structured light face recognition system under test (such as projection angle, pattern type, light intensity, etc.) are set, and the known coded pattern Ip(xp,yp) of the system under test is entered into the virtual projector. The projector is then controlled to project this pattern onto the virtual face model at a fixed position and angle. Finally, a virtual model of the infrared imaging camera was constructed in the modeling software. Referring to the actual installation position relationship between the camera and the projector in the system under test, the virtual camera and the virtual projector were set to the same preset angle to ensure that the virtual camera could completely capture the pattern area projected on the virtual face model, providing a complete scene basis for subsequent ray tracing and data extraction.

[0036] In step S2, a physically based ray tracing rendering engine (such as V-Ray, Arnold, etc.) is used. This engine can accurately simulate the physical behaviors of light propagation, reflection, and refraction. First, the rendering engine is started to simulate the operation of the structured light projector in the system under test. A pre-defined coded pattern beam is emitted from the virtual structured light projector. After the beam propagates to the surface of the virtual face model, it is reflected according to the previously set optical parameters of the skin material. The reflected beam carries the 3D contour and detail information of the face model and continues to propagate. Subsequently, the rendering engine tracks the transmission path of the reflected beam to ensure that the beam accurately enters the lens of the virtual infrared camera. During this process, the engine calculates parameters such as the propagation distance and direction changes of the light in real time, eliminating the influence of ambient light and other interference factors to ensure the accuracy of the light transmission simulation. Finally, the rendering engine outputs a test infrared image Ii(xc,yc) under ideal conditions, observed from the infrared camera perspective of the structured light face recognition system under test. This image completely preserves the key information of the coded pattern after the face deformation, which is an important foundation for subsequent 3D data extraction and reverse ray path calculation.

[0037] In step S3, firstly, the known coded pattern Ip(xp,yp) projected by the virtual projector in step S1 and the ideal test infrared image Ii(xc,yc) output by ray tracing in step S2 are obtained. Corresponding feature points in the two images are precisely matched using a feature matching algorithm, and the deformation amount d of the corresponding point in the pattern is calculated. The calculation of the deformation amount needs to eliminate image noise interference to ensure data accuracy. Subsequently, the depth information of the target face is recovered based on the principle of triangulation. The core of triangulation is to use the baseline distance B between the projector and the camera, the camera focal length f, and the deformation amount d to calculate the depth z of a certain feature point using the formula z = (f × B) / d (where f is the camera focal length, determined by the camera hardware parameters; B is the baseline distance between the projector and the camera, i.e., the straight-line distance between them on the mounting plane, which is a preset fixed value). After calculating the depth information of all feature points, the intrinsic parameters of the virtual camera (such as pixel size and imaging range) are combined to convert them into two-dimensional coordinates in the x and y directions in the actual space, thereby completely constructing the three-dimensional light field information. This information contains all the spatial geometric features of the target face, providing an accurate data source for subsequent reverse optical path calculation.

[0038] In step S4, using the three-dimensional light field information extracted in step S3 as input, the light wavelength information is first sampled and quantized using the point source method, converting continuous light wave information into discrete digital signals for computer processing. Subsequently, based on the propagation formula of light waves in free space, and combined with the optical parameters of the illumination projection module (such as lens focal length, polarizer characteristics, and PBS splitting ratio), the discrete light wave information is further calculated to simulate the propagation process of light waves in an actual optical system, including polarization state changes, light intensity attenuation, and phase shift. Through iterative calculations, the pixel information of the encoded pattern is continuously adjusted, ultimately obtaining the encoding of the complex amplitude light wave field of the object light wave on the hologram, i.e., the coded pattern IS(xs,ys) recognizable by the spatial light modulation device. After transmission through the optical system, this encoded pattern can form a modulated light field highly consistent with the ideal image features on the photosensitive surface of the structured light face recognition system under test.

[0039] In step S5, firstly, the control and calculation module loads the coded pattern IS(xs,ys) generated in step S4 into the spatial light modulator. The loading process must ensure the integrity and synchronization of the pattern transmission to avoid distortion or delay. Subsequently, the infrared laser source in the illumination projection module is activated. The emitted beam is sequentially modulated into S-polarized light by the first polarizer, then processed by a homogenizing lens to form a uniform parallel beam. This S-polarized parallel beam is reflected after being incident on a polarizing beam splitter and continues to be transmitted to the second polarizer for further polarization modulation, finally illuminating the spatial light modulator loaded with the coded pattern. The spatial light modulator modulates the phase and amplitude of the incident parallel beam according to the coded pattern. The modulated beam is reflected and passes through the second polarizer again, converting into P-polarized light. This P-polarized light undergoes transmission when incident on the PBS, and finally, after being focused and oriented by the projection lens, it is precisely projected onto the photosensitive surface of the structured light face recognition system under test, completing the simulated projection of the 3D mask light field information. The entire projection process must ensure the stability and accuracy of the light field, maintaining a high degree of consistency with the ideal scene.

[0040] In step S6, after dynamic projection is completed, the raw data output by the structured light face recognition system under test is first collected, including the actual image captured by the infrared camera, the reconstructed 3D depth map, and other key information. The data collection process must ensure no data loss and no interference. Subsequently, the reconstructed 3D depth map is compared point-by-point with the simulated 3D face model input in step S1. The root mean square error (RMSE) is used to calculate the reconstruction error index. RMSE accurately reflects the degree of deviation between the actual reconstruction result and the ideal model, and the error calculation range covers all facial feature points, ensuring comprehensive evaluation. The control and calculation module feeds back the error to the reverse optical path calculation stage in step S4 based on the magnitude and distribution of the reconstruction error, and adjusts the relevant algorithm parameters in the holographic principle, such as sampling accuracy, number of iterations, and correction coefficients for light wave propagation parameters. Based on the adjusted algorithm parameters, the reverse optical path calculation is performed again to generate a new coded pattern. Through multiple iterations, the reconstruction error is continuously reduced, and the fidelity of the coded pattern is continuously improved, ensuring the accuracy and reliability of subsequent tests.

[0041] In the final step S7, firstly, the face recognition confidence score output by the structured light face recognition system under test is collected. This score is the identity matching evaluation value given by the structured light face recognition system after recognizing the received light field, directly reflecting the anti-spoofing capability of the structured light face recognition system. Then, a reasonable confidence score threshold is set (this threshold is determined based on the application scenario and security level of the system under test), and the actual detection score is analyzed to see if it exceeds this threshold: if the score exceeds the threshold, it indicates that the structured light face recognition system failed to identify the simulated attack, posing a security risk; if it does not exceed the threshold, it indicates that the structured light face recognition system possesses the corresponding anti-spoofing capability. Simultaneously, a comprehensive testing and evaluation system is constructed by combining the subjective visual comparison results obtained in step S6 (i.e., the visual difference between the actual reconstructed depth map and the ideal model) and objective indicators such as root mean square error. During the evaluation process, the results under different test scenarios and attack intensities need to be classified and statistically analyzed. The performance of the structured light face recognition system under test in various situations should be comprehensively analyzed to clarify its advantages and disadvantages in anti-spoofing capabilities. Finally, a complete test evaluation report should be formed, and the three-dimensional mask simulation test of the structured light face recognition system under test should be completed to provide a scientific basis for the optimization and improvement of the structured light face recognition system.

[0042] It should be noted that the above embodiments are merely illustrative examples, and the technical solutions of each embodiment can be combined, all of which are within the protection scope of this invention.

[0043] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0044] The testing apparatus and method for the structured light face recognition system provided by this invention have been described in detail above. Any obvious modifications made by those skilled in the art without departing from the essence of this invention will constitute an infringement of the patent rights of this invention and will incur corresponding legal liability.

Claims

1. A testing device for a structured light face recognition system, characterized in that... It includes a face recognition module under test, an illumination projection module, and a dynamic simulated light field generation module; The illumination projection module emits an optical beam to the dynamic simulated light field generation module; the dynamic simulated light field generation module receives the optical beam and generates a modulated light field, which is then projected onto the photosensitive surface of the face recognition module under test through the optical transmission path of the illumination projection module; the face recognition module under test processes the received modulated light field to generate a face recognition confidence score and a reconstructed 3D depth map, and feeds the face recognition confidence score and the reconstructed 3D depth map back to the dynamic simulated light field generation module.

2. The testing apparatus as described in claim 1, characterized in that... The face recognition module under test includes: a structured light projector, an infrared imaging camera, and a target face 3D reconstruction module; The structured light projector is used to emit invisible light of a specific wavelength and project a preset coded optical pattern onto the test target area; The infrared imaging camera is installed at a preset angle to the structured light projector to capture the optical pattern modulated by the test light field. The target face 3D reconstruction module analyzes the pattern captured by the infrared imaging camera, calculates the depth information and constructs the corresponding 3D model. Then, it compares the model with the target face templates stored in the database, outputs the face recognition confidence score, and completes the identity authentication simulation process.

3. The testing apparatus as described in claim 2, characterized in that... The projection angle, pattern type, and light intensity parameters of the invisible light are configured according to a preset simulated attack scenario to generate a test light field corresponding to different three-dimensional mask optical effects.

4. The testing apparatus as described in claim 1, characterized in that... The lighting projection module includes: a light source, a first polarizer, a homogenizing lens, a projection lens, a second polarizer, and a polarizing beam splitter. The light source is used to emit infrared beams; The first polarizer is used to modulate the polarization state of the light beam emitted by the light source, converting natural light into S-polarized light and ensuring the polarization consistency of the light beam. The homogenizing lens is used to homogenize the S-polarized light that has passed through the first polarizer, eliminate the uneven light intensity of the beam, form a uniform parallel beam, and avoid affecting the test accuracy due to differences in light intensity. The polarizing beam splitter is used to separate and control the transmission of S-polarized light and P-polarized light. When an S-polarized parallel beam is incident, the polarizing beam splitter reflects it to the direction of the second polarizer. When P-polarized light, which is subsequently modulated, is incident, the polarizing beam splitter allows it to pass through, thus achieving orderly switching of the optical path. The second polarizer is used to further polarize the reflected S-polarized light, and at the same time receive the reflected beam and convert it into P-polarized light to ensure that the beam polarization state meets the subsequent transmission requirements. The projection lens is used to project the optically processed and modulated light beam onto the photosensitive surface of the structured light face recognition system under test.

5. The testing apparatus as described in claim 4, characterized in that... The light source is an infrared laser in the same wavelength band as the structured light face recognition system under test.

6. The testing apparatus as described in claim 1, characterized in that... The dynamic simulated light field generation module includes a control and calculation module and a spatial light modulation device; wherein... The control and calculation module is used to generate holograms of corresponding test scenarios, simulate three-dimensional light field information under different facial features and different attack scenarios, and complete the simulation of light transmission path, three-dimensional data extraction, reverse light path calculation and encoding pattern optimization. It also receives data from the face recognition module under test, performs error analysis and algorithm parameter adjustment, and realizes dynamic optimization of the test process. The spatial light modulation device adjusts the phase and amplitude parameters of the incident light wave in real time according to the instructions of the control and calculation module, modulates the light beam, and generates an coded response pattern. After being transmitted through the optical system of the illumination projection module, it forms a three-dimensional light field in space that is consistent with the optical effect of a real 3D mask.

7. The testing apparatus as described in claim 6, characterized in that... The spatial light modulation device is any one of a reflective liquid crystal spatial light modulator, a transmissive spatial light modulator, or a digital micromirror array.

8. A testing method for a structured light face recognition system, implemented based on the testing apparatus described in any one of claims 1 to 7, characterized in that... Includes the following steps: S1: Construct a 3D face mesh model in simulation software, add optical parameters, and obtain the structured light pattern of the structured light face recognition system to be tested; S2: From the perspective of the infrared imaging camera in the structured light face recognition system under test, render the infrared image captured by the infrared imaging camera when the virtual face is illuminated by structured light. S3: Extract three-dimensional data, and recover depth information based on the projection pattern and the infrared image captured by the infrared imaging camera using triangulation. S4: Calculate a set of coded patterns such that the difference between the image formed on the sensor of the infrared imaging camera after the coded pattern is modulated by the spatial light modulation device and the infrared image captured by the infrared imaging camera in step S2 is minimized. S5: The encoded pattern is loaded onto the spatial light modulation device, and the modulated light wave is generated by the parallel light wave illumination after beam expansion and collimation, and transmitted to the sensor of the infrared imaging camera. S6: Obtain the 3D depth image reconstructed by the structured light face recognition system under test, perform error calculation with the 3D face mesh model constructed in step S1, adjust the preset algorithm parameters, and generate a coding pattern with higher fidelity. S7: Combine subjective visual comparison results and objective indicators to form the test evaluation results, and complete the three-dimensional mask simulation test of the structured light face recognition system under test.

9. The test method as described in claim 8, characterized in that... In step S3, the formula for triangulation is: Where z is the depth of a point; f is the focal length of the infrared imaging camera; B is the baseline distance between the projector and the infrared imaging camera; and d is the deformation of the corresponding point in the pattern calculated using a feature matching algorithm.

Images