Information processing system, information processing method, and program

Abstract

[Problem] To realize a more convenient watermark. [Solution] Provided is an information processing system comprising: an acquisition unit that acquires a partial region of a watermark image comprising a plurality of watermark tiles encoded on the basis of a cover image and prescribed information and arranged in a tile pattern; and a processing unit comprising a decoder that decodes the prescribed information from the partial region.

Description

Information processing system, information processing method, and program

[0001] This disclosure relates to an information processing system, an information processing method, and a program.

[0002] In recent years, watermark technology, which embeds metadata into the pixel values ​​of an image, has been utilized. Watermark technology makes it possible to detect unauthorized use and leakage of images. Furthermore, as disclosed in Non-Patent Document 1, for example, watermark technology using DNN (Deep Neural Network) has also been developed.

[0003] Tu Bui, et al., "TrustMark: Universal Watermarking for Arbitrary Resolution Images," November 30, 2023, [Online], [Retrieved March 17, 2025], Internet<https: / / arxiv.org / abs / 2311.18297>

[0004] However, in the technology disclosed in Non-Patent Document 1, it is difficult to read (decode) the watermark unless the area in which the watermark is embedded, i.e., the encoded area, and the area to be read, i.e., the decoded area, are not nearly identical.

[0005] According to one aspect of this disclosure, an information processing system is provided, comprising: an acquisition unit that acquires a portion of a watermark image consisting of a plurality of watermark tiles encoded and arranged in a tile-like manner based on a cover image and predetermined information; and a processing unit that includes a decoder that decodes predetermined information from the portion of the image.

[0006] Furthermore, in another aspect of this disclosure, an information processing method is provided in which a processor acquires a portion of a watermark image consisting of a plurality of watermark tiles encoded based on a cover image and predetermined information and arranged in a tile-like manner, and decodes predetermined information from the portion of the image.

[0007] Furthermore, according to another aspect of this disclosure, a program is provided that causes a computer to function as an information processing device, comprising a processing unit that includes a decoder for decoding predetermined information from a portion of a watermark image consisting of a plurality of watermark tiles encoded and arranged in a tile-like manner based on a cover image and predetermined information.

[0008] This figure illustrates an overview of an information processing method according to one embodiment of the present disclosure. The watermark tile R according to the same embodiment. ｉ and Decode area A Ｄ This is a diagram illustrating the comparison with 1 to 4. This is a block diagram illustrating an example of the functional configuration of the information processing system 1 according to the same embodiment. This is a flowchart illustrating an example of the first learning flow according to the same embodiment. This is a diagram illustrating the parameters that the shift transform network incorporated into the decoder 141 according to the same embodiment learns and infers. This is a diagram illustrating the forward propagation of the encoder 131 and the generation of the watermark image S according to the same embodiment. This is a diagram illustrating cropping of a part of the region and the forward propagation of the decoder 141 according to the same embodiment. This is a diagram illustrating an example of the second learning flow according to the same embodiment. This is a diagram illustrating an example of processing by the extension module 145 according to the same embodiment. This is a diagram illustrating an example of processing by the extension module 145 according to the same embodiment. This is a diagram illustrating an example of processing by the extension module 145 according to the same embodiment. This is a diagram illustrating the generation of a watermark image S using the encoder 132 that has undergone the second learning according to the same embodiment. This is a diagram illustrating the decoding of predetermined information using the decoder 142 that has undergone the second learning according to the same embodiment. This is a watermark image S generated by the information processing method according to the same embodiment. ３６０ This figure shows an example. The watermark image S generated by the information processing method according to this embodiment. Ｃ This figure shows an example. The watermark image S generated by the information processing method according to this embodiment. ＵＰThis figure shows an example. This figure shows an example of applying the watermark image S according to the same embodiment to a 3D asset. This figure illustrates a watermark image S generation service using the information processing method according to the same embodiment. This figure illustrates a decoding service using the information processing method according to the same embodiment. This is a block diagram illustrating an example of the hardware configuration of the information processing device 90 according to the same embodiment. This figure illustrates an overview of the watermark technology disclosed in Non-Patent Literature 1. Encoding area A in the prior art Ｅ and Decode area A Ｄ This is a diagram illustrating an example. It is a diagram intended to explain the parameters that the original STN learns and infers.

[0009] Preferred embodiments of this disclosure will be described in detail below with reference to the attached drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0010] Furthermore, in this specification and drawings, when describing multiple identical components to distinguish them, letters or other symbols may be added to the end of the reference numerals. On the other hand, when there is no need to distinguish multiple identical components, the letters or other symbols may be omitted, and a description common to all identical components may be provided.

[0011] The explanation will be presented in the following order: 1. Embodiments 1.1. Overview 1.2. Example Functional Configuration 1.3. First Learning 1.4. Second Learning 1.5. Application Examples 2. Example Hardware Configuration 3. Summary

[0012] <1. Embodiments> <<1.1. Overview>> First, an overview of the information processing system 1 according to one embodiment of this disclosure will be described.

[0013] In recent years, watermarking technology, which embeds meta information such as the copyright holder, licensee, content ID, and whether copying is permitted into an image (still image or moving image) and links the meta information to the image in an inseparable format, has been utilized. According to this technology, it becomes possible to detect unauthorized use, leakage, and the like of images (content that uses images), and thus it is possible to protect the interests of copyright holders and other parties.

[0014] Furthermore, as disclosed in Non-Patent Document 1, for example, watermarking technology utilizing DNN has also been developed.

[0015] FIG. 21 is a diagram for explaining an overview of the watermarking technology disclosed in Non-Patent Document 1.

[0016] In the watermarking technology disclosed in Patent Document 1, first, a cover image C with an arbitrary aspect ratio ０ having a shape (H0, W0) is resized to the input shape (H1, W1) of a neural network to generate a cover image C1.

[0017] Subsequently, a watermark component W is generated based on the cover image C1 and a message to be embedded １ After calculating , the watermark component W １ is resized to the shape (H0, W0) of the cover image C ０ to obtain a watermark component W ０ .

[0018] Thereafter, the watermark component W ０ is applied to the cover image C ０ to generate a watermarked image.

[0019] According to the above-described technology, a watermarked image compatible with any aspect ratio can be generated, and high robustness against changes such as deformation and rotation can be achieved by using a neural network.

[0020] However, in conventional watermarking technologies including the technology disclosed in Non-Patent Document 1, it is difficult to read the watermark if the encoding region and the decoding region do not substantially match.

[0021] Figure 22 shows the encoding region A in the conventional technology. Ｅ and Decode area A Ｄ This is an example shown in the figure.

[0022] As shown in Figure 22, in the case of the conventional technology, encoding region A is generally Ｅ The area (indicated by the thick dashed line) is the same size as the entire watermark image S.

[0023] Meanwhile, decode region A Ｄ The area (shown by a thick solid line) could be, for example, an image taken by a camera or an image captured by a computer. However, as illustrated in Figure 22, the decoding area A Ｄ Encoded area A Ｅ If all of these are not included, it is difficult to read the watermark using conventional technology.

[0024] Therefore, when reading the watermark using conventional technology, encoding area A Ｅ Decode region A to include all of Ｄ It is required to obtain this.

[0025] However, in cases where a watermark is encoded across the entire content (image), such as in vertically scrolling manga content or VP (Virtual Production) content, encoding area A Ｅ Decode region A, which includes all of the above. Ｄ It may be difficult to photograph or capture the image.

[0026] In particular, in video production (VP), it is expected that there will be obstructions such as performers, props, and set pieces between the watermark image and the camera, making it more difficult to read the watermark.

[0027] An information processing method according to one embodiment of this disclosure was conceived with the above-mentioned points in mind, and it enables a more convenient watermark while supporting any aspect ratio.

[0028] Figure 1 is a diagram illustrating an overview of an information processing method according to one embodiment of the present disclosure.

[0029] An example of a watermark image S according to this embodiment is shown on the left side of Figure 1. Here, the watermark image S is an image generated by embedding a watermark into an arbitrary cover image C.

[0030] Furthermore, the right side of Figure 1 illustrates a tile set R according to this embodiment. The tile set R according to this embodiment consists of a plurality of watermark tiles R encoded and arranged in a tile pattern based on a cover image C and predetermined information (for example, arbitrary metadata). ｉ One of its characteristics is that it consists of [this].

[0031] In the information processing method according to this embodiment, the tile set R is applied to the cover image C, that is, the watermark tile R is applied to the cover image C. ｉ A watermark image S is generated by arranging these elements in a tile pattern.

[0032] In the watermark image S shown on the left side of Figure 1, each of the rectangles indicated by thin lines is a watermark tile R. ｉ It corresponds to.

[0033] Furthermore, on the left side of Figure 1 is the decoded area A, which is captured by a camera or computer. Ｄ Examples 1 through 4 are given.

[0034] Figure 2 shows the watermark tile R. ｉ and Decode area A Ｄ This diagram is intended to explain the comparison with 1-4.

[0035] The first row of Figure 2 shows Watermark Tile R. ｉ and Decode area A Ｄ 1 is given as an example. Here, the decode region A Ｄ 1. Coincidentally, the size and position of the watermark tile R ｉ The image must match.

[0036] In this case, decode area A Ｄ1. Watermark Tile R ｉ Since it is completely contained, decode region A Ｄ 1 = Watermark Tile R ｉ It is possible to decode the specified information from it.

[0037] Also, in the second row of Figure 2, there is a watermark tile R. ｉ and Decode area A Ｄ Examples 2 and 3 are given. Here, the decoding region A Ｄ 2 is Watermark Tile R ｉ They are the same size and multiple watermark tiles R ｉ The image may also include parts of each of them.

[0038] The information processing method according to this embodiment, for example, learns using data augmentation related to two-axis shift described later, thereby enabling the watermark tile R ｉ Perform the corresponding alignment process and decode area A Ｄ This enables decoding of the specified information from 2.

[0039] Also, in the third row of Figure 2, there is a watermark tile R. ｉ and Decode area A Ｄ Three examples are given. Here, the decoding region A Ｄ 3 is Watermark Tile R ｉ Smaller in size, with multiple watermark tiles R ｉ The image may contain parts of each of them.

[0040] The information processing method according to this embodiment learns using data augmentation related to two-axis shifting and data augmentation related to cropping and resizing, as described later, to decode area A Ｄ This enables decoding of the specified information from 3.

[0041] Also, in the fourth row of Figure 2, there is a watermark tile R. ｉ and Decode area A Ｄ 4 is given as an example. Here, the decode region A Ｄ 4 is Watermark Tile R ｉLarger in size, with multiple watermark tiles R ｉ The image may include parts or all of each of them.

[0042] The information processing method according to this embodiment learns using data augmentation related to two-axis shifting and data augmentation related to cropping and resizing, as described later, to decode area A Ｄ This enables decoding of the specified information from 4.

[0043] <<1.2. Example of Functional Configuration>> Next, an example of the functional configuration of the information processing system 1 that executes the information processing method according to this embodiment will be described. Figure 3 is a block diagram showing an example of the functional configuration of the information processing system 1 according to this embodiment.

[0044] As shown in Figure 3, the information processing system 1 according to this embodiment includes at least an acquisition unit 110, a learning unit 120, an encoding processing unit 130, a decoding processing unit 140, and a storage unit 150.

[0045] (Acquisition unit 110) The acquisition unit 110 according to this embodiment acquires various types of information used in the learning unit 120, the encoding processing unit 130, and the decoding processing unit 140, respectively.

[0046] The various types of information mentioned above include the cover image C, predetermined information (such as arbitrary metadata), and a portion of the watermark image S (i.e., the decoded region A). Ｄ Examples include:

[0047] For example, the acquisition unit 110 is a watermark tile R ｉ The cover image C used in its generation and predetermined information may be acquired based on user input. In this case, the acquisition unit 110 includes various input devices to accept such input.

[0048] Furthermore, for example, the acquisition unit 110 is a watermark tile R ｉ Cover image C and predetermined information used in the generation, and decoding region A ＤThe data may also be obtained from a user terminal or the like used by the user. In this case, the acquisition unit 110 includes a communication device that communicates with the user terminal or the like. In this case, the user terminal can also be considered an example of an acquisition unit.

[0049] Furthermore, for example, the acquisition unit 110 is the decode area A Ｄ The data may be acquired by photography. In this case, the acquisition unit 110 may be equipped with a camera.

[0050] (Learning Unit 120) The learning unit 120 according to this embodiment learns the encoder and decoder, which will be described later.

[0051] The learning unit 120 according to this embodiment is composed of, for example, an integrated circuit such as a processor.

[0052] (Encoding Processing Unit 130) The encoding processing unit 130 (sometimes referred to as the generation unit) according to this embodiment uses the encoder that has been trained by the learning unit 120 to generate watermark tiles R ｉ The generation process is performed.

[0053] The encoding processing unit 130 according to this embodiment is composed of, for example, an integrated circuit such as a processor.

[0054] (Decode Processing Unit 140) The Decode Processing Unit 140 (sometimes simply referred to as the processing unit) according to this embodiment uses the decoder, which has been trained by the learning unit 120, to decode area A Ｄ Decode the specified information from it.

[0055] The decoding processing unit 140 according to this embodiment applies a watermark tile R to the decoding area. ｉ One of its features is that it performs alignment processing based on and decodes predetermined information.

[0056] The decoding processing unit 140 according to this embodiment is composed of, for example, an integrated circuit such as a processor.

[0057] (Storage Unit 150) The storage unit 150 according to this embodiment stores various types of information used in the learning unit 120, the encoding processing unit 130, and the decoding processing unit 140, respectively.

[0058] Furthermore, the storage unit 150 stores the results of encoding by the encoding processing unit 130 (i.e., the watermark tile R ｉ The decoding result (i.e., predetermined information) by the decoding processing unit 140 (e.g., watermark image S) is stored.

[0059] The above describes an example of the functional configuration of the information processing system 1 according to this embodiment. Note that the above functional configuration explained using Figure 3 is merely an example, and the functional configuration example of the information processing system 1 according to this embodiment is not limited to this example.

[0060] For example, the information processing system 1 according to this embodiment may further include a display unit or the like.

[0061] Furthermore, the information processing system 1 according to this embodiment may be implemented by a single device or by multiple devices.

[0062] The functional configuration of the information processing system 1 according to this embodiment can be flexibly modified according to specifications, operation, etc.

[0063] <<1.3. First Learning>> Next, the first learning according to this embodiment and the processing using the encoder and decoder after the first learning will be described.

[0064] Figure 4 is a flowchart showing an example of the first learning flow according to this embodiment.

[0065] In the example shown in Figure 4, the model definition is performed first (S101).

[0066] In step S101, the definitions of each network, such as the encoder and decoder, may be performed.

[0067] In particular, one of the features of the first learning process according to this embodiment is that a decoder 141 (see Figure 7) is defined that incorporates a network capable of acquiring shift transformation capabilities for input images through learning (hereinafter also referred to as a shift transformation network). An example of such a network is STN (Spatial Transformer Networks).

[0068] Figure 23 is a diagram illustrating the parameters that the original STN learns and infers.

[0069] As shown in Figure 23, the original STN uses the parameter (θ) of the 6-dimensional affine transform applied to the input. １１ , θ １２ , θ １３ , θ ２１ , θ ２２ , θ ２３ It learns and infers ).

[0070] On the other hand, Figure 5 is a diagram illustrating the parameters that the shift transform network incorporated into the decoder 141 according to this embodiment learns and infers.

[0071] On the left side of Figure 5 is the decode region A according to this embodiment. Ｄ An example of an input corresponding to this is shown. Decode area A according to this embodiment Ｄ This is the same size watermark tile R ｉ This could be a portion of the watermark image S, which is arranged in a tile-like pattern.

[0072] In this case, decode area A Ｄ θ cyclically in the X-axis direction ｘ By shifting only the watermark tile R ｉ This allows the pixels in the X-axis direction to be aligned. Similarly, decode region A Ｄ θ cyclically in the Y-axis direction ｙ By shifting only the watermark tile R ｉ This allows the pixels in the Y-axis direction to be aligned.

[0073] Therefore, the shift conversion network incorporated in the decoder 141 according to this embodiment uses two-dimensional parameters (θ) corresponding to the shift amount in the X-axis direction and the shift amount in the Y-axis direction. ｘ , θ ｙ It learns and infers ).

[0074] Based on the learning and reasoning described above, the decoding region A Ｄ Watermark tile R ｉThe alignment processing based on [the foregoing] and decoding can be implemented by a single decoder 141.

[0075] With reference again to FIG. 4, the description of the first learning process continues below.

[0076] After model definition in step S101, preparation of training images (S102) and acquisition of a mini-batch by the learning unit 120 (S103) are performed.

[0077] Next, the learning unit 120 performs forward propagation of the encoder (S104) and generation of the watermark image S (S105).

[0078] FIG. 6 is a diagram for explaining forward propagation of the encoder 131 and generation of the watermark image S according to the present embodiment.

[0079] As shown in FIG. 6, the encoder 131 according to the present embodiment performs learning to output a watermark tile R ｉ based on the input cover image C and predetermined information M.

[0080] Furthermore, by tiling the watermark tile R output by the encoder 131 ｉ onto the cover image C, a watermark image S is generated. That is, the watermark image S is generated by arranging a plurality of identical watermark tiles R ｉ in a tiled pattern on the cover image C.

[0081] Note that the above method for generating the watermark image S is merely an example, and the method for generating the watermark image S according to the present embodiment is not limited to this example.

[0082] For example, the encoder 131 according to the present embodiment may encode input predetermined information M to output a watermark R ｉ . In this case, a watermark density image D generated based on a tile set R formed by combining watermark R ｉ and the cover image C is generated, and the watermark image S may be generated by satisfying S = C + D*R.

[0083] Here, the watermark density image D is an image that determines the strength of a watermark, and may be applied collectively to the entire cover image C, or may be applied for each watermark R ｉ . According to such a watermark density image D, it is possible to adjust the strength of the watermark and the similarity with the cover image C according to the position of the watermark image S.

[0084] Referring again to FIG. 4, the description of the first learning flow will be continued.

[0085] After generating the watermark image S in step S105, the learning unit 120 performs cropping of a partial region from the watermark image S (S106) and forward propagation of the decoder 141 (S107).

[0086] FIG. 7 is a diagram for explaining cropping of a partial region and forward propagation of the decoder 141 according to the present embodiment.

[0087] As shown in FIG. 7, a partial region is cropped from the watermark image S to obtain a cropped image CR. The cropped image CR is a decode region A input to the decoder 141 in decoding using the trained decoder 141 Ｄ This is an image corresponding to

[0088] Further, the decoder 141 outputs (decodes) predetermined information M based on the input cropped image CR Ｄ .

[0089] Here, the cropping position from the watermark image S may be random. By randomizing the cropping position, decoding with high robustness can be realized with respect to the position of the decode region A Ｄ with respect to the watermark image S.

[0090] Further, the crop size from the watermark image S is not limited to the same size as the watermark tile R ｉ , and is not limited to a size smaller than the watermark tile R ｉ , and the watermark tile R ｉIt may include sizes larger than the specified limit. The upper and lower limits of the crop size should be appropriately designed based on the desired decoding accuracy.

[0091] Watermark Tile R ｉ By training the decoder 141 to decode based on the CR of a cropped image that is randomly cropped to a size smaller than the size of the watermark tile R, the decoder 141 that has undergone this training will be able to decode the watermark tile R. ｉ Smaller size decoding area A Ｄ From the specified information M Ｄ It becomes possible to decode it.

[0092] Also, Watermark Tile R ｉ By training the decoder 141 to decode based on the CR of a cropped image that is randomly cropped to a size larger than the size of the watermark tile R, the decoder 141 that has undergone this training will be able to decode the watermark tile R. ｉ Larger size decoding area A Ｄ From the specified information M Ｄ It becomes possible to decode it.

[0093] Let's continue explaining the first learning process by referring to Figure 4 again.

[0094] After the forward propagation of the decoder 141 in step S107, the learning unit 120 performs loss calculation (S108), backpropagation (S109), and weight update (S110).

[0095] In step S108, the learning unit 120 determines the loss related to the cover image C and the watermark image S, and the predetermined information M input to the encoder 131 and the predetermined information M output by the decoder 141. Ｄ Calculate the losses related to that.

[0096] The learning unit 120 may calculate the loss between the cover image C and the watermark image S using a loss function that is trained to increase the similarity between images, such as MSE (Mean Squared Error), PSNR (Peak Signal-to-Noise Ratio), LPIPS (Learned Perceptual Image Patch Similarity), or FFT (Fast Fourier Transformation).

[0097] Furthermore, the learning unit 120 learns a predetermined information M input to the encoder 131 and a predetermined information M output by the decoder 141 using a loss function that is learned to match messages such as BCE (Binary Cross-Entropy). Ｄ You may also calculate the losses related to that.

[0098] Next, the learning unit 120 performs a convergence determination (S111). If the learning unit 120 determines that the learning has not converged (S111: NO), it returns to step S103.

[0099] On the other hand, if the learning unit 120 determines that the learning has converged (S111: YES), it terminates the series of processes.

[0100] The first learning process according to this embodiment has been described above. By generating a watermark image S using the encoder 131 that has undergone the first learning process, and decoding it using the decoder 141 that has undergone the first learning process, the decoding region A is much smaller than that of the watermark image S. Ｄ It becomes possible to decode specific information from this.

[0101] Furthermore, by generating a watermark image S using the encoder 131 that has undergone the first training, and decoding it using the decoder 141 that has undergone the first training, the watermark tile R ｉ Decode region A that is larger / smaller than Ｄ It becomes possible to decode specific information from this.

[0102] <<1.4. Second Learning>> Next, the second learning according to this embodiment and the processing using the encoder and decoder after the second learning will be described.

[0103] In the following, we will mainly explain the differences between the first and second learning processes, and will omit detailed explanations of processes common to both the first and second learning processes.

[0104] The first learning described above uses a decoder 141 incorporating a shift transform network to decode region A Ｄ Watermark tile R ｉ One of its features is that it performs alignment processing and decoding based on [a specific method / framework].

[0105] In contrast, the second learning according to this embodiment involves the watermark tile R ｉ A center position estimation module that estimates the center position of the watermark tile R ｉ The system may include an alignment module that performs alignment processing based on the center position of the system.

[0106] In other words, one of the features of the second learning according to this embodiment is that it performs alignment more clearly compared to the first learning.

[0107] Furthermore, in the second learning according to this embodiment, watermark tile R ｉ The watermark components for each item do not necessarily have to be the same.

[0108] Figure 8 is a diagram illustrating an example of the second learning flow according to this embodiment.

[0109] In the second learning stage, the encoder 132 generates a watermark tile R based on the predetermined input information M and cover image C. ｉ The program will perform learning that produces output.

[0110] Furthermore, the encoder 132 may also receive input such as a mask image MSK indicating the center position (CX, CY) of the cover image C, or the center position (CX, CY) itself. This is expected to improve the efficiency of the alignment process described later. However, inputting the above-mentioned information regarding the center position (CX, CY) to the encoder 132 is not mandatory.

[0111] Next, processing is performed by the Augmentation module 145.

[0112] The expansion module 145 has multiple watermark tiles R ｉ Combine the combined watermark tiles R ｉ Random cropping or a clear shift (translation) is then applied to create an augmented image R. Ａ Generates an extended image R. Ａ In decoding using the trained decoder 142, the decoding region A input to the decoder 142 is... Ｄ This is the corresponding image.

[0113] Figures 9 to 11 illustrate an example of processing performed by the extension module 145.

[0114] For example, as shown in the example in Figure 9, the expansion module 145 has multiple identical watermark tiles R ｉ Combine the combined watermark tiles R ｉ Watermark Tile R ｉ By cropping an area of ​​the same size (shown by a thick line) at random positions, an expanded image R is created. Ａ You may generate this.

[0115] Furthermore, as shown in Figure 10, for example, the extension module 145 divides the cover image C into four rectangular regions of the same size and generates four watermark tiles R from each rectangular region. ｉ Combine the combined watermark tiles R ｉ Watermark Tile R ｉ By cropping an area of ​​the same size (shown by a thick line) at random positions, an expanded image R is created. ＡYou may generate this.

[0116] Furthermore, as shown in Figure 11, for example, the extension module 145 generates multiple watermark tiles R from each of the multiple cover images C within the badge. ｉ Combine the combined watermark tiles R ｉ Watermark Tile R ｉ By cropping an area of ​​the same size (shown by a thick line) at random positions, an expanded image R is created. Ａ You may generate this.

[0117] Generated augmented image R Ａ This is input to the localization module 146.

[0118] The localization module 146 according to this embodiment is an example of a center position estimation module.

[0119] The localization module 146 receives the input extended image R Ａ Based on Watermark Tile R ｉ Estimate the central position of [the object].

[0120] For example, if the shift amount by the extension module is K = (Δa, Δb), the extended image R Ａ The center position can be expressed as (CX + Δa, CY + Δb). In this case, the localization module 146 may output the estimated shift amount, K' = (Δa', Δb').

[0121] Watermark tile R estimated by localization module 146 ｉ Information regarding the central position, and the expanded image R Ａ This is input to the registration module 147.

[0122] The registration module 147 according to this embodiment is an example of an alignment module.

[0123] The registration module 147 according to this embodiment is a watermark tile R ｉ Based on information regarding the central position, the expanded image R ＡWatermark Tile R included ｉ By performing a shift process so that the center position becomes the center of the image, the alignment image R Ｒ Obtain it.

[0124] Next, the decoder 142 receives the input alignment image R Ｒ Based on the predetermined information M Ｄ The program will perform learning that produces output.

[0125] The above describes the processing flow corresponding to steps S104 to S107 in the first learning process for the second learning process.

[0126] In the second learning phase, the learning unit 120 then performs loss calculation, backpropagation, weight update, and convergence determination.

[0127] In the second learning stage, the learning unit 120 learns the cover image C and the watermark tile R. ｉ Loss related to (LOSS 1), loss related to the shift amount by the expansion module 145 and the shift amount estimated by the expansion module 145 (LOSS 2), and predetermined information M input to the encoder 132 and predetermined information M output by the decoder 142. Ｄ Calculate the loss related to (LOS 3).

[0128] The learning unit 120 may calculate the LOSS 1 relating to the cover image C and the watermark image S using a loss function that is learned to increase the similarity between images, such as MSE, PSNR, LPIPS, or FFT.

[0129] Alternatively, the learning unit 120 may calculate LOSS 3 using a loss function that is learned to match messages such as BCE.

[0130] Furthermore, the learning unit 120 may calculate LOSS 2 using a loss function that is learned to match output values ​​such as MSE. L2.

[0131] Furthermore, the loss shown as LOSS 2, which relates to the shift amount by the expansion module 145 and the shift amount estimated by the expansion module 145, is due to the watermark tile R included in a portion of the input to the expansion module 145. ｉ Information regarding the center position and the watermark tile R output by the expansion module 145 ｉ This is an example of information regarding the central position and the associated losses.

[0132] Furthermore, the expansion module 145 adds rotation, scaling, and other noise to the expanded image R in addition to shifting. Ａ You may generate this.

[0133] Furthermore, the localization module 146 uses technologies such as Keypoint Detection and Object Detection to create watermark tiles R. ｉ The central position may be estimated.

[0134] Thus, the data augmentation and center position estimation in the second learning process are flexibly adaptable.

[0135] Next, the generation of a watermark image S using the encoder 132 that has undergone a second training process will be explained. Figure 12 is a diagram illustrating the generation of a watermark image S using the encoder 132 that has undergone a second training process according to this embodiment.

[0136] The encoding processing unit 130 first processes the cover image C into multiple patches C of a predetermined size (H x W). ｉ Divide it into parts.

[0137] Next, the encoding processing unit 130 processes multiple patches C ｉ Each of these, along with the predetermined information, is input to the encoder 132.

[0138] The encoder 132 generates multiple patches C based on the input information. ｉ Multiple watermark tiles R corresponding to each of them ｉ (The illustration is omitted in Figure 12) is output.

[0139] Note that the encoder 132 has patch C ｉA mask image MSK1 indicating the center position may also be input. Using the mask image MSK1 is expected to improve the accuracy of the center position estimation performed by the decoding processing unit 140.

[0140] Furthermore, the encoder 132 has patch C ｉ A mask image MSK2 showing the peripheral region may also be input. According to the mask image MSK2, it is possible to prevent a watermark from being encoded in the peripheral region, and consequently, the watermark tile R ｉ It is expected to have the effect of making the boundary less noticeable.

[0141] However, inputting mask images MSK1 and MSK2 to encoder 132 is not mandatory.

[0142] The encoding processing unit 130 processes multiple watermark tiles R output by the encoder 132. ｉ The two elements are combined to generate a watermark image S.

[0143] Next, the decoding of predetermined information using the decoder 142 after the second learning process will be described. Figure 13 is a diagram illustrating the decoding of predetermined information using the decoder 142 after the second learning process according to this embodiment.

[0144] The decoding processing unit 140 obtains the decoded area A by taking or capturing the watermark image S. Ｄ This is input into the localization module 146.

[0145] The localization module 146 receives the input decode area A Ｄ Watermark Tile R included ｉ Estimate the central position of [the object].

[0146] In this example, watermark tile R ｉ Let Ki = (Δai, Δbi) be the center position.

[0147] Decode area A Ｄ Watermark Tile R ｉThere may be multiple center positions Ki. In this case, the localization module 146 may estimate each of the multiple center positions Ki.

[0148] In the example shown in Figure 13, the localization module 146 is a watermark tile R １ The center position K1 = (Δa1, Δb1), and the watermark tile R ２ The central positions K2 = (Δa2, Δb2) are estimated respectively.

[0149] The decoding processing unit 140 controls the decoding area A Ｄ The localization module 146 estimates the center position Ki, which is then input to the registration module 147.

[0150] The registration module 147 decodes region A based on the input center position Ki. Ｄ Alignment processing is performed on the image R, and the aligned image R is created. Ｒ Outputs.

[0151] Here, if multiple center positions Ki are input, the registration module 147 selects the center position Ki (K) that is closer to the center in the watermark image S from among the multiple center positions Ki. Ｃｅｎｔｅｒ Alignment processing may be performed based on the above.

[0152] In the example shown in Figure 13, the registration module 147 selects the center position K1 = (Δa1, Δb1) that is closer to the center in the watermark image S from the input center positions K1 = (Δa1, Δb1) and center positions K2 = (Δa2, Δb2). Ｃｅｎｔｅｒ Alignment processing is performed.

[0153] The above process minimizes the loss of information associated with the alignment process.

[0154] Furthermore, if multiple center positions Ki are input, the registration module 147 may calculate the scale, rotation angle, etc., based on the relationship between the multiple center positions Ki, and perform alignment processing based on the calculated scale, rotation angle θ, etc.

[0155] However, in this case, the watermark tile R when generating the watermark image S ｉ Assume that the distance D between the center positions Ki of the elements is known.

[0156] For example, the registration module 147 is K Ｃｅｎｔｅｒ Fix K Ｃｅｎｔｅｒ Select the closest Ki, K Ｃｅｎｔｅｒ The distance d between the selected Ki and the registration module 147 is calculated. The registration module 147 can determine the scale based on the known distance D and the calculated distance d.

[0157] Furthermore, the registration module 147 is K Ｃｅｎｔｅｒ After moving it to the center of the image, K Ｃｅｎｔｅｒ The angle between the line connecting the selected Ki and the X-axis (or Y-axis), i.e., the rotation angle θ, is calculated.

[0158] By performing the above-described process, it is possible to achieve more accurate alignment by performing an affine transformation based on the calculated scale and rotation angle θ.

[0159] Furthermore, the registration module 147 may perform alignment processing based on other information, such as the printing direction of the special printed material on which the watermark image S is printed.

[0160] The decoding processing unit 140 processes the alignment image R output by the registration module 147. Ｒ This is input to decoder 142.

[0161] Decoder 142 receives the input alignment image R Ｒ Based on the predetermined information M Ｄ Decode it.

[0162] The second learning process according to this embodiment has been described above. By generating a watermark image S using the encoder 132 that has undergone the second learning process, and decoding it using the decoder 142 that has undergone the second learning process, the decoding region A is much smaller than that of the watermark image S. Ｄ It becomes possible to decode specific information from this.

[0163] Furthermore, by generating a watermark image S using the encoder 132 that has undergone a second training, and decoding it using the decoder 142 that has undergone a second training, the watermark tile R ｉ Decode region A that is larger / smaller than Ｄ It becomes possible to decode specific information from this.

[0164] <<1.5. Application Examples>> Next, specific examples of applications of the information processing method according to this embodiment will be explained.

[0165] For example, the information processing method according to this embodiment may be used to encode a watermark onto a so-called 360° image.

[0166] A 360° image is an image that records the surrounding 360° view from a specific point. Viewers can adjust the angle to see the scene at a predetermined field of view.

[0167] Figure 14 shows a watermark image S generated by the information processing method according to this embodiment. ３６０ This is an example shown in the figure.

[0168] Figure 14 shows a watermark image S generated by the information processing method according to this embodiment. ３６０ , and decoding area A Ｄ Examples are given.

[0169] Watermark image S ３６０ This is the cover image C, which is a 360° image. ３６０ (Illustration omitted) This is a 360° image in which predetermined information M has been encoded.

[0170] Due to its characteristics, it is difficult to capture the entire 360° image with a conventional camera. Therefore, decoding region A Ｄ As shown in the illustration, watermark image S ３６０ It remains within a portion of the whole.

[0171] Conventionally, decoding watermarks was difficult in the situations described above, but according to the information processing method of this embodiment, it becomes possible to encode watermarks into 360° images and decode watermarks from 360° images.

[0172] Furthermore, for example, the information processing method according to this embodiment may be used to encode watermarks on various types of identification photographs.

[0173] Figure 15 shows a watermark image S generated by the information processing method according to this embodiment. Ｃ This is an example shown in the figure.

[0174] Figure 15 shows a watermark image S generated by the information processing method according to this embodiment. Ｃ , and decoding area A Ｄ Examples are given.

[0175] Watermark image S Ｃ This is the cover image C, which is the identification photograph affixed to the passport. Ｃ (Illustration omitted) This is an image in which the specified information M has been encoded.

[0176] During immigration procedures, passports may be photographed. However, in such cases, close-up photography may only capture a small area, or the photographer may not be able to clearly understand the relationship between the photographed area and the passport's position.

[0177] Therefore, conventionally, decoding watermarks from passport photos was sometimes difficult, but according to the information processing method of this embodiment, encoding watermarks on passport photos and decoding watermarks from passport photos can be easily achieved.

[0178] Furthermore, for example, the information processing method according to this embodiment may be used to encode watermarks on various video content such as anime, dramas, movies, and TV programs.

[0179] Figure 16 shows a watermark image S generated by the information processing method according to this embodiment. ＵＰ This is an example shown in the figure.

[0180] Figure 16 shows a watermark image S generated by the information processing method according to this embodiment. ＵＰ , and decoding area A Ｄ Examples are given.

[0181] Watermark image S ＵＰ This is the video content, cover image C. ｕｐ (Illustration omitted) This is an image in which the specified information M has been encoded.

[0182] Furthermore, the watermark image S illustrated in Figure 16 is also shown. ＵＰ This is video content that was illegally uploaded to a video site by a third party.

[0183] When video content is uploaded illegally, various objects O such as logos, titles, and characters may be added to the video content, or parts of the video content may be missing L, in order to evade detection.

[0184] Conventionally, decoding watermarks was difficult in the situations described above, but according to the information processing method of this embodiment, the decoding area A does not include object O, missing L, etc. Ｄ If the data can be obtained, it can be decoded, making it possible to detect illegally uploaded video content, etc.

[0185] Furthermore, for example, the watermark image S generated by the information processing method according to this embodiment may be applied to various 3D assets.

[0186] Figure 17 shows an example of applying the watermark image S according to this embodiment to a 3D asset.

[0187] Figure 17 shows an example in which a watermark image S is attached to a 3D asset 80, which is an information processing device.

[0188] The watermark image S according to this embodiment can be adjusted in shape and size to match the shape and size of the 3D asset 80.

[0189] Furthermore, by using the texture image of the surface to be attached in the 3D asset 80 as the cover image C, it is possible to generate a watermark image S that does not interfere with the design of the 3D asset 80.

[0190] Next, we will describe the watermark image S generation service (encoding service) and decoding service using the information processing method according to this embodiment.

[0191] Figure 18 is a diagram illustrating a watermark image S generation service using the information processing method according to this embodiment.

[0192] Figure 18 shows a user terminal 70 and an encoding processing unit 130 used by a user enjoying the service.

[0193] The user terminal 70 may be, for example, an information processing device such as a smartphone, tablet, PC (Personal Computer), or wearable device.

[0194] Furthermore, the encoding processing unit 130 may be provided on, for example, a server installed on the cloud.

[0195] In this case, the user uses the user terminal 70 to send an arbitrary message (an example of predetermined information M) and a cover image C to the encoding processing unit 130.

[0196] The encoding processing unit 130 generates a watermark image S based on the received message and cover image C, and sends the generated watermark image S back to the user terminal.

[0197] Figure 19 is a diagram illustrating a decoding service using the information processing method according to this embodiment.

[0198] Figure 19 shows a user terminal 70 and a decoding processing unit 140 used by a user enjoying the service.

[0199] The decoding processing unit 140 may be provided, for example, on a server installed on the cloud.

[0200] In this case, the user takes a picture of the watermark image S using the user terminal 70, and the decoded area A obtained by the picture is taken. Ｄ The data is sent to the decoding processing unit 140. The user terminal 70 can be considered an example of an acquisition unit.

[0201] The decoding processing unit 140 processes the received decoding area A Ｄ Message from (specified information M) Ｄ Decode an example of this and send the message back to the user's terminal.

[0202] The watermark image S generation service and decoding service using the information processing method according to this embodiment have been described above.

[0203] With the services described above, users can easily obtain watermark images S and use the obtained watermark images S for any purpose they choose.

[0204] Furthermore, with the services described above, it becomes possible to obtain messages and other information from watermark images S taken by the user.

[0205] In the above example, the encoding processing unit 130 and the decoding processing unit 140 are provided on a server located on the cloud, but the functions of the encoding processing unit 130 and the decoding processing unit 140 may also be provided to the user terminal 70 as an application.

[0206] <2. Hardware Configuration Example> Next, a hardware configuration example of an information processing device 90 which may include an acquisition unit 110, a learning unit 120, an encoding processing unit 130, or a decoding processing unit 140 according to one embodiment of the present disclosure will be described.

[0207] Figure 20 is a block diagram showing an example of the hardware configuration of an information processing device 90 according to one embodiment of the present disclosure.

[0208] As shown in Figure 20, the information processing device 90 includes, for example, a processor 871, a ROM 872, a RAM 873, a host bus 874, a bridge 875, an external bus 876, an interface 877, an input device 878, an output device 879, a storage device 880, a drive 881, a connection port 882, and a communication device 883. Note that the hardware configuration shown here is just an example, and some of the components may be omitted. Furthermore, the information processing device 90 may include components other than those shown here.

[0209] (Processor 871) The processor 871 functions, for example, as an arithmetic processing unit or a control unit, and controls the overall operation or part thereof of each component based on various programs recorded in the ROM 872, RAM 873, storage 880, or removable storage medium 901. The processor 871 is an example of an integrated circuit, which is a collection of electronic circuits composed of various electronic components and their interconnections.

[0210] (ROM 872, RAM 873) ROM 872 is a means for storing programs loaded into the processor 871 and data used for calculations. RAM 873 temporarily or permanently stores, for example, programs loaded into the processor 871 and various parameters that change as needed when executing those programs.

[0211] (Host bus 874, bridge 875, external bus 876, interface 877) The processor 871, ROM 872, and RAM 873 are interconnected via, for example, the host bus 874, which is capable of high-speed data transmission. On the other hand, the host bus 874 is connected to the external bus 876, which has a relatively low data transmission speed, via, for example, the bridge 875. The external bus 876 is also connected to various components via the interface 877.

[0212] (Input device 878) The input device 878 may include, for example, a mouse, keyboard, touch panel, buttons, switches, and levers. Furthermore, the input device 878 may also include a remote controller (hereinafter referred to as a remote control) capable of transmitting control signals using infrared rays or other radio waves. The input device 878 may also include an audio input device such as a microphone.

[0213] (Output device 879) The output device 879 is a device that can visually or audibly notify the user of acquired information, such as a display device such as a CRT (Cathode Ray Tube), LCD, or organic EL, an audio output device such as a speaker or headphones, a printer, a mobile phone, or a facsimile. The output device 879 according to this disclosure also includes various vibration devices capable of outputting tactile stimuli.

[0214] (Storage 880) Storage 880 is a device for storing various types of data. Examples of storage devices used for storage 880 include magnetic storage devices such as hard disk drives (HDDs), semiconductor storage devices, optical storage devices, or magneto-optical storage devices.

[0215] (Drive 881) Drive 881 is a device that reads information recorded on a removable storage medium 901, such as a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, or writes information to the removable storage medium 901.

[0216] (Removable storage medium 901) The removable storage medium 901 is, for example, DVD media, Blu-ray® media, HD DVD media, various semiconductor storage media, etc. Of course, the removable storage medium 901 may also be, for example, an IC card equipped with a contactless IC chip, or an electronic device, etc.

[0217] (Connection port 882) Connection port 882 is a port for connecting external devices 902, such as a USB (Universal Serial Bus) port, an IEEE 1394 port, a SCSI (Small Computer System Interface), an RS-232C port, or an optical audio terminal.

[0218] (External connected device 902) The external connected device 902 is, for example, a printer, a portable music player, a digital camera, a digital video camera, or an IC recorder.

[0219] (Communication device 883) The communication device 883 is a communication device for connecting to a network, and is, for example, a communication card for wired or wireless LAN, Bluetooth®, or WUSB (Wireless USB), a router for optical communication, a router for ADSL (Asymmetric Digital Subscriber Line), or a modem for various types of communication.

[0220] <3. Summary> As explained above, the information processing system 1 encodes a plurality of watermark tiles R based on the cover image C and predetermined information M, and arranges them in a tile-like manner. ｉ An acquisition unit 110 acquires a portion of the watermark image S, and a predetermined information M is obtained from the portion of the image S. Ｄ The system includes a decoding processing unit 140 equipped with a decoder for decoding the data.

[0221] The above configuration makes it possible to create a more convenient watermark.

[0222] While preferred embodiments of the present disclosure have been described in detail above with reference to the attached drawings, the technical scope of the present disclosure is not limited to such examples. It is clear to any person with ordinary skill in the art of the present disclosure that various modifications or alterations may be conceived within the scope of the technical ideas described in the claims, and these will naturally also fall within the technical scope of the present disclosure.

[0223] For example, in the above example, Watermark Tile R ｉWhile the case where it is rectangular was given as the main example, Watermark Tile R ｉ It can take on various shapes, such as squares, rectangles, rhombuses, triangles, and other polygons.

[0224] Also, Watermark Tile R ｉ The size of the watermark tile R should be appropriately designed according to the size of the predetermined information M to be encoded. ｉ If the size is reduced, a smaller decoding area A is created. Ｄ From the specified information M Ｄ It becomes possible to decode it.

[0225] Furthermore, each step of the processing described in this disclosure does not necessarily have to be processed chronologically in the order shown in the flowchart or sequence diagram. For example, each step of the processing for each device may be processed in an order different from the order described, or may be processed in parallel.

[0226] Furthermore, the series of processes performed by each device described in this disclosure may be implemented by a program stored on a non-transitory computer-readable storage medium. Each program is, for example, loaded into RAM when executed by a computer and executed by a processor such as a CPU. The storage medium is, for example, a magnetic disk, an optical disk, a magneto-optical disk, or flash memory. Alternatively, the program may be distributed without using a storage medium, for example, via a network.

[0227] Furthermore, the effects described herein are merely descriptive or illustrative and not limiting. In other words, the technology relating to this disclosure may produce other effects that are obvious to those skilled in the art from the description herein, in addition to or instead of the effects described herein.

[0228] The following configurations also fall within the technical scope of this disclosure: (1) An information processing system comprising: an acquisition unit that acquires a portion of a watermark image consisting of a plurality of watermark tiles encoded based on a cover image and predetermined information and arranged in a tile-like manner; and a processing unit that includes a decoder for decoding predetermined information from the portion of the image. (2) The information processing system according to (1), wherein the processing unit performs alignment processing based on the watermark tiles on the portion of the image and decodes predetermined information. (3) The information processing system according to (2), wherein the watermark image is generated by arranging a plurality of identical watermark images in a tile-like manner on the cover image. (4) The information processing system according to (3), wherein the processing unit performs the alignment processing using the decoder which incorporates a shift transformation network. (5) The information processing system according to (4), wherein the processing unit performs the alignment processing using the decoder which has learned the shift amount in the X-axis direction and the shift amount in the Y-axis direction. (6) The information processing system according to any one of (3) to (5), wherein the processing unit decodes predetermined information from the partial region which is smaller in size than the watermark tile, using the decoder which has learned to decode based on the partial region which is randomly cropped to a size smaller than the watermark tile. (7) The information processing system according to any one of (3) to (6), wherein the processing unit decodes predetermined information from the partial region which is larger in size than the watermark tile, using the decoder which has learned to decode based on the partial region which is randomly cropped to a size larger than the watermark tile. (8) The information processing system according to (2), wherein the watermark image is generated by combining a plurality of watermark tiles which are encoded based on each of a plurality of partial cover images obtained by dividing the cover image and predetermined information.(9) The information processing system according to (8), wherein the processing unit comprises an alignment module that performs the alignment process based on the center position of the watermark tile included in the partial region. (10) The information processing system according to (9), wherein the processing unit comprises a center position estimation module that estimates the center position of the watermark tile included in the partial region. (11) The information processing system according to (10), wherein, if the partial region includes multiple center positions of the watermark tile, the processing unit performs the alignment process based on the center position of the watermark tile that is closer to the center in the watermark image. (12) The information processing system according to (11), wherein, if the partial region includes multiple center positions of the watermark tile, the processing unit performs the alignment process based on at least one of a scale and a rotation angle calculated based on the relationship between the multiple center positions. (13) The information processing system according to any one of (1) to (12), further comprising: a generation unit for generating the watermark image, wherein the generation unit generates the watermark image using an encoder that has undergone learning based on the loss relating to the cover image and the watermark image. (14) The information processing system according to (13), wherein the decoder and the encoder perform learning based on the loss relating to predetermined information input to the encoder and predetermined information output by the decoder. (15) The information processing system according to (10), wherein the center position estimation module performs learning based on the loss relating to information relating to the center position of the watermark tile included in the input partial region and information relating to the center position of the watermark tile output. (16) The information processing system according to any one of (1) to (15), wherein the watermark image includes a 360° image. (17) The information processing system according to any one of (1) to (16), wherein the watermark image includes a passport photograph.(18) The watermark image is an information processing system according to any one of (1) to (17) above, wherein the watermark image includes video content. (19) The watermark image is applied to a 3D asset by the information processing system according to any one of (1) to (15) above. (20) An information processing method comprising: a processor acquiring a portion of a watermark image consisting of a plurality of watermark tiles encoded based on a cover image and predetermined information and arranged in a tile-like manner; and decoding predetermined information from the portion of the image. (21) A program that causes a computer to function as an information processing device comprising: a processing unit having a decoder for decoding predetermined information from a portion of a watermark image consisting of a plurality of watermark tiles encoded based on a cover image and predetermined information and arranged in a tile-like manner.

[0229] 1 Information Processing System 110 Acquisition Unit 120 Learning Unit 130 Encoding Processing Unit 131 Encoder 132 Encoder 140 Decoding Processing Unit 141 Decoder 142 Decoder 145 Expansion Module 146 Localization Module 147 Registration Module C Cover Image R ｉ Watermark tile S Watermark image A Ｄ Decode area

Claims

1. An information processing system comprising: an acquisition unit that acquires a portion of a watermark image consisting of a plurality of watermark tiles encoded and arranged in a tile pattern based on a cover image and predetermined information; and a processing unit that includes a decoder that decodes predetermined information from the portion of the image.

2. The information processing system according to claim 1, wherein the processing unit performs alignment processing based on the watermark tile on the partial area and decodes predetermined information.

3. The information processing system according to claim 2, wherein the watermark image is generated by arranging a plurality of identical watermark images in a tile-like manner on the cover image.

4. The information processing system according to claim 3, wherein the processing unit performs the alignment process using the decoder, which incorporates a shift conversion network.

5. The information processing system according to claim 4, wherein the processing unit performs the alignment process using the decoder which has learned the shift amount in the X-axis direction and the shift amount in the Y-axis direction.

6. The information processing system according to claim 3, wherein the processing unit decodes predetermined information from the partial region which is smaller in size than the watermark tile, using the decoder which has learned to decode based on the partial region which is randomly cropped to a size smaller than the size of the watermark tile.

7. The information processing system according to claim 3, wherein the processing unit decodes predetermined information from the portion of the watermark tile that is larger in size than the watermark tile, using the decoder which has learned to decode based on the portion of the watermark tile that is randomly cropped to a size larger than the watermark tile.

8. The information processing system according to claim 2, wherein the watermark image is generated by combining a plurality of watermark tiles encoded based on each of a plurality of partial cover images obtained by dividing the cover image and predetermined information.

9. The information processing system according to claim 8, wherein the processing unit includes an alignment module that performs the alignment process based on the center position of the watermark tile included in the partial region.

10. The information processing system according to claim 9, wherein the processing unit comprises a center position estimation module for estimating the center position of the watermark tile included in the partial region.

11. The information processing system according to claim 10, wherein, if the partial region includes multiple center positions of the watermark tile, the processing unit performs the alignment process based on the center position of the watermark tile that is closer to the center in the watermark image.

12. The information processing system according to claim 11, wherein, if the partial region includes multiple center positions of the watermark tile, the processing unit performs the alignment process based on at least one of the scale and rotation angle calculated based on the relationship between the multiple center positions.

13. The information processing system according to claim 1, further comprising: a generation unit for generating the watermark image, wherein the generation unit generates the watermark image using an encoder that has undergone learning based on the loss relating to the cover image and the watermark image.

14. The information processing system according to claim 13, wherein the decoder and the encoder perform learning based on the loss relating to predetermined information input to the encoder and predetermined information output by the decoder.

15. The information processing system according to claim 10, wherein the center position estimation module performs learning based on the loss relating to the information relating to the center position of the watermark tile included in the input partial region and the information relating to the center position of the watermark tile output.

16. The information processing system according to claim 1, wherein the watermark image includes a 360° image.

17. The information processing system according to claim 1, wherein the watermark image includes a passport photograph.

18. The information processing system according to claim 1, wherein the watermark image includes video content.

19. An information processing method comprising: a processor acquiring a portion of a watermark image consisting of a plurality of watermark tiles encoded based on a cover image and predetermined information and arranged in a tile-like manner; and decoding predetermined information from the portion of the image.

20. A program that causes a computer to function as an information processing device, comprising a processing unit equipped with a decoder that decodes predetermined information from a portion of a watermark image consisting of a plurality of watermark tiles encoded based on a cover image and predetermined information and arranged in a tile-like manner.

Images