Determination device, determination method, determination program, output device, output method, and output program

Abstract

Provided are a determination device, a determination method, a determination program, an output device, an output method, and an output program that improve the accuracy of determining fake images. [Solution] The information processing device 10 as a judgment device has a calculation unit 41 that calculates the degree of deviation between a predetermined index based on a real image showing a predetermined object and a predetermined index based on a judgment target image showing the predetermined object, and a judgment unit 42 that judges that the judgment target image is a fake image (for example, an image generated using generation AI (Artificial Intelligence) and not actually taken) if the degree of deviation calculated by the calculation unit 41 is equal to or greater than a predetermined threshold.

Description

[Technical Field]

[0001] The present invention relates to a determination device, a determination method, a determination program, an output device, an output method, and an output program. [Background technology]

[0002] Conventionally, there are known techniques for distinguishing between images that have not actually been taken (fake images), such as images generated by synthesizing images or images generated using artificial intelligence (AI), and images that have actually been taken (real images). As an example of such a technique, there is a forensic tool that can detect the cutting and pasting of fake images. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Digital Image Forensic Analyzer<URL:https: / / www.imageforensic.org / > [Non-patent document 2] Top 5 Free Picture Analysis Tools for Forensic Investigations <URL:https: / / medium.com / @theboyscyber / top-5-free-picture-analysis-tools-for-forensic-investigations-882df4eca8ef> Summary of the Invention [Problem to be solved by the invention]

[0004] However, the above-mentioned techniques may not be accurate in determining whether an image is fake.

[0005] For example, with the above-mentioned technology, if the image to be judged is generated by cutting and pasting, it can be judged to be a fake image, but if the entire image to be judged is generated by a generation AI, it is difficult to judge that it is a fake image.

[0006] There is also a known technology that uses a machine learning-based judgment model to determine whether a target image is a fake image generated by a generative AI. However, with such technology, the judgment model must be trained using images from a known generative AI. Therefore, if the target image is created by an unknown generative AI, it is difficult to determine that it is a fake image (i.e., it is vulnerable to zero-day attacks).

[0007] The present application has been made in consideration of the above, and aims to improve the accuracy of determining fake images. [Means for solving the problem]

[0008] The determination device according to the present application is characterized by having a calculation unit that calculates the degree of deviation between a predetermined index based on a real image showing a predetermined object and a predetermined index based on a determination target image showing the predetermined object, and a determination unit that determines that the determination target image is a fake image if the degree of deviation calculated by the calculation unit is equal to or greater than a predetermined threshold value. [Effects of the Invention]

[0009] According to one aspect of the embodiment, it is possible to accurately determine whether an image to be determined is a fake image. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram illustrating an example of information processing according to an embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of the index. [Figure 3]FIG. 3 is a diagram illustrating an example of a determination process using a plurality of indices. [Figure 4] FIG. 4 is a diagram showing an example of the configuration of the information processing device 10 according to the embodiment. [Figure 5] FIG. 5 is a diagram showing an example of the image information database 31. As shown in FIG. [Figure 6] FIG. 6 is a flowchart (1) showing an example of the procedure of information processing according to the embodiment. [Figure 7] FIG. 7 is a flowchart (2) showing an example of the procedure of information processing according to the embodiment. [Figure 8] FIG. 8 is a hardware configuration diagram showing an example of a computer that realizes the functions of the information processing device 10. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, a detailed description will be given of a determination device, a determination method, a determination program, an output device, an output method, and an output program (hereinafter referred to as "embodiments") according to the present application, with reference to the drawings. Note that the determination device, the determination method, the determination program, the output device, the output method, and the output program according to the present application are not limited to these embodiments. Furthermore, the same components in the following embodiments are denoted by the same reference numerals, and duplicated descriptions will be omitted.

[0012] 1. Embodiment Information processing implemented by the determination device, output device, etc. of this embodiment will be described using Fig. 1. Fig. 1 is a diagram showing an example of information processing according to this embodiment. In Fig. 1, it is assumed that the information processing according to this embodiment is implemented by an information processing device 10, which is an example of the determination device and output device according to this embodiment.

[0013] As shown in Fig. 1, an information processing system 1 according to the embodiment includes an information processing device 10 and a user terminal 100. The information processing device 10 and the user terminal 100 are connected to each other via a network N (see Fig. 4, for example) so as to be able to communicate with each other via a wired or wireless connection. The network N is, for example, a wide area network (WAN) such as the Internet. Note that the information processing system 1 shown in Fig. 1 may include a plurality of information processing devices 10 and a plurality of user terminals 100.

[0014] 1 is an information processing device that performs information processing, and is realized by, for example, a server device, a cloud system, etc. For example, the information processing device 10 provides the user terminal 100 with an application for determining whether an image is a fake image (hereinafter, may be referred to as a "determination application").

[0015] The information processing device 10 may be a device that distributes information to be displayed in a determination application installed in the user terminal 100 to the user terminal 100. The information processing device 10 may also be a server that distributes application data itself.

[0016] Furthermore, the information processing device 10 may function as a distribution device that distributes control information to the user terminal 100. Here, the control information is written in, for example, a script language such as JavaScript (registered trademark) or a style sheet language such as CSS (Cascading Style Sheets). Note that the determination application itself distributed from the information processing device 10 may be regarded as the control information.

[0017] The user terminal 100 shown in Fig. 1 is an information processing device used by a user. The user terminal 100 is realized, for example, by a smartphone, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, a PDA (Personal Digital Assistant), or the like. The user terminal 100 displays information distributed by the information processing device 10 or the like using a web browser or an application. In the example shown in Fig. 1, the user terminal 100 is a notebook PC.

[0018] The information processing performed by the information processing device 10 will be described below with reference to FIG. 1. In the following description, it is assumed that the user terminal 100 is used by a user (user U1) identified by the user ID "UID#1." In the following description, the user terminal 100 may be considered to be the same as the user U1. In other words, in the following description, the user U1 may also be read as the user terminal 100.

[0019] First, the information processing device 10 receives an image to be determined as a fake image (hereinafter, may be referred to as a "determination target image") from the user terminal 100 (step S1). For example, the information processing device 10 receives a determination target image #1 showing a person via a determination application.

[0020] Next, the information processing device 10 calculates the degree of deviation between a predetermined index based on a real image showing a person (for example, an image actually captured by a camera or the like) and a predetermined index based on the judgment target image #1 (step S2). Next, the information processing device 10 determines whether the judgment target image #1 is a fake image based on the degree of deviation (step S3).

[0021] Here, a score-based generative model (hereinafter sometimes referred to as a "generative model"), which is an extension of the diffusion model to continuous time, is capable of calculating the likelihood indicating how familiar an image is. In other words, a generative model that has learned real images is capable of calculating the likelihood indicating the possibility that an image is a real image. However, such likelihood may have low reliability in determining whether an image is fake. Therefore, the information processing device 10 breaks down such likelihood into multiple parts and uses each of them as an index related to the likelihood to determine whether an image is fake.

[0022] For example, the information processing device 10 determines whether the image to be determined #1 is a fake image by calculating the degree of deviation between an index relating to the likelihood of a real image in a generative model that has been trained using a training image (real image) showing a person and an index relating to the likelihood of the image to be determined #1 in the generative model.

[0023] Here, an index relating to the likelihood of an image will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of the index.

[0024] As shown in Figure 2, the likelihood of an image (X0) before Gaussian noise is added in the generative model is expressed by the left side of Equation E1. Here, the left side of Equation E1 is decomposed into two items as shown on the right side, which are index #1 and index #2, respectively. For example, index #1 is the likelihood of an image transitioning to complete Gaussian noise after going through the diffusion process (X T ) in the generative model. Also, index #2 indicates the likelihood of the point X0 in the probability distribution at time 0 to the point X in the probability distribution at time T. T This shows the increase or decrease in unit volume due to the variable transformation up to the point.

[0025] As an example, the information processing device 10 calculates index #1 for each of the following data sets: data set #1 of real images of people; data set #2 of fake images generated by method #1 (e.g., images of people generated by a generation AI); data set #3 of fake images generated by method #2 (e.g., images of people generated by a generation AI different from method #1); data set #4 of images in which Gaussian noise has been added a predetermined number of times (e.g., 10 times) to real images of people; and data set #5 of images in which Gaussian noise has been added a predetermined number of times (e.g., 30 times) to real images of people. Then, the information processing device 10 normalizes index #1 for each data set to obtain the probability distribution shown in graph G1.

[0026] As shown in graph G1, the data set #1 of real images forms a different mode (mountain) from the data set of fake images and the data set of images to which noise has been added. Therefore, the information processing device 10 can determine whether the image #1 to be determined is a fake image by calculating the degree of deviation #1 between the index #1 in the data set #1 of real images and the index #1 of the image #1 to be determined.

[0027] For example, the information processing device 10 derives a Gaussian distribution #1 from index #1 of a data set #1 of real images, and calculates a degree of deviation #1 between the Gaussian distribution #1 and index #1 of the determination target image #1. Then, if the degree of deviation #1 is equal to or greater than a predetermined threshold, the information processing device 10 determines that the determination target image #1 is a fake image. As an example, the information processing device 10 calculates the likelihood of index #1 of the determination target image #1 in the Gaussian distribution #1, and if the likelihood is equal to or less than a predetermined threshold (in other words, if the degree of deviation #1 is equal to or greater than a predetermined threshold), determines that the determination target image #1 is a fake image. Furthermore, if the calculated likelihood is higher than a predetermined threshold (in other words, if the degree of deviation #1 is less than a predetermined threshold), the information processing device 10 determines that the determination target image #1 is a real image.

[0028] Furthermore, the information processing device 10 calculates index #2 for each of the data sets #1 to #5. Then, the information processing device 10 normalizes index #2 for each data set to obtain the probability distribution shown in graph G2.

[0029] As shown in graph G2, the data set #1 of real images forms a different mode from the data set of fake images and the data set of images to which noise has been added. Therefore, the information processing device 10 can determine whether the image #1 to be determined is a fake image by calculating the degree of deviation #2 between the index #2 in the data set #1 of real images and the index #2 of the image #1 to be determined.

[0030] For example, the information processing device 10 derives a Gaussian distribution #2 from index #2 of a data set #1 of real images, and calculates the degree of deviation #2 between the Gaussian distribution #2 and index #2 of the determination target image #1. Then, if the degree of deviation #2 is equal to or greater than a predetermined threshold, the information processing device 10 determines that the determination target image #1 is a fake image. As an example, the information processing device 10 calculates the likelihood of index #2 of the determination target image #1 in the Gaussian distribution #2, and if the likelihood is equal to or less than a predetermined threshold (in other words, if the degree of deviation #2 is equal to or greater than a predetermined threshold), determines that the determination target image #1 is a fake image. Furthermore, if the calculated likelihood is higher than a predetermined threshold (in other words, if the degree of deviation #2 is less than a predetermined threshold), the information processing device 10 determines that the determination target image #1 is a real image.

[0031] The information processing device 10 may determine whether the determination target image #1 is a fake image using multiple indices related to the dataset #1 of real images. For example, the information processing device 10 calculates a deviation degree #1 and a deviation degree #2. Then, the information processing device 10 determines that the determination target image #1 is a real image if the deviation degrees #1 and #2 are less than a predetermined threshold, and determines that the determination target image #1 is a fake image if either the deviation degrees #1 and #2 are equal to or greater than the predetermined threshold, or if the deviation degrees #1 and #2 are equal to or greater than the predetermined threshold.

[0032] Furthermore, the information processing device 10 may derive a Gaussian mixture distribution using multiple indices related to the dataset #1 of real images. Then, the information processing device 10 may calculate the likelihood of the target image #1 in the Gaussian mixture distribution based on the multiple indices of the target image #1, and determine that the target image #1 is a fake image if the likelihood is equal to or less than a predetermined threshold. Furthermore, the information processing device 10 may determine that the target image #1 is a real image if the calculated likelihood is higher than a predetermined threshold.

[0033] An example of determination using indicators #1 and #2 will now be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of determination processing using a plurality of indicators.

[0034] In the example of FIG. 3, the information processing device 10 obtains a two-dimensional vector based on indices #1 and #2 for each of the data set #1 of real images. Then, the information processing device 10 derives a mixed Gaussian distribution D11 based on each of the two-dimensional vectors of the data set #1 of real images. Furthermore, the information processing device 10 obtains a two-dimensional vector P11 based on indices #1 and #2 for the determination target image #1. Then, the information processing device 10 calculates the likelihood of the two-dimensional vector P11 in the mixed Gaussian distribution D11, and if the likelihood is equal to or less than a predetermined threshold, determines that the determination target image #1 is a fake image. Furthermore, if the calculated likelihood is higher than a predetermined threshold, the information processing device 10 determines that the determination target image #1 is a real image.

[0035] By using multiple indices, for example, even if the image to be judged #1 is a fake image and index #1 of the image to be judged #1 follows the probability distribution of index #1 in the dataset #1 of real images as shown in Figure 3, it can be judged to be a fake image if index #2 of the image to be judged #1 does not follow the probability distribution of index #2 in the dataset #1 of real images.

[0036] The index is not limited to the above, and any index may be used. For example, it is known that the likelihood of an image in a generative model is affected by the complexity of the image. Therefore, the information processing device 10 may determine whether the target image #1 is a fake image based on index #3 indicating the complexity of the image. For example, the information processing device 10 calculates index #3 indicating the magnitude of variance of the pixel values ​​of the image, the length of the bit string when encoded using lossless compression, etc.

[0037] As a specific example, the information processing device 10 determines whether the determination target image #1 is a fake image by calculating the degree of deviation #3 between the index #3 in the data set #1 of real images and the index #3 of the determination target image #1. As an example, the information processing device 10 derives a Gaussian distribution #3 from the index #3 in the data set #1 of real images, and calculates the degree of deviation #3 between the Gaussian distribution #3 and the index #3 of the determination target image #1. Then, if the degree of deviation #3 is equal to or greater than a predetermined threshold, the information processing device 10 determines that the determination target image #1 is a fake image. As an example, the information processing device 10 calculates the likelihood of the index #3 of the determination target image #1 in the Gaussian distribution #3, and if the likelihood is equal to or less than a predetermined threshold (in other words, if the degree of deviation #3 is equal to or greater than a predetermined threshold), determines that the determination target image #1 is a fake image. Furthermore, the information processing device 10 determines that the determination target image #1 is a real image when the calculated likelihood is higher than a predetermined threshold (in other words, when the degree of deviation #3 is less than a predetermined threshold).

[0038] Furthermore, when a fake image generated by a generation AI or the like is transformed into the frequency domain (Fourier transform), artifacts may appear in the high-frequency band that is imperceptible to the human eye. Therefore, the information processing device 10 may determine whether the target image #1 is a fake image based on index #4 indicating the probability distribution in the frequency domain of the image.

[0039] For example, the information processing device 10 obtains a probability distribution (index #4) for the real image dataset #1 by averaging and normalizing the frequency domain data converted from each of the real image datasets #1. The information processing device 10 also obtains a probability distribution (index #4) for the determination target image #1 by converting the determination target image #1 into the frequency domain and normalizing the data. The information processing device 10 then calculates the inter-distribution distance (degree of deviation #4) between the probability distribution for the real image dataset #1 and the probability distribution for the determination target image #1, and determines that the determination target image #1 is a fake image if the inter-distribution distance is equal to or greater than a predetermined threshold. The information processing device 10 also determines that the determination target image #1 is a real image if the inter-distribution distance is less than a predetermined threshold. Here, the inter-distribution distance may be any distance, such as the Kolmogorov-Smirnov distance or the FID (Fréchet Inception Distance).

[0040] The information processing device 10 may determine that the determination target image #1 is a fake image if any of the deviation degrees #1 to #4 is equal to or greater than a predetermined threshold. In other words, the information processing device 10 may determine that the determination target image #1 is a real image if all of the deviation degrees #1 to #4 are less than a predetermined threshold.

[0041] Returning to FIG. 1, the explanation will be continued. Next, the information processing device 10 generates a vector space indicating vectors of the real image dataset #1 and vectors of the determination target image #1 based on a plurality of indices (step S4). For example, the information processing device 10 generates a three-dimensional vector space using three indices from the indices #1 to #4. To give a specific example, the information processing device 10 generates the three-dimensional vector space based on indices from the indices #1 to #4 of the determination target image #1 whose degree of deviation from those of the real image dataset #1 satisfies a predetermined condition. To give a more specific example, the information processing device 10 generates a three-dimensional vector space using indices from the indices #1 to #4 of the determination target image #1 whose degree of deviation from those of the real image dataset #1 is equal to or greater than a predetermined threshold.

[0042] In the example of Fig. 1, it is assumed that, among the deviation degrees #1 to #4, deviation degree #2 is equal to or greater than a predetermined threshold. In such a case, the information processing device 10 generates a three-dimensional vector space F1 using three indices including index #2. In the example of Fig. 1, an example is shown in which a three-dimensional vector space F1 is generated using indexes #1 to #3, but the indices other than index #2 may be set arbitrarily.

[0043] For example, the information processing device 10 obtains a three-dimensional vector based on the indices #1 to #3 for each of the real image data set #1. Then, the information processing device 10 derives a Gaussian mixture distribution D1 using each of the three-dimensional vectors of the real image data set #1. Furthermore, the information processing device 10 obtains a three-dimensional vector P1 based on the indices #1 to #3 for the determination target image #1. Then, the information processing device 10 generates a three-dimensional vector space F1 onto which the Gaussian mixture distribution D1 and the three-dimensional vector P1 are projected.

[0044] Next, the information processing device 10 outputs the three-dimensional vector space F1 to the user terminal 100 (step S5). For example, the information processing device 10 outputs the three-dimensional vector space F1 via a determination application. As a specific example, the information processing device 10 outputs the three-dimensional vector space F1 together with the indexes used to generate the three-dimensional vector space F1 and information indicating that the degree of deviation #2 is equal to or greater than a predetermined threshold.

[0045] When the information processing device 10 receives an operation from the user U1 specifying a combination different from the indices #1 to #3, the information processing device 10 may generate and output a three-dimensional space based on the received combination. This allows the user U1 to understand how much each indices of the judgment target image #1 deviates from the real image.

[0046] Furthermore, the information processing device 10 may output a determination result for the determination target image #1 based on the indices #1 to #4. For example, the information processing device 10 outputs a determination result as to whether the determination target image #1 is a fake image for each of the indices #1 to #4. As a specific example, when the degree of deviation #2 is equal to or greater than a predetermined threshold, the information processing device 10 outputs information indicating that the determination target image #1 is determined to be a fake image based on the index #2.

[0047] Furthermore, the information processing device 10 may output a determination result for the determination target image #1 based on a plurality of indices among the indices #1 to #4. For example, if any of the deviation degrees #1 to #4 is equal to or greater than a predetermined threshold, the information processing device 10 outputs information indicating that the determination target image #1 is determined to be a fake image.

[0048] As described above, the information processing device 10 according to the embodiment can determine that an image generated by a generation AI is a fake image, thereby improving the accuracy of determining fake images.

[0049] Furthermore, the information processing device 10 according to the embodiment can detect fake images without using a judgment model trained using images from the generation AI, thereby improving vulnerability to zero-day attacks.

[0050] Furthermore, the information processing device 10 according to the embodiment can visualize the degree of deviation of each index between the image to be judged and the real image, in other words, the basis for the judgment of the object to be judged. Therefore, for example, if the judgment is incorrect, it is possible to consider the cause and improve the judgment.

[0051] [2. Other processing examples] The above-described process is merely an example, and the information processing device 10 may perform various processes using various information. In this regard, examples are listed below.

[0052] [2-1. Indicators used to generate vector spaces] In the example of FIG. 1, the information processing device 10 may generate a three-dimensional vector space using indices #1 to #4 of the determination target image #1 whose degree of deviation from those of the real image data set #1 is less than a predetermined threshold. Here, it is assumed that, of the degrees of deviation #1 to #4, the degree of deviation #2 is less than the predetermined threshold. In such a case, the information processing device 10 generates a three-dimensional vector space using three indices including the index #2.

[0053] Furthermore, the number of indices used by the information processing device 10 to generate a vector space is not limited to three and may be any number. For example, the information processing device 10 may generate a two-dimensional vector space using two indices from among the indices #1 to #4.

[0054] Furthermore, the indexes are not limited to indexes #1 to #4 and may be any number. For example, the number of indexes may be three or less, or may be five or more.

[0055] [2-2. Presentation of the degree of deviation] 1, the information processing device 10 may present the indexes #1 to #4 and the deviation degrees #1 to #4 to the user U1. For example, the information processing device 10 causes the user terminal 100 to display a table showing the indexes #1 to #4 and the corresponding deviation degrees #1 to #4.

[0056] The information processing device 10 may present an index among the degrees of deviation #1 to #4 whose value is equal to or greater than a predetermined threshold, or may present an index whose value is less than the predetermined threshold. For example, when the value of degree of deviation #1 is equal to or greater than a predetermined threshold and the determination target image #1 is determined to be a fake image based on index #1, the information processing device 10 presents index #1 and degree of deviation #1 to the user. Furthermore, when the value of degree of deviation #2 is less than a predetermined threshold and the determination target image #1 is determined to be a real image based on index #2, the information processing device 10 presents index #2 and degree of deviation #2 to the user. That is, the information processing device 10 may present an index that is the basis for determining that the determination target image #1 is a fake image or an index that is the basis for determining that the determination target image #1 is a real image, along with the degree of deviation of the index.

[0057] Furthermore, the information processing device 10 may present a table showing the indices #1 to #4 and the corresponding deviation degrees #1 to #4, and may accept an operation from the user U1 to set (select) three combinations of the indices #1 to #4. Thereafter, the information processing device 10 may generate and output a three-dimensional vector space based on the set combinations.

[0058] [2-3. About images] 1, the image to be determined is an image showing a person, but the present invention is not limited to this example and may be an image showing any object. For example, the image to be determined may be an image posted by a general user on a predetermined service (for example, a social networking service (SNS), a weather forecast service, etc.).

[0059] In such a case, the information processing device 10 estimates the object indicated by the determination target image by performing image analysis processing on the determination target image and morphological analysis on comments posted together with the determination target image. The information processing device 10 then calculates the degree of deviation between the indices #1 to #4 based on a dataset of real images indicating the estimated object and the indices #1 to #4 based on the determination target image, determines whether the determination target image is a fake image, and generates a three-dimensional vector space. For example, if the object indicated by the determination target image is estimated to be "weather," the information processing device 10 calculates the degree of deviation between the indices #1 to #4 based on a dataset of real images indicating "weather" and the indices #1 to #4 based on the determination target image based on a dataset of real images indicating "weather" and a generative model trained on the dataset of real images indicating "weather," determines whether the determination target image is a fake image, and generates a three-dimensional vector space.

[0060] That is, the information processing device 10 performs processing related to the above-described embodiment on the target indicated by the determination target image based on a data set of real images prepared in advance.

[0061] 3. Configuration of Information Processing Device Next, the configuration of the information processing device 10 will be described with reference to Fig. 4. Fig. 4 is a diagram showing an example of the configuration of the information processing device 10 according to an embodiment. As shown in Fig. 4, the information processing device 10 has a communication unit 20, a storage unit 30, and a control unit 40.

[0062] (Regarding the communication unit 20) The communication unit 20 is realized by, for example, a network interface card (NIC), etc. The communication unit 20 is connected to the network N by wire or wirelessly, and transmits and receives information to and from the user terminal 100, etc.

[0063] (Regarding the storage unit 30) The storage unit 30 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. As shown in FIG. 4 , the storage unit 30 has an image information database 31 and a model database 32.

[0064] (About Image Information Database 31) The image information database 31 stores various types of information related to real images. An example of the information stored in the image information database 31 will now be described with reference to FIG. 5. FIG. 5 is a diagram showing an example of the image information database 31. In the example of FIG. 5, the image information database 31 has items such as "image ID," "target information," and "image."

[0065] "Image ID" indicates identification information for identifying a real image. "Target information" is information about the target indicated by the real image, such as "people" or "weather." "Image" indicates the real image identified by "Image ID."

[0066] That is, FIG. 5 shows an example in which the target information of the real image "Image #1" identified by the image ID "CID#1" is "Target Information #1."

[0067] (About Model Database 32) The model database 32 stores generative models. For example, the model database 32 stores, for each predetermined object, a generative model that has been trained using real images of a predetermined object. For example, the model database 32 stores multiple generative models, such as a generative model that has been trained using real images of a "person" and a generative model that has been trained using real images of "weather."

[0068] (Regarding the control unit 40) The control unit 40 is a controller, and is realized, for example, by a CPU (Central Processing Unit) or an MPU (Micro Processing Unit) executing various programs stored in a storage device inside the information processing device 10 using RAM as a work area. The control unit 40 is also a controller, and is realized, for example, by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array). As shown in FIG. 4 , the control unit 40 according to the embodiment has a calculation unit 41, a determination unit 42, a generation unit 43, an output unit 44, and a presentation unit 45, and realizes or executes the functions and actions of information processing described below.

[0069] (Regarding the calculation unit 41) The calculation unit 41 calculates the degree of deviation between a predetermined index based on a real image showing a predetermined object and a predetermined index based on a determination target image showing the predetermined object. For example, in the example of Fig. 1, the calculation unit 41 refers to the storage unit 30 (e.g., the image information database 31) and calculates the degree of deviation between a predetermined index based on a real image showing a person and a predetermined index based on determination target image #1.

[0070] Furthermore, the calculation unit 41 may calculate the degree of deviation between an index related to the likelihood of a real image in a generative model trained using training images showing a predetermined object and an index related to the likelihood of a determination target image in the generative model. For example, in the example of Fig. 1, the calculation unit 41 refers to the storage unit 30 (e.g., the image information database 31 or the model database 32) and calculates the degree of deviation between an index related to the likelihood of a real image in a generative model trained using training images showing a person and an index related to the likelihood of determination target image #1 in the generative model.

[0071] Furthermore, the calculation unit 41 may calculate the degree of deviation between the likelihood of a real image after the diffusion process in the generative model and the likelihood of a target image after the diffusion process. For example, in the example of FIG. 1 , the calculation unit 41 calculates, for each of a data set #1 of real images showing a person and a target image #1, an index #1 indicating the likelihood in the generative model in a state in which the image has transitioned to complete Gaussian noise after the diffusion process. Then, the calculation unit 41 calculates the degree of deviation #1 between the index #1 in the data set #1 of real images and the index #1 of the target image #1.

[0072] Furthermore, the calculation unit 41 may calculate the degree of deviation between the increase or decrease in unit volume of the probability distribution in the diffusion process of the real image and the increase or decrease in unit volume of the probability distribution in the diffusion process of the determination target image. For example, in the example of FIG. 1 , the calculation unit 41 calculates, for each of a data set #1 of real images showing a person and the determination target image #1, an index #2 indicating the increase or decrease in unit volume of the probability distribution until the image transitions to complete Gaussian noise through the diffusion process. Then, the calculation unit 41 calculates the degree of deviation #2 between the index #2 in the data set #1 of real images and the index #2 of the determination target image #1.

[0073] Furthermore, the calculation unit 41 may calculate the degree of deviation between the complexity of a real image and the complexity of the determination target image. For example, in the example of Fig. 1, the calculation unit 41 calculates an index #3 indicating the image complexity for each of a data set #1 of real images showing a person and the determination target image #1. Then, the calculation unit 41 calculates the degree of deviation #3 between the index #3 in the data set #1 of real images and the index #3 of the determination target image #1.

[0074] Furthermore, the calculation unit 41 may calculate the degree of deviation between the probability distribution when a real image is converted into the frequency domain and the probability distribution when an image to be determined is converted into the frequency domain. For example, in the example of FIG. 1 , the calculation unit 41 calculates an index #4 indicating the probability distribution in the frequency domain for each of a data set #1 of real images showing a person and the image to be determined #1. Then, the calculation unit 41 calculates the degree of deviation #4 between the index #4 in the data set #1 of real images and the index #4 of the image to be determined #1.

[0075] The calculation unit 41 may also calculate the degree of deviation between a Gaussian distribution of a predetermined index based on a real image and a predetermined index based on the image to be determined. For example, in the example of FIG. 1 , the calculation unit 41 derives a Gaussian distribution #1 from index #1 of a data set #1 of real images, and calculates a degree of deviation #1 between the Gaussian distribution #1 and index #1 of the image to be determined. The calculation unit 41 also derives a Gaussian distribution #2 from index #2 of the data set #1 of real images, and calculates a degree of deviation #2 between the Gaussian distribution #2 and index #2 of the image to be determined. The calculation unit 41 derives a Gaussian distribution #3 from index #3 of the data set #1 of real images, and calculates a degree of deviation #3 between the Gaussian distribution #3 and index #3 of the image to be determined. Furthermore, the calculation unit 41 derives a Gaussian mixture distribution using a plurality of indices related to the dataset #1 of real images, and calculates the likelihood of the target image #1 in the Gaussian mixture distribution based on the plurality of indices of the target image #1.

[0076] Furthermore, the calculation unit 41 may calculate the degree of deviation between corresponding indices between a plurality of predetermined indices based on real images and a plurality of predetermined indices based on the determination target image. For example, in the example of Fig. 1, the calculation unit 41 calculates the degree of deviation between corresponding indices between indices #1 to #4 in data set #1 of real images and indices #1 to #4 in determination target image #1.

[0077] (Regarding the determination unit 42) The determination unit 42 determines that the determination target image is a fake image when the degree of deviation calculated by the calculation unit 41 is equal to or greater than a predetermined threshold. For example, in the example of FIG. 1, the determination unit 42 determines that the determination target image #1 is a fake image when the degree of deviation #1 is equal to or greater than a predetermined threshold. Furthermore, the determination unit 42 determines that the determination target image #1 is a fake image when the degree of deviation #2 is equal to or greater than a predetermined threshold. Furthermore, the determination unit 42 determines that the determination target image #1 is a fake image when the degree of deviation #3 is equal to or greater than a predetermined threshold. Furthermore, the determination unit 42 determines that the determination target image #1 is a fake image when the degree of deviation #4 is equal to or greater than a predetermined threshold.

[0078] Furthermore, the determination unit 42 may determine that the image to be determined is a fake image if any of the degrees of deviation is equal to or greater than a predetermined threshold. For example, in the example of Fig. 1, the determination unit 42 determines that the image to be determined #1 is a fake image if any of the degrees of deviation #1 to #4 is equal to or greater than a predetermined threshold.

[0079] (Regarding the generation unit 43) The generation unit 43 generates a vector space in which the real image and the target image are projected, based on a predetermined index based on the real image showing a predetermined object and a predetermined index based on the target image showing the predetermined object. For example, in the example of Fig. 1, the generation unit 43 generates a three-dimensional vector space indicating the vectors of the dataset #1 of the real image based on the predetermined index and the vectors of the target image #1 based on the predetermined index.

[0080] Furthermore, the generation unit 43 may generate a vector space based on indices set by a user from among predetermined indices. For example, in the example of Fig. 1, the generation unit 43 generates a three-dimensional vector space using indices #1 to #4 and an index set by user U1 from among the indices #1 to #4.

[0081] Furthermore, the generating unit 43 may generate a vector space based on an index, among a plurality of predetermined indices, whose degree of deviation satisfies a predetermined condition. For example, in the example of Fig. 1, the generating unit 43 generates a three-dimensional vector space based on an index, among the indices #1 to #4 of the determination target image #1, whose degree of deviation from those of the real image data set #1 satisfies a predetermined condition.

[0082] Furthermore, the generating unit 43 may generate a vector space based on at least one of a plurality of predetermined indices whose degree of deviation is equal to or greater than a predetermined threshold. For example, in the example of Fig. 1, the generating unit 43 generates a three-dimensional vector space using indices #1 to #4 of the determination target image #1 whose degree of deviation from those of the real image data set #1 is equal to or greater than a predetermined threshold.

[0083] Furthermore, the generating unit 43 may generate a vector space based on at least one of a plurality of predetermined indices whose degree of deviation is less than a predetermined threshold. For example, in the example of Fig. 1, the generating unit 43 generates a three-dimensional vector space using indices #1 to #4 of the determination target image #1 whose degree of deviation from those of the real image data set #1 is less than a predetermined threshold.

[0084] (Regarding the output unit 44) The output unit 44 outputs the vector space generated by the generation unit 43. For example, in the example of FIG.

[0085] (Regarding presentation unit 45) The presenting unit 45 presents the degree of deviation between a plurality of predetermined indices based on the real image and a plurality of predetermined indices based on the determination target image. For example, in the example of Fig. 1, the presenting unit 45 presents indices #1 to #4 and the degrees of deviation #1 to #4.

[0086] [4. Information processing flow] The information processing procedure of the information processing device 10 according to the embodiment will be described with reference to Fig. 6. Fig. 6 is a flowchart (1) showing an example of the information processing procedure according to the embodiment.

[0087] 6, the information processing device 10 calculates the degree of deviation between a predetermined index based on a real image showing a predetermined object and a predetermined index based on a determination target image showing the predetermined object (step S101). Subsequently, the information processing device 10 determines whether the degree of deviation is equal to or greater than a predetermined threshold (step S102). If the degree of deviation is equal to or greater than the predetermined threshold (step S102; Yes), the information processing device 10 determines that the determination target image is a fake image (step S103) and ends the process.

[0088] On the other hand, if the degree of deviation is not equal to or greater than the predetermined threshold (step S102; No), the information processing device 10 determines that the determination target image is a real image (step S104), and ends the process.

[0089] Next, the procedure of information processing of the information processing device 10 according to the embodiment will be described with reference to Fig. 7. Fig. 7 is a flowchart (2) showing an example of the procedure of information processing according to the embodiment.

[0090] 7, the information processing device 10 determines whether or not a determination target image has been received (step S201). If a determination target image has not been received (step S201; No), the information processing device 10 waits until a determination target image is received.

[0091] On the other hand, if the determination target image is received (step S201; Yes), the information processing device 10 generates a vector space in which the real image and the determination target image are projected based on a predetermined index based on the real image showing a predetermined object and a predetermined index based on the determination target image showing a predetermined object (step S202). Subsequently, the information processing device 10 outputs the vector space (step S203) and ends the process.

[0092] [5. Modifications] The above-described embodiment is merely an example, and various modifications and applications are possible.

[0093] [5-1. Processing mode] Of the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, and conversely, all or part of the processes described as being performed manually can be performed automatically using known methods. In addition, the information, including the processing procedures, specific names, various data, and parameters shown in the above text and drawings, can be changed as desired unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown in the drawings.

[0094] Furthermore, the components of each device shown in the figure are conceptual functional components and do not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

[0095] Furthermore, the above-described embodiments can be combined as appropriate within the scope of not causing any contradiction in the processing content.

[0096] [6. Effects] As described above, the information processing device 10 according to the embodiment includes a calculation unit 41 and a determination unit 42. The calculation unit 41 calculates the degree of deviation between a predetermined index based on a real image showing a predetermined object and a predetermined index based on a determination target image showing the predetermined object. The calculation unit 41 also calculates the degree of deviation between corresponding indices, between a plurality of predetermined indices based on the real image and a plurality of predetermined indices based on the determination target image. The determination unit 42 determines that the determination target image is a fake image when the degree of deviation calculated by the calculation unit 41 is equal to or greater than a predetermined threshold. The determination unit 42 also determines that the determination target image is a fake image when any degree of deviation is equal to or greater than a predetermined threshold.

[0097] As a result, the information processing device 10 according to the embodiment can determine that an image generated by the generation AI is a fake image, thereby improving the accuracy of determining fake images.

[0098] Furthermore, in the information processing device 10 according to the embodiment, for example, the calculation unit 41 calculates the degree of deviation between an index related to the likelihood of a real image in a generative model trained using learning images representing a predetermined target and an index related to the likelihood of a determination target image in the generative model. The calculation unit 41 also calculates the degree of deviation between the likelihood of a real image after a diffusion process in the generative model and the likelihood of a determination target image after the diffusion process. The calculation unit 41 also calculates the degree of deviation between the increase or decrease in unit volume of the probability distribution in the diffusion process of the real image and the increase or decrease in unit volume of the probability distribution in the diffusion process of the determination target image. The calculation unit 41 also calculates the degree of deviation between the complexity of the real image and the complexity of the determination target image. The calculation unit 41 also calculates the degree of deviation between the probability distribution when the real image is transformed into the frequency domain and the probability distribution when the determination target image is transformed into the frequency domain. Furthermore, the calculation unit 41 calculates the degree of deviation between the Gaussian distribution of the predetermined index based on the real image and the predetermined index based on the determination target image.

[0099] As a result, the information processing device 10 according to the embodiment can determine whether the image to be judged is a fake image or not based on the degree of deviation of various indicators between the image to be judged and a real image, thereby improving the accuracy of the judgment.

[0100] The information processing device 10 according to the embodiment also includes a generation unit 43, an output unit 44, and a presentation unit 45. The generation unit 43 generates a vector space in which a real image showing a predetermined object and a target image showing a predetermined object are projected, based on the real image and the target image. The output unit 44 outputs the vector space generated by the generation unit 43. The presentation unit 45 presents the degree of deviation between corresponding indices, between the plurality of indices based on the real image and the plurality of indices based on the target image.

[0101] As a result, the information processing device 10 according to the embodiment can visualize the degree of deviation between the image to be judged and each index of the real image, in other words, the basis for the judgment of the object to be judged, so that if the judgment is incorrect, it is possible to consider the cause and improve the judgment.

[0102] In addition, in the information processing device 10 according to the embodiment, for example, the generation unit 43 generates a vector space based on an index set by a user among predetermined indexes.

[0103] As a result, the information processing device 10 according to the embodiment can generate and output a vector space based on an index arbitrarily set by a user, thereby improving convenience.

[0104] In the information processing device 10 according to the embodiment, for example, the generation unit 43 generates a vector space based on an index among a plurality of predetermined indices whose deviation degree satisfies a predetermined condition. The generation unit 43 also generates a vector space based on at least one index among the plurality of predetermined indices whose deviation degree is equal to or greater than a predetermined threshold. The generation unit 43 also generates a vector space based on at least one index among the plurality of predetermined indices whose deviation degree is less than a predetermined threshold.

[0105] As a result, the information processing device 10 according to the embodiment can generate and output a vector space according to the degree of deviation of a predetermined index, thereby improving convenience.

[0106] [7. Hardware Configuration] The information processing device 10 according to each of the above-described embodiments is realized, for example, by a computer 1000 configured as shown in Fig. 8. The following description will be given taking the information processing device 10 as an example. Fig. 8 is a hardware configuration diagram showing an example of a computer that realizes the functions of the information processing device 10. The computer 1000 has a CPU 1100, a ROM 1200, a RAM 1300, an HDD 1400, a communication interface (I / F) 1500, an input / output interface (I / F) 1600, and a media interface (I / F) 1700.

[0107] The CPU 1100 operates and controls each unit based on programs stored in the ROM 1200 or the HDD 1400. The ROM 1200 stores a boot program executed by the CPU 1100 when the computer 1000 starts up, programs that depend on the hardware of the computer 1000, and the like.

[0108] The HDD 1400 stores programs executed by the CPU 1100, data used by such programs, etc. The communication interface 1500 receives data from other devices via a communication network 500 (corresponding to the network N in the embodiment) and sends the data to the CPU 1100, and also transmits data generated by the CPU 1100 to other devices via the communication network 500.

[0109] The CPU 1100 controls output devices such as a display and a printer, and input devices such as a keyboard and a mouse, via the input / output interface 1600. The CPU 1100 acquires data from the input devices via the input / output interface 1600. The CPU 1100 also outputs generated data to the output devices via the input / output interface 1600.

[0110] Media interface 1700 reads a program or data stored in recording medium 1800 and provides it to CPU 1100 via RAM 1300. CPU 1100 loads the program or data from recording medium 1800 onto RAM 1300 via media interface 1700 and executes the loaded program. Recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase Change Rewritable Disc), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory.

[0111] For example, when the computer 1000 functions as the information processing device 10, the CPU 1100 of the computer 1000 executes programs loaded onto the RAM 1300 to realize the functions of the control unit 40. The HDD 1400 also stores various data in the storage device of the information processing device 10. The CPU 1100 of the computer 1000 reads and executes these programs from the recording medium 1800, but as another example, the CPU 1100 may obtain these programs from another device via a predetermined communication network.

[0112] [8. Other] Although some of the embodiments of the present application have been described in detail above with reference to the drawings, these are merely examples, and the present invention can be implemented in other forms that include the embodiments described in the Disclosure of the Invention section and that have undergone various modifications and improvements based on the knowledge of those skilled in the art.

[0113] Furthermore, the information processing device 10 described above can flexibly change its configuration, for example, by calling an external platform or the like using an API (Application Programming Interface) or network computing, depending on the function.

[0114] Furthermore, the term "unit" in the claims can be read as "means" or "circuit," etc. For example, a generating unit can be read as generating means or generating circuit. [Explanation of symbols]

[0115] 10. Information processing equipment 20 Communications Department 30 Storage section 31 Image Information Database 32 Model Database 40 Control Unit 41 Calculation section 42 Judgment section 43 Generation part 44 Output section 45 Presentation section 100 user terminals

Claims

1. a calculation unit that calculates a degree of deviation between a predetermined index based on a real image showing a predetermined object and a predetermined index based on a determination target image showing the predetermined object; a determination unit that determines that the image to be determined is a fake image when the degree of deviation calculated by the calculation unit is equal to or greater than a predetermined threshold; A determination device comprising:

2. The calculation unit Calculating the degree of deviation between an index relating to the likelihood of the real image in a generative model trained using a learning image representing the predetermined target and an index relating to the likelihood of the determination target image in the generative model.

2. The determination device according to claim 1.

3. The calculation unit Calculating the degree of deviation between the likelihood of the real image after the diffusion process and the likelihood of the determination target image after the diffusion process in the generative model.

3. The determination device according to claim 2.

4. The calculation unit Calculating the degree of deviation between the increase or decrease in unit volume of the probability distribution in the diffusion process of the real image and the increase or decrease in unit volume of the probability distribution in the diffusion process of the determination target image 3. The determination device according to claim 2.

5. The calculation unit Calculating the degree of deviation between the complexity of the real image and the complexity of the determination target image 2. The determination device according to claim 1.

6. The calculation unit Calculating the degree of deviation between the probability distribution when the real image is transformed into the frequency domain and the probability distribution when the determination target image is transformed into the frequency domain 2. The determination device according to claim 1.

7. The calculation unit Calculating the degree of deviation between the Gaussian distribution of the predetermined index based on the real image and the predetermined index based on the determination target image 2. The determination device according to claim 1.

8. The calculation unit calculating a degree of deviation between corresponding indices based on the real image and a plurality of indices based on the determination target image; The determination unit If any of the degrees of deviation is equal to or greater than a predetermined threshold, the image to be determined is determined to be a fake image.

2. The determination device according to claim 1.

9. A computer-implemented determination method comprising: a calculation step of calculating a degree of deviation between a predetermined index based on a real image showing a predetermined object and a predetermined index based on a determination target image showing the predetermined object; a determination step of determining that the image to be determined is a fake image when the degree of deviation calculated by the calculation step is equal to or greater than a predetermined threshold; A determination method comprising:

10. a calculation step of calculating a degree of deviation between a predetermined index based on a real image showing a predetermined object and a predetermined index based on a determination target image showing the predetermined object; a determination step of determining that the image to be determined is a fake image when the degree of deviation calculated by the calculation step is equal to or greater than a predetermined threshold; A determination program that causes a computer to execute the above.

11. a generation unit that generates a vector space in which the real image and the determination target image are projected based on a predetermined index based on a real image showing a predetermined object and a predetermined index based on a determination target image showing the predetermined object; an output unit that outputs the vector space generated by the generation unit; An output device comprising:

12. The generation unit The vector space is generated based on an index set by a user among predetermined indexes.

12. The output device according to claim 11.

13. a calculation unit that calculates the degree of deviation between corresponding indices based on the real image and the corresponding indices based on the determination target image; and The generation unit The vector space is generated based on an index whose deviation degree satisfies a predetermined condition among a plurality of predetermined indexes.

12. The output device according to claim 11.

14. The generation unit The vector space is generated based on at least one of the plurality of predetermined indices whose deviation is equal to or greater than a predetermined threshold.

14. The output device according to claim 13.

15. The generation unit The vector space is generated based on at least one of the plurality of predetermined indices whose deviation is less than a predetermined threshold.

14. The output device according to claim 13.

16. a presentation unit that presents a degree of deviation between corresponding indices based on the real image and the target image; 12. The output device of claim 11, further comprising:

17. 1. A computer-implemented output method comprising: a generation step of generating a vector space in which a real image showing a predetermined object and a determination target image showing a predetermined object are projected based on the real image and the determination target image, based on the predetermined index based on the real image showing the predetermined object; an output step of outputting the vector space generated by the generation step; An output method comprising:

18. a generation procedure for generating a vector space in which a real image showing a predetermined object and a determination target image showing the predetermined object are projected, based on a predetermined index based on the real image and the determination target image; an output step for outputting the vector space generated by the generation step; An output program characterized by causing a computer to execute the above.

Images