Vectorization of image generated by artificial intelligence

Abstract

To allow for changing size while preventing deterioration of image quality, regarding images generated by artificial intelligence, or the like.SOLUTION: In an information processing method executed by an information processing device having a processor, the processor executes: generating a first image in raster format using a generative adversarial network that includes a generator and a discriminator; converting the first image into a second image in vector format; inputting a third image obtained by converting the second image into raster format, or the second image, to the discriminator, to obtain discrimination results; and storing the second image in a storage unit, if the discrimination results are true, or if the discrimination results are the same as results discriminated by the discriminator regarding the first image.SELECTED DRAWING: Figure 6

Description

Technical Field

[0005] , , , , , , ,

[0001] The present invention relates to an information processing method, an information processing apparatus, and a program for vectorizing an image generated by artificial intelligence into a vector image.

Background Art

[0002] Conventionally, it is known that high-quality images are generated by using generative adversarial networks (GANs) (see, for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When an image is generated using artificial intelligence (AI), it is often the case that the image is generated using a generative adversarial network (GAN). Although the images generated by GANs are of high quality, most of them are raster format images, and thus are not suitable for changing the image size. For example, even for a high-resolution raster image, if it is significantly enlarged, blurring will inevitably occur in parts such as curves, and the image quality will deteriorate. Therefore, the present invention provides an information processing method, an information processing apparatus, and a program that enable size change while preventing deterioration of image quality even for an image generated by artificial intelligence (AI) or the like.

[0005] ​​​​​​​​​[Means for solving the problem]

[0006] An information processing method according to one aspect of the present invention is executed by an information processing device having a processor. An information processing method wherein the processor includes a generator and a discriminator, and To generate a first image in raster format using twerking, and to convert the first image in vector format Converting to a second image, a third image obtained by converting the second image to a raster format, or the previous The second image is input to the classifier and the identification result is obtained. If the identification result is true, , or if the identification result is the same as the identification result of the first image by the classifier, The program executes the following: store two images in the memory unit. [Effects of the Invention]

[0007] According to the present invention, even images generated by artificial intelligence (AI), etc., do not suffer from image quality degradation. It prevents this while allowing resizing. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example of a system configuration according to the embodiment. [Figure 2] This figure shows an example of the physical configuration of the information processing device according to the embodiment. [Figure 3] This figure shows an example of a processing block in an information processing apparatus according to the embodiment. [Figure 4] This figure shows an example of a saved image related to Example 1. [Figure 5] This figure shows an example of noise associated with the image conversion algorithm according to Example 1. [Figure 6] This flowchart shows an example of processing performed by the information processing device according to Example 1. [Figure 7] This figure shows an example of a processing block for an information processing device according to Example 2. [Figure 8] It is a flowchart showing a processing example of the information processing apparatus according to the second embodiment.

Mode for Carrying Out the Invention

[0009] Embodiments of the present invention will be described with reference to the accompanying drawings. In each figure, those assigned the same reference numerals have the same or similar configurations.

[0010] <System Configuration> FIG. 1 is a diagram showing an example of a system configuration according to the embodiment. In the example shown in FIG. 1, the server 10 and each information processing apparatus 20A, 20B, 20C, 20D are connected via a network so that data can be transmitted and received. When not distinguishing the information processing apparatuses individually, they are also referred to as the information processing apparatus 20.

[0011] The server 10 is an information processing apparatus capable of collecting and analyzing data, and may be composed of one or more information processing apparatuses. The information processing apparatus 20 is an information processing apparatus capable of executing machine learning, such as a smartphone, a personal computer, a tablet terminal, a server, a connected car, etc. Note that the information processing apparatus 20 may be directly or indirectly connected to an invasive or non-invasive electrode that senses brain waves, and may be a device capable of analyzing and transmitting / receiving brain wave data.

[0012] In the system shown in FIG. 1, the server 10 generates an image by, for example, GANs, converts the generated image into a raster format (format), and stores the vector format image. At this time, instead of simply converting the raster format image generated by GANs into a vector format, the conversion error when converting the raster format image into a vector format image is taken into consideration That is, taking into account the conversion error from raster format to vector format, When a vector image is displayed on a screen, etc., the original raster image is displayed. A vector-format image that displays almost identically to the original is saved.

[0013] Server 10, in response to a request for a predetermined image from each information processing device 20, uses GANs. A predetermined image is generated using this method, and based on the generated predetermined image, the vector format predetermined as described above is used. The system generates an image and transmits this predetermined vector image to the requested information processing device 20. You may do that.

[0014] Each information processing device 20 sends a request to the server 10 to generate a predetermined image by specifying conditions, etc. It transmits to the server. In addition, each information processing device 20 receives a predetermined vector image from the server 10. You may believe and display a predetermined image in the virtual space.

[0015] As a result, a predetermined raster image generated by AI takes conversion errors into account. Because it has been converted to a vector format that can be identified with a raster image, this given image is Even if the size is changed, the quality can be prevented from degrading. For example, when generated by AI If the image is sold in a virtual space (for example, on the metaverse), even if the size is changed... This will make it possible to sell images without any loss of quality.

[0016] <Hardware Configuration> Figure 2 shows an example of the physical configuration of the information processing device 10 according to the embodiment. The processing device 10 consists of a CPU (Central Processing Unit) 10a which corresponds to the calculation unit, and a storage unit. This corresponds to RAM (Random Access Memory) 10b and the storage unit corresponds to ROM (Read O It has a memory unit 10c, a communication unit 10d, an input unit 10e, and a display unit 10f. These configurations are connected to each other via a bus, enabling data transmission and reception between them.

[0017] This embodiment describes the case where the information processing device 10 is composed of a single computer. To clarify, the information processing device 10 is a combination of multiple computers or multiple arithmetic units. This may be implemented. Also, the configuration shown in Figure 1 is just one example, and the information processing device 10 may be other than these. It may have these configurations, or it may not have some of these configurations.

[0018] CPU10a is an example of a processor, and is stored in RAM10b or ROM10c. This is a control unit that performs control related to the execution of the program, as well as calculations and processing of data. CPU10 a is, for example, a program that performs learning using a predetermined learning model (learning program). This is the processing unit. The CPU 10a receives various data from the input unit 10e and the communication unit 10d. The data is received, and the calculation results are displayed on the display unit 10f or stored in RAM 10b. ru.

[0019] RAM10b is a memory unit that allows data to be rewritten, for example, semiconductor memory It may be composed of memory elements. RAM10b is the program executed by CPU10a, each learning It stores data about the model, the parameters of each learning model, and the data used for training. This is also acceptable. Note that these are examples, and RAM10b may store data other than those listed above. They may be included, or some of them may not be remembered.

[0020] ROM10c is a part of the storage unit from which data can be read, for example, semiconductor memory It may consist of memory elements. ROM10c may contain, for example, a learning program or a program that can be rewritten. It is acceptable to store data that does not exist.

[0021] The communication unit 10d is an interface for connecting the information processing device 10 to other devices. The signal unit 10d may be connected to a communication network such as the Internet.

[0022] The input unit 10e accepts data input from the user, for example, keyboard input. It may include a keyboard and a touch panel.

[0023] The display unit 10f visually displays the calculation results from the CPU 10a, for example, It may be composed of an LCD (Liquid Crystal Display). The display unit 10f displays the calculation result. Displaying this information can contribute to XAI (eExplainable AI). Display unit 10 f may, for example, display learning results or data related to learning.

[0024] The learning program can be read by the computer, such as RAM10b or ROM10c. It may be stored and provided on a non-temporary recording medium, or connected by the communication unit 10d. It may be provided via a communication network. The information processing device 10 provides one or more C The execution of the program by PU10a enables various operations described later. These physical configurations are illustrative and do not necessarily have to be independent configurations. For example The information processing device 10 is an LSI that integrates the CPU 10a and RAM 10b and ROM 10c. It may also have (Large-Scale Integration). In addition, the information processing device 10 may have a GPU (Graphical Processing Unit) and ASIC (Application Specific Integrated Circuit) ) may also be equipped with.

[0025] The configuration of the information processing device 20 is the same as that of the information processing device 10 shown in Figure 2. Therefore, I will omit that explanation. Also, the information processing device 10 and the information processing device 20 are related to data processing. It is sufficient to have a basic configuration such as a CPU 10a and RAM 10b, and input section 1 0e and the display unit 10f do not need to be provided. Also, the input unit 10e and the display unit 10f are external. The unit may be connected using an interface.

[0026] <Processing Configuration> Figure 3 shows an example of a processing block of the information processing device (server device) 10 according to the embodiment. This is a diagram. The information processing device 10 consists of an acquisition unit 102, an image generation unit 104, and a database (DB). )110, first image conversion unit 112, second image conversion unit 114, determination unit 116, storage unit 118 The information processing device 20 is equipped with an output unit 120. The information processing device 20 may be composed of a quantum computer or the like. stomach.

[0027] The acquisition unit 102 and the output unit 120 can be implemented, for example, by the communication unit 10d shown in Figure 2. Image generation unit 104, first image conversion unit 112, second image conversion unit 114, and determination unit 116 This can be realized, for example, by the CPU 10a shown in Figure 2. DB110 is shown, for example, in Figure 2. This can be implemented by ROM10c and / or RAM10b. The storage unit 118 is, for example, shown in Figure This can be achieved by the RAM 10b and / or ROM 10c shown in 2.

[0028] The acquisition unit 102 acquires the image generation request. The image generation request is processed by each information processing unit. This could be a request sent from device 20, or an operation of the user managing information processing device 10. Requests obtained according to the task are also acceptable. For example, an image generation request may be an image data request. This includes the conditions for image generation or text for image generation, and the image is generated by GANs. The data entered may also be for the purpose of generating the image. The acquisition unit 102 responds to the acquisition of the image generation request. Then, the image generation unit 104 is instructed to generate an image.

[0029] The image generation unit 104, for example, executes a Generative Adversarial Network (GANs). The image is then generated. Specifically, the image generation unit 104 consists of a generator 106 and a discriminator. Using a Generative Adversarial Network (GAN) that includes 108, the first image in raster format is generated. To generate. For example, GANs can use one of the following methods: (1) pix2pix By learning the relationships between images from conditional images and image pairs, we can learn the relationships between images from a single image. An image is generated that complements the image relationships of A. (2) CycleGAN Using two sets of images, generate one image from the other, and convert the other image back to the first. The model is trained to improve accuracy when it is cycled. (3) CGAN (Conditional GAN) This is called a conditional GAN, and it involves giving additional information to the generator and discriminator in addition to the image data. Then, training is performed to enable conditioning. (4)DCGAN (Deep Convolutional GAN) DCGAN is a type of CGAN in a broad sense, and the main difference from the original GAN ​​is that it is raw The generator and discriminator networks each have a fully connected layer. The image is generated using convolutional layers and transposed convolutional layers. (5)PGGAN(Progressive Growing GAN) Unlike PGGAN and DCGAN, it gradually increases the resolution of the training data, and In addition, the generator and discriminator networks also have layers added while maintaining their target structure to increase resolution. By increasing the value, an image is generated. (6) BigGAN The generator uses orthogonal normalization and conditionally generates high-resolution images up to 512x512 pixels. The image is generated using the model that is being created. (7) StyleGAN Start with low-resolution training and gradually add layers that support higher resolutions to the model as you train. By using progressive growing, an image is generated. (8) StackGAN By using a multi-stage GAN architecture, the initial stage GAN captures the overall picture, resulting in a lower resolution image. A low-resolution image is generated, and in subsequent stages of the GAN, a higher-resolution image is generated. (9) AttnGAN By paying attention to individual words in textual descriptions, such as image captions, The image is generated by combining details from various sub-regions of the image. The method described above is just one example of a GAN; other GAN methods may also be used.

[0030] The image generation unit 104 executes predetermined GANs to respond to the image generation request. An image is generated. For example, the image generation unit 104 generates noise and the noise generator 1 Input to 06. The generator 106 receives the input noise or is included in the image generation request. An image is generated using the data. The noise can be random numbers. For example, generator 106 generates an image using the above data. Even if a neural network having any of the aforementioned GANs with a predetermined structure is used good.

[0031] The discriminator 108 uses the first image in raster format generated by the generator 106, and / or The correct image stored in DB110 is retrieved. The classifier 108 determines if the input image is generated by generator 1 It determines whether the image was generated by 06 or is a ground truth image. For example, classifier 10 8 is a neural network among the GANs described above that has a predetermined structure corresponding to the generator 106. A network may be used. Note that the correct image is not necessarily stored in DB110. It does not need to be a specific image, but an image that satisfies the conditions specified in the image generation request. That's good too.

[0032] The image generation unit 104 distinguishes from the generator 106 based on the authenticity determination result by the classifier 108. The parameters for the instrument 108 are updated. For example, the image generation unit 104 receives input from the classifier 108. To correctly determine the authenticity of an image, backpropagation is used. The parameters of the classifier 108 are updated. Also, the image generation unit 104 updates the parameters of the classifier 108. The backpropagation method is used to determine that the image generated by generator 106 is the correct image. The parameters of the generator 106 are updated using this. The image generation unit 104 then generates the final generated The image is output to the first image conversion unit 112.

[0033] The first image conversion unit 112 converts the final image (first image) generated by the image generation unit 104. The first image conversion unit 112 obtains the image and converts it into a vector image (second image). Regarding the conversion method from star format to vector format, any known method can be used. For example, a tool for converting from raster format to vector format is Vectorizer. io, Vector Magic, Autotracer.org, Convertio Examples include pngtosvg.com. The first image conversion unit 112 converts the second vector format. The image is output to the second image conversion unit 114 or the classifier 108.

[0034] The second image conversion unit 114 converts the vector format generated by the first image conversion unit 112 into a second image. Convert the image to a raster image (third image). For example, the second image conversion unit 114, The process of displaying a vector image on the screen is executed, and from the state where the image is displayed... You can also convert it to raster format. Alternatively, you can convert it to SVG (Scalable Vector Graphics) format. If it is an image, it can be displayed on the screen by the browser, and it will be displayed on the screen. The image in this form may be converted into a third image in raster format. The second image conversion unit 114, Any known method can be used for converting to vector format. For example, The tool for converting from raster format to svgtopng is available at svgtopng.com, Adobe. Examples include Photoshop Express. The second image conversion unit 114 is raster type The third image of the equation is output to the classifier 108.

[0035] The discriminator 108 receives the second image in vector format output by the first image conversion unit 112, and The system inputs a third image in raster format output by the second image conversion unit 114, and the identification result Obtain it.

[0036] When a third image is input to classifier 108, classifier 108 performs a model-independent metaphysical analysis. Xi(Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks: https: / / a You can also run rxiv.org / abs / 1703.03400). The discriminator 108 will process data with different formats. By performing meta-learning, which allows classification learning even in data, raster images and data can be used. This makes it possible to classify the authenticity of images in image format and other formats using the same learning model.

[0037] The determination unit 116 uses the identification result of the second or third image output by the classifier 108, The identification result of the first image, which is the source of the second image, is obtained, and it is determined whether the identification results of both are the same. This determines whether the first image is true or false. For example, the identification result of the first image is the same as the correct image, i.e., true (Tr It is assumed that it is identified as ue). Therefore, the determination unit 116 is determined by the classifier 108 The system determines whether the identification result of the outputted second or third image is true (the same as the correct image). It may be determined. If the determination result is positive, the determination unit 116 will determine the second image in vector format. The memory unit 118 is instructed to store the information.

[0038] The memory unit 118 stores the second image that has been determined to be positive by the determination unit 116. This occurs when the identification result of the second or third image output by the classifier 108 is true (same as the correct image). It is determined that it is (i), or the second or third image output by the classifier 108. This means that the image recognition result and the recognition result of the first image are determined to be the same.

[0039] Furthermore, when determining authenticity, the discriminator 108 uses the correct images registered in DB110, and the images The predetermined image specified by the generation request is identified as true, but the second or third image If this is entered, the first image may be treated as true to determine authenticity.

[0040] The output unit 120 sends the image generation request to the information processing device 20, which sent the image generation request, and the storage unit 11 The second image stored in 8 in vector format may be output. Also, the output unit 120 may output the second The information processing device 20 that sent the image acquisition request may also output this information.

[0041] Through the above processing, the predetermined raster image generated by the AI ​​takes conversion errors into consideration. Because it has been converted to vector format, even if the size of this given image is changed, This prevents image quality degradation. For example, if an image generated by AI is in a virtual space (for example) When sold (on the metaverse), it is possible to sell images without any loss of quality even when resized. It will become.

[0042] Here, due to the conversion error from raster image to vector image, the authenticity determination is incorrect (F If it is determined to be (alse), or if the identification result of the classifier 108 is different before and after the conversion, When this is the result, it would be good to generate a vector image that takes into account how it will be displayed. At this time, when updating vector images, use raster images to create vectors. There are two examples: Example 1, which updates an image in a specific format, and Example 2, which directly updates an image in a vector format. This will be explained below.

[0043] [Example 1] In Example 1, as described above, a vector image was modified using a raster image. The image generation unit 104 uses GANs to generate a new first image (update). The block configuration shown in Figure 3 corresponds to the block configuration of Example 1.

[0044] In Example 1, the determination unit 116 determines if the identification result of the second or third image is false. ) or if the identification result of the second or third image and the first image are input to the classifier If the identification result differs from the one obtained previously, noise is added to the generator 106 to generate a new image. Instruct them to do so.

[0045] The generator 106 generates an image using noise, and the discriminator 108 receives the generated by the generator 106. The authenticity of the selected image is determined. The image that is judged to be authentic here is the updated first image. The output is sent to the first image conversion unit 112, and the updated first image is converted into the second image. Then, the second image, or the third image obtained by transforming the second image, is input to the classifier 108. The identification result is output to the determination unit 116. In the determination unit 116, if the result of the classifier 108 is true The processes of image generation, authenticity detection, transformation, and determination are repeated until a determination is made. If the determination unit 116 determines that the result is positive, the second image in vector format is stored in the storage unit 118. It will be done.

[0046] This means that, taking conversion errors into account, the classifier 108 will determine it to be true when displayed. It will be possible to save images in vector format. This will allow the image quality to be preserved when scaling. This makes it possible to maintain the image quality of images generated by image generation AI without any degradation.

[0047] Furthermore, when the determination unit 116 instructs the generator 106 to generate an image, it determines the raster format. Modify the noise according to the conversion algorithm from one image to a second image in vector format. It may include the following: For example, the information processing device 10 converts from raster format to vector format. The relationship between the algorithm and noise is learned using a learning model, and which noise is applied. It is good to know how quickly it will converge. At this time, the determination unit 116 determines the conversion algorithm The first image conversion unit 112 stores information that associates noise with each element, and uses this information to perform Select the noise corresponding to the conversion algorithm used and output it to the generator 106. That's fine.

[0048] This speeds up the convergence time until the determination unit 116 makes a positive determination, and the information processing system This makes it possible to improve the processing efficiency of the device 10.

[0049] <Data Example> Figure 4 shows an example of a saved image according to Example 1. In the example shown in Figure 4, the storage unit 1 An example of an image stored in 18 is shown. For example, a vector image is assigned an image ID. The image ID "0001" is associated with vector data as "Img01". Vector data, for example, is the numerical value of multiple points that are control points. These are curves derived from mathematical formulas, etc.

[0050] Figure 5 shows an example of noise associated with the image conversion algorithm according to Example 1. The information shown in Figure 5 is held, for example, by the determination unit 116. For example, raster type In the case of a conversion algorithm that converts from the formula "PNG" to the vector format "SVG", Noi The code "N01" is associated with it. The relationship between the conversion algorithm and noise shown in Figure 5 is: It may be trained using a learning model.

[0051] <Example of processing> Figure 6 is a flowchart showing an example of processing by the information processing device 10 according to Example 1. In step S102, the image generation unit 104 includes a generator 106 and a discriminator 108. A raster-formatted first image is generated using a generative network.

[0052] In step S104, the first image conversion unit 112 generates the image generated by the image generation unit 104. The first image is converted into a second image in vector format. At this time, the image generation unit 104 performs the following: The first image that is finally generated (and determined to be true by the classifier 108) is processed by the first image conversion unit 11 Outputting to 2 would be best.

[0053] In step S106, the first image conversion unit 112 converts the second image to the discriminator 10 Input to 8, or the second image conversion unit 114 converts the second image into a third image in raster format. The converted third image is input to the classifier 108. The classifier 108 then selects the second image or the third image The system performs authenticity verification on the item and obtains the verification result.

[0054] In step S108, the determination unit 116 determines if the identification result of the second image or the third image is true. Either this identification result is the identification result of the first image which will be the basis for the second image by classifier 108. Determine if it is the same as [the other item]. If the result of the determination in step S108 is positive, the process continues. Proceed to step S110. If the result of the judgment in step S108 is negative, return to step S102. Then, the first image is updated by the image generation unit 104.

[0055] In step S110, if the result of the determination unit 116 is positive, the memory unit 118 The second image, in vector format, is stored when a positive result is obtained.

[0056] Through the above process, the first statement determined to be true using a Generative Adversarial Network (GAN) For an image, update the image using the same GANs, taking into account the conversion error to vector format. This makes it possible to save the image as a vector image. The second image in vector format, which has been confirmed by the determination unit 116, is displayed when this second image is shown. In this case, the raster image, which was judged as true by classifier 108, remains unchanged in quality and size. The image quality does not degrade even at times.

[0057] [Example 2] In Example 2, as described above, the vector image is updated directly, but for example, DeepSVG can generate images in a ter format (https: / / proceedings.neurips.cc / paper / 2 Using the generative model of (020 / file / bcf9d6bd14a2095866ce8c950b702341-Paper.pdf), vector Update the image in the specified format. Note that any model that generates vector images will work. The model may be used.

[0058] Figure 7 shows an example of a processing block of the information processing device 10 according to Example 2. Blocks that perform the same processing as shown in the diagram are denoted by the same reference numeral. The second image generation shown in Figure 7. Section 202 uses a generative model (e.g., DeepSVG) capable of generating vector-format images. A second adversarial system including a second generator 204 and a discriminator 206 that generate a vector-format image. A generative network is constructed. Here, an adversarial generative network generates raster images. The workpiece will be referred to as the first image generation unit 104.

[0059] The determination unit 208 uses the classifier 206 of the second image converted by the first image conversion unit 112 to determine the second image. If the identification result is not true, the second generator 204 will generate an image in vector format. Give instructions.

[0060] The second generator 204 inputs the vector image generated using the generative model to the discriminator 206. Here, the second image generation unit 202 has an enemy generator 204 and a discriminator 206. A pair generative network is used to generate high-quality vector images. Here, classifier 20 Step 6 performs the meta-learning described above in order to determine authenticity using vector-based images. It has a learning model that does this.

[0061] The determination unit 208 acquires the identification result from the classifier 206 of the second image generation unit 202 and identifies If the result is false, the second generator 204 is instructed to generate an image, and if the identification result is true, the vector The second image in tar format is output for storage in the storage unit 118.

[0062] Through the above process, a second generator 204 generates a vector-format image, and a vector-format A new generative adversarial network is constructed using a discriminator 206 that distinguishes between genuine and counterfeit images. This makes it possible to process the image while it remains in vector format. This reduces the conversion process from raster format to vector format images. This makes it possible to reduce the processing load of the processing device 10.

[0063] <Example of processing> Figure 8 is a flowchart showing an example of processing by the information processing device 10 according to Example 2. For the process related to Example 2, steps S102 to S106 shown in Figure 6 are the same. Since this is a process, the processing is omitted. Figure 8 shows the processing from step S108 onwards.

[0064] In step S108, the determination unit 208 determines if the identification result of the second image or the third image is true. Either this identification result is the identification result of the first image which will be the basis for the second image by classifier 206. Determine if it is the same as [the other item]. If the result of the determination in step S108 is positive, the process continues. Proceed to step S206. If the result of the judgment in step S108 is negative, proceed to step S202. nothing.

[0065] In step S202, the second generator 204 generates an image in vector format. The model is used to generate (update) a new second image.

[0066] In step S204, the determination unit 208 determines that the identification result of the updated second image is true. Determine whether it is true. If the result of the determination in step S204 is positive, the process proceeds to step S206. Proceed to step S204, and if the result of the judgment in step S204 is negative, return to step S202.

[0067] In step S206, if the result of the determination unit 208 is positive, the memory unit 118 The second image, in vector format, is stored when a positive result is obtained.

[0068] Through the above process, a second generator 204 generates a vector-format image, and a vector-format A new generative adversarial network is constructed using a discriminator 206 that distinguishes between genuine and counterfeit images. This makes it possible to process the image while it remains in vector format. This reduces the conversion process from raster format to vector format images. This makes it possible to reduce the processing load of the processing device 10. In Example 2, the first stroke generation unit 10 The raster image generated by a method other than 4 is treated as the true image, and the first image conversion unit 112 and the second Using the image generation unit 202 and the determination unit 208, the vector-format image described above is generated. That's good too.

[0069] Furthermore, as described above, the second image stored in the storage unit 118 of Example 1 or Example 2 , it may be used in a virtual space. Raster images are raster images, vector images The image may also be called a vector image.

[0070] The embodiments described above are provided to facilitate understanding of the present invention and do not limit the invention. This is not intended to provide an explanation. It describes each element of the embodiment, as well as its arrangement, materials, conditions, and shape. The size and other specifications are not limited to those exemplified and can be changed as appropriate. Furthermore, it is possible to partially substitute or combine the configurations shown in different embodiments. ru. [Explanation of Symbols]

[0071] 10, 20... Information processing device, 10a... CPU, 10b... RAM, 10c... ROM, 10d ...Communication unit, 10e...Input unit, 10f...Display unit, 102...Acquisition unit, 104...Image generation unit (No. 1 image generation unit), 106...generator (first generator), 108...discriminator, 110...DB, 11 2...First image conversion unit, 114...Second image conversion unit, 116...Determination unit, 118...Storage unit, 12 0...Output unit, 202...Second image generation unit, 204...Second generator, 206...Discriminator, 208... Judgment section

Claims

1. An information processing method executed by an information processing device having a processor, comprising: The processor, Converting a first image in raster format into a second image in vector format; inputting the second image into the classifier that regards the first image as true, the classifier being included in a generative adversarial network including a generator and a classifier, and obtaining a classification result; storing the second image in a storage unit if the classification result is true or if the classification result is the same as the classification result of the first image by the classifier; An information processing method for performing the above.

2. The processor, updating the second image using the generator if the classification result is false or if the classification result differs from the classification result of the first image by the classifier; The information processing method according to claim 1 , further comprising the steps of:

3. An information processing device having a processor, The processor, Converting a first image in raster format into a second image in vector format; inputting the second image into the classifier that regards the first image as true, the classifier being included in a generative adversarial network including a generator and a classifier, and obtaining a classification result; storing the second image in a storage unit if the classification result is true or if the classification result is the same as the classification result of the first image by the classifier; An information processing device that executes the above.

4. A processor included in the information processing device Converting a first image in raster format into a second image in vector format; inputting the second image into the classifier that regards the first image as true, the classifier being included in a generative adversarial network including a generator and a classifier, and obtaining a classification result; storing the second image in a storage unit if the classification result is true or if the classification result is the same as the classification result of the first image by the classifier; A program that executes the following.

Images