Secure authentication

Abstract

Disclosed is a method for measuring a blood perfusion of a subject, the method comprising: receiving an image generated when the subject is illuminated by coherent infrared light; generating a plurality of partial images based on the image; the plurality of partial images are provided to a data-driven model to determine material property metrics based on locations of the plurality of partial images within the image, where the data-driven model is trained based on historical partial images and corresponding material property metrics to determine material property metrics. The historical partial images are provided to the data-driven model based on the positions of the historical partial images; material property metrics are provided.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to secure authentication and provides a method for allowing an object to access a resource, a method for measuring blood perfusion of an object, use of material properties, use of multiple partial images, a non-transitory computer-readable storage medium, use of a data-driven model, a device and / or system for allowing an object to access a resource. BACKGROUND

[0002] Authentication processes can be spoofed by masks, images, etc. that represent characteristics of a user. Spoofing items are becoming more realistic, so it is becoming more difficult to distinguish between a spoofed object and a human.

[0003] Therefore, there is a need for reliably distinguishing between a human and a spoofed item.

[0004] It is an object of the present disclosure to provide a robust, reliable and secure method for authenticating a human. SUMMARY

[0005] In an aspect, the present disclosure relates to a method for allowing an object to access a resource, the method comprising: receiving a request to access a resource, in response to receiving the request to access the resource, triggering illuminating the object with coherent infrared light, and triggering generating an image of the object while the object is illuminated by the coherent infrared light; generating a plurality of partial images based on the image; providing the plurality of partial images to a data-driven model based on locations of the plurality of partial images within the image to determine whether the object is a living organism, wherein the data-driven model is trained based on a plurality of historical partial images to determine whether one or more objects associated with the plurality of historical partial images are living organisms; allowing the object to access the resource based on determining that the object is a living organism.

[0006] In another aspect, the present disclosure relates to a method for measuring blood perfusion of an object, the method comprising: receiving an image generated while the object is illuminated by coherent infrared light; generating a plurality of partial images based on the image; providing the plurality of partial images to a data-driven model based on locations of the plurality of partial images within the image to determine a blood perfusion measure, wherein the data-driven model is trained based on historical partial images and corresponding blood perfusion measures; optionally, receiving the blood perfusion measure from the data-driven model; providing the blood perfusion measure.

[0007] In an aspect, the present disclosure relates to a method for measuring blood perfusion of an object, the method comprising:

[0008] receiving an image of the object under illumination by coherent infrared light,

[0009] generating a plurality of partial images based on the image,

[0010] determining a blood perfusion measure from the partial images and the locations of the partial images within the image by providing the plurality of partial images to a data driven model trained on the basis of historical partial images and corresponding blood perfusion measures, the historical partial images being provided to the data driven model on the basis of locations of the historical partial images,

[0011] providing the blood perfusion measure.

[0012] In another aspect, the disclosure relates to a method for allowing a subject to access a resource, the method comprising:

[0013] receiving a request to access the resource,

[0014] in response to receiving the request to access the resource, triggering illuminating the subject with coherent infrared light, and

[0015] triggering generating an image of the subject while the subject is illuminated by the coherent infrared light,

[0016] generating a plurality of partial images on the basis of the image,

[0017] determining whether the subject is a living organism from the partial images and the locations of the partial images within the image by providing the plurality of partial images to a data driven model trained on the basis of a plurality of historical partial images to determine whether one or more subjects associated with the plurality of historical partial images are living organisms, the plurality of historical partial images being provided on the basis of locations of the plurality of historical partial images,

[0018] allowing the subject to access the resource on the basis of determining that the subject is a living organism.

[0019] In another aspect, the disclosure relates to a use of an indication of a material property generated as described herein for allowing a subject to access a resource.

[0020] In another aspect, the disclosure relates to a method for measuring blood perfusion of a subject, the method comprising: receiving an image of the subject under illumination with coherent infrared light,

[0021] generating a plurality of partial images from the image; providing the plurality of partial images to a data driven model on the basis of locations of the plurality of partial images within the image to determine a blood perfusion measure, wherein the data driven model is trained with historical partial images and corresponding blood perfusion measures, the historical partial images being provided to the data driven model on the basis of locations of the historical partial images; providing the blood perfusion measure.

[0022] In another aspect, this disclosure relates to a method for allowing an object to access a resource, the method comprising: receiving a request to access the resource; in response to receiving the request to access the resource, triggering the object to be illuminated with coherent infrared light, and triggering the generation of an image of the object when the object is illuminated by the coherent infrared light; generating a plurality of partial images based on the image; providing the plurality of partial images to a data-driven model based on the positions of the plurality of partial images within the image to determine whether the object is a living organism, wherein the data-driven model is trained using a plurality of historical partial images to determine whether one or more objects associated with the plurality of historical partial images are living organisms, the plurality of historical partial images being provided based on the positions of the plurality of historical partial images; and allowing the object to access the resource based on the determination that the object is a living organism.

[0023] On the other hand, this disclosure relates to a method for allowing an object to access resources, the method comprising: receiving an image generated when the object is irradiated with coherent infrared light; generating a plurality of partial images based on the image; providing the plurality of partial images to a data-driven model based on the positions of the plurality of partial images within the image to determine whether the object is a living organism, wherein the data-driven model is trained based on a plurality of historical partial images to determine whether one or more objects associated with the plurality of historical partial images are living organisms; and allowing the object to access resources based on the determination that the object is a living organism.

[0024] On the other hand, this disclosure relates to the use of multiple partial images generated based on an image generated when an object is irradiated with coherent infrared light for allowing the object to access resources in such a way that the multiple partial images are provided to a data-driven model based on the positions of the multiple partial images within the image to determine whether the object is a living organism based on the multiple partial images, wherein the data-driven model is trained based on historical partial images, wherein the data-driven model is trained based on multiple historical partial images to determine whether one or more objects associated with the multiple historical partial images are living organisms.

[0025] On the other hand, this disclosure relates to a non-transitory computer-readable storage medium including instructions that, when processed by a computer, configure the computer to perform any of the methods described herein.

[0026] On the other hand, this disclosure relates to a device and / or system for allowing objects to access resources, the device and / or system comprising: a processor; and a memory storing instructions that, when executed by the processor, configure the device and / or system to perform any of the methods described herein.

[0027] On the other hand, this disclosure relates to the use of a data-driven model adapted to receive multiple partial images based on their positions within a generated image, wherein the image is generated when an object is illuminated by coherent infrared light, and wherein the partial images are generated based on the image to the data-driven model, and to determine whether the object is a living organism based on the multiple partial images, wherein the data-driven model is trained based on multiple historical partial images to determine whether one or more objects associated with the multiple historical partial images are living organisms.

[0028] On the other hand, this disclosure relates to a method for measuring the material properties of an object, the method comprising:

[0029] Receive an image of the object under coherent infrared light illumination.

[0030] Multiple partial images are generated based on this image.

[0031] Based on the positions of these multiple partial images within the image, they are provided to a data-driven model to determine material property measures. This data-driven model is trained based on historical partial images and corresponding historical material property measures, and these historical partial images are provided to the data-driven model based on their positions.

[0032] Provide a measure of the material's properties. Example

[0033] Any disclosures, embodiments, and examples described herein relate to the methods, systems, devices, and computer-readable storage media listed above and below. Advantageously, the benefits provided by any embodiments and examples also apply to all other embodiments and examples.

[0034] In the following sections, the terminology and / or technical fields used herein and / or the scope of this disclosure will be outlined by way of definition and / or examples. Where examples are given, it should be understood that this disclosure is not limited to those examples.

[0035] This disclosure provides an apparatus for a method of efficiently, robustly, stably, and reliably authenticating objects. Commercial authentication systems can be easily fooled by hyper-realistic deception objects such as silicon masks. Blood perfusion can be used as an effective feature to distinguish between deception masks and real people. Blood perfusion can be detected by providing an image of the object generated when it is irradiated with coherent infrared light to a data-driven model. The data-driven model can reliably distinguish between living organisms and deception objects. In particular, providing multiple partial images to the data-driven model based on the location of multiple partial images within an image to determine whether the object is a living organism and / or an indication of material properties allows for efficient and secure authentication. Partial images are easier to process, thus enabling secure authentication to be implemented using fewer computational resources. Providing partial images based on location allows processing all image data associated with the image while improving computational efficiency in terms of hardware requirements and computation time. To determine an indication of material properties, the context associated with the complete image data and therefore the relationships between the partial images improve the accuracy of authentication. Therefore, secure authentication can be performed based on low-cost and readily available hardware. This enables secure authentication of mobile devices with limited space and battery capacity, such as smartphones. Furthermore, providing partial images based on their location within an image allows for even more reliable authentication, as the context of the partial image within the image can be taken into account. Therefore, spoofed objects can be reliably detected. Ultimately, the security of authentication, particularly image-based authentication, can be improved. This enables the implementation of authentication systems on low-cost hardware for high-security use cases such as border control, access control, and engine start control.

[0036] These and other objectives are addressed by the subject matter of the independent claims, and will become apparent upon reading the following description. The dependent claims relate to embodiments of the content of this disclosure.

[0037] In embodiments, these methods may further include receiving material property measures from a data-driven model. Providing multiple partial images to the data-driven model based on the positions of multiple partial images within an image to determine whether an object is a living organism may refer to providing multiple partial images to the data-driven model based on the positions of multiple partial images within an image to determine material property measures. Allowing an object to access resources based on determining that the object is a living organism may refer to allowing the object to access resources based on blood perfusion measures received from the data-driven model indicating that the object may be a living organism. Blood perfusion measures can indicate whether an object can be a living organism. Therefore, blood perfusion measures can indicate that the object can be a living organism, or that the object can be a deceiving object. The object can be a person and / or a user, such as a user of a device. The device can be configured to perform any of the steps described herein.

[0038] In this embodiment, the coherent infrared light may be patterned coherent infrared light.

[0039] Coherent infrared light can include near-infrared, mid-infrared, and / or far-infrared light. Near-infrared light can range from 780 nm to 3000 nm, excluding the 3000 nm value. Mid-infrared light can range from 3 µm to 15 µm, excluding the 15 µm value. Far-infrared light can range from 15 µm to 1000 µm. Infrared light can be associated with wavelengths between 700 nm and 800 nm and / or between 1000 nm and 1200 nm.

[0040] In this embodiment, the wavelength of the coherent infrared light can be between 700 nm and 800 nm and / or between 1100 nm and 1200 nm. This is advantageous because sunlight may be relatively weak in that range. Therefore, using light within this range improves the performance of security authentication under ambient light conditions.

[0041] Patterned coherent infrared light can include multiple beams. Illuminating an object with patterned coherent infrared light can include projecting multiple beams associated with the patterned coherent infrared light onto the object. Projecting multiple beams associated with the patterned coherent infrared light onto the object can include projecting multiple spots of light associated with the multiple beams onto the object.

[0042] In this embodiment, a partial image may refer to a portion of an image. A portion of an image may be smaller than the image itself. A partial image may include a portion of an image smaller than the image. Multiple partial images and / or segments may have the same shape and / or size. The data-driven model may be parameterized and / or trained to receive multiple partial images of equal size and / or shape. This enables efficient parameterization of the data-driven model, thereby saving resources used for implementing secure authentication, even on mobile devices with limited battery and storage capacity.

[0043] In embodiments, the image may show multiple spots associated with illumination by patterned coherent infrared light. Multiple partial images may show at least one spot among the multiple spots in each partial image. Specifically, a partial image may be associated with one spot. Further, the multiple partial images may show the background associated with the multiple spots. Specifically, the multiple partial images may show the background associated with at least one spot associated with at least one of the multiple partial images. Therefore, multiple partial images can be generated to associate, in particular, at least one spot in each partial image. Optionally, multiple partial images can be generated to associate, in particular, the background associated with at least one spot in each partial image. In embodiments, at least one partial image may be associated with fewer than 10, preferably fewer than 5, and even more preferably fewer than two spots. By doing so, the contribution of each spot and its relationship to other spots, particularly surrounding spots, can be evaluated to measure the blood perfusion and / or material properties of the object.

[0044] In one embodiment, the region of interest associated with an image can be selected based on a floodlight image showing the object. This may include identifying one or more portions of the image associated with the object by classifying one or more portions of the image associated with the object. An indication of the region of interest may be received. Multiple partial images may be generated based on the selected region of interest.

[0045] In an embodiment, a data-driven model can refer to a model suitable for describing one or more nonlinear relationships between input data and output data. Input data can refer to data to be provided to the data-driven model and / or data received by the data-driven model. Output data can be data to be received from the data-driven model and / or data to be provided by the data-driven model. Therefore, the data-driven model can determine output data based on transforming input data via one or more nonlinear relationships. In this context, input data can be, for example, multiple partial images. Output data can be, for example, an indication that the object associated with the image is either a living organism or a deceptive object.

[0046] The data-driven model can be trained based on multiple historical partial images to determine whether an object, preferably an object associated with the multiple historical partial images, can be a living organism. This can mean that the data-driven model can be trained based on a first set and a second set. The first set includes multiple first partial images and corresponding indicators of whether an object associated with the first set can be a living organism, and the second set includes multiple second partial images and corresponding indicators of whether an object associated with the second set can be a living organism.

[0047] In embodiments, the data-driven model may include one or more embedding layers. The embedding layers may be adapted to transform provided data (such as partial images) into a machine-processable representation. The machine-processable representation may include one or more numerical values. The machine-processable representation may include one or more first-order tensors and / or may be second-order tensors. In embodiments, the machine-processable representation of multiple partial images may be a compressed representation of the multiple partial images. Any of these methods may further include mapping the multiple partial images to the machine-processable representation of the multiple partial images. Preferably, each partial image may generate one machine-processable representation.

[0048] Furthermore, the data-driven model may include one or more model blocks, such as transformer blocks or graph neural network blocks. One or more model blocks may be adapted to transform machine-processable representations into context tensors. The context tensor may be a second-order tensor. Further, the data-driven model may include one or more classification layers. The classification layers may be adapted to transform the context tensor into an indication of whether an object can be a living organism.

[0049] In this embodiment, the object can be a living organism and / or a deceptive object. The deceptive object can be presented for authenticating non-living objects. The deceptive object may include one or more characteristics associated with an authorized user.

[0050] In an embodiment, a historical partial image can be a partial image. Embodiments applied to partial images can be similarly applied to historical partial images. A historical partial image can be a portion of training data used to train a data-driven model. A historical partial image can be associated with a corresponding blood perfusion metric. A historical partial image can be associated with an indication that the object associated with the historical partial image is a living organism. Therefore, a historical partial image may have been classified based on whether the object associated with the historical partial image is a living organism.

[0051] In embodiments, providing multiple partial images to a data-driven model based on their positions within an image to determine whether an object can be a living organism, wherein the data-driven model can be trained based on multiple historical partial images to determine whether one or more objects associated with the multiple historical partial images can be living organisms, can refer to providing multiple partial images to the data-driven model based on their positions to determine whether an object can be a living organism based on the multiple partial images, wherein the data-driven model can be trained based on historical partial images and corresponding historical indications of whether historical objects associated with these historical partial images can be living organisms. These methods may further include receiving an indication from the data-driven model whether an object can be a living organism. Based on receiving an indication that an object can be a living organism, access to resources can be permitted for the object. Allowing access to resources based on determining that an object can be a living organism can refer to allowing access to resources based on receiving an indication that an object can be a living organism. During training, the indication regarding whether an object can be a living organism may include one or more labels associated with historical images. Further, the indication regarding whether an object can be a living organism may indicate whether an object associated with an image can be a living organism. The indication regarding whether an object can be a living organism may include numerical and / or Boolean values. If the value is within a predefined range, the blood perfusion measurement indicates that the object associated with the image can be a living organism. Otherwise, the blood perfusion measurement indicates that the object associated with the image can be a spoofed object. A Boolean value can indicate whether the object associated with the image can be a spoofed object or a living organism.

[0052] In an embodiment, providing the data-driven model with the partial images based on their positions can refer to providing the data-driven model with multiple partial images, wherein providing the multiple partial images can indicate the positions of the multiple partial images within the image. In an embodiment, providing the data-driven model with the partial images based on their positions includes providing the data-driven model with data structures associated with the multiple partial images and their positions within the image, wherein the data-driven model can be parameterized and / or trained to be provided with the data structures associated with the multiple partial images and their positions within the image and / or to provide the data-driven model with sequences of multiple partial images based on their positions within the image, wherein the data-driven model can be parameterized and / or trained to be provided with sequences of multiple partial images. The sequence of multiple partial images can indicate the distance between two or more partial images among the multiple partial images. This can also be applied to historical partial images. The multiple partial images can be provided in sequence, wherein the sequence depends on the positions of the multiple partial images within the image. The sequence can involve a sequence of positions of the multiple partial images within the image. The sequence can be determined by the positions of the multiple partial images within the image. In embodiments, these methods may further include mapping a plurality of partial images to data structures associated with the plurality of partial images and their locations within an image. The data structures associated with the plurality of partial images and their locations within an image can be obtained by mapping the plurality of partial images to the data structures associated with the plurality of partial images and their locations within an image.

[0053] In an embodiment, allowing an object to access a resource may include allowing the object to perform at least one operation using a device, computing device, and / or system. The resource may be a device, system, computing device, functionality of a computing device, functionality of a device, functionality of a system, and / or entity. Additionally and / or alternatively, allowing an object to access a resource may include allowing the object to access an entity. An entity may be a physical entity and / or a virtual entity. A virtual entity may be, for example, a database. A physical entity may be an access-restricted area. An access-restricted area may be one of the following: a secure area, a room, an apartment, a vehicle, a portion of the previously mentioned examples, etc.

[0054] In embodiments, these methods may further include receiving a floodlight image of an object generated when the object can be illuminated with infrared light. The floodlight image may show and / or indicate the outline of the object. Further, these methods may include receiving a template image of an authorized user. The template image may show and / or indicate the outline of an authorized user. Further, these methods may include providing the floodlight image and the template image to an authentication data-driven model to determine whether an object corresponds to an authorized user. The authentication data-driven model may be trained based on multiple historical floodlight images and multiple historical template images to determine whether one or more objects associated with historical floodlight images correspond to one or more authorized users associated with historical template images. For this purpose, the authentication data-driven model may include one or more embedding layers. The one or more embedding layers may be configured to reduce the dimensionality of the floodlight image. The one or more embedding layers may transform the floodlight image and the template image into tensors, particularly two-dimensional tensors or one-dimensional tensors. Alternatively, receiving the template image may include receiving a tensor associated with the template image. The tensor associated with the template image may be obtained by providing the template image to one or more embedding layers of, for example, a second data-driven model. Additionally or alternatively, the authentication data-driven model may include one or more classification layers. These classification layers may be configured to receive a tensor associated with the floodlight image and a tensor associated with the template image, and / or to classify the tensor associated with the floodlight image and the tensor associated with the template image based on whether the floodlight image and the template image can be associated with the same authorized user. Additionally or alternatively, the authentication data-driven model may include one or more mathematical relationships, particularly Euclidean distance, and / or cosine similarity, for determining the distance between the tensor associated with the floodlight image and the tensor associated with the template image.

[0055] In embodiments, these methods may further include receiving a floodlight image of the object generated when the object can be illuminated with infrared light. The floodlight image may show the outline of the object, and receiving a template image of an authorized user. The template image may show the outline of an authorized user. These methods may further include providing the floodlight image and the template image to a data-driven model to determine whether the object corresponds to an authorized user. The data-driven model may be trained based on multiple historical floodlight images and multiple historical template images to determine whether one or more objects associated with historical floodlight images correspond to one or more authorized users associated with historical template images.

[0056] In one embodiment, providing multiple partial images to a data-driven model based on their positions within an image can include providing the multiple partial images sequentially based on their positions within the image. The data-driven model can be further parameterized and / or trained to provide the multiple partial images sequentially based on their positions within the image. By doing so, even more reliable authentication can be achieved because the context of the partial images within the image can be taken into account. Therefore, deceitful objects can be reliably detected.

[0057] In an embodiment, generating multiple partial images based on an image includes dividing the image into multiple segments. Further, generating multiple partial images based on an image includes cropping the image according to the multiple segments to generate multiple partial images. The multiple partial images and / or segments may have the same shape and / or size.

[0058] In an embodiment, the sequence based on the positions of multiple partial images within an image can be a linear sequence, starting with the partial image associated with the top left segment, followed by other partial images associated with top segments from left to right, and then other partial images associated with other top segments from left to right. The data-driven model can include a transformer-based architecture. The data-driven model can be parameterized and / or trained to apply self-attention, particularly multi-head self-attention, to multiple partial images, particularly linear sequences and / or machine-processable representations of linear sequences. Machine-processability of the linear sequence can be achieved by passing the linear sequence through one or more embedding layers. Linear sequences allow the use of components already available in the data-driven model. Therefore, resources and time can be saved for implementing highly reliable security authentication.

[0059] In an embodiment, providing multiple partial images to a data-driven model based on their positions within an image may include providing a graph data structure representing the multiple partial images to the data-driven model. The graph data structure may include multiple nodes and multiple edges. A node in the graph data structure may represent one of the multiple partial images, and an edge in the graph data structure may represent the distance between at least two of the multiple partial images. The data-driven model may be parameterized and / or trained to be provided with the graph data structure representing the multiple partial images. The number of nodes may be equal to the number of partial images. The multiple partial images may be represented by nodes in the graph. The number of edges may be equal to the number of pairs of partial images. Nodes may include node vectors. Edges may include edge vectors. The graph data structure may be a linear graph data structure representing a sequence of partial images. The data structure can be a graph data structure associated with multiple partial images and their positions within an image. This graph data structure can include multiple nodes and multiple edges. A node of the graph data structure can be associated with one of the partial images, and an edge of the graph data structure can be associated with a sequence of two or more partial images and / or the distance between them. The data-driven model can be parameterized such that the provided data structure is parameterized as a graph data structure. By doing so, more reliable and robust detection of deception targets is achieved, thus enabling secure authentication. The graph data structure can be a directed graph data structure. A directed graph data structure can be associated with a sequence of multiple partial images within an image. The direction associated with the directed graph data structure can indicate the position of the multiple partial images within the image.

[0060] In an embodiment, one or more of the plurality of partial images and / or segments may include at least a portion of a light spot generated by patterning coherent infrared light illuminating the object. One or more of the plurality of partial images and / or segments may further include at least a portion surrounding the light spot within a predefined distance from its center. By doing so, background suppression is achieved, thereby focusing the attention of the data-driven model on determining whether the object is a living organism or a spoofed object. Thus, secure authentication is achieved.

[0061] In this embodiment, partial images may be associated with a grid of images, and the sequence may be defined by a predefined sequence path that connects partial images in the grid. One or more portions of the grid may be associated with two or more partial images. The first partial image in the sequence may be a predefined partial image, and subsequent partial images may be the nearest neighbor and / or second nearest neighbor of the previous partial image. Additionally or alternatively, subsequent partial images may be partial images adjacent to the previous partial images. In particular, adjacent partial images may be associated with the same portion of the image as the preceding partial image in the sequence.

[0062] In embodiments, these methods may further include preprocessing the image by detecting one or more features in the image and enhancing the image based on the detected features. The detected features may be light spots. Enhancing the image based on the detected features may include cropping the image based on the one or more detected features. Preferably, cropping the image based on the one or more detected features may include eliminating at least a portion of the image features around them within a predefined distance from the feature center. In embodiments, the one or more image features may be spatial features of an object. Preferably, cropping the image based on the one or more detected image features may include eliminating at least a portion of the one or more detected image features or eliminating at least a portion of the image features around them. By doing so, background suppression is achieved, thereby focusing the attention of the data-driven model on determining whether the object is a living organism or a deceitful object. Therefore, secure authentication is achieved.

[0063] In embodiments, these methods may further include preprocessing multiple partial images by applying one or more image enhancement techniques to multiple partial images. Applying one or more image enhancement techniques to multiple partial images can produce multiple partial images of equal size and / or shape. In embodiments, image enhancement techniques may include at least one of the following: scaling, cropping, pruning, rotating, blurring, distorting, shearing, resizing, folding, changing contrast, changing brightness, adding noise, multiplying by at least a portion of pixel values, filtering, adjusting color, applying convolution, imprinting, sharpening, flipping, averaging pixel values, etc. Preferably, applying one or more image enhancement techniques may include cropping and / or shearing images based on one or more image features associated with the image. Cropping images may include changing the distance between one or more first image features associated with the image and one or more second image features, and optionally changing the distance between one or more first image features associated with the image and one or more third image features.

[0064] By doing so, background suppression is achieved, thus focusing the attention of the data-driven model on determining whether the object is a living organism or a deceitful object. Therefore, secure authentication is achieved.

[0065] In embodiments, material property measures, particularly blood perfusion measures, may be suitable for determining whether an object associated with an image can be a living organism, and / or may indicate whether an object associated with an image can be a living organism. Material property measures, particularly blood perfusion measures, may include one or more numerical values ​​and / or Boolean values. The numerical value may indicate a confidence score associated with determining whether an object associated with an image can be a living organism. The Boolean value may indicate whether an object associated with an image can be a living organism. If one or more numerical values ​​are within a predefined range, the material property measure, particularly blood perfusion measures, may indicate that the object associated with the image can be a living organism. Otherwise, the one or more numerical values ​​may indicate that the object associated with the image can be a deceitful object.

[0066] In this embodiment, it can be determined whether an object is a living organism based on multiple partial images.

[0067] In embodiments, these methods may be computer-implemented methods.

[0068] In an embodiment, "model driven by training data" can refer to using training data to drive the model. In an embodiment, "generating multiple partial images based on an image" can refer to generating multiple partial images based on the image.

[0069] In embodiments, material property measurements may be associated with material type and / or blood perfusion measurements, particularly including material type and / or blood perfusion measurements. Material types may include biological or non-biological materials, translucent or non-translucent materials, metallic or non-metallic materials, skin or non-skin, latex or non-latex, silicone or non-silicone, fabric or non-fabric, reflective or non-reflective, specular or non-specular, foam or non-foam, hair or non-hair, roughness groups, etc. Material types may include further specifications of the material type; for example, in a biological category, the type may be human skin, or in a non-biological category, the type may be plastic, glass, or metal. Further subtypes may be assigned to each type. Indications of material properties may be material property measurements. Material properties may indicate the type of material and / or whether the object associated with the material property can be a living organism. Attached Figure Description

[0070] The disclosure will be further described below with reference to the accompanying drawings. In the drawings and the disclosure, the same reference numerals are intended to refer to the same or similar elements, components and / or portions.

[0071] Figure 1AAn embodiment of a device for allowing objects to access resource 102 is shown.

[0072] Figure 1B An example of a system for allowing objects to access resources is shown.

[0073] Figure 2 An example of a method for allowing objects to access resources is shown.

[0074] Figure 3 An embodiment of generating multiple partial images 324 based on image 322 is shown.

[0075] Figure 4 An embodiment is shown in which multiple partial images 432 are provided to a data-driven model to determine whether an object is associated with image 430.

[0076] Figure 5 An example of a method for allowing objects to access resources is shown.

[0077] Figure 6 An embodiment of generating multiple partial images 324 based on image 322 is shown.

[0078] Figure 7 An embodiment is shown in which multiple partial images 324 are provided to a data-driven model to determine whether an object is associated with image 322.

[0079] Figure 8 Examples of multiple partial images are shown. Detailed Implementation

[0080] The following embodiments are merely examples for implementing the methods, systems, or application devices disclosed herein and should not be considered limiting.

[0081] Figure 1A An embodiment of a device for allowing objects to access resource 102 is shown.

[0082] Device 102 may include an illumination source 108, a camera 110 including a sensor 112, a processor 104, and / or a memory 114. Illumination source 108 may emit coherent light, preferably coherent infrared light, toward object 106. Infrared light may be undetectable to object 106. Sensor 112 of camera 110 may be sensitive to the light emitted from illumination source 108. Therefore, sensor 112 may be adapted to generate an image of object 106 when a user is illuminated by light emitted from illumination source 108. Processor 104 may receive the image of object 106. Processor 104 may process the image of object 106. By doing so, processor 104 may determine whether object 106 corresponds to a living organism. This may include determining whether the image of object 106 corresponds to a living organism. For this purpose, processor may execute instructions stored in memory 114. Execution of instructions by the processor may cause the following to occur: Figure 2 , Figure 3 and Figure 6 Any of the methods described herein.

[0083] Alternatively, the illumination source 108 and / or camera 110 may be included in the second and / or third device. In this example, the processor 104 may be communicatively coupled to the second and / or third device to trigger the illumination source 108 to emit light and trigger the sensor 112 to generate an image of the object 106 when it is illuminated.

[0084] Processor 104 can receive requests to access resources, such as unlocking a device. For this purpose, the device may include a user interface. Object 106 can request access to resources through the user interface. For example, the device could be a telephone. Object 106 may wish to control the telephone. For this purpose, object 106 may need to authenticate. Object 106 can request access to resources and / or authenticate by using the device's touchscreen display. This can trigger a signal to the device's processor 104. Based on the received signal, processor 104 triggers illumination source 108 to emit light.

[0085] Figure 1B An example of a system for allowing objects to access resources is shown.

[0086] The system may include a first device 130 and a second device 128. In an embodiment, the first device 130 may include a processor 104 and a memory 114. The first device 130 may be communicatively connected to the second device 128, which includes an illumination source 108 and a camera 122 (including a sensor 112). For example, the first device 130 may be connected to the second device 128 by means of a cloud service. In particular, the processor 104 and / or the memory 114 may be part of a cloud service. The second device 128 may be configured to provide an image generated by the sensor 112 to the first device 130. Receiving a request to access a resource may refer to receiving a signal that triggers the illumination source 108 to emit light at the illumination source 108. In response to receiving this signal, the illumination source 108 may be triggered to illuminate the object 106 with light. Further, the camera 110 may receive a signal that triggers the generation of an image of the object 106. Receiving a request to access a resource may further include receiving a signal at the camera 110.

[0087] Processor 104 can determine whether an object 106 associated with an image is a living organism by generating multiple partial images based on an image and providing these partial images to a data-driven model. Based on the indication that object 106 is a living organism, processor 104 can provide a signal to grant access rights to object 106. Providing a signal to grant access rights to object 106 can be referred to as allowing object 106 to access resources. This can be as follows: Figure 2 , Figure 3 and / or Figure 6 As described in the context.

[0088] Figure 2 An example of a method for allowing objects to access resources is shown.

[0089] Requests to access resources can be received 202. For example, a user may wish to unlock a device such as a smartphone and / or perform actions on the device that may require authentication, such as making a payment. For this purpose, the user can request access to resources, particularly the device. Additionally or alternatively, the user can request access to an area and / or item. This request can be triggered by the user entering the area and / or interacting with the user interface. The request can be received at the device, at a security point associated with the area, and / or at a control unit associated with the storage of the item.

[0090] A request for access to a resource can trigger the illumination of an object with coherent infrared light. The object may be oriented towards a device, area, and / or item. The object may be a user. Receiving the request may include receiving a signal indicating a request to access the resource. Receiving the signal indicating the request may cause the generation and / or provision of a signal indicating the triggering of coherent infrared light illumination. The object may be located near the illumination source for coherent infrared light illumination. Preferably, the illumination source may be oriented towards the object. Therefore, the object can be illuminated with coherent infrared light. The coherent infrared light may interact with the material associated with the object. Depending on the material and the resulting interaction, different interference patterns can be obtained by illuminating the object with coherent infrared light. An image may illustrate the interaction between the coherent infrared light and the material of the object. In particular, the coherent infrared light may be infrared coherent infrared light. Infrared light may be invisible to humans. Therefore, this enables authentication of people independently of their awareness of the authentication process. Thus, secure authentication is achieved. Furthermore, infrared light can penetrate deeper into the skin (e.g., the dermis). The dermis may include blood vessels. Blood vessels may be filled with blood. The amount of blood within a portion of a blood vessel may depend on the cardiac cycle. Therefore, the movement of blood within the skin region of a living organism can vary over time, preferably periodically. Coherent infrared light can interact with blood vessels and blood. Based on this interaction, at least a portion of the coherent infrared light can leave the skin. This departing light can be collected in an image. The faster the blood flows, the more the light is deflected. Therefore, the presence of moving blood (so-called blood perfusion) can cause the coherent infrared light in the generated image to become blurred.

[0091] Furthermore, the coherent infrared light can be patterned coherent infrared light. This allows for the illumination of a smaller area of ​​an object while permitting secure authentication of authorized users and detection of spoofed objects. Patterned coherent infrared light can include multiple beams. Illuminating the patterned coherent infrared light can cause multiple beams to be projected onto the object. Thus, multiple light spots can be projected onto the object. In an embodiment, the patterned coherent infrared light can include fewer than 4000 beams. In an embodiment, projecting patterned coherent infrared light onto an object can cause the projection of fewer than 4000 light spots. In an embodiment, the patterned coherent infrared light can include fewer than 3000 beams, preferably fewer than 2000 beams, and most preferably fewer than 1000 beams. In an embodiment, projecting patterned coherent infrared light onto an object can cause the projection of fewer than 3000 light spots, preferably fewer than 2000 light spots, and most preferably fewer than 1000 light spots.

[0092] Therefore, image 206 can be generated when the object is illuminated by coherent infrared light. A signal indicating the triggering of coherent infrared light illumination can trigger the generation of an image of the object when it is likely to be illuminated by coherent infrared light.

[0093] Image preprocessing 208 can be performed by enhancing the image. Image enhancement can include applying one or more image enhancement techniques to the image. For example, image preprocessing can include detecting one or more image features in the image 218. When the coherent infrared light may be patterned coherent infrared light, the image features can include light spots. Therefore, detecting one or more image features in the image can refer to detecting one or more light spots. Further, the image can be enhanced based on the detected image features, preferably light spots. Enhancing the image based on the detected image features can refer to cropping the image based on the detected image features. The image can include image features and a background. The background may be irrelevant to the blood perfusion of the object. Therefore, background removal may be advantageous. Enhancing the image based on the detected image features can include removing at least a portion of the background. Cropping the image based on the image features can include removing at least a portion of the background based on the detected image features. For example, the background located within a predefined distance from the center of one or more image features can be removed by cropping the image. Examples of preprocessed images may include... Figure 3 and Figure 6 As shown.

[0094] It can be as follows Figure 3 and Figure 6 As described in the context, multiple partial images are generated based on the image 210. These partial images can be preprocessed by enhancing them 212. This can further improve the reliability of authenticating authorized users.

[0095] The correlation between images and material property measures of objects, particularly blood perfusion measures, can be represented by a data-driven model. Therefore, a data-driven model can be trained to obtain the correlation between images and material property measures, particularly blood perfusion measures. The data-driven model can be trained based on multiple historical partial images to determine whether one or more objects associated with multiple historical partial images are living organisms. For this purpose, the data-driven model can be initialized, and one or more sets of historical partial images can be provided to the data-driven model. Historical partial images can be, for example, in... Figure 3 and Figure 6 As described in the context, processing is performed through a data-driven model. The processing of partial images can be analogous to the processing of historical partial images. By processing historical images, the data-driven model can generate indications as to whether an object associated with a historical image and / or historical partial image can be a living organism, as in... Figure 3 and Figure 6As described in the context. Historical partial images can be generated based on historical images. One or more sets of historical partial images can be generated based on one or more historical images. Therefore, a set of historical partial images can be generated based on a single historical image. A set of historical partial images can be associated with historical indicators of whether the objects associated with the historical partial images are living organisms. For example, a set of historical partial images can be labeled based on whether the objects associated with the historical partial images are living organisms or deceptive objects. Multiple partial images can be provided to the data-driven model 214.

[0096] Based on determining that an object can be a living organism by providing multiple partial images to a data-driven model, the object may be allowed access to resources 216. This may include unlocking a device (e.g., a smartphone), allowing the object to perform a requested action on and / or using the device, providing the object with access to an area and / or providing the object with access to an item and / or providing the object with an item.

[0097] Figure 3 An embodiment of generating multiple partial images 324 based on image 322 is shown.

[0098] Image 322 can be derived from, for example Figure 1A and / or Figure 1B The image 322 can be generated from the device and / or system described in the context of the image. Figure 2 It is generated as described in the context. Further, image 322 can be as follows: Figure 2 The preprocessing results are described in the context of [the previous sentence]. Multiple partial images 324 can be generated based on a preprocessed image 322. Generating multiple partial images may include dividing image 322 into multiple segments and generating multiple partial images 324 associated with the multiple segments. These segments may be associated with geometric shapes, such as squares, rectangles, triangles, or any other geometric shapes. These segments may have equal shapes and / or sizes, and / or may differ in shape and / or size. Figure 3 In this embodiment, image 322 can be divided into four segments of equal shape and size. Based on these segments, partial image 324 can be generated. In this embodiment, two of the multiple partial images may include the same region associated with image 322. For example, these segments may define non-overlapping segments associated with image 322. Partial image 324 may be generated based on these segments by including a portion of another segment.

[0099] Image 322 may show multiple image features. Image features may be light spots. Partial image 324 may show one or more image features. Preferably, partial image 324 may show one image feature and its surrounding area. The surrounding area of ​​the image feature may be a region within a predefined distance from the center of the image feature. Therefore, the number of image features in image 322 may be equal to the number of generated partial images 324.

[0100] A portion of image 324 can be provided to the data-driven model based on its position within image 322. Figure 3 In this model, partial images 324 can be provided by flattening the generated partial images. Flattening a partial image can refer to generating a linear sequence 336 of partial images based on the position of partial image 324 within image 322. Therefore, providing partial images 324 to a data-driven model based on their positions within image 322 can include providing a sequence 336 of partial images to the data-driven model, particularly when the data-driven model can include an attention mechanism (such as a transformer architecture). For example, partial images 324 can be provided in the following sequence: starting with the top left partial image 326, followed by other top partial images 332 by increasing the distance between subsequent partial images and previous partial images 326. These images can be followed by a second top left partial image 328, and further by increasing the distance between subsequent partial images and previous partial images 328, followed by a second top partial image 330.

[0101] Figure 4 An embodiment is shown in which multiple partial images 432 are provided to a data-driven model to determine whether an object is associated with image 430.

[0102] Partial image sequences 444 can be provided to the data-driven model, specifically to partial image embeddings 412.

[0103] The data-driven model may include partial image embeddings 412, positional embeddings 414, transformer blocks 416, fully connected layers 418, softmax or sigmoid functions 420, or combinations thereof. The data-driven model may be a visual transformer. Providing multiple partial images 432 may include embedding the partial images 432 via partial image embeddings 412 and positional embeddings 414. Embedding the partial images 432 via partial image embeddings 412 can produce machine-processable representations of the partial images 432 and the relationships between them. The machine-processable representation of the partial images 432 may be a tensor, particularly a second-order tensor.

[0104] Applying positional embedding 414 can refer to adding positional factors to the machine-processable representation obtained via partial image embedding 412. Preferably, the input data can specify the relationship between partial images, particularly sequences of partial images. Positional factors It can indicate the location of a portion of the image within the image. For example, the location factor can be obtained based on the following equation. :

[0105]

[0106]

[0107] Here, pos can refer to the position of the partial image within the image, i can refer to the dimension associated with the partial image embedding 412, and d can refer to the dimension of the model (e.g., a transformer decoder, transformer encoder, or transformer encoder-decoder). This can be referred to as absolute position embedding. Alternatively, position embedding 414 can be based on Rotated Position Embedding (RoPE). Position embedding 414 is advantageous because it enables the processing of sequential data without requiring additional dimensions indicating the position of each partial image. Subsequently, position embedding 414 reduces the computational resources required to embed the input data.

[0108] A machine-processable representation of part of image 432 can be provided to transformer block 416, for example in Figure 5The encoder block 552 and / or decoder block 554 are described in the context. The transformer block 416 can be adapted to apply normalization, multi-head self-attention, residual connections, stitching, multilayer perceptrons, or combinations thereof to a machine-processable representation of a portion of image 432. This can produce a context tensor. The transformer block can transform the machine-processable representation of the portion of image 432 into a context tensor. The context tensor can be a machine-processable representation of the portion of image 432 and the relationships between the portions of image 432. The context tensor can be a tensor, particularly a second-order tensor. The context tensor can be provided to a fully connected layer 418. The fully connected layer 418 can transform the context tensor into a tensor indicating whether image 430 and / or portion of image 432 can be associated with a living organism. The tensor can then be provided to a softmax function or a sigmoid function 420 to determine whether an object associated with image 430 and / or portion of image 432 can be a living organism. This indication can be provided by a data-driven model. The indication can include one or more numerical values. These numerical values ​​can indicate a confidence score, which is associated with determining whether the object associated with image 430 is a living organism. This indication can include a vector. The vector can include multiple numerical values. To determine whether an object is a living organism, the data-driven model can classify partial images 432. Classifying partial images 432 can refer to simultaneously receiving multiple partial images 432 and providing an indication of whether an object is a living organism based on the received multiple partial images 432. Therefore, the fully connected layer 418 and the softmax or sigmoid function 420 can be adapted to classify the tensor generated by the transformer block 416, partial image embeddings 412, and location embeddings 414. For this purpose, the data-driven model can be trained based on historical partial images to embed historical partial images, transforming the historically embedded partial images into a historical context tensor. Based on the historical context tensor, the classification portion of the data-driven model, including the softmax or sigmoid function 420 and the fully connected layer 418, can classify historical partial images to determine whether the object associated with the historical partial images is a living organism.

[0109] In the example, image 430 could show a living organism under coherent infrared light illumination. Therefore, providing multiple partial images 432 to a data-driven model could lead to the classification of multiple partial images 432 as "skin." Thus, the data-driven model may have identified the object as a living organism.

[0110] Figure 5 An embodiment of transformer block 416 is shown.

[0111] A transformer block may include one or more encoder blocks 552 and / or one or more decoder blocks 554. The transformer block can receive an embedded partial image sequence. Therefore, the embedded input data can be an embedded partial image sequence. The embedded partial image sequence can be processed by applying methods such as... Figure 4 The location embedding 414 and / or partial image embedding 412 described in the context are used to obtain the information.

[0112] The embedded input data can be processed by the encoder block. The embedded input data can be provided to layer normalization 424 via residual connections. Multi-head self-attention 422 can be applied to the embedded input data. Multi-head self-attention 422 can include both multi-head and self-attention components. Self-attention can be understood as a filter applied to the embedded input data. By applying a filter to the embedded input data, elements associated with the embedded input data that contribute to the output data to be generated can be identified. Elements associated with the embedded input data can refer to multiple partial images associated with the sequence. Therefore, a filter can represent the degree of contribution of elements associated with the embedded input data to the output data to be generated. Applying a filter can be referred to as weighting the elements associated with the embedded input data. This is particularly advantageous for long element sequences. Filters can be learned and improved during training by learning to identify the contributions of elements associated with the embedded input data. Self-attention can refer to attention generated based on input data. Therefore, filters can be determined based on the input data, preferably the embedded input data. The embedded input data can be used as the query Q, key K, and value V for the self-attention operation. Self-attention can refer to attention based on the received input data. Therefore, a filter can be computed by inserting a corresponding tensor based on the embedded input data, using the following formula:

[0113]

[0114] in, The dimension corresponding to the key.

[0115] To further improve the efficiency of encoder block 552, multiple heads are used to apply filters, resulting in multi-head self-attention 522. Multi-head self-attention 522 can include applying filters to two or more portions of the embedded input data. Therefore, a tensor can be split into two or more parts, and filters can be applied to each of the two or more parts separately via two or more heads according to the following equation:

[0116] Wherein, parameter matrix Where i can refer to the number of heads. , and It can refer to the value, key, and query dimension.

[0117] The results of two or more heads can be cascaded according to the following equation:

[0118] in, And h can refer to the number of heads.

[0119] The embedded input data can be transformed into a context tensor via multi-head self-attention 522. The context tensor can represent a sequence of elements in the input data and the relationships between two or more elements. The context tensor can be a second-order tensor and / or may include one or more first-order tensors. After multi-head self-attention 522, layer normalization 524 can be applied based on the context tensor and / or the embedded input data from the residual connections. Applying layer normalization 524 can refer to normalizing the context tensor. Normalizing the context tensor reduces the values ​​of the entries in the context tensor. This reduces the computational cost associated with processing the context tensor. Furthermore, it improves training by promoting loss convergence and preventing instability.

[0120] After layer normalization 524, the context tensor can be passed to feedforward layer 526 again, followed by layer normalization 528 based on the residual connections to the context tensor and / or the output of feedforward layer 526. Feedforward layer 526 can be a feedforward neural network. The feedforward neural network can include multiple fully connected neurons. Passing the context tensor through the feedforward neural network can result in a linear transformation of the context tensor. Additionally or alternatively, the neural network can include one or more activation functions, such as rectified linear units (ReLU). Thus, the neural network can be configured to perform one or more nonlinear operations on the context tensor and / or nonlinearly transform the context tensor. After the context tensor has been transformed and / or normalized by feedforward layer 526 and layer normalization 528, the context tensor can be provided to one or more additional encoder blocks 552. Passing the context tensor through feedforward layer 526 can adapt the context tensor to the processing of the attention layers of one or more further encoder blocks 552 in order to apply a self-attention filter, preferably a multi-head self-attention filter 522. The context vector, after being transformed by layer normalization 528 and feedforward layer 526, can be referred to as the hidden state. This hidden state (also known as the output of encoder block 552) can be provided as follows: Figure 4 One or more fully connected layers 418 described in the context of the context.

[0121] Alternatively or additionally, the embedded input data can be provided to the decoder block 554.

[0122] Decoder block 554 may include layer normalization 532, masked multi-head self-attention 530, feedforward layer 534, and / or layer normalization 536. Embedded input data is processed as follows: Figure 4 The input data is obtained through the embedding layer as described in the context of the above. The embedded input data can be provided to the layer normalization 532 via residual connections. Further, masked multi-head self-attention 530 can be applied to the embedded input data. Masked multi-head self-attention 530 corresponds to multi-head self-attention 522 as described in the context of encoder block 552, wherein the portion of the embedded input data associated with elements in the sequence that are later than the element to be generated is additionally masked. Additionally or alternatively, the portion of the input data associated with elements in the sequence that are later than the element to be generated may not be received and / or transformed into the embedded input data. Therefore, the transformer decoder can be adapted to generate subsequent elements of the sequence, while the transformer encoder can be adapted to generate missing elements within a sequence and / or between two or more sequences. Thus, the transformer encoder can be configured for classification tasks. The transformer decoder can be configured for element generation.

[0123] Similar to encoder block 552, a context tensor can be generated by applying masked multi-head self-attention 530 and layer normalization 532. The context tensor can be provided to layer normalization 532 via residual connections. Further, feedforward layer 534 and layer normalization 536 can be similar to feedforward layer 526 and layer normalization 528. The context tensor can be provided to one or more additional decoder blocks 554.

[0124] Figure 6 An embodiment of generating multiple partial images 324 based on image 322 is shown.

[0125] Image 640 can be derived from, for example Figure 1A and / or Figure 1B The image 640 can be generated from the device and / or system described in the context of the image. Figure 2 It is generated as described in the context. Further, image 322 can be as follows: Figure 2 The preprocessing results are described in the context of [the preceding text]. This can be as follows: Figure 3 As described in the context, multiple partial images 636 are generated based on the preprocessed image 640.

[0126] In an embodiment, the data-driven model may include a graph neural network, and / or some images 636 may be represented as a graph. The graph data structure may be a data representation including nodes and edges. Some images 636 may be represented as nodes and / or associated with nodes, and the relationships between some images 636 may be represented as edges and / or associated with edges. The relationships between some images 636 may refer to the distance between two or more some images 636. For example, distant some images 636 may be loosely correlated, while nearby some images 636 may be closely correlated.

[0127] Figure 7 An embodiment is shown in which multiple partial images 324 are provided to a data-driven model to determine whether an object is associated with image 322.

[0128] In an embodiment, the data-driven model may include a graph neural network, and / or a portion of the image 736 may be represented as a graph data structure. Therefore, the data-driven model may receive a graph data-driven model structure at one or more embedding layers. These one or more embedding layers may be adapted to embed one or more nodes associated with the graph data structure and / or embed one or more edges associated with the graph data structure and / or embed the graph data structure. Thus, node embedding, edge embedding, and graph embedding 614 can be applied by passing the graph representation 634 of the portion of the image through one or more embedding layers. The embedding layers may be configured to perform node embedding, edge embedding, and graph embedding 614.

[0129] The data-driven model may include one or more embedding layers for applying node embedding, edge embedding, and / or graph embedding 714. Node embedding, edge embedding, and / or graph embedding 714 can transform a graph representation 734 of a portion of an image into one or more node vectors, one or more edge vectors, and one or more graph vectors. The number of node vectors may be equal to the number of nodes, and therefore equal to the number of portions of the image 736. The number of edge vectors may be equal to the number of edges, and therefore equal to the number of pairs of portions of the image 736. The number of graph vectors may be equal to the number of graphs, and therefore equal to the number of images 740. Node vectors, edge vectors, and graph vectors can be transformed separately. Therefore, the data-driven model may include one or more functions for transforming node vectors, one or more functions for transforming edge vectors, and one or more functions for transforming graph vectors. For this purpose, the data-driven model may include one or more graph neural network layers 716. The graph neural network layers may include one or more functions for transforming node vectors, one or more functions for transforming edge vectors, and one or more functions for transforming graph vectors.

[0130] A graph neural network layer can receive one or more node vectors 744 and transform them into one or more updated node vectors 750 by applying one or more node functions 756. The one or more node functions 756 can be determined by one or more parameters. Training the graph neural network may include updating one or more parameters associated with the one or more node functions 756.

[0131] Furthermore, the graph neural network layer can receive one or more edge vectors 746 and transform them into one or more updated edge vectors 752 by applying one or more edge functions 758. The one or more edge functions 758 can be determined by one or more parameters. Training the graph neural network may include updating one or more parameters associated with the one or more edge functions 758.

[0132] Furthermore, the graph neural network layer can receive one or more graph vectors 748 and transform them into one or more updated graph vectors 754 by applying one or more graph functions 760. The one or more graph functions 760 can be determined by one or more parameters. Training the graph neural network may include updating one or more parameters associated with the one or more graph functions 760.

[0133] By transforming the graph representation 734 of a portion of the image, the graph parameters of specified nodes, edges, and the entire graph can be updated. Updating the graph parameters produces a second graph. This second graph may differ from the graph representation 734 of the portion of the image provided to the data-driven model.

[0134] The second graph can be classified based on whether image 740 represents a living organism. Whether it is a living organism or a deceptive object can be an attribute of image 740, and therefore also an attribute of the graph. Therefore, the second graph can be provided to one or more classification layers to classify the graph. Image 740 can be associated with objects that are living organisms. Thus, a data-driven model can provide indications that an object can be a living organism. This is in Figure 7 The middle can be regarded as the "skin".

[0135] Figure 8 Examples of multiple partial images are shown.

[0136] This disclosure has also been described in conjunction with various preferred embodiments and examples. However, by studying the accompanying drawings, this disclosure, and the claims, those skilled in the art, as well as those practicing the claimed subject matter, will understand and implement other variations. It is particularly noteworthy that any steps presented can be performed in any order; that is, this disclosure is not limited to a specific order of these steps. Furthermore, it is not required that different steps be performed at a specific location or node in a distributed system; that is, each step can be performed on different nodes using different devices / data processing.

[0137] As used herein, "determine" also includes "initiating or causing determination," "generate" also includes "initiating and / or causing generation," and "provide" also includes "initiating or causing determination, generation, selection, sending, and / or receiving." "Initiating or causing an action" includes any processing signal that triggers a computing node or device to perform a corresponding action.

[0138] In the claims and specification, the word "comprising" or "including" or similar wording does not exclude other elements or steps and should not be construed as limiting oneself to the listed elements or steps. The indefinite article "a" or "an" does not exclude multiple. A single element or other unit may perform the function of several entities or items recited in the claims. The fact that certain measures are recited only in mutually different dependent claims does not indicate that a combination of these measures cannot be used in advantageous implementations or that additional elements may be included.

[0139] Within the scope of this disclosure, provision may include any interface configured to provide data. This may include application programming interfaces, human-machine interfaces (such as displays), and / or software module interfaces. Provision may include transmitting or submitting data to the interface, particularly displaying data to a user or having data used by a receiving entity.

[0140] Any disclosures and embodiments described herein relate to the methods, systems, devices, and computer program elements listed above, and vice versa. Advantageously, the benefits provided by any embodiment and example also apply to all other embodiments and examples, and vice versa.

Claims

1. A method for measuring the material properties of an object, the method comprising: Receive an image of the object under coherent infrared illumination. Multiple partial images are generated based on this image. Based on the positions of these multiple partial images within the image, they are provided to a data-driven model to determine material property measures. This data-driven model is trained based on historical partial images and corresponding historical material property measures, and these historical partial images are provided to the data-driven model based on their positions. Provide a measure of the material's properties.

2. A method for allowing an object to access a resource, the method comprising: Receive requests to access resources. In response to receiving the request to access the resource, trigger the illumination of the object with coherent infrared light, and The image of the object is generated when the object is illuminated by the coherent infrared light. Multiple partial images are generated based on this image. Based on the positions of multiple partial images within an image, these partial images are provided to a data-driven model to determine whether an object is a living organism. The data-driven model is trained on multiple historical partial images to determine whether one or more objects associated with these historical partial images are living organisms. These historical partial images are provided based on their positions. Access to the resource is permitted based on the determination that the object is a living organism.

3. The method of claim 2, further comprising receiving a floodlight image generated when the object is illuminated by infrared light and a template image generated when an authorized user is illuminated by infrared light, and providing the floodlight image and the template image to an authentication data-driven model to determine whether the object corresponds to the authorized user, wherein, The authentication data-driven model is trained on multiple historical floodlight images and multiple historical template images to determine whether one or more objects associated with these historical floodlight images correspond to one or more authorized users associated with these historical template images.

4. The method according to any one of claims 1 to 3, wherein, Providing the data-driven model with the partial images based on their positions includes providing the data-driven model with a sequence of the partial images based on their positions within the image, wherein the data-driven model is parameterized to receive the sequence of the partial images as input.

5. The method of claim 4, wherein, The sequence of multiple partial images includes multiple partial images arranged in order of their positions within the image.

6. The method according to any one of claims 1 to 5, wherein, Providing the partial images to the data-driven model based on their positions includes providing the data-driven model with data structures associated with the partial images and their positions within the image, wherein the data-driven model is parameterized to be provided with data structures associated with the partial images and their positions within the image.

7. The method of claim 6, wherein, The data structure is a graph data structure associated with the plurality of partial images and the positions of the plurality of partial images within the image, wherein the graph data structure includes a plurality of nodes and a plurality of edges, wherein a node of the graph data structure is associated with one of the partial images, and wherein an edge of the graph data structure is associated with a sequence of two or more partial images and / or the distance between them, and the data-driven model is parameterized to provide the data structure means that the data-driven model is parameterized to provide these graph data structures.

8. The method of claim 7, wherein, The graph data structure is a directed graph data structure, wherein the direction associated with the directed graph data structure indicates the position of the plurality of partial images within the image.

9. The method as claimed in claims 4 to 5 or any one of claims 7 to 8, wherein, These partial images relate to a grid of the image, and the sequence is defined by a predefined sequence path that connects the partial images in the grid.

10. The method according to any one of claims 1 to 8, wherein, Generating the multiple partial images based on the image includes dividing the image into multiple segments and / or cropping the image based on the multiple segments to generate the multiple partial images.

11. The method according to any one of claims 1 to 10, wherein, The coherent infrared light is patterned coherent infrared light, wherein the patterned coherent infrared light comprises fewer than 4,000 beams, and / or wherein projecting the patterned coherent infrared light onto an object causes fewer than 4,000 light spots to be projected onto the object.

12. The method of claim 11, wherein, One or more of the plurality of partial images and / or segments include at least a portion of a light spot generated by illuminating the object with the patterned coherent infrared light.

13. Use of a plurality of partial images generated based on an image generated when an object is irradiated with coherent infrared light in the method according to any one of claims 1 to 12.

14. Use of a data-driven model configured to receive a plurality of partial images based on their positions within an image in the method according to any one of claims 1 to 12.

15. A device and / or system for allowing objects to access resources, the device and / or system comprising: processor; as well as A memory storing instructions that, when executed by the processor, configure the device and / or system to perform the method as described in any one of claims 1 to 12.

Images