Information processing device, information processing method, and recording medium
Patent Information
- Application Number
- JP2024511141
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2024-12-04
- Estimated Expiration
- 2042-03-31
AI Technical Summary
Current authentication systems are inadequate in evaluating and mitigating risks from adversarial samples that can mislead face recognition and authentication processes, as existing methods focus on class classification rather than similarity-based authentication processing.
An information processing device and method that calculates similarity and gradient information between feature amounts of authentic and adversarial information, determines perturbation positions, applies perturbations to generate adversarial samples, and evaluates authentication risks based on these samples.
Effectively generates adversarial samples suitable for authentication processing, enabling the appropriate evaluation of risks in authentication systems and preventing incorrect authentication results.
Abstract
Description
Information processing device, information processing method, and recording medium
[0001] The present disclosure relates to the technical fields of an information processing device, an information processing method, and a recording medium.
[0002] As a system of this type, there is known a system that evaluates an authentication device that performs authentication processing. For example, Patent Literature 1 discloses a system that calculates a quantitative evaluation value of robustness against adversarial samples for an authentication model that performs face recognition or the like.
[0003] As another related technique, for example, Patent Literature 2 discloses that by using a feature vector of a face image to obtain candidate adversarial samples that are likely to cause errors in face recognition by a face recognition device, candidate adversarial samples can be obtained. Patent Literature 3 discloses that an attacker generates adversarial inputs so as to be misidentified in face recognition.
[0004] International Publication No. 2021 / 038788 International Publication No. 2021 / 144857 Special Publication No. 2021-501414
[0005] This disclosure aims to improve upon the techniques disclosed in the prior art documents.
[0006] One aspect of the information processing device disclosed herein comprises a similarity calculation means for calculating a similarity between a feature of first information and a feature of second information, a gradient information calculation means for calculating gradient information indicating a gradient of the similarity, a perturbation position determination means for determining an element in the first information to be perturbed based on the gradient information, a perturbation addition means for adding a perturbation to the element in the first information to be perturbed, and a risk assessment means for assessing a risk in the authentication process based on a result of an authentication process that compares the first information to which the perturbation has been added with the second information.
[0007] One aspect of the information processing method disclosed herein includes using at least one computer to calculate a similarity between a feature of first information and a feature of second information, calculate gradient information indicating a gradient of the similarity, determine an element in the first information to be perturbed based on the gradient information, perturb the element in the first information to be perturbed, and evaluate a risk in the authentication process based on a result of an authentication process that compares the first information to which the perturbation has been applied with the second information.
[0008] One aspect of the recording medium of this disclosure is a recording medium having recorded thereon a computer program that causes at least one computer to execute an information processing method, which calculates a similarity between a feature of first information and a feature of second information, calculates gradient information indicating a gradient of the similarity, determines an element in the first information to be subjected to a perturbation based on the gradient information, applies a perturbation to the element in the first information to be subjected to the perturbation, and evaluates a risk in the authentication process based on a result of an authentication process that compares the first information to which the perturbation has been applied with the second information.
[0009] 1 is a block diagram showing a hardware configuration of an information processing device according to a first embodiment. FIG. 2 is a block diagram showing a functional configuration of an information processing device according to the first embodiment. FIG. 3 is a flowchart showing a flow of operation of an information processing device according to the first embodiment. FIG. 4 is a block diagram showing a functional configuration of an information processing device according to a second embodiment. FIG. 5 is a flowchart showing a flow of operation of an information processing device according to the second embodiment. FIG. 6 is a flowchart showing a flow of operation of determining a perturbation-adding position by an information processing device according to a third embodiment. FIG. 7 is a flowchart showing a flow of operation of determining a perturbation-adding position by an information processing device according to a fourth embodiment. FIG. 8 is a flowchart showing a flow of operation of determining a perturbation-adding position by an information processing device according to a fifth embodiment. FIG. 9 is a flowchart showing a flow of operation of risk assessment by an information processing device according to a sixth embodiment. FIG. 10 is a block diagram showing a functional configuration of an information processing device according to a seventh embodiment. A conceptual diagram showing an example of an attack on a face recognition gate at an airport. FIG. 11 is a flowchart showing a flow of operation of an information processing device according to the seventh embodiment. FIG. 12 is a flowchart showing a flow of operation of determining a perturbation-adding position by an information processing device according to an eighth embodiment.
[0010] Hereinafter, embodiments of an information processing device, an information processing method, and a recording medium will be described with reference to the drawings.
[0011] First Embodiment An information processing apparatus according to a first embodiment will be described with reference to FIGS. 1 to 3. FIG.
[0012] (Hardware Configuration) First, the hardware configuration of the information processing apparatus according to the first embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the hardware configuration of the information processing apparatus according to the first embodiment.
[0013] 1 , an information processing device 10 according to the first embodiment includes a processor 11, a RAM (Random Access Memory) 12, a ROM (Read Only Memory) 13, and a storage device 14. The information processing device 10 may further include an input device 15 and an output device 16. The processor 11, RAM 12, ROM 13, storage device 14, input device 15, and output device 16 are connected via a data bus 17.
[0014] The processor 11 loads a computer program. For example, the processor 11 is configured to load a computer program stored in at least one of the RAM 12, the ROM 13, and the storage device 14. Alternatively, the processor 11 may load a computer program stored in a computer-readable storage medium using a storage medium reading device (not shown). The processor 11 may acquire (i.e., load) the computer program from a device (not shown) located outside the information processing device 10 via a network interface. The processor 11 controls the RAM 12, the storage device 14, the input device 15, and the output device 16 by executing the loaded computer program. In particular, in this embodiment, when the processor 11 executes the loaded computer program, a functional block for evaluating the risk of the authentication process is realized within the processor 11. In other words, the processor 11 may function as a controller that executes each control in the information processing device 10.
[0015] The processor 11 may be configured as, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field-Programmable Gate Array), a DSP (Demand-Side Platform), or an ASIC (Application Specific Integrated Circuit). The processor 11 may be configured as one of these, or may be configured to use multiple processors in parallel.
[0016] The RAM 12 temporarily stores computer programs executed by the processor 11. The RAM 12 temporarily stores data that the processor 11 temporarily uses while it is executing the computer programs. The RAM 12 may be, for example, a dynamic random access memory (D-RAM) or a static random access memory (SRAM). Alternatively, other types of volatile memory may be used instead of the RAM 12.
[0017] The ROM 13 stores computer programs executed by the processor 11. The ROM 13 may also store fixed data. The ROM 13 may be, for example, a PROM (Programmable Read Only Memory) or an EPROM (Erasable Read Only Memory). Alternatively, other types of non-volatile memory may be used instead of the ROM 13.
[0018] The storage device 14 stores data that is to be saved long-term by the information processing device 10. The storage device 14 may operate as a temporary storage device for the processor 11. The storage device 14 may include, for example, at least one of a hard disk device, a magneto-optical disk device, an SSD (Solid State Drive), and a disk array device.
[0019] The input device 15 is a device that receives input instructions from a user of the information processing device 10. The input device 15 may include, for example, at least one of a keyboard, a mouse, and a touch panel. The input device 15 may be configured as a mobile terminal such as a smartphone or a tablet. The input device 15 may also be, for example, a device that includes a microphone and is capable of voice input.
[0020] The output device 16 is a device that outputs information related to the information processing device 10 to the outside. For example, the output device 16 may be a display device (e.g., a display) that can display information related to the information processing device 10. The output device 16 may also be a speaker or the like that can output information related to the information processing device 10 as audio. The output device 16 may be configured as a mobile terminal such as a smartphone or a tablet. The output device 16 may also be a device that outputs information in a format other than an image. For example, the output device 16 may be a speaker that outputs information related to the information processing device 10 as audio.
[0021] 1 illustrates an example of an information processing device 10 including multiple devices, but all or some of the functions may be realized by a single device. In this case, the information processing device may be configured to include only the processor 11, RAM 12, and ROM 13 described above, and the other components (i.e., the storage device 14, input device 15, and output device 16) may be provided by an external device connected to the information processing device. Furthermore, some of the calculation functions of the information processing device may be realized by an external device (e.g., an external server, a cloud, etc.).
[0022] (Functional Configuration) Next, the functional configuration of the information processing device 10 according to the first embodiment will be described with reference to Fig. 2. Fig. 2 is a block diagram showing the functional configuration of the information processing device according to the first embodiment.
[0023] 2, the information processing apparatus 10 according to the first embodiment includes, as components for realizing its functions, a similarity calculation unit 110, a gradient information calculation unit 120, a perturbation application position determination unit 130, a perturbation application unit 140, and a risk assessment unit 150. Each of the similarity calculation unit 110, the gradient information calculation unit 120, the perturbation application position determination unit 130, the perturbation application unit 140, and the risk assessment unit 150 may be a processing block realized by, for example, the above-described processor 11 (see FIG. 1). Furthermore, each of the similarity calculation unit 110, the gradient information calculation unit 120, the perturbation application position determination unit 130, the perturbation application unit 140, and the risk assessment unit 150 may be configured to include a neural network.
[0024] The similarity calculation unit 110 is configured to receive input of a feature amount of the first information (hereinafter referred to as "first information feature amount" as appropriate) and a feature amount of the second information (hereinafter referred to as "second information feature amount" as appropriate). The similarity calculation unit 110 is then configured to calculate the similarity between the input first information feature amount and second information feature amount. Note that the method for calculating the similarity is not particularly limited, and existing technology may be appropriately adopted. The similarity may be a matching score obtained by matching the first information feature amount and the second information feature amount. Specific examples of the first information and the second information will be described in detail in other embodiments described later.
[0025] The gradient information calculation unit 120 is configured to be able to calculate gradient information indicating the gradient of the similarity calculated by the similarity calculation unit 110. Note that the method for calculating the gradient information is not particularly limited, and existing techniques may be appropriately adopted. The gradient information may be information including the Jacobian of the similarity. For example, a ) and the second information feature f(X t ) is expressed as L{f(X a ), f(X t )}, the gradient information ∇L(X a , X t ) may be calculated as in the following formula (1):
[0026] Here, M is the number of dimensions of X.
[0027] The perturbation-imparting position determining unit 130 is configured to be able to determine an element in the first information to be perturbed based on the gradient information calculated by the gradient information calculating unit 120. The perturbation-imparting position determining unit 130 determines, for example, at least one element to which a perturbation is to be imparted from among a plurality of elements included in the first information. Note that the "perturbation" here refers to noise imparted to increase the similarity between the first information and the second information. For example, the gradient information shown in the above-mentioned formula (1) being positive means that the element X a This means that the similarity between the first information and the second information is increased by applying a perturbation to the first information. A more specific method for determining the element to be perturbed will be described in detail in another embodiment described later.
[0028] The perturbation applying unit 140 is configured to apply a perturbation to the element determined by the perturbation applying position determining unit 130. That is, the perturbation applying unit 140 is configured to apply a perturbation to some elements of the first information, thereby generating perturbed first information (hereinafter referred to as an "adversarial sample" as appropriate). When the adversarial sample generated in this manner is used, for example, the similarity between the first information and the second information is higher than when no perturbation is applied. That is, the adversarial sample generated by the perturbation applying unit 140 is information that is likely to be erroneously recognized as the second information in the authentication process.
[0029] The risk assessment unit 150 is configured to be able to assess risks in the authentication process (in other words, risks potential in the authentication device or authentication model that executes the authentication process). More specifically, the risk assessment unit 150 assesses risks in the authentication process based on the results of the authentication process using the adversarial sample generated by the perturbation addition unit 140. For example, the risk assessment unit 150 may assess the possibility that the generated adversarial sample (i.e., the first information to which the perturbation has been added) will be recognized as the second information. A specific assessment method by the risk assessment unit 150 will be described in detail in another embodiment described later.
[0030] The risk assessment unit 150 may be provided separately from the device that generates the adversarial samples. For example, the adversarial sample generation device including the similarity calculation unit 110, the gradient information calculation unit 120, the perturbation position determination unit 130, and the perturbation addition unit 140 and the risk assessment device including the risk assessment unit 150 may be configured as separate devices.
[0031] The authentication process may be performed by an authentication device provided separately from the information processing device 10 according to this embodiment. In this case, the adversarial sample generated by the perturbation adding unit 140 may be output to the authentication device, and the risk assessment unit 150 may assess the risk using the authentication result input from the authentication device. Alternatively, the risk assessment unit 150 may have a function for executing the authentication process. In other words, the risk assessment unit 150 may be configured to itself execute the authentication process and assess the risk of the authentication process based on the authentication result.
[0032] (Operation Flow) Next, the overall operation flow of the information processing device 10 according to the first embodiment will be described with reference to Fig. 3. Fig. 3 is a flowchart showing the operation flow of the information processing device according to the first embodiment.
[0033] 3 , when the operation of the information processing device 10 according to the first embodiment starts, the similarity calculation unit 110 first acquires a first information feature amount and a second information feature amount (step S101). Then, the similarity calculation unit 110 calculates the similarity between the acquired first information feature amount and the second information feature amount (step S102). Information related to the similarity calculated by the similarity calculation unit 110 is output to the gradient information calculation unit 120.
[0034] Next, the gradient information calculation unit 120 calculates gradient information indicating the gradient of the similarity calculated by the similarity calculation unit 110 (step S103). The gradient information calculated by the gradient information calculation unit 120 is output to the perturbation application position determination unit 130.
[0035] Next, the perturbation applying position determining unit 130 determines an element to which a perturbation is to be applied in the first information based on the gradient information calculated by the gradient information calculating unit 120 (step S104). Information on the element determined by the perturbation applying position determining unit 130 is output to the perturbation applying unit 140.
[0036] Next, the perturbation applying unit 140 applies a perturbation to the element determined by the perturbation applying position determining unit 130 (step S105). That is, the perturbation is applied to the first information to generate an adversarial sample. The adversarial sample generated by the perturbation applying unit 140 is used in the authentication process.
[0037] Next, the risk assessment unit 150 assesses the risk in the authentication process based on the authentication result of the authentication process using the adversarial sample generated by the perturbation application unit 140 (step S106). The risk assessment unit 150 may output the risk assessment result.
[0038] (Technical Effects) Next, technical effects obtained by the information processing device 10 according to the first embodiment will be described.
[0039] As described with reference to FIGS. 1 to 3 , in the information processing device 10 according to the first embodiment, adversarial samples are generated by applying perturbations based on the similarity between two pieces of information. Then, the risk of the authentication process is evaluated based on the results of the authentication process using the generated adversarial samples. In this manner, the risk of the authentication process against a hostile input can be appropriately evaluated. Specifically, it becomes possible to evaluate the risk of the authentication process against attacks that are intended to intentionally obtain erroneous authentication results.
[0040] Note that JSMA (Jacobian-based Saliency Map Attack) is known as a method for generating adversarial samples, but this method is intended for classification processing (i.e., processing that obtains a classification probability vector as a processing result), and therefore cannot be directly applied when generating adversarial samples for authentication processing (i.e., processing that obtains a similarity as a processing result). However, the information processing device according to this embodiment generates adversarial samples suitable for authentication processing, and therefore can appropriately evaluate the risks in the authentication processing.
[0041] Second Embodiment An information processing device 10 according to a second embodiment will be described with reference to Figures 4 and 5. Note that the second embodiment differs from the first embodiment described above only in some configurations and operations, and other parts may be the same as the first embodiment. Therefore, the following will describe in detail the parts that differ from the first embodiment already described, and will omit explanations of other overlapping parts as appropriate.
[0042] (Functional Configuration) First, the functional configuration of the information processing device 10 according to the second embodiment will be described with reference to Fig. 4. Fig. 4 is a block diagram showing the functional configuration of the information processing device according to the second embodiment. Note that in Fig. 4, the same elements as those shown in Fig. 2 are denoted by the same reference numerals.
[0043] As shown in Fig. 4, the information processing device 10 according to the second embodiment is configured to include, as components for realizing its functions, a similarity calculation unit 110, a gradient information calculation unit 120, a perturbation application position determination unit 130, a perturbation application unit 140, a risk assessment unit 150, and a feature extraction unit 160. That is, the information processing device 10 according to the second embodiment further includes the feature extraction unit 160 in addition to the configuration of the first embodiment (see Fig. 2). The feature extraction unit 160 may be a processing block realized by, for example, the above-described processor 11 (see Fig. 1).
[0044] The feature extraction unit 160 is configured to receive a first image, which is a specific example of first information, and a second image, which is a specific example of second information. The first image is an image including a first living organism, and the second image is an image including a second living organism. The first image and the second image may be, for example, a facial image including the face of a living organism or an iris image including the iris. The feature extraction unit 160 is configured to extract features from the first image and the second image. That is, the feature extraction unit 160 is configured to extract features related to the first living organism included in the first image and features related to the second living organism included in the second image. The features extracted by the feature extraction unit 160 are input to the similarity calculation unit 110 as first information features and second information features.
[0045] (Operation Flow) Next, the overall operation flow of the information processing device 10 according to the second embodiment will be described with reference to Fig. 5. Fig. 5 is a flowchart showing the operation flow of the information processing device according to the second embodiment. Note that in Fig. 5, the same processes as those shown in Fig. 3 are denoted by the same reference numerals.
[0046] 5 , when the operation of the information processing device 10 according to the second embodiment starts, the feature extraction unit 160 first acquires a first image and a second image (step S201). Then, the feature extraction unit 160 extracts a first information feature from the first image and a second information feature from the second image (step S202). Information related to the feature extracted by the feature extraction unit 160 is output to the similarity calculation unit 110.
[0047] Next, the similarity calculation unit 110 calculates the similarity between the first information feature amount and the second information feature amount extracted by the feature amount extraction unit 160 (step S102). Information related to the similarity calculated by the similarity calculation unit 110 is output to the gradient information calculation unit 120.
[0048] Next, the gradient information calculation unit 120 calculates gradient information indicating the gradient of the similarity calculated by the similarity calculation unit 110 (step S103). The gradient information calculated by the gradient information calculation unit 120 is output to the perturbation application position determination unit 130.
[0049] Next, the perturbation-imparting-position determining unit 130 determines an element in the first information to which a perturbation is to be added, based on the gradient information calculated by the gradient information calculating unit 120 (step S104). Note that the element here may be a pixel in the first image. The perturbation-imparting-position determining unit 130 may perform a process of determining the position of a pixel to which a perturbation is to be added, among a plurality of pixels included in the first image, for example. Information regarding the element determined by the perturbation-imparting-position determining unit 130 is output to the perturbation adding unit 140.
[0050] Next, the perturbation applying unit 140 applies a perturbation to the element determined by the perturbation applying position determining unit 130 (step S105). That is, the perturbation applying unit 140 applies a perturbation to the pixel of the first image determined by the perturbation applying position determining unit 130, thereby generating an adversarial sample. The adversarial sample generated by the perturbation applying unit 140 is used in the authentication process.
[0051] Next, the risk assessment unit 150 assesses the risk in the authentication process based on the authentication result of the authentication process using the adversarial sample generated by the perturbation application unit 140 (step S106). The risk assessment unit 150 may output the risk assessment result.
[0052] (Technical Effects) Next, technical effects obtained by the information processing device 10 according to the second embodiment will be described.
[0053] 4 and 5 , in the information processing device 10 according to the second embodiment, feature amounts are extracted from each of the first image and the second image, and perturbations are applied based on the similarity between the first image and the second image to generate adversarial samples. This allows appropriate assessment of the risk of authentication processing using images. For example, it is possible to appropriately assess the risk of adversarial input for face authentication using face images or iris authentication using iris images.
[0054] Third Embodiment An information processing device 10 according to a third embodiment will be described with reference to Fig. 6. Note that the third embodiment describes a specific example of the operation when determining the perturbation application position in the first and second embodiments described above (i.e., the operation corresponding to step S104 in Fig. 3), and other parts may be the same as those in the first and second embodiments. Therefore, the following will describe in detail parts that differ from the embodiments already described, and will omit explanations of other overlapping parts as appropriate.
[0055] (Perturbation Addition Position Determining Operation) First, the flow of the perturbation addition position determining operation (i.e., the operation when determining the element to add perturbation) by the information processing device 10 according to the third embodiment will be described with reference to Fig. 6. Fig. 6 is a flowchart showing the flow of the perturbation addition position determining operation by the information processing device according to the third embodiment.
[0056] 6 , in the perturbation-imparting-position determining operation by the information processing device 10 according to the third embodiment, the perturbation-imparting-position determining unit 130 first acquires gradient information (i.e., gradient information of the similarity between the first information feature amount and the second information feature amount) calculated by the gradient information calculating unit 120 (step S301). Then, the perturbation-imparting-position determining unit 130 searches for one element with the maximum gradient information based on the gradient information calculated by the gradient information calculating unit 120 (step S302).
[0057] Next, the perturbation position determination unit 130 determines one element with the maximum gradient information obtained as a result of the search as the element to be perturbed (i.e., the perturbation position) (step S303). Note that if there are multiple elements with the maximum gradient information, the perturbation position determination unit 130 may select one element from the multiple elements and determine all of the elements to be perturbed as the elements to be perturbed. Thereafter, the perturbation position determination unit 130 outputs information about the elements to be perturbed to the perturbation unit 140 (step S304).
[0058] (Technical Effects) Next, technical effects obtained by the information processing device 10 according to the third embodiment will be described.
[0059] As described with reference to FIG. 6 , the information processing device 10 according to the third embodiment determines one element with the maximum gradient information as the perturbation target. In this way, the perturbation position can be easily and appropriately determined based on the gradient information. Therefore, it is possible to appropriately generate adversarial samples and evaluate the risk of the authentication process.
[0060] Fourth Embodiment An information processing device 10 according to a fourth embodiment will be described with reference to Fig. 7 . Note that, like the third embodiment described above, the fourth embodiment describes a specific example of a perturbation-imparting position determination operation, and other parts may be the same as those of the first and second embodiments. Therefore, the following will describe in detail parts that differ from the embodiments already described, and will omit explanations of other overlapping parts as appropriate.
[0061] (Perturbation Applying Position Determining Operation) First, the flow of the perturbation applying position determining operation by the information processing device 10 according to the fourth embodiment will be described with reference to Fig. 7. Fig. 7 is a flowchart showing the flow of the perturbation applying position determining operation by the information processing device according to the fourth embodiment.
[0062] 7 , in the perturbation-imparting-position determining operation by the information processing device 10 according to the fourth embodiment, the perturbation-imparting-position determining unit 130 first acquires gradient information (i.e., gradient information of the similarity between the first information feature amount and the second information feature amount) calculated by the gradient information calculating unit 120 (step S401). Then, the perturbation-imparting-position determining unit 130 sorts the elements in descending order of the gradient information calculated by the gradient information calculating unit 120 (step S402).
[0063] Next, the perturbation-imparting-position determining unit 130 determines a predetermined number of elements in descending order of gradient information as elements to which a perturbation is to be applied (i.e., perturbation-imparting positions) (step S403). Note that the "predetermined number" here refers to the number of elements selected as perturbation-imparting positions, and may be a value that can be arbitrarily set by, for example, a user. For example, if the predetermined number is set to "3," the perturbation-imparting-position determining unit 130 may determine the element with the highest, second highest, and third highest gradient information as perturbation-imparting positions. Thereafter, the perturbation-imparting-position determining unit 130 outputs information regarding the elements to which a perturbation is to be applied to the perturbation applying unit 140 (step S404).
[0064] (Technical Effects) Next, technical effects obtained by the information processing device 10 according to the fourth embodiment will be described.
[0065] As described in FIG. 7 , the information processing device 10 according to the fourth embodiment determines a predetermined number of elements as perturbation targets in descending order of gradient information. In this way, it is possible to easily and appropriately determine perturbation positions based on the gradient information. Therefore, it is possible to appropriately generate adversarial samples and evaluate the risk of authentication processing.
[0066] Fifth Embodiment An information processing device 10 according to a fifth embodiment will be described with reference to Fig. 8 . Note that the fifth embodiment, like the third and fourth embodiments described above, describes a specific example of a perturbation-imparting position determination operation, and other parts may be the same as those of the first and second embodiments. Therefore, the following will describe in detail parts that differ from the embodiments already described, and will omit explanations of other overlapping parts as appropriate.
[0067] (Perturbation Addition Position Determining Operation) First, the flow of the perturbation addition position determining operation by the information processing device 10 according to the fifth embodiment will be described with reference to Fig. 8. Fig. 8 is a flowchart showing the flow of the perturbation addition position determining operation by the information processing device according to the fifth embodiment.
[0068] 8 , in the perturbation-imparting position determination operation by the information processing device 10 according to the fifth embodiment, the perturbation-imparting position determination unit 130 first acquires gradient information calculated by the gradient information calculation unit 120 (i.e., gradient information of the similarity between the first information feature amount and the second information feature amount) (step S501). Then, the perturbation-imparting position determination unit 130 compares the gradient information calculated by the gradient information calculation unit 120 with a predetermined threshold (step S502). Note that the "predetermined threshold" here is a threshold that is set in advance to determine the perturbation-imparting position.
[0069] Next, the perturbation position determination unit 130 determines elements whose gradient information is higher than a predetermined threshold as elements to be perturbed (i.e., perturbation positions) (step S503). Therefore, for example, if a relatively small number of elements are to be determined as perturbation positions, the predetermined threshold can be set to a higher value. Conversely, if a relatively large number of elements are to be determined as perturbation positions, the predetermined threshold can be set to a lower value. Thereafter, the perturbation position determination unit 130 outputs information about the elements to be perturbed to the perturbation unit 140 (step S404).
[0070] In the process of step S503, if there is no gradient information lower than the predetermined threshold (i.e., if all gradient information is lower than the predetermined threshold), the predetermined threshold may be reset to a lower value, and the processes of steps S502 and S503 may be performed again. Alternatively, if there is no gradient information lower than the predetermined threshold, the perturbation application position may be determined by the method already described in the third and fourth embodiments.
[0071] (Technical Effects) Next, technical effects obtained by the information processing device 10 according to the fifth embodiment will be described.
[0072] As described with reference to FIG. 8 , the information processing device 10 according to the fifth embodiment determines an element whose gradient information is greater than a predetermined threshold as a perturbation target. In this way, the perturbation position can be easily and appropriately determined based on the gradient information. Therefore, it is possible to appropriately generate adversarial samples and evaluate the risk of the authentication process.
[0073] Sixth Embodiment An information processing device 10 according to a sixth embodiment will be described with reference to Fig. 9. The sixth embodiment differs from the first to fifth embodiments described above only in some structural operations, and other parts may be the same as the first to fifth embodiments. Therefore, the following will describe in detail the parts that differ from the embodiments already described, and will omit a description of other overlapping parts as appropriate.
[0074] (Risk Assessment Operation) First, the flow of the risk assessment operation (i.e., the operation when assessing risk using generated adversarial samples) by the information processing device 10 according to the sixth embodiment will be described with reference to Fig. 9. Fig. 9 is a flowchart showing the flow of the risk assessment operation by the information processing device according to the sixth embodiment.
[0075] 9 , in the risk assessment operation by the information processing device 10 according to the sixth embodiment, the risk assessment unit 150 first acquires the result of the authentication process using the adversarial sample (i.e., the first information to which perturbation has been added) (step S601). Then, the risk assessment unit 150 calculates the probability of an erroneous authentication based on the acquired authentication result (step S602). The erroneous authentication probability is the probability that an erroneous authentication result will be obtained, and can be calculated, for example, by dividing the number of times that erroneous authentications have occurred by the total number of times that authentications have been performed.
[0076] Next, the risk assessment unit 150 assesses the risk of the authentication process based on the calculated false authentication probability (step S603). For example, the risk assessment unit 150 may determine that the higher the false authentication probability, the higher the risk. Thereafter, the risk assessment unit 150 outputs the assessment result (step S604). The risk assessment unit 150 may output the false authentication probability together with the assessment result.
[0077] (Technical Effects) Next, technical effects obtained by the information processing device 10 according to the sixth embodiment will be described.
[0078] 9, in the information processing device 10 according to the sixth embodiment, the risk is evaluated based on the erroneous authentication probability calculated from the authentication result. In this way, it is possible to appropriately evaluate the risk that an erroneous authentication result will be output in response to a hostile input.
[0079] Seventh Embodiment An information processing device 10 according to a seventh embodiment will be described with reference to Figures 10 to 12. The seventh embodiment differs only in part of the configuration and operation from the first to sixth embodiments described above, and other parts may be the same as the first to sixth embodiments. Therefore, the following will describe in detail the parts that differ from the embodiments already described, and will omit a description of other overlapping parts as appropriate.
[0080] (Functional Configuration) First, the functional configuration of the information processing device 10 according to the seventh embodiment will be described with reference to Fig. 10. Fig. 10 is a block diagram showing the functional configuration of the information processing device according to the seventh embodiment. Note that in Fig. 10, the same elements as those shown in Fig. 2 are denoted by the same reference numerals.
[0081] 10 , the information processing device 10 according to the seventh embodiment is configured to include, as components for realizing its functions, a similarity calculation unit 110, a gradient information calculation unit 120, a perturbation application position determination unit 130, a perturbation application unit 140, and a risk assessment unit 150. In particular, the similarity calculation unit 110 according to the seventh embodiment is configured to receive a third information feature amount in addition to the first information feature amount and the second information feature amount.
[0082] The similarity calculation unit 110 according to the seventh embodiment is configured to receive, as input, a feature amount of third information (hereinafter referred to as "third information feature amount") in addition to the feature amount of first information and the feature amount of second information. The similarity calculation unit 110 is configured to be able to calculate the similarity between the first information feature amount and the third information feature amount (hereinafter referred to as "second similarity") in addition to the similarity between the first information feature amount and the second information feature amount already described (hereinafter referred to as "first similarity").
[0083] The gradient information calculation unit 120 according to the seventh embodiment is configured to be able to calculate gradient information of the second similarity (hereinafter referred to as "second gradient information") in addition to the gradient information of the first similarity (hereinafter referred to as "first gradient information") already described. For example, a ) and the third information feature f(X s ) is expressed as L{f(X a ), f(X s )}, the gradient information ∇L(X a , X s ) may be calculated as in the following formula (2):
[0084]
[0085] The perturbation-imparting position determiner 130 according to the seventh embodiment is configured to be able to determine an element to be perturbed in the first information, based on the first gradient information and the second gradient information calculated by the gradient information calculator 120. That is, the perturbation-imparting position determiner 130 determines an element to be perturbed, based on two pieces of gradient information. Note that a method for determining an element to be perturbed using two pieces of gradient information will be described in detail in another embodiment described later.
[0086] Note that the information processing device 10 according to the seventh embodiment may include the feature extraction unit 160 described in the second embodiment in addition to the configuration described above. In this case, the feature extraction unit 160 may receive a third image (i.e., an image including a third living organism) in addition to the first and second images. The feature extraction unit 160 may be configured to extract a first information feature from the first image, a second information feature from the second image, and a third information feature from the third image.
[0087] (Example of Possible Attack) Next, an example of an attack that is possible on the information processing device 10 according to the seventh embodiment will be described with reference to Fig. 11. Fig. 11 is a conceptual diagram showing an example of an attack on a face recognition gate at an airport.
[0088] As shown in Fig. 11 , the information processing device 10 according to the seventh embodiment may evaluate risks in an airport face authentication system. For example, suppose there is person A, an accomplice, and person B, a terrorist. In this case, person A first submits a photograph to apply for a passport. The submitted photograph appears to the human eye to be that of person A, but to the authentication device, it resembles both person A and person B.
[0089] Person A then hands over his passport to Person B. Person B then presents the passport handed over by Person A and attempts to pass through an unmanned airport gate (a gate that permits passage through using facial recognition). At this time, the gate performs authentication processing using the photograph (registered image) submitted when applying for the passport, but as already explained, this registered image also resembles Person B. Therefore, the unmanned gate succeeds in authenticating Person B (i.e., it mistakenly identifies Person A), resulting in Person B fraudulently passing through.
[0090] As described above, in authentication processing, photos that resemble those of multiple users may pose a greater threat. Therefore, when assessing the risk of authentication processing, it is preferable to use the similarity to multiple users. In contrast, the information processing device 10 according to the seventh embodiment takes into account the similarity between two pieces of information (i.e., the similarity between the first information and the second information, and the similarity between the first information and the third information). This allows for risk assessment assuming the above-described example of an attack by multiple users.
[0091] (Operation Flow) Next, the overall operation flow of the information processing device 10 according to the seventh embodiment will be described with reference to Fig. 12. Fig. 12 is a flowchart showing the operation flow of the information processing device according to the seventh embodiment. Note that in Fig. 12, the same processes as those shown in Fig. 3 are denoted by the same reference numerals.
[0092] 12 , when the operation of the information processing apparatus 10 according to the seventh embodiment is started, the similarity calculation unit 110 first acquires a first information feature amount, a second information feature amount, and a third information feature amount (step S701). The similarity calculation unit 110 then calculates a first similarity, which is the similarity between the first information feature amount and the second information feature amount, and a second similarity, which is the similarity between the first information feature amount and the third information feature amount (step S702). Information regarding the first and second similarities calculated by the similarity calculation unit 110 is output to the gradient information calculation unit 120.
[0093] Next, the gradient information calculation unit 120 calculates first gradient information indicating the gradient of the first similarity calculated by the similarity calculation unit 110 and second gradient information indicating the gradient of the second similarity (step S703). The first and second gradient information calculated by the gradient information calculation unit 120 are output to the perturbation application position determination unit 130.
[0094] Next, the perturbation-imparting-position determining unit 130 determines an element to which a perturbation is to be imparted in the first information, based on the first gradient information and the second gradient information calculated by the gradient information calculating unit 120 (step S704). Information on the element determined by the perturbation-imparting-position determining unit 130 is output to the perturbation imparting unit 140.
[0095] Next, the perturbation applying unit 140 applies a perturbation to the element determined by the perturbation applying position determining unit 130 (step S105). That is, the perturbation is applied to the first information to generate an adversarial sample. The adversarial sample generated by the perturbation applying unit 140 is used in the authentication process.
[0096] Next, the risk assessment unit 150 assesses the risk in the authentication process based on the authentication result of the authentication process using the adversarial sample generated by the perturbation application unit 140 (step S106). The risk assessment unit 150 may output the risk assessment result.
[0097] (Technical Effects) Next, technical effects obtained by the information processing device 10 according to the seventh embodiment will be described.
[0098] As described with reference to Figures 10 to 12, in the information processing device 10 according to the seventh embodiment, adversarial samples are generated by taking into account the third information in addition to the first information and the second information. Therefore, compared to the case where only the first information and the second information are taken into account, adversarial samples can be generated more appropriately. For example, adversarial samples can be generated by taking into account both the source image and the target image described above. Therefore, it is possible to more appropriately evaluate the risk of the authentication process.
[0099] Eighth Embodiment An information processing device 10 according to an eighth embodiment will be described with reference to Fig. 13. The eighth embodiment describes a specific example of the perturbation-imparting position determination operation in the seventh embodiment described above, and other parts may be the same as those in the first to seventh embodiments. Therefore, the following will describe in detail parts that differ from the embodiments already described, and will omit explanations of other overlapping parts as appropriate.
[0100] (Perturbation Applying Position Determining Operation) First, the flow of the perturbation applying position determining operation by the information processing device 10 according to the eighth embodiment will be described with reference to Fig. 13. Fig. 13 is a flowchart showing the flow of the perturbation applying position determining operation by the information processing device according to the eighth embodiment.
[0101] As shown in FIG. 13, in the perturbation position determination operation by the information processing device 10 according to the eighth embodiment, the perturbation position determination unit 130 first acquires the first gradient information (i.e., gradient information of the similarity between the first information feature and the second information feature) and the second gradient information (i.e., gradient information of the similarity between the first information feature and the third information feature) calculated by the gradient information calculation unit 120 (step S801).
[0102] Next, the perturbation-imparting-position determining unit 120 calculates an index value from the first gradient information and the second gradient information calculated by the gradient information calculating unit 120 (step S802). The "index value" here is a value used as an index for determining the perturbation-imparting position. The index value may be, for example, a value calculated as the product of the first gradient information and the second gradient information. Alternatively, the index value may be a weighted sum of the first gradient information and the second gradient information. Alternatively, the index value may be the sum of the absolute value of the first gradient information and the absolute value of the second gradient information.
[0103] Next, the perturbation-adding-position determining unit 130 determines a position in the first information to which a perturbation is to be added based on the calculated index value (step S803). For example, the perturbation-adding-position determining unit 130 may determine one element with the largest index value as the perturbation-adding position (see the third embodiment). Alternatively, the perturbation-adding-position determining unit 130 may determine a predetermined number of elements in descending order of index value as the perturbation-adding positions (see the fourth embodiment). Alternatively, the perturbation-adding-position determining unit 130 may determine an element with an index value greater than a predetermined threshold as the perturbation-adding position (see the fifth embodiment). Thereafter, the perturbation-adding-position determining unit 130 outputs information about the element to which a perturbation is to be added to the perturbation adding unit 140 (step S804).
[0104] (Technical Effects) Next, technical effects obtained by the information processing device 10 according to the eighth embodiment will be described.
[0105] 13 , in the information processing device 10 according to the eighth embodiment, the element to be perturbed is determined based on the index value calculated from the first gradient information and the second gradient information. In this way, the position to be perturbed can be determined by taking into consideration the similarity between the first information and the second information and the similarity between the first information and the third information. Therefore, it is possible to appropriately generate adversarial samples and evaluate the risk of the authentication process.
[0106] The scope of each embodiment also includes a processing method in which a program that operates the configuration of each embodiment to realize the functions of the above-described embodiments is recorded on a recording medium, the program recorded on the recording medium is read as code, and the program is executed on a computer. In other words, a computer-readable recording medium is also included in the scope of each embodiment. Furthermore, each embodiment includes not only a recording medium on which the above-described program is recorded, but also the program itself.
[0107] Examples of recording media that can be used include floppy disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, magnetic tapes, non-volatile memory cards, and ROMs. Furthermore, the scope of each embodiment is not limited to programs that execute processing by themselves, but also includes programs that execute processing by operating on an OS in cooperation with other software or functions of an expansion board. Furthermore, the program itself may be stored on a server, and part or all of the program may be downloaded from the server to a user terminal.
[0108] <Supplementary Notes> The above-described embodiment may be further described as in the following supplementary notes, but is not limited to the following.
[0109] (Supplementary Note 1) The information processing device described in Supplementary Note 1 is an information processing device that includes a similarity calculation means that calculates a similarity between a feature of first information and a feature of second information, a gradient information calculation means that calculates gradient information that indicates a gradient of the similarity, a perturbation application position determination means that determines an element in the first information to be perturbed based on the gradient information, a perturbation application means that applies a perturbation to the element in the first information to be perturbed, and a risk assessment means that evaluates a risk in the authentication process based on a result of an authentication process that compares the first information to which the perturbation has been applied with the second information.
[0110] (Appendix 2) The information processing device described in Appendix 2 is the information processing device described in Appendix 1, wherein the first information is a first image including a first living organism, the second information is a second image including a second living organism, and the similarity calculation means calculates the similarity between a feature related to the first living organism extracted from the first image and a feature related to the second living organism extracted from the second image.
[0111] (Supplementary Note 3) The information processing device described in Supplementary Note 3 is the information processing device described in Supplementary Note 1 or 2, wherein the perturbation position determination means determines one element having the maximum gradient information as the element to be perturbed.
[0112] (Supplementary Note 4) The information processing device described in Supplementary Note 4 is the information processing device described in Supplementary Note 1 or 2, wherein the perturbation position determination means determines a predetermined number of elements in descending order of gradient information as elements to be subjected to the perturbation.
[0113] (Supplementary Note 5) The information processing device described in Supplementary Note 5 is the information processing device described in Supplementary Note 1 or 2, wherein the perturbation position determination means determines an element whose gradient information is greater than a predetermined threshold as the element to be perturbed.
[0114] (Supplementary Note 6) The information processing device described in Supplementary Note 6 is the information processing device described in any one of Supplements 1 to 5, wherein the risk assessment means calculates a probability of false authentication in the authentication process and assesses the risk in the authentication process based on the false authentication probability.
[0115] (Supplementary Note 7) The information processing device described in Supplementary Note 7 is the information processing device described in any one of Supplements 1 to 6, wherein the similarity calculation means calculates a similarity between a feature of the first information and a feature of third information in addition to a similarity between a feature of the first information and a feature of the second information; the gradient information calculation means calculates gradient information of a similarity between a feature of the first information and a feature of third information in addition to gradient information of a similarity between the feature of the first information and a feature of the second information; and the perturbation application position determination means determines an element to be applied with a perturbation based on gradient information of a similarity between the feature of the first information and a feature of third information and gradient information of a similarity between the feature of the first information and a feature of third information.
[0116] (Appendix 8) The information processing device described in Appendix 8 is the information processing device described in Appendix 7, wherein the perturbation position determination means determines the element to be subjected to the perturbation using at least one of the product, weighted sum, and sum of absolute values of gradient information of the similarity between the feature of the first information and the feature of the third information and gradient information of the similarity between the feature of the first information and the feature of the third information.
[0117] (Supplementary Note 9) The information processing method described in Supplementary Note 9 is an information processing method that, using at least one computer, calculates a similarity between a feature of first information and a feature of second information, calculates gradient information indicating a gradient of the similarity, determines an element in the first information to be subjected to a perturbation based on the gradient information, applies a perturbation to the element in the first information to be subjected to the perturbation, and evaluates a risk in the authentication process based on a result of an authentication process that compares the first information to which the perturbation has been applied with the second information.
[0118] (Supplementary Note 10) The recording medium described in Supplementary Note 10 is a recording medium having recorded thereon a computer program for causing at least one computer to execute an information processing method of calculating a similarity between a feature of first information and a feature of second information, calculating gradient information indicating a gradient of the similarity, determining an element in the first information to be subjected to a perturbation based on the gradient information, perturbing the element in the first information to be subjected to the perturbation, and evaluating a risk in the authentication process based on a result of the authentication process of comparing the first information to which the perturbation has been applied with the second information.
[0119] (Supplementary Note 11) The computer program described in Supplementary Note 11 is a computer program that causes at least one computer to execute an information processing method that calculates a similarity between a feature of first information and a feature of second information, calculates gradient information indicating a gradient of the similarity, determines an element in the first information to be subjected to a perturbation based on the gradient information, applies a perturbation to the element in the first information to be subjected to the perturbation, and evaluates a risk in the authentication process based on a result of the authentication process that compares the first information to which the perturbation has been applied with the second information.
[0120] (Supplementary Note 12) The information processing system described in Supplementary Note 12 is an information processing system including: a similarity calculation means for calculating a similarity between a feature of first information and a feature of second information; a gradient information calculation means for calculating gradient information indicating a gradient of the similarity; a perturbation application position determination means for determining an element in the first information to be perturbed based on the gradient information; a perturbation application means for applying a perturbation to the element in the first information to be perturbed; and a risk assessment means for assessing a risk in the authentication process based on a result of an authentication process that compares the first information to which the perturbation has been applied with the second information.
[0121] This disclosure may be modified as appropriate within the scope that does not contradict the gist or idea of the invention that can be read from the claims and the entire specification, and information processing devices, information processing methods, and recording media that involve such modifications are also included in the technical idea of this disclosure.
[0122] REFERENCE SIGNS LIST 10 Information processing device 11 Processor 110 Similarity calculation unit 120 Gradient information calculation unit 130 Perturbation application position determination unit 140 Perturbation application unit 150 Risk evaluation unit 160 Feature extraction unit
Claims
1. a similarity calculation means for calculating a similarity between a feature amount of the first information and a feature amount of the second information; gradient information calculation means for calculating gradient information indicating a gradient of the similarity; a perturbation applying position determining means for determining an element to be subjected to perturbation in the first information based on the gradient information; a perturbation adding means for adding a perturbation to the element to be the perturbation adding target in the first information; a risk assessment means for assessing a risk in the authentication process based on a result of the authentication process of comparing the first information to which the perturbation has been applied with the second information; An information processing device comprising:
2. the first information is a first image including a first living body, the second information is a second image including a second living body, the similarity calculation means calculates a similarity between a feature amount related to the first living body extracted from the first image and a feature amount related to the second living body extracted from the second image; The information processing device according to claim 1 .
3. the applying position determining means determines one element having the maximum gradient information as an element to be subjected to the perturbation; 3. The information processing device according to claim 1 or 2.
4. the applying position determining means determines a predetermined number of elements in descending order of gradient information as elements to which the perturbation is to be applied; 3. The information processing device according to claim 1 or 2.
5. The applying position determination means determines an element whose gradient information is greater than a predetermined threshold as the element to be subjected to the perturbation.
3. The information processing device according to claim 1 or 2.
6. the risk assessment means calculates a probability of an authentication error in the authentication process, and assesses a risk in the authentication process based on the probability of an authentication error. The information processing device according to claim 1 .
7. the similarity calculation means calculates a similarity between the feature amount of the first information and the feature amount of the second information, and also a similarity between the feature amount of the first information and the feature amount of the third information; the gradient information calculation means calculates gradient information of a similarity between the feature amount of the first information and the feature amount of the second information, in addition to gradient information of a similarity between the feature amount of the first information and the feature amount of the third information; the perturbation applying position determining means determines the element to be subjected to the perturbation based on gradient information of a similarity between the feature amount of the first information and the feature amount of the second information and gradient information of a similarity between the feature amount of the first information and the feature amount of the third information. The information processing device according to claim 1 .
8. the perturbation applying position determining means determines the element to be subjected to the perturbation using at least one of a product, a weighted sum, and a sum of absolute values of gradient information of a similarity between the feature amount of the first information and the feature amount of the second information and gradient information of a similarity between the feature amount of the first information and the feature amount of the third information. The information processing device according to claim 7.
9. by at least one computer, Calculating a similarity between a feature amount of the first information and a feature amount of the second information; Calculating gradient information indicating a gradient of the similarity; determining an element to be perturbed in the first information based on the gradient information; imparting a perturbation to the element to be subjected to the perturbation in the first information; evaluating a risk in the authentication process based on a result of the authentication process of matching the first information to which the perturbation has been applied with the second information; Information processing methods.
10. At least one computer Calculating a similarity between a feature amount of the first information and a feature amount of the second information; Calculating gradient information indicating a gradient of the similarity; determining an element to be perturbed in the first information based on the gradient information; imparting a perturbation to the element to be subjected to the perturbation in the first information; evaluating a risk in the authentication process based on a result of the authentication process of matching the first information to which the perturbation has been applied with the second information; A computer program for executing an information processing method.
