Method and system for distinguishing between real and fake people based on a spectral camera

Abstract

This application relates to the field of spectral imaging technology, and provides a method and system for identifying real and fake people based on a spectral camera. The method utilizes a spectral camera to collect spectral data of the object, including static spectral data and continuous dynamic spectral data. Static spectral features are extracted from the static spectral data, and dynamic temporal features are extracted from the continuous dynamic spectral data to train a classification and recognition model. This trained model is then used to identify real and fake people, improving the accuracy of identification and avoiding false positives. Furthermore, it enhances the anti-attack capability of the classification and recognition model, making it suitable for scenarios such as security access control, financial payments, intelligent robots, border inspection, and unmanned retail.

Description

Technical Field

[0001] This application relates to the field of spectral imaging technology, and in particular to a method and system for identifying real and fake people based on a spectral camera. Background Technology

[0002] Currently, RGB cameras, near-infrared cameras, and depth cameras are widely used for liveness detection and identity recognition in fields such as security access control, financial payments, intelligent robots, unmanned retail, and identity authentication. Existing technologies mainly determine whether a target is a real person through two-dimensional textures, facial structure, motion coordination, and differences in infrared brightness.

[0003] Currently, common liveness detection solutions mainly suffer from the following problems: insufficient resistance to highly realistic prosthetics; and for materials such as silicone dummies, 3D printed head molds, and latex masks, ordinary RGB or near-infrared solutions are prone to misjudgment. Summary of the Invention

[0004] In view of this, the purpose of this application is to propose a method for identifying real and fake people based on a spectral camera, which is beneficial to improving the recognition accuracy.

[0005] To achieve the above objectives, one embodiment of this application discloses a method for identifying real and fake people based on a spectral camera, which includes:

[0006] A spectral database is constructed by acquiring spectral data of a known object within a target wavelength range using a spectral camera; wherein the known object includes real people and dummy people, the spectral data includes spectral data of real people and spectral data of dummy people, and the spectral data includes static spectral data and continuous dynamic spectral data; Static spectral features are extracted from the static spectral data, and dynamic temporal features are extracted from the continuous dynamic spectral data; wherein, the static spectral features are the characteristic absorption and scattering features of the known object within the band, and the dynamic temporal features are the spectral fluctuation features corresponding to the periodic physiological activities of the known object that change over time; The static spectral features and the dynamic temporal features are used as model inputs, and the identification result of whether it is a real person or a fake person is used as model output to train the classification and identification model. The trained classification and recognition model is used to identify whether an unknown object is a real person or a fake person.

[0007] In one embodiment, the static spectral features include at least one of the following: oxyhemoglobin absorption features near 540 nm wavelength, deoxyhemoglobin absorption features near 560 nm wavelength, blood oxygen difference features near 760 nm wavelength, oxyhemoglobin absorption features near 850 nm wavelength, and water absorption features near 940 nm wavelength. Alternatively, the static spectral characteristics can be obtained by calculating the reflectance difference, ratio, or spectral line slope between different characteristic bands.

[0008] In one embodiment, the dynamic temporal physiological characteristics include at least one of the following: spectral fluctuation characteristics corresponding to weak pulse changes, spectral fluctuation characteristics corresponding to dynamic changes in blood oxygenation, and spectral fluctuation characteristics corresponding to periodic microcirculation changes.

[0009] In one embodiment, before acquiring spectral data of a known object within the target wavelength range using a spectral camera and constructing a spectral database, the method further includes: A monochromator outputs monochromatic light of multiple known wavelengths within the target band to perform wavelength calibration on the spectroscopic camera, establishing the correspondence between detector pixel positions and actual wavelengths; and a standard reference response is used to calibrate the spectroscopic camera's spectral response, resulting in a calibrated spectroscopic camera.

[0010] In one embodiment, after acquiring spectral data of a known object within a target wavelength range using a spectral camera and constructing a spectral database, the method further includes extracting static spectral features from the static spectral data and dynamic temporal features from the continuous dynamic spectral data; prior to this, the method also includes: The spectral data is preprocessed; wherein the preprocessing includes noise removal and normalization.

[0011] In one embodiment, the acquisition area of ​​the spectral data includes at least one of the following regions of the known object: the face region, the forehead region, the bridge of the nose region, the cheek region, the periorbital region, and the hand region.

[0012] In one embodiment, the static spectral data includes static spectral data under different ambient light and different background light, and the continuous dynamic spectral data includes continuous dynamic spectral data under different ambient light and different background light.

[0013] In one embodiment, the dummy includes at least one of the following: silicone dummy, resin model, latex mask, wax figure, screen copy object, and photograph object.

[0014] In one embodiment, the target wavelength range is 400-1000nm.

[0015] Based on the same inventive concept, another embodiment of this application discloses a real / dummy person identification system based on a spectral camera, which includes: The data acquisition module is used to acquire spectral data of a known object within a target wavelength range using a spectral camera and to construct a spectral database; wherein the known object includes real people and dummy people, the spectral data includes spectral data of real people and spectral data of dummy people, and the spectral data includes static spectral data and continuous dynamic spectral data; The feature extraction module is used to extract static spectral features from the static spectral data and dynamic temporal features from the continuous dynamic spectral data; wherein, the static spectral features are the characteristic absorption and scattering characteristics of the known object in the band, and the dynamic temporal features are the spectral fluctuation characteristics corresponding to the periodic physiological activities of the known object over time; The model training module is used to train the classification and recognition model by taking the static spectral features and the dynamic temporal features as model inputs and the recognition result of whether it is a real person or a fake person as model output. The recognition module is used to identify whether an unknown object is a real person or a fake person using a trained classification and recognition model.

[0016] This application provides a method for identifying real and fake people based on a spectral camera. The method utilizes a spectral camera to collect spectral data of the object, including static and continuous dynamic spectral data. Static spectral features are extracted from the static spectral data, and dynamic temporal features are extracted from the continuous dynamic spectral data to train a classification and recognition model. This trained model is then used to distinguish between real and fake people, improving the accuracy of identification and avoiding false positives. Furthermore, it enhances the anti-attack capability of the classification and recognition model, making it suitable for scenarios such as security access control, financial payments, intelligent robots, border inspection, and unmanned retail. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for identifying real and fake people based on a spectral camera, provided as an embodiment of this application; Figure 2 A flowchart of a method for identifying real and fake people based on a spectral camera, provided as another embodiment of this application; Figure 3 A flowchart of a method for identifying real and fake people based on a spectral camera, provided as another embodiment of this application; Figure 4 This is a schematic diagram of a real / dummy human identification system based on a spectral camera, provided as another embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Through long-term research, the inventors discovered that with the development of biomimetic materials and forgery techniques, traditional liveness detection technologies have gradually revealed the following problems: 1. Attacks using two-dimensional photographs and high-definition screen captures can easily bypass ordinary RGB recognition systems; 2. Highly realistic materials such as silicone dummies, resin head molds, latex masks, and wax figures already possess high similarity to human facial textures and structures, making them difficult to distinguish effectively using traditional depth detection or ordinary infrared detection; 3. Most existing liveness detection methods are based on texture differences at the image level, rather than identifying from the spectral characteristics of human tissue itself, thus limiting their resistance to attacks on highly realistic prosthetics; 4. Most existing spectral detection technologies only utilize a single near-infrared band or simple multi-band analysis, lacking comprehensive analysis of characteristics such as hemoglobin, water absorption, and subcutaneous scattering in human tissue. This results in low recognition accuracy and a high likelihood of misjudgment.

[0021] Based on this, this application provides a real / dummy person identification scheme based on a spectral camera to solve the above-mentioned technical problems.

[0022] like Figure 1 As shown, one embodiment of this application provides a method for identifying real and fake people based on a spectral camera, which includes the following steps: Step S10: Use a spectral camera to collect spectral data of known objects within the target wavelength range and construct a spectral database; wherein, known objects include real people and dummy people, spectral data include spectral data of real people and spectral data of dummy people, and spectral data include static spectral data and continuous dynamic spectral data.

[0023] Step S20: Extract static spectral features from static spectral data and extract dynamic temporal features from continuous dynamic spectral data; wherein, static spectral features are the characteristic absorption and scattering features of a known object within a band, and dynamic temporal features are the spectral fluctuation features corresponding to the periodic physiological activities of a known object that change over time.

[0024] Step S30: Train the classification and recognition model by using static spectral features and dynamic temporal features as model inputs and the recognition result of whether it is a real person or a fake person as model output.

[0025] Step S40: Use the trained classification and recognition model to identify whether the unknown object is a real person or a fake person.

[0026] The real-dummy human identification method based on a spectral camera provided in this application uses a spectral camera to collect spectral data of the object. The spectral data includes static spectral data and continuous dynamic spectral data. Static spectral features are extracted from the static spectral data and dynamic temporal features are extracted from the continuous dynamic spectral data to train a classification and recognition model. The trained classification and recognition model is used to identify real people and fake people, thereby improving the accuracy of real-person and fake-person identification and avoiding misjudgment.

[0027] The real and fake human identification method based on spectral camera provided in this application can be used to identify real and fake people on the road in the field of autonomous driving. For example, if there is a human-shaped advertising sign on the road, it can be identified as a fake person.

[0028] The real / dummy person identification method based on spectral cameras provided in this application improves the anti-attack capability of the classification and recognition model by extracting static spectral features from static spectral data and dynamic temporal features from continuous dynamic spectral data. It is applicable to scenarios such as security access control, financial payment, intelligent robots, border inspection, and unmanned retail.

[0029] Among them, the spectroscopic camera can simultaneously capture the two-dimensional spatial image of the target and the complete spectral information of each pixel. It decomposes the incident light into dozens to hundreds of continuous narrow bands through beam splitting elements such as filter arrays, gratings or prisms, forming a three-dimensional data cube containing spatial coordinates and wavelength. Unlike ordinary RGB cameras that can only record three wide bands of red, green and blue, it can resolve the unique spectral fingerprints of different substances and infer their composition, structure and dynamic state.

[0030] Optionally, the target wavelength range is 400-1000nm, covering the visible and near-infrared bands.

[0031] Optionally, the spectroscopic camera is an array snapshot spectroscopic camera, capable of acquiring the entire spectral band in a single exposure, making it particularly suitable for capturing weak dynamic physiological signals such as pulse and microcirculation. Its structure includes a CMOS imaging sensor, a two-dimensional graded filter array, and a compound eye lens. Optionally, the array snapshot filter array contains 64 channels, covering the 400nm-1000nm spectral band, with each channel having a spectral resolution of approximately 10nm.

[0032] Optionally, the dummy includes at least one of the following: silicone dummy, resin model, latex mask, wax figure, screen-reproduced object, and photographic object.

[0033] Optionally, real people include real people of different skin colors and ages.

[0034] In one embodiment, the static spectral features include at least one of the following: oxyhemoglobin absorption features near 540 nm wavelength, deoxyhemoglobin absorption features near 560 nm wavelength, blood oxygen difference features near 760 nm wavelength, oxyhemoglobin absorption features near 850 nm wavelength, and water absorption features near 940 nm wavelength.

[0035] Among them, the characteristic bands of 540nm, 560nm, 760nm, 850nm and 940nm were selected as static spectral features because they accurately cover the core optical features of human skin in the 400-1000nm visible-near infrared band that are most biologically specific and difficult to be simulated by biomimetic materials: 540nm and 560nm correspond to the characteristic absorption peaks of oxyhemoglobin and deoxyhemoglobin, respectively. The combination of the two can not only independently characterize the concentration difference of the two hemoglobins, but also stably reflect the total hemoglobin content through the difference in reflectance, thus establishing the essential distinction between real people and bloodless prostheses from the perspective of blood components. The 760nm and 850nm wavelengths form a standard complementary pair for blood oxygen detection. The inverse difference in the extinction coefficients of oxygenated and deoxygenated hemoglobin at these two wavelengths is used to calculate blood oxygen saturation, further verifying the unique blood oxygenation state of living organisms. The 940nm wavelength corresponds to the characteristic absorption peak of skin moisture, which can effectively distinguish between real human biological tissues rich in water and biomimetic materials with low water content such as silicone and resin. The five wavelengths form a complementary verification system from three dimensions: blood composition, oxygenation state, and tissue moisture. At the same time, they all avoid strong interference wavelengths and are within the engineering range that is easily achieved by spectral cameras, which can provide stable, reliable, and difficult-to-forge material-level characteristic evidence for the identification of real and fake people.

[0036] In one embodiment, static spectral characteristics are obtained by calculating the reflectance difference, ratio, or spectral line slope between different characteristic bands.

[0037] The reflectance difference, calculated by subtracting the reflectance of two wavelength bands, eliminates interference from factors such as light source intensity fluctuations and system response differences. For example, the difference between 540nm and 560nm stably reflects total hemoglobin content, while the difference between 940nm and 850nm accurately characterizes skin moisture content. The reflectance ratio, calculated by dividing the reflectance of two wavelength bands, further offsets the effects of non-specific factors such as uneven illumination and skin pigmentation. For instance, the ratio of 760nm to 850nm directly reflects blood oxygen saturation, a core quantitative indicator of the oxygenation status of living blood. The spectral slope of the 500nm-600nm band, by fitting the rate of change of reflectance with wavelength within this band, comprehensively reflects the overall absorption characteristics of hemoglobin, supplementing the limitations of single-band features. Combining these three methods allows for the extraction of material-level optical characteristics of human tissue from different dimensions, further mitigating environmental interference and providing a stable and reliable quantitative basis for identifying real and fake human bodies.

[0038] In one embodiment, the dynamic temporal physiological characteristics include at least one of the following: spectral fluctuation characteristics corresponding to weak pulse changes, spectral fluctuation characteristics corresponding to dynamic changes in blood oxygenation, and spectral fluctuation characteristics corresponding to periodic microcirculation changes.

[0039] Among them, the spectral fluctuation characteristics corresponding to the subtle pulse changes, the spectral fluctuation characteristics corresponding to the dynamic changes in blood oxygen, and the spectral fluctuation characteristics corresponding to the periodic microcirculation changes all originate from the unique blood circulation and autonomic nervous system regulation physiological activities of living organisms. These are essential living identifiers that cannot be simulated by any inanimate prosthesis such as silicone, resin, or wax figures.

[0040] Specifically, the heartbeat drives the blood flow in subcutaneous capillaries to increase or decrease periodically at 0.8-1.5 Hz, which in turn causes synchronous and weak fluctuations in the reflectance of hemoglobin-sensitive bands such as 540 nm and 560 nm, forming the spectral signal corresponding to the pulse. The coupling effect of respiration and blood circulation brings about small periodic changes in blood oxygen saturation, which manifests as regular fluctuations in the ratio of 760 nm to 850 nm reflectance. Meanwhile, the automatic contraction and relaxation of vascular smooth muscle and the low-frequency regulation of autonomic nerves produce microcirculatory blood flow fluctuations at 0.01-0.5 Hz, corresponding to slow periodic changes in the reflectance of the entire spectrum. These three factors cover multi-scale physiological rhythms from high-frequency heartbeats to low-frequency neural regulation, which can verify the bioactivity of the target from different dimensions. This effectively compensates for the deficiency that static spectral features may be partially simulated by highly realistic materials, and significantly improves the anti-attack capability and reliability of real and fake human identification.

[0041] Reference Figure 2 As shown, in one embodiment, before acquiring spectral data of a known object within the target wavelength range using a spectral camera and constructing a spectral database, the following steps are also included: S01. A monochromator outputs monochromatic light of multiple known wavelengths within the target band to calibrate the spectrophotometer, establishing the correspondence between detector pixel positions and actual wavelengths. A standard reference response is then used to calibrate the spectrophotometer's spectral response, resulting in a calibrated spectrophotometer. This ensures the accuracy of subsequently acquired data.

[0042] Specifically, a monochromator outputs monochromatic light within the wavelength range of 400nm-1000nm, sampling in 5nm increments at wavelengths of 400nm, 405nm, 410nm, 415nm, and 420nm. This allows for the acquisition of more data, thereby improving calibration accuracy. A spectroscopic camera sequentially acquires the response images corresponding to each monochromatic light. Gaussian fitting is used to determine the actual center wavelength of each filter channel, establishing the correspondence between detector pixel positions and actual wavelengths, thus completing wavelength calibration.

[0043] Then, dark current correction, system response correction, and flat field correction are performed sequentially. Dark current correction is performed by acquiring dark field images under completely dark conditions and subtracting the dark current value from subsequently acquired images; system response correction uses a standard reflective white plate to acquire a reference spectrum and normalizes the camera's spectral response curve; flat field correction is used to eliminate the influence of uneven response among sensor pixels.

[0044] Reference Figure 3 As shown, in one embodiment, after step S10, which involves acquiring spectral data of a known object within the target wavelength range using a spectral camera and constructing a spectral database, step S10 further includes extracting static spectral features from static spectral data and extracting dynamic temporal features from continuous dynamic spectral data; prior to this, the following steps are also included: Step S02: Preprocess the spectral data; the preprocessing includes noise removal and standardization to improve the comparability between different samples.

[0045] Optionally, preprocessing may include whiteboard correction, dark current removal, Savitzky-Golay smoothing, multi-derivative processing, normalization, SNV normalization, and MSC multivariate scattering correction.

[0046] Among them, whiteboard calibration uses a standard reflective whiteboard as a reference benchmark to convert the original spectral intensity into relative reflectivity, eliminate the overall intensity deviation caused by light source fluctuations, system response differences and ambient light changes, unify the spectral benchmark under different acquisition conditions, and ensure that the spectral data only reflects the optical characteristics of the target itself.

[0047] Dark current removal eliminates interference unrelated to incident light, such as CMOS or CCD thermal noise and leakage current, by collecting the sensor's inherent noise signal under no-light conditions and subtracting it pixel by pixel from the actual collected spectrum. This effectively improves the signal-to-noise ratio of the spectral signal in low-light environments and prevents noise from masking weak physiological spectral fluctuations.

[0048] Savitzky-Golay smoothing uses a sliding window polynomial fitting method to smooth the spectral curve. While effectively filtering out high-frequency random noise and photon noise, it fully preserves the position, shape, and feature details of the absorption peaks, avoiding peak distortion caused by simple smoothing and ensuring the accuracy of subsequent feature extraction.

[0049] Multi-derivative processing eliminates the overall spectral shift caused by baseline drift, background scattering, and pigment interference by performing first or second derivatives on the spectral curve, enhances the resolution of characteristic absorption peaks, highlights subtle spectral differences, effectively separates overlapping absorption features, and improves the ability of the spectrum to identify the composition of substances.

[0050] Normalization maps spectral reflectance to a fixed numerical range, eliminating the overall brightness differences caused by different light intensities and acquisition distances, bringing all spectral data to a uniform level, facilitating feature comparison between different samples, and improving the stability and convergence of subsequent model training.

[0051] SNV standardization eliminates spectral shifts and scaling caused by sample surface roughness, scattering differences, and optical path variations by normalizing the mean and variance of individual spectra, thereby enhancing the comparability of spectral shapes and effectively suppressing interference from skin surface scattering.

[0052] MSC multivariate scattering correction uses the average spectrum of the dataset as a reference and performs linear regression correction on the spectrum of each sample. At the same time, it eliminates the multivariate scattering effects such as baseline shift and spectral scaling. Compared with SNV correction, it more comprehensively suppresses the interference caused by subcutaneous scattering and skin thickness differences, so that the spectrum more realistically reflects the intrinsic composition characteristics of the tissue.

[0053] In one embodiment, the spectral data acquisition area includes at least one of the following regions of a known subject: the face, forehead, bridge of the nose, cheeks, periorbital area, and hands. These areas have relatively thin skin, a more uniform stratum corneum, and abundant subcutaneous capillaries. Their spectral responses to changes in blood oxygenation, microcirculation, and moisture are sensitive and stable, and they are less affected by hair, thick stratum corneum, or bone obstruction. This allows for the effective capture of characteristic absorption and dynamic fluctuation information of biological tissues such as hemoglobin and water, while also facilitating rapid, non-contact data acquisition, thus balancing data reliability and ease of collection.

[0054] In one embodiment, the static spectral data includes static spectral data under different ambient light and different background light conditions, while the continuous dynamic spectral data includes continuous dynamic spectral data under different ambient light and different background light conditions. By covering various real-world lighting combinations, including indoor and outdoor environments, strong and weak light, and complex backgrounds, the system can fully simulate lighting changes and background interference in real-world applications. This comprehensively verifies the stability, robustness, and anti-interference ability of spectral features in complex environments, avoids data bias caused by single lighting conditions, ensures that the trained recognition model can adapt to various complex environments, and significantly improves the system's reliability and generalization ability in real-world scenarios.

[0055] Among them, the classification and recognition models can be Support Vector Machine (SVM), Principal Component Analysis (PCA), Partial Least Squares Discriminant Analysis (PLS-DA), Random Forest, Convolutional Neural Network (CNN), or Transformer models. SVM, PLS-DA, and Random Forest are traditional machine learning models that are computationally efficient and have good generalization, making them suitable for classification and recognition of small sample spectral data. PCA is used for dimensionality reduction and feature compression of spectral data, which can remove redundant information and improve the computational efficiency of the model. CNN neural networks can automatically extract deep local features of the spectrum and are suitable for one-dimensional time-series spectral or two-dimensional spectral image data. The Transformer model captures long-distance spectral dependencies with its self-attention mechanism, which can effectively integrate static spectral features and dynamic time-series features, enhancing the recognition ability of complex samples. The above models can be used individually or in combination to meet different data scales and recognition accuracy requirements, ensuring the accuracy and reliability of the real and fake person identification results.

[0056] Optionally, the identification results may also include suspected cases, in which case manual identification can be performed through human intervention.

[0057] Reference Figure 4 As shown, based on the same inventive concept, another embodiment of this application discloses a real / dummy person identification system based on a spectral camera, which includes the following modules: The data acquisition module is used to acquire spectral data of known objects within the target wavelength range using a spectral camera and to build a spectral database. The known objects include real people and dummies, and the spectral data includes spectral data of real people and spectral data of dummies. The spectral data includes static spectral data and continuous dynamic spectral data. The feature extraction module is used to extract static spectral features from static spectral data and dynamic temporal features from continuous dynamic spectral data. Static spectral features are the characteristic absorption and scattering characteristics of the object within the band, while dynamic temporal features are the spectral fluctuation characteristics corresponding to the periodic physiological activities of the object over time. The model training module is used to train the classification and recognition model by taking static spectral features and dynamic temporal features as model inputs and the recognition result of whether it is a real person or a fake person as the model output. The recognition module is used to identify whether an unknown object is a real person or a fake person using a trained classification and recognition model.

[0058] The real / dual person identification system based on a spectral camera in this embodiment has the beneficial effects of the system embodiments described above, and will not be repeated here.

[0059] In one embodiment, the real / sham human identification system based on a spectral camera further includes the following modules: The calibration module is used to perform wavelength calibration on the spectrophotometer by outputting monochromatic light of multiple known wavelengths within the target band from a monochromator, establishing the correspondence between detector pixel positions and actual wavelengths. It also uses a standard reference response to calibrate the spectrophotometer's spectral response, resulting in a calibrated spectrophotometer. This ensures the accuracy of subsequently acquired data.

[0060] In one embodiment, the real / sham human identification system based on a spectral camera further includes the following modules: The preprocessing module is used to preprocess the spectral data; this preprocessing includes noise removal and standardization to improve the comparability between different samples.

[0061] It should be noted that "a certain body" or "a certain part" can be a portion of the corresponding "component," meaning that "a certain body" or "a certain part" is integrally formed and manufactured with the "other parts of the component"; or it can be an independent component that can be separated from the "other parts of the component," meaning that "a certain body" or "a certain part" can be manufactured independently and then combined with the "other parts of the component" to form a whole. The expression of "a certain body" or "a certain part" in this application is only one embodiment for ease of reading, and is not intended to limit the scope of protection of this application. Any technical solution that includes the above features and has the same function should be understood as an equivalent technical solution of this application.

[0062] It should be noted that some embodiments of this application have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that in the above embodiments and still achieve the desired result. In addition, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired result.

[0063] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; under the concept of this application, the above embodiments or technical features of different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0064] The embodiments described herein are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments described herein should be included within the protection scope of this application.

Claims

1. A method for identifying real and fake people based on a spectral camera, characterized in that, include: A spectral database is constructed by acquiring spectral data of a known object within a target wavelength range using a spectral camera; wherein the known object includes real people and dummy people, the spectral data includes spectral data of real people and spectral data of dummy people, and the spectral data includes static spectral data and continuous dynamic spectral data; Static spectral features are extracted from the static spectral data, and dynamic temporal features are extracted from the continuous dynamic spectral data; wherein, the static spectral features are the characteristic absorption and scattering features of the known object within the band, and the dynamic temporal features are the spectral fluctuation features corresponding to the periodic physiological activities of the known object that change over time; The static spectral features and the dynamic temporal features are used as model inputs, and the identification result of whether it is a real person or a fake person is used as model output to train the classification and identification model. The trained classification and recognition model is used to identify whether an unknown object is a real person or a fake person.

2. The method for identifying real and fake people based on a spectral camera according to claim 1, characterized in that, The static spectral characteristics include at least one of the following: oxyhemoglobin absorption characteristics near 540 nm wavelength, deoxyhemoglobin absorption characteristics near 560 nm wavelength, blood oxygen difference characteristics near 760 nm wavelength, oxyhemoglobin absorption characteristics near 850 nm wavelength, and water absorption characteristics near 940 nm wavelength. Alternatively, the static spectral characteristics can be obtained by calculating the reflectance difference, ratio, or spectral line slope between different characteristic bands.

3. The method for identifying real and fake people based on a spectral camera according to claim 1, characterized in that, The dynamic temporal physiological characteristics include at least one of the following: spectral fluctuation characteristics corresponding to weak pulse changes, spectral fluctuation characteristics corresponding to dynamic changes in blood oxygenation, and spectral fluctuation characteristics corresponding to periodic microcirculation changes.

4. The method for identifying real and fake people based on a spectral camera according to claim 1, characterized in that, Before acquiring spectral data of a known object within the target wavelength range using a spectral camera and constructing a spectral database, the process also includes: A monochromator outputs monochromatic light of multiple known wavelengths within the target band to perform wavelength calibration on the spectroscopic camera, establishing the correspondence between detector pixel positions and actual wavelengths; and a standard reference response is used to calibrate the spectroscopic camera's spectral response, resulting in a calibrated spectroscopic camera.

5. The method for identifying real and fake people based on a spectral camera according to claim 1, characterized in that, After acquiring spectral data of a known object within the target wavelength range using a spectral camera and constructing a spectral database, the steps before extracting static spectral features from the static spectral data and extracting dynamic temporal features from the continuous dynamic spectral data include: The spectral data is preprocessed; wherein the preprocessing includes noise removal and normalization.

6. The method for identifying real and fake people based on a spectral camera according to claim 1, characterized in that, The spectral data acquisition area includes at least one of the following regions of the known object: facial region, forehead region, bridge of nose region, cheek region, periorbital region, and hand region.

7. The method for identifying real and fake people based on a spectral camera according to claim 1, characterized in that, The static spectral data includes static spectral data under different ambient light and different background light, and the continuous dynamic spectral data includes continuous dynamic spectral data under different ambient light and different background light.

8. The method for identifying real and fake people based on a spectral camera according to claim 1, characterized in that, The dummy includes at least one of the following: silicone dummy, resin model, latex mask, wax figure, screen copy object, and photograph object.

9. The method for identifying real and fake people based on a spectral camera according to claim 1, characterized in that, The target wavelength range is 400-1000nm.

10. A system for identifying real and fake people based on a spectral camera, characterized in that, include: The data acquisition module is used to acquire spectral data of a known object within a target wavelength range using a spectral camera and to construct a spectral database; wherein the known object includes real people and dummy people, the spectral data includes spectral data of real people and spectral data of dummy people, and the spectral data includes static spectral data and continuous dynamic spectral data; The feature extraction module is used to extract static spectral features from the static spectral data and dynamic temporal features from the continuous dynamic spectral data; wherein, the static spectral features are the characteristic absorption and scattering characteristics of the known object in the band, and the dynamic temporal features are the spectral fluctuation characteristics corresponding to the periodic physiological activities of the known object over time; The model training module is used to train the classification and recognition model by taking the static spectral features and the dynamic temporal features as model inputs and the recognition result of whether it is a real person or a fake person as model output. The recognition module is used to identify whether an unknown object is a real person or a fake person using a trained classification and recognition model.

Images