Camera, electronic device, electronic signing device, electronic extraction device, system and method

Abstract

A camera is configured to capture an image and sign the captured image with a signature based on changing the image pixels to generate a signed image, wherein the difference between the signed image and the captured image is imperceptible to human visual perception.

Description

[0001] Sony Group Corporation et al.

[0002] CAMERA, ELECTRONIC DEVICE, ELECTRONIC SIGNING DEVICE, ELECTRONIC EXTRACTION DEVICE, SYSTEM AND METHOD

[0003] TECHNICAL FIELD

[0004] The present disclosure generally pertains to tracking copyrighted works and their use in generative artificial intelligence (Al) models.

[0005] TECHNICAL BACKGROUND

[0006] The unauthorized use of copyrighted works by generative Al models presents a legal and ethical challenge, as such models often scrape, process, and reproduce copyrighted works without proper authorization from the original creators. This results in potential copyright infringement and economic harm to artists and rights holders.

[0007] Although there exist techniques for tracking copyrighted works, it is generally desirable to improve on existing techniques.

[0008] SUMMARY

[0009] According to a first aspect, the present disclosure provides a camera configured to capture an image, and sign the captured image with a signature based on changing the image pixels to generate a signed image, wherein the difference between the signed image and the captured image is imperceptible to human visual perception.

[0010] According to a second aspect, the present disclosure provides an electronic device configured to extract a signature from a target image generated based on the signed image output by the camera.

[0011] According to a third aspect, the present disclosure provides an electronic signing device configured to sign a first signal with a signature to generate a signed signal, such that the difference between the signed signal and the first signal is imperceptible to human perception, provide the signed signal with a signal identification, and save the signature and the signal identification in a distributed ledger.

[0012] According to a fourth aspect, the present disclosure provides an electronic extraction device configured to retrieve the signature stored by the electronic signing device on the distributed ledger from the distributed ledger, and to extract a signature from data generated based on the signed signal output by the electronic signing device.Sony Group Corporation et al.

[0013] According to a fifth aspect, the present disclosure provides a system comprising a distributed ledger configured to store a signature, the signature enabling signing of a first signal to generate a signed signal, such that the difference between the signed signal and the first signal is imperceptible to human perception.

[0014] According to a sixth aspect, the present disclosure provides a method for controlling a camera comprising the steps of capturing an image, and signing the captured image with a signature based on changing the image pixels to generate a signed image, wherein the difference between the signed image and the captured image is imperceptible to human visual perception.

[0015] According to a seventh aspect, the present disclosure provides a method comprising the step of extracting a signature from a target image generated based on the signed image output by the camera.

[0016] According to an eight aspect, the present disclosure provides a method comprising the steps of signing a first signal with a signature to generate a signed signal, such that the difference between the signed signal and the first signal is imperceptible to human perception, providing the signed signal with a signal identification, and saving the signature and the signal identification in a distributed ledger.

[0017] According to a ninth aspect, the present disclosure provides a method comprising the steps of retrieving the signature stored by the electronic signing device on the distributed ledger from the distributed ledger, and extracting a signature from data generated based on the signed signal output by the electronic signing device.

[0018] Further aspects are set forth in the dependent claims, the drawings and the following description.

[0019] BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Embodiments are explained by way of example with respect to the accompanying drawings, in which:

[0021] Fig. 1 shows a signing process configured to sign captured images with a signature;

[0022] Fig. 2 shows a process of associating a signed image with an image ID;

[0023] Fig. 3 shows Al-assisted image generation of a new image based on the signed image of Fig. 1 that is published;

[0024] Fig. 4 shows a process for identifying a publicly available new image and comparing it to signed images stored in a database;Sony Group Corporation et al.

[0025] Fig. 5 shows an extraction process of a signature from a newly generated image;

[0026] Fig. 6 shows a matching process of an extracted signature to multiple signatures of a database; Fig. 7 shows a matching process of an extracted signature to one camera-specific signature; Fig. 8 shows a process of identifying the signed image 3 used during the image generation of new image 7;

[0027] Fig. 9 shows a database generation for signed audio signals and a content identification of an audio signal based on the generated database;

[0028] Fig. 10 shows a diagram of a system configured to host a camera-specific signature for verifying the ownership of the camera and images connected to the camera-specific signature;

[0029] Fig. 11 shows in more detail an exemplifying communication flow related to the system of Fig. 10;

[0030] Fig. 12 shows an exemplifying communication flow related to the system of Fig. 10;

[0031] Fig. 13 shows a block diagram of a camera configured to perform a signing process of a captured image; and

[0032] Fig. 14 shows a block diagram of an electronic device configured to perform any one of the processes of Figs. 1, 2, 4 to 9, 11 and 12.

[0033] DETAILED DESCRIPTION OF EMBODIMENTS

[0034] Before a detailed description of the embodiments under reference of Fig. 1 is given, general explanations are made.

[0035] As mentioned at the outset, intellectual property infringement poses a significant challenge. It has been found that this issue affects digitally generated and modified works, such as audio signals and images, but also data captured by sensors, such as microphones and image sensors, particularly when competitors use this data to train machine learning models.

[0036] Therefore, some embodiments pertain to a camera configured to capture an image and sign the captured image with a signature based on changing the image pixels to generate a signed image, wherein the difference between the signed image and the captured image is imperceptible to human visual perception.

[0037] The camera may be any type of camera for capturing images, such as an RGB camera, a time-of-flight (ToF) camera (e.g., dToF, iToF), or an event-based camera or the like.Sony Group Corporation et al.

[0038] The camera may store or register the signature in a database.

[0039] The signature may be a binary code or message that is embedded into the captured image to generate the signed image. The binary message may be represented as a bit string.

[0040] The signature embedded in the captured image to generate the signed image may refer to a perturbation input in the captured image, which can be used to encode information invisibly within the image. That is, the image pixels of the captured image may be modified to encode the signature.

[0041] For example, the binary message represented as a bit string may be incorporated into the signed image by modifying the image pixels of the captured image in a way that is imperceptible to human visual perception.

[0042] Thus, it may be ensured that the captured image and the signed image are visually (by humans) indistinguishable from each other.

[0043] Adding the signature to the captured image to generate the signed image may be considered a watermarking technique.

[0044] The camera may add a signature to each image after capture, i.e., to multiple images.

[0045] One advantage of a camera that signs images directly is the enhanced security and authenticity of the captured data. By embedding an imperceptible signature into each image at the point of capture, the camera ensures that the data is traceable and verifiable. This direct signing process makes it extremely difficult for unauthorized parties to alter or remove the signature, thereby protecting the intellectual property rights of the image owner.

[0046] Additionally, this method provides a robust mechanism for detecting unauthorized use of the data captured by the camera in machine learning models, as the embedded signature can be reliably identified even after the data has been processed or transformed. This helps in maintaining the integrity of the data and provides a clear path for accountability and traceability, especially when linked to blockchain technology, as explained below in more detail.

[0047] Signing the captured image may be based on a neural network. For example, as explained in Zhu, Jiren, et al. "Hidden: Hiding data with deep networks." Proceedings of the European conference on computer vision (ECCV), 2018; or as explained in Fernandez, Pierre, et al. "The stable signature: Rooting watermarks in latent diffusion models." Proceedings of the IEEE / CVF International Conference on Computer Vision, 2023.Sony Group Corporation et al.

[0048] For example, the neural network (e.g., artificial neural network) may include a Feed-Forward Network, a Residual Network (ResNet), a Recurrent Neural Network (RNN), a Convolutional Neural Network (CNN), a Generative Adversarial Network (GAN), a Transformer Neural Network and / or any other suitable neural network architecture. The skilled person may find a suitable architecture for the artificial neural network based on his expert knowledge.

[0049] The neural network may sign the captured image by using an encoder-decoder framework where the encoder embeds a binary message as a signature into a captured image, creating an encoded signed image that appears visually indistinguishable from the original. This encoded signed image may then be transmitted or stored. The message may be encoded such that the decoder network may be able to retrieve the hidden signature from the encoded image. The signing may include making tiny, imperceptible perturbations to the image, ensuring that the hidden signature is robust to various alterations and distortion, such as blurring, cropping, and compression, but also alterations such as found during Al assisted image generation.

[0050] An adversarial network may also be employed to provide feedback to the encoder during training and thereby improve the quality of the encoded signed images, making them harder to detect by unauthorized parties. The adversarial network may attempt to distinguish between original training images and output signed images, thereby encouraging the encoder to produce signed images that are visually indistinguishable from the originals.

[0051] For example, the adversarial loss, may penalizes the encoder when the discriminator of the adversarial network successfully identifies a signed image, and may thereby drive the encoder to refine its embedding process, making the signature less detectable.

[0052] The neural network may be trained based on training images and training signatures used as input, as explained for example in Zhu, Jiren, et al. "Hidden: Hiding data with deep networks." Proceedings of the European conference on computer vision (ECCV), 2018; or as explained in Fernandez, Pierre, et al. "The stable signature: Rooting watermarks in latent diffusion models." Proceedings of the IEEE / CVF International Conference on Computer Vision, 2023.

[0053] For example, during training, after initializing the encoder, decoder and the adversarial discriminator network with random weights, a training image and training signature of a set of training images and training signatures may be provided as input to the encoder who outputs an encoded image. Noise may be applied to simulate real-world distortions (e.g., Gaussian blurring, cropping, JPEG compression etc.). The noised encoded image may then be passed through the decoder to recover the training signature.Sony Group Corporation et al.

[0054] Furthermore, the encoded image and the training image may be passed through the adversarial discriminator to classify whether the image is an original training image or a modified encoded image, i.e., to predict whether the image contains an embedded signature.

[0055] During the training process, several loss functions may be calculated and minimized during the backward pass to enhance the performance of the encoder, decoder, and adversarial discriminator. These loss functions may include:

[0056] Image Distortion Loss: This loss measures the distance between the cover image and the encoded image, ensuring minimal visual differences.

[0057] Signature Distortion Loss: This loss measures the distance between the original signature and the recovered signature, ensuring accurate message retrieval.

[0058] Adversarial Loss: This loss is based on the discriminator’s ability to accurately detect the encoded image. Minimizing this loss results in less detectable signature embedding, encouraging the encoder to perform more subtle perturbations.

[0059] Discriminator Loss: This classification loss measures the discriminator's accuracy in predicting whether an image is an original training image or an encoded image.

[0060] During the backward pass, these loss functions may be minimized, thereby training the encoder and decoder to improve their performance in embedding and recovering signatures.

[0061] Simultaneously, the adversarial discriminator may be trained to enhance its detection capabilities, which penalizes the encoder when it successfully detects encoded images. This adversarial feedback encourages the encoder to make more subtle modifications for signature encoding, resulting in higher-quality, less detectable encoded images.

[0062] After sufficient iterative training, the encoder-decoder network will be capable of embedding signatures into images with high visual quality and robustness, thereby generating a signed image from a captured image. The adversarial discriminator may ensure that the embedded signatures are difficult to detect and visually imperceptible to humans.

[0063] Signing the captured image with the signature may therefore include adding a hidden perturbation to the image as the signature.

[0064] Signing the captured image with the signature may include adding the signature to the captured image such that it can be extracted from the signed image. For example, the signature is embedded into the captured image using the encoder-decoder framework as explained above.Sony Group Corporation et al.

[0065] Thus, the decoder may be trained to recognize and extract the embedded signature from the signed image accurately.

[0066] The encoder may modify the captured image in a way that the changes are imperceptible to the human eye but can be detected and decoded by the trained decoder network. During training, the network may therefore learn to balance the visibility of the encoded signature with the robustness against various distortions such as Gaussian blurring, pixel-wise dropout, cropping, and JPEG compression, Al-assisted modification or the like.

[0067] The neural network may include a latent diffusion model (LDM) with its own encoder-decoder pair. For example, the neural network may include a fine-tuned LDM decoder for embedding the signature in the captured image and generating the signed image.

[0068] Thus, training may include pre-training an extractor. The pre-trained extractor that can recover embedded signatures from images may be based on the decoder of the above-mentioned trained encoder-decoder framework. Training may also include fine-tuning the LDM’s latent decoder to ensure that all generated signed images inherently contain the signature, based on the pre-trained extractor, which may check if the signature is correctly embedded. This fine-tuning may be achieved by optimizing the LDM decoder with a combination of two losses: a perceptual image loss, which ensures that the signed images remain visually similar to non-signed ones, and a message loss, which ensures that the signature is reliably embedded and retrievable (see e.g., Fernandez, Pierre, et al. "The stable signature: Rooting watermarks in latent diffusion models." Proceedings of the IEEE / CVF International Conference on Computer Vision. 2023). The signature may be encoded such that it can be reliably extracted even after the image has been altered, for example during artificial intelligence-assisted image generation with generative Al, providing a secure and effective method for signing and verifying image authenticity and ownership.

[0069] The signature within the signed image may therefore be configured to be resilient to image modification. For example, such that after artificial intelligence assisted image generation based on the signed image the generated image includes the signature at least partly.

[0070] The signature within the signed image may be configured to be resilient, such that the signature can at least partly be extracted from an image (e.g., target image) generated by artificial intelligence assisted image generation based on the signed image. For example, the signature within the signed image may be configured to be resilient, such that the signature can be extracted from an image (e.g., target image) generated by artificial intelligence assisted imageSony Group Corporation et al.

[0071] generation based on the signed image with statistical confidence. Statistical confidence may refer to a predefined confidence threshold. Thus, the signature may be extracted from a target image generated based on the signed image according to a predefined confidence threshold. For example, the signature extracted from an image generated with artificial intelligence based on the signed image may be and / or match the signature of the signed image according to a predefined confidence threshold, i.e., with statistical confidence. The extracted signature may be recognizable as the signature embedded in the signed image.

[0072] The signature may be device-specific or image-specific.

[0073] The signature may be a camera-specific signature associated with a cryptographic key. For example, the signature may be a device-specific cryptographic key assigned to the camera or the signature may be coupled to a device-specific cryptographic key.

[0074] The cryptographic key may be unique to the camera. The camera-specific signature and / or the cryptographic key may be securely stored in a blockchain, linking the ownership of the device to an individual or entity, e.g., through their public key.

[0075] For example, the camera may be associated to a private and public key pair and the cameraspecific signature may be stored during a transaction on the blockchain together with the public key. The blockchain transaction may be signed with the private key, e.g., by the camera.

[0076] The camera may be further configured to provide an image identification assigned to the signed image and to save the image identification and the signature associated with each other to a distributed ledger.

[0077] A distributed ledger may refer to a decentralized digital record-keeping system that securely records, synchronizes, and shares transactions across multiple nodes in a network. Each transaction is time-stamped, immutable, and transparent to authorized participants. Distributed ledgers form the foundation for blockchain technology.

[0078] The image identification may be a calculated hash of the image, generated using a cryptographic hashing algorithm such as SHA-256. The calculated hash serves as a unique identifier for the image, ensuring that even minor alterations to the image will result in a different hash value. The signature, along with the image hash, may be stored in the distributed ledger by creating a new transaction. This transaction may include the image hash, the camera-specific signature, and the public key associated with the camera. The transaction may then be broadcast to the network of nodes participating in the distributed ledger.Sony Group Corporation et al.

[0079] Upon receiving the transaction, each node may validate the transaction by verifying the cryptographic signature using the public key. Once validated, the transaction is added to the distributed ledger, ensuring that the image hash and the camera-specific signature are immutably recorded. The distributed ledger thus maintains a secure and transparent record of the image hash and the associated signature, linking them to the camera's public key.

[0080] When an individual or entity claims ownership of a target image and / or a signed image they may need to possess the private key corresponding to the public key. The claimant may, therefore, use their private key to sign a message or a challenge provided by a verifier.

[0081] The verifier can then use the public key stored in the distributed ledger to verify the signature on the message or challenge. If the signature is valid, it confirms that the claimant possesses the private key associated with the public key recorded in the ledger, thereby verifying ownership. Additionally, the verifier can recalculate the hash of the signed image and compare it with the hash stored in the distributed ledger to ensure the image has not been altered.

[0082] Also, the transaction to store the signature and / or the public key and / or the image hash may be digitally signed by the cryptographic private key of the camera corresponding to the public key. Thereby the verifier, can directly verify the cryptographic signature on the transaction without the claimant providing an additional signed message or challenge.

[0083] All processes regarding the storage of the signature and / or the public key and / or the image identification on the distributed ledger may be performed by the camera.

[0084] Some embodiments pertain to an electronic device configured to extract a signature from a target image generated based on the signed image output by the camera.

[0085] The electronic device may may include a processor, a memory (RAM, ROM or the like), a storage, input means (mouse, keyboard, camera, etc.), output means (display (e.g. liquid crystal, (organic) light emitting diode, etc.), loudspeakers, etc., a (wireless) interface, etc., as it is generally known for electronic devices (computers, smartphones, etc.). Moreover, it may include sensors for sensing audio signals (e.g., a microphone), for sensing still image or video image data (image sensor, camera sensor, video sensor, etc.), for sensing a fingerprint, for sensing environmental parameters (e.g. radar, humidity, light, temperature), etc.

[0086] The target image from which the signature is extracted may be an artificial intelligence generated image, for example based on the signed image. Thus, the target image may include the signature of the signed image, i.e., the signature used for signing the signed image, e.g., at least partly.Sony Group Corporation et al.

[0087] The electronic device may acquire the target image from any published source. The electronic device may thus be configured to scan public sources for images with digital signatures or watermarks.

[0088] Extraction may be implemented as discussed above, for example based on the decoder of a trained encoder-decoder framework. Thus, extraction may be based on a trained neural network. During the extraction the target image, which may be one of the scanned images, may be processed to identify patterns that correspond to the hidden signature.

[0089] For example, during the extraction a set of values, which represent the likelihood of each bit in the signature being either 0 or 1 may be generated and output. A binary thresholding operation may be applied to these values, where each extracted bit may be classified as 0 or 1 based on its confidence score.

[0090] The electronic device may be further configured to match the extracted signature of the image to the known signature of the signed image. The known signature may refer to the signature used for signing the captured image for generating the signed image.

[0091] During the matching the signatures may be compared. Matching may include a statistical test. Matching may include determining whether the number of matching bits of the extracted signature compared to the known signature (e.g., the signature of the signed image) exceeds a predefined threshold.

[0092] For example, in a detection process, the extracted signature may be compared against the known signature of the signed image. If it closely matches, e.g., according to a predefined threshold, the target image may be determined to be matching, and for example flagged as matching. The electronic device may therefore determine the target image to be Al-generated based on the signed image.

[0093] For example, in an identification process, multiple signatures of different signed images may be used (each corresponding to different users, cameras or model instances). The extracted signature may be compared across multiple known signatures of different signed images. The best match may then be determined as matching. In this way, the target image may be attributed to the most likely source, identifying the specific model, camera or owner of the signed image which was used to generate the tested image.

[0094] The electronic device may be configured to be able to access a database in which a signature of a at least one signed image corresponding to the camera or in which multiple camera-specificSony Group Corporation et al.

[0095] signatures corresponding to different cameras are stored or registered, e.g., registered by the camera.

[0096] The electronic device may be configured to monitor images published on public sources, searching for examples with registered digital signatures. For this purpose, the electronic device may conduct the extraction and matching process as described above.

[0097] If an image with a registered signature is found, the electronic device may determine if the image was taken by the camera corresponding to the signature. For this purpose, the electronic device may query a database of signed images corresponding to the signature.

[0098] If the image does not match any of the signed images of the camera, it may be determined that an artificial intelligence (Al) system was trained using data from the camera, suggesting potential intellectual property infringement. Then, a user may be notified of such.

[0099] If the electronic device determines that the image was taken by the camera, the user may be notified of such. Then, the user who may be the author may confirm whether the usage was fair and not infringing on intellectual property.

[0100] The electronic device may be further configured to verify the ownership of the extracted signature matched to the signature of the signed image. Verifying the ownership may include retrieving the signature of the signed image from a distributed ledger. The verification process may be conducted by the electronic device as described in the following.

[0101] Ownership of the target image may be verified by querying the distributed ledger for the signature of the signed image determined to be matching the extracted signature of the target image.

[0102] The corresponding transaction may be retrieved which may reveal the camera-specific signature and the public key. To verify ownership, the public key may be compared to the known public key of the individual or entity claiming ownership.

[0103] Furthermore, the hash of the signed image may be retrieved from the distributed ledge, for example if the transaction includes the hash of the signed image.

[0104] Additionally, the integrity of the image may be confirmed by recalculating the hash of the signed image and ensuring it matches the hash stored in the distributed ledger.

[0105] Comparing the hash of the signed image stored in the distributed ledger to the recalculated hash of the image may include acquiring the signed image from a database. The signed image itself may be stored directly in the distributed ledger, e.g., by the camera, or due to storage constraintsSony Group Corporation et al.

[0106] the image may be stored in an external storage system (e.g., cloud sever), and only its hash may be recorded in the distributed ledger.

[0107] Thus, the electronic device may retrieve the signed image with the signature that matches the extracted signature of the target image from a storage (e.g., the distributed ledger or an external storage) and calculate a hash from the signed image.

[0108] If multiple signed images were signed with the same signature, the electronic device may conduct a pre-filtering by comparing the multiple signed images to the target image and determine the matching signed image, for example, the signed image most likely used for image generation, e.g., based on a pre-defined threshold.

[0109] Then, the ownership of the image may be confirmed by the electronic device by recalculating the hash of the signed image used for generating the target image and ensuring it matches the hash stored in the distributed ledger.

[0110] Additionally or alternatively, the electronic device may use the public key stored and retrieved from the distributed ledger to verify the signature on a generated challenge. Thus, the electronic device may provide a generated challenge to a claimant and retrieve the challenge signed with the private key from the claimant. If the signature on the challenge is valid (e.g., as confirmed by the public key retrieved from the distributed ledger), it confirms that the claimant possesses the private key associated with the public key recorded in the ledger, thereby verifying ownership. Additionally or alternatively, the transaction to store the signature and / or the public key and / or the image hash may be digitally signed by the cryptographic private key of the camera corresponding to the public key. Thereby the electronic device which may retrieve the transaction may directly verify the cryptographic signature on the transaction with the retrieved public key, thereby verifying ownership.

[0111] If the device-specific cryptographic key is stored in a blockchain, linking the ownership of the device to a person, whose identity does not need to be disclosed as only their public key needs to be registered, may be achieved by the electronic device.

[0112] In this way, unauthorized use of intellectual property, e.g., a signed image used for training machine learning models to generate the target image, may be detected and confirmed by the electronic device.

[0113] Some embodiments pertain to an electronic signing device configured to sign a first signal with a signature to generate a signed signal, such that the difference between the signed signal and theSony Group Corporation et al.

[0114] first signal is imperceptible to human perception, provide the signed signal with a signal identification, and save the signature and the signal identification in a distributed ledger.

[0115] The electronic signing device may include a processor, a memory (RAM, ROM or the like), a storage, input means (mouse, keyboard, camera, etc.), output means (display (e.g. liquid crystal, (organic) light emitting diode, etc.), loudspeakers, etc., a (wireless) interface, etc., as it is generally known for electronic devices (computers, smartphones, etc.). Moreover, it may include sensors for sensing audio signals (e.g., a microphone), for sensing still image or video image data (image sensor, camera sensor, video sensor, etc.), for sensing a fingerprint, for sensing environmental parameters (e.g. radar, humidity, light, temperature), etc.

[0116] The first signal may for example be an image or an audio signal. The first signal may be a recording or capture of a real -world event, e.g., image capture by a camera or audio capture by a microphone, or it may be digitally generated.

[0117] The electronic signing device may be the camera described in this specification.

[0118] The signing process and the signature may include any of the features of the signing process and the signature as explained with regard to the camera.

[0119] For example, signing the first signal may be based on a neural network. For example, signing the first signal with the signature may include adding a hidden perturbation to the first signal as the signature. For example, signing the first signal with the signature may include adding the signature to the first signal such that it can be extracted from the signed signal.

[0120] For example, the signature within the signed signal may be configured to be resilient to signal modification, e.g., such that after artificial intelligence assisted modification based on the signed signal the generated data includes the signature at least partly.

[0121] For example, the signature within the signed signal may be configured to be resilient to signal modification, such that the signature can at least partly be extracted from data generated by artificial intelligence assisted data generation based on the signed signal. For example, the signature within the signed image may be configured to be resilient, such that the signature can be extracted from data generated by artificial intelligence assisted data generation based on the signed signal with statistical confidence. As explained above, statistical confidence may refer to a predefined confidence threshold. Thus, the signature may be extracted from target data generated based on the signed signal according to a predefined confidence threshold. For example, the signature extracted from data generated with artificial intelligence based on theSony Group Corporation et al.

[0122] signed signal may be and / or match the signature of the signed signal according to a predefined confidence threshold, i.e., with statistical confidence.

[0123] The signal identification may correspond and therefore include any of the features described with regard to the image identification described in this specification.

[0124] Saving the signal identification and the signature in the distributed ledger may include any feature described with regard to saving the (camera-specific) signature and image identification described in this specification.

[0125] Thus, the electronic signing device may be configured to sign the first signal and generate a hash from the signed signal, which may be associated to the electronic device’s secret key, and stored on the blockchain as described above with regard to the camera, such that ownership of the signature and / or signed signal and / or any signal including the signature can be verified as described above.

[0126] Concerning the audio signal as a first signal, the signature may be imperceptibly embedded into the audio signal, ensuring that the signed signal as an audio signal (signed audio signal) remains indistinguishable from the original first signal to human auditory perception. Signing the audio signal may be performed for example as explained in Singh, Mayank Kumar, et al.

[0127] "SilentCipher: Deep Audio Watermarking." arXiv preprint arXiv 2406 (2024): 03822, or in Li, Pengcheng, et al. "IDEAW: Robust Neural Audio Watermarking with Invertible DualEmbedding." arXiv preprint arXiv:2409.19627 (2024).

[0128] The signed audio signal may be signed using a neural network-based encoder-decoder framework, e.g., similar to the process described for images. The encoder network may embed the signature (e.g., binary message) into the audio signal by making subtle perturbations that are imperceptible to human hearing. This process may involve modifying the audio samples in a way that preserves the original sound quality while embedding the signature. The decoder network may be trained to extract the embedded signature from the signed audio signal accurately. Thus, the neural network model may make use of the limitations of the human auditory system.

[0129] Signing may include frequency masking, where a low-energy signal becomes inaudible in the presence of a high-energy signal at a nearby frequency. Thus, the signature may be placed in frequency bands where the original audio signal has high energy. Signing may include phase manipulation, for example setting the phase of the embedded signature phase shifted (e.g., by 7t) relative to the original audio signal for each frequency bin. Signing may include magnitude constraints, for example ensuring that the magnitude of each frequency bin of the signature isSony Group Corporation et al.

[0130] less than or equal to that of the original audio signal, which may prevent the signature from introducing noticeable changes in the audio signal's spectral content. Signing may include temporal masking, for example, embedding the signature during periods of high amplitude or complex audio segments where the perturbations are less likely to be noticed due to the masking effects of louder or more intricate sounds. Signing may include spectral shaping, for example distributing the energy of the signature in a manner that mimics the spectral characteristics of the original audio signal. Signing may include subtle amplitude modulation, for example, introducing slight variations in the amplitude of the audio signal that are below the threshold of human perception. These variations can encode the binary message without altering the perceived loudness or quality of the audio.

[0131] During training, the neural network is provided with a set of training audio signals and corresponding training signatures. The training audio signal may be any type of audio signal from any source. The encoder modifies the training audio signals to embed the signatures, creating signed audio signals. These signed audio signals may then be subjected to various distortions and transformations, such as noise addition, compression, and equalization, to simulate real-world conditions.

[0132] Some embodiments pertain to an electronic extraction device configured to retrieve the signature stored by the electronic signing device on the distributed ledger from the distributed ledger, and to extract a signature from data (e.g., target data) generated based on the signed signal output by the electronic signing device.

[0133] The electronic extraction device may include any feature of the electronic device described in this specification, wherein the electronic extraction device may extract from any type of target data. For example, data may refer to an image and / or an audio signal etc.

[0134] For example, the electronic extraction device may be configured to match the extracted signature of the data (e.g., target data) to the signature of the signed signal. For example, the electronic extraction device may be configured to verify the ownership of the extracted signature based on the retrieved signature.

[0135] Some embodiments pertain to a system comprising a distributed ledger configured to store a signature, the signature enabling signing of a first signal to generate a signed signal, such that the difference between the signed signal and the first signal is imperceptible to human perception. Signing of a first signal may refer to any of the signing processes described in this specification, for example it may refer to a watermarking process.Sony Group Corporation et al.

[0136] The distributed ledger may include any of the features regarding the distributed ledger described in this specification. The storage of the signature may include any of the features for storing the signature in a blockchain or distributed ledger described in this specification. The signature for signing the first signal may include any of the features of the signature, e.g., the camera-specific signature, described in this specification. The first signal may include any of the features of the first signal or the captured image described in this specification. The signed signal may include any of the features of the signed signal or the signed image described in this specification.

[0137] The system may include the camera, the electronic device, the electronic signing device and / or the electronic extraction device, or any feature of these devices described in this specification. Some embodiments pertain to a method for controlling the camera comprising the steps of capturing an image, signing the captured image with a signature based on changing the image pixels to generate a signed image, wherein the difference between the signed image and the captured image is imperceptible to human visual perception.

[0138] The method may include any feature described with regard to the camera described in this specification.

[0139] Some embodiments pertain to a method comprising the step of extracting a signature from a target image generated based on the signed image output by the camera.

[0140] The method may include any feature described in regard to the electronic device.

[0141] Some embodiments pertain to a method comprising the steps of signing a first signal with a signature to generate a signed signal, such that the difference between the signed signal and the first signal is imperceptible to human perception, providing the signed signal with a signal identification and saving the signature and the signal identification in a distributed ledger.

[0142] The method may include any feature described in regard to the electronic signing device described in this specification.

[0143] Some embodiments pertain to a method comprising the steps of retrieving the signature stored by the electronic signing device on the distributed ledger from the distributed ledger, and extracting a signature from data generated based on the signed signal output by the electronic signing device. The method may include matching the extracted signature of the data to the signature used for signing the signed signal.

[0144] The method may include any feature described in regard to the electronic extraction device described in this specification.Sony Group Corporation et al.

[0145] The methods as described herein are also implemented in some embodiments as a computer program causing a computer and / or a processor to perform the method, when being carried out on the computer and / or processor. In some embodiments, also a non-transitory computer-readable recording medium is provided that stores therein a computer program product, which, when executed by a processor, such as the processor described above, causes the methods described herein to be performed.

[0146] Returning to Fig. 1, a signing process configured to sign captured images with a signature is shown.

[0147] Initially, captured image 1 and camera-specific signature 2 are provided to the signing process 14. The signing process 14 signs captured image 1 based on the camera-specific signature 2. Signing process 14 includes adding a hidden perturbation based on camera-specific signature 2 to the image signal of the captured image 1, thereby generating signed image 3 that includes the camera-specific signature 2 encoded within. That is, the image information, of at least some or all of the pixels, is changed based on the hidden perturbation, i.e., the cameraspecific signature 2.

[0148] The camera-specific signature 2 encoded within the signed image 3 by the signing process 14 is not perceptible by human eyes. That is, the difference between captured image 1 and signed image 3, which includes the captured image 1 with the encoded camera-specific signature, and which is output from the signing process 14, is not visually perceptible by humans.

[0149] The camera-specific signature 2 is associated with the camera (e.g., camera 10, Fig. 9). Thus, based on the camera-specific signature 2 the camera can be identified as the source of the signed image 3. The signed image 3 is further associated with an image identification as described in Fig. 2.

[0150] The signing process 14 is conducted by a trained neural network (e.g., as explained in the articles Zhu, Jiren, et al. "Hidden: Hiding data with deep networks." Proceedings of the European conference on computer vision (ECCV). 2018.; Fernandez, Pierre, et al. "The stable signature: Rooting watermarks in latent diffusion models." Proceedings of the IEEE / CVF International Conference on Computer Vision. 2023.) included in the circuitry of camera 1.

[0151] The camera-specific signature 2 within the signed image 3 is encoded such that it can be extracted from the signed image 3 (see e.g., 8, Fig. 5).

[0152] The camera-specific signature within the signed image 3 is a modification-resilient signature, such that if the signed image 2 is modified, for example when it is used during an artificialSony Group Corporation et al.

[0153] intelligence (Al) assisted image generation (see Fig. 3), the camera-specific signature 2 can still be extracted at least in part, such that it is recognizable with statistical reliability based on a predefined threshold when compared to camera-specific signature 2 of signed image 3. (Thus, the signing process 14 is conducted by a trained neural network as for example explained in Zhu, Jiren, et al. "Hidden: Hiding data with deep networks." Proceedings of the European conference on computer vision (ECCV). 2018; or as explained in Fernandez, Pierre, et al. "The stable signature: Rooting watermarks in latent diffusion models." Proceedings of the IEEE / CVF International Conference on Computer Vision. 2023.).

[0154] The signing process 14 is conducted by the camera that captures the captured image 1. Thus, the camera provides the captured image 2, the camera-specific signature 3 and the signing process 14.

[0155] Alternatively, the signing process 14 may not be conducted by the camera but by another electronic device which may obtain the captured image 1 and the camera-specific signature, e.g., form the camera or from a database.

[0156] After signing the signed image 3 may be published.

[0157] Instead of a captured image 2 also other information, such as an audio signal, e.g., captured by a microphone and / or produced by an electronic device (e.g., in a studio), may be signed by the signing process 14. In this case the camera-specific signature 2 may be a signature specific to the microphone, the electronic device, the studio, the owner and / or the user capturing, producing and / or generating the audio signal.

[0158] Fig. 2 shows a process of associating a signed image 3 with an image ID.

[0159] At 15 the signed image 2 of Fig. 1 is assigned an image ID 4 for identification.

[0160] At 16 the signed image 3 and the image ID 4 as well as the camera-specific signature 2 are stored.

[0161] The signed image 3, the image ID 4 and the camera-specific signature 2 may be stored associated with each other. Alternatively, only the image ID 4 and the camera-specific signature 2 may be stored associated with each other. For example, the camera specific signature 2 may be stored on a blockchain (see BC, Figs. 11 and 12).

[0162] The image ID 4 may be generated, for example, by computing a hash of the signed image 3.Sony Group Corporation et al.

[0163] Thus, the image ID 4, which may be a hash, of the signed image 3 may be stored on a blockchain. The camera-specific signature 2 may be stored together or associated with the image ID 4 on the blockchain (see Fig. 12).

[0164] The signed image 3 may be stored in a database. The signed image 3 may be stored on a blockchain or it may be stored in an off-chain database. This allows for flexibility in storage solutions while ensuring the integrity and traceability of the image data.

[0165] To associate the off-chain stored signed image 3 with the on-chain hash, the image ID 4 may be used as a reference. When retrieving the signed image 3 from the off-chain storage, the image ID 4 may be recalculated from the signed image 3 and matched against the hash (the image ID 4) stored on the blockchain.

[0166] Fig. 3 shows how signed image 3 of Fig. 1 that is published is used in Al-assisted image generation to generate a new image.

[0167] If signed image 3 (e.g., 3 of Fig. 1) is publicly available it can be used, even without permission from the creator, in Al-assisted image generation (e.g., by DALL-E, MidJoumey, Stable Diffusion, and other similar Al platforms etc.).

[0168] The Al-image generator 6 selects multiple images, including images 5a to 5d and signed image 3, as input sources for generating a new image 7 from a publicly available database (e.g., based on web scraping or the like). The Al-image generator 6 processes the selected images 3, 5a-5d, combining various elements from each to generate a new image 7. During this process, parts of signed image 3 are incorporated into the new image 7.

[0169] Since signed image 3 contains a camera-specific signature 2 encoded within, elements of this signature 2 are also embedded within the new image 7. The signature's robustness ensures that it remains identifiable even after the Al-image generator 6 modifies the original image 3.

[0170] The presence of the camera-specific signature 2 in the new image 7 allows for verification of the original source data. By analyzing the signature 2 elements within the new image 7, it is possible to trace back to signed image 3, confirming its contribution to the Al-generated content.

[0171] Fig. 4 shows a process for identifying a publicly available new image and comparing it to signed images stored in a database.

[0172] If new image 7 is published, it becomes easily accessible to the public.

[0173] At 22, new image 7 is acquired from a public source. This acquisition can be performed through various means, such as web scraping, API access, or manual download.Sony Group Corporation et al.

[0174] At 23, it is determined whether the new image 7 includes a signature 9.

[0175] At 24 new image 7 is compared to a database of signed images 2 (e.g., signed image 2 of Fig. 1-3).

[0176] The comparison may be based on a predefined similarity threshold. As the new image 7 is an image generated by the Al-image generator (e.g., 6, Fig. 3) it does not match any of the signed images 3 stored in the database completely, e.g., above the similarity threshold.

[0177] Alternatively, if the image does match the signed image 3, e.g., above the similarity threshold, it may still be determined if the publisher of the image corresponding to signed image 3 is allowed to publish signed image 3.

[0178] If it is determined that the new image 7 does not match one of the signed images 3, any of the processes of Figs. 5 to 8 may proceed.

[0179] Alternatively, the determination at step 23 may be made using an extraction process 8 as described in Fig. 5, which analyzes the new image 7 to detect the presence of the embedded signature 9.

[0180] Alternatively, at step 24 instead or additionally to the image comparison, signature 9 may be matched to the signature(s) 3, e.g., stored in a database, as described in Figs 6 and 7.

[0181] The signed images 3 used for comparison may be signed with the camera-specific signature 2 as determined during the matching process of Figs. 6 or 7. Thus at 24, the new image 7 may be compared to only the signed images 3 with a signature matching the signature 9. Also, at 24 it may be determined which of the signed images 3 with matching signature was used during the image generation of new image 7, e.g., based on the similarity threshold.

[0182] An image ID 4 of the signed image 3 used during image generation of new image 7 may therefore be determined.

[0183] Fig. 5 shows an extraction process of the signature from the newly generated image of Fig. 3. Image 7 is generated based on the process of Fig. 3 and incorporates parts of signed image 3 of Figs. 1 and 3. Image 7 undergoes an extraction process 8 to retrieve the signature 9. The extraction process 8 is conducted by a neural network, as explained in “Fernandez, Pierre, et al. "The stable signature: Rooting watermarks in latent diffusion models." Proceedings of the IEEE / CVF International Conference on Computer Vision. 2023.”

[0184] The extracting process decodes and outputs signature 9 from the image 7.Sony Group Corporation et al.

[0185] A matching process may be conducted to determine whether the extracted signature 9 matches the original camera-specific signature 2 as explained in Figs. 6 and 7.

[0186] Fig. 6 shows a matching process of extracted signature 9 to multiple signatures 2 of a database. Signature 9, extracted from new image 7 during the extraction process 8 of Fig. 5, is provided as input to the matching process 13. Additionally, multiple signatures 2 stored in database 11, each associated with their respective image IDs 4 (as shown in Fig. 2), are provided as input to the matching process 13.

[0187] The matching process 13 involves comparing the extracted signature 9 to the signatures 2 stored in database 11. This comparison is performed using a statistical test to determine whether the extracted signature 9 matches any of the signatures 2 in the database. The test relies on the number of matching bits between the extracted signature 9 and the signatures 2 being tested. The matching process 13 outputs the camera-specific signature 2 as it is the signature 2 with the most matching bits. The matching process may additionally rely on a predefined threshold, wherein output camera-specific signature 2 must include a number of matching bits exceeding a predefined threshold.

[0188] Additionally, to the camera-specific signature 2 also all image IDs 4 corresponding to the camera-specific signature may be output based on the database 11.

[0189] It may be possible that each signature 2 is only associated with one image ID 4. In this case the image ID 4 may be retrieved from a storage, e.g., a blockchain (e.g., BC, Fig. 12), which may be used to confirm ownership of signed image 3 of image ID 4 (see e.g., Fig. 12). Also, ownership of the signature 2 and therefore of new image 7 including the signature 2 may be verified (see e.g., Fig. 11).

[0190] However, in case of multiple image IDs 4 of signed images 3 encoded with the same cameraspecific signature 2, the specific signed image 3 used for the image generation of new image 7 may still be determined, e.g., as described in Fig. 8.

[0191] Alternatively, the camera-specific signature 2 may be retrieved from a database, e.g., blockchain, for confirmation that new image 7 used intellectual property of the owner of the camera-specific signature 2 (see e.g. Fig. 11).

[0192] Fig. 7 shows a matching process of extracted signature to one camera-specific signature.Sony Group Corporation et al.

[0193] Signature 9, extracted from new image 7 during the extraction process 8 of Fig. 5, is provided as input to the matching process 13a. Additionally, camera-specific signature 2 is provided as input to the matching process 13a.

[0194] As describe in Fig. 6 with regard to matching process 13, the matching process 13a involves comparing the extracted signature 9 to the camera-specific signature 2.

[0195] This comparison is performed using a statistical test to determine whether the extracted signature 9 matches the signature 2 in the database. The test relies on the number of matching bits between the extracted signature 9 and the camera-specific signature 2.

[0196] If the number of matching bits exceeds a predefined threshold, the extracted signature 9 is determined to contain the embedded camera-specific signature 2, which is the case in this example.

[0197] Therefore, matching process 13a outputs the matching result 14 “Yes”, as the number of matching bits exceeds the predefined threshold. On this basis new image 7 may be flagged as incorporating the camera-specific signature. Alternatively, the matching result 14 would indicate “No”.

[0198] As explained with regard to Fig. 6, also the image ID(s) 4 corresponding to the camera-specific signature 2 may be determined and output, e.g., based on a database of image IDs 4 associated with the camera-specific signature 2 (e.g., 11 of Fig. 6, see also Fig. 2).

[0199] It may be possible that each signature 2 is only associated with one image ID 4, thereby easily identifying the image 3 used in the image generation process of new image 7.

[0200] In this case the image ID 4 may be retrieved from a storage, e.g., a blockchain (BC, Fig. 12), and ownership of the image ID 4 and therefore of new image 7 including the signature 2 of signed image 3 with image ID 4 may be verified (see e.g., Fig. 12). Also, ownership of signature 2 and therefore of new image 7 including the signature 2 may be verified (see e.g., Fig. 11).

[0201] However, in case of multiple image IDs 4 of signed images 3 encoded with the same cameraspecific signature 2, the specific signed image 3 used for the image generation of new image 7 may still be determined, e.g., as described in Fig. 8.

[0202] Alternatively, the camera-specific signature 2 may be retrieved from a database, e.g., blockchain, for confirmation that new image 7 used intellectual property of the owner of the camera-specific signature 2 (see e.g. Fig. 11).Sony Group Corporation et al.

[0203] Fig. 8 shows the process of identifying the signed image 3 used during the image generation of new image 7.

[0204] In 25 new image 7 is acquired as well as signed images 3 with image IDs 4 corresponding to camera-specific signature 2 matched to extracted signature 9 of new image 7, e.g., as described in Figs. 6 or 7.

[0205] In 26 new image 7 is compared to signed images 3 with image IDs 4. Based on a predetermined similarity threshold the signed image 3 most similar to the new image 7 may be determined as the signed image 3 used in the image generation of new image 7.

[0206] If signed image 3 is identified as the image used during generation of new image 7, the image ID 4 may be retrieved from a blockchain (see Fig. 12) to confirm ownership of signed image 3 of image ID 4. Also, the storage date may be used to confirm that signed image 3 of image ID 4 was generated prior to new image 7.

[0207] Alternatively, the camera-specific signature 2 may be retrieved from a database, e.g., blockchain, for confirmation that new image 7 used intellectual property of the owner of the camera-specific signature 2 (see e.g. Fig. 11).

[0208] The processes of Figs. 2 to 8 which are adapted to images, may in the same vein be also used with regard to audio signals, wherein the signed image is a signed audio signal, the signature is the signature for the device capturing or generating the audio signal or a signature associated with the owner, e.g., the studio or the user of the microphone or the electronic device, and wherein the image ID 4 is a audio ID of the audio signal. An example of such is illustrated in Fig. 9.

[0209] Fig. 9 shows a database generation for signed audio signals and a content identification of an audio signal based on the generated database.

[0210] During database generation 30 multiple signed audio signals 3a to 3c, which are signed with signatures 2a to 2c, are stored via storing process 17 in the database.

[0211] As explained with regard to Fig. 1, the signatures 2a to 2c of the signed audio signals 3a to 3c are imperceptible to human perception as compared to the respective audio signal before signing. Furthermore, the signatures 2a to 2c are robust to modification of the signed audio signals (A signature robust to modification may be embedded as explained for example in Li, Pengcheng, et al. "IDEAW: Robust Neural Audio Watermarking with Invertible Dual -Embedding." arXiv preprint arXiv:2409.19627 (2024).).Sony Group Corporation et al.

[0212] During the storing process 17 the signature 2a of signed audio signal la which includes subsignatures 2al to 2a3 as components of the signature 2a is stored under the audio signal ID 4a. Similarly, the signature 2b of signed audio signal 3b which includes sub-signatures 2b 1 to 2b3 as components of the signature 2b is stored under the audio signal ID 4b and the signature 2c of signed audio signal 3 c which includes sub-signatures 2c 1 to 2c3 as components of the signature 2c is stored under the audio signal ID 4c.

[0213] The storing process 17 includes an extraction of the signature 2a and of sub-signatures 2a 1 to 2a3 with an artificial neural network.

[0214] During the content identification 40, (corresponding to e.g., Figs. 5 to 7) an audio signal 7a, which is publicly available, is provided to the extraction process 8, which extracts signature 9 as well as sub-signature 9a from the audio signal 7a (corresponding to e.g., Fig. 5).

[0215] The extracted signature 9 is then provided to the matching process 13 which compares the extracted signature 9 to the signatures 2a to 2c of the database generated during the database generation 30 (corresponding to e.g., Fig. 6).

[0216] The matching process 13 outputs audio signal ID 4a corresponding to audio signal 2 as an answer from the queried database as well as the sub-signatures 2al to 2a3. The sub-signatures 2al to 2a3 output from the matching process 13 are compared in a second matching process 13b with the extracted sub-signature 9a (see e.g., Fig. 6). Subsequently, the matching process 13b outputs the sub-signature 2al as result. In this way, the second matching process 13b confirms the first result that audio signal 7a was generated based on the audio signal 3a with audio signal ID 4a, e.g., with an Al generator using the audio signal 3 as a basis (corresponding to e.g., Fig. 3).

[0217] Signing of the audio signals 3a-3c may be performed congruently to the signing process 14 of Fig. 1, wherein signing may be conducted by a trained neural network model as for example explained in Li, Pengcheng, et al. "IDEAW: Robust Neural Audio Watermarking with Invertible Dual -Embedding." arXiv preprint arXiv:2409.19627 (2024); or in Singh, Mayank Kumar, et al. "SilentCipher: Deep Audio Watermarking." arXiv preprint arXiv 2406 (2024): 03822.

[0218] Fig. 10 shows a diagram of a system that is configured to host a camera-specific signature for verifying the ownership of the camera and images connected to the camera-specific signature.Sony Group Corporation et al.

[0219] The system includes a distributed computation platform 50 which hosts a blockchain BC, a camera 10, which includes a camera-specific signature (e.g., 2, Fig. 1) and a client computer 100 which is configured to conduct the processes of Figs. 5 to 8.

[0220] The camera 10 stores the camera-specific signature on the blockchain to confirm ownership of images encoded, e.g., at least partly, with the camera-specific signature of the owner of camera 10. For example, images encoded with the camera-specific signature may include images including the camera-specific signature embedded within the respective image, e.g., such that it can be extracted with statistical confidence. This may occur for example when images are generated based on signed images which include the camera-specific signature embedded within the pixel information.

[0221] Alternatively, instead of a camera 10 the camera-specific signature or a signature for signing any type of information, e.g., an audio signal as described in Fig. 9, may be transmitted from a client computer.

[0222] Fig. 11 shows in more detail an exemplifying communication flow related to the system of Fig. 10.

[0223] At 301, the camera 10 generates a keypair comprising a public key pk and a corresponding private key sk.

[0224] At 302, the camera 10 publishes the camera-specific signature 2 and the public key pk signed by the private key sk, on a blockchain BC. This publication process involves creating a transaction on the blockchain BC that records the camera-specific signature 2 at a specific timestamp. The blockchain BC provides an immutable and transparent ledger, ensuring that the camera-specific signature 2 is securely stored and publicly accessible. This step establishes the camera's identity on the blockchain BC, allowing for future verification of images captured by the camera 10. The camera 10 uses the private key sk to sign the transaction. This cryptographic signature provides proof that the transaction was created by the owner of camera 10, i.e., by the owner of the private key sk.

[0225] At step 303, a client computer 100 extracts signature 9 from new image 7, as illustrated in Fig. 5. The extraction process involves analysing the image to detect the embedded signature 9. The client computer 100 uses a trained extractor (e.g., 8, Fig. 5) to decode the hidden message within the image and identify the signature 9. Once extracted, the client computer 100 matches the extracted signature 9 to the camera-specific signature 2 of camera 10, as depicted in Fig. 6 orSony Group Corporation et al.

[0226] Fig. 7. This matching process involves comparing the extracted signature 9 to the cameraspecific signature 2 to verify the camera's identity.

[0227] At step 304, the client computer 100 searches the blockchain BC for the camera-specific signature 2. This search involves querying the blockchain BC to locate the transaction that recorded the camera-specific signature 2.

[0228] At 305, the client computer 100 uses blockchain APIs or other blockchain access methods to retrieve the camera-specific signature 2 transaction which also includes the timestamp of the transaction and the public key pk.

[0229] Furthermore, metadata of the camera, such as the camera's model and serial number, may be stored with the camera-specific signature 2 at 302. Then the client computer 100 may also retrieve the stored metadata during 305.

[0230] At 306, the client computer 100 verifies the ownership of the camera-specific signature 2, and thereby ownership of all images that include camera-specific signature 2 at least partly embedded (e.g., new image 7, Fig. 3), by performing a cryptographic verification process based on the published public key pk and the transaction which was signed by the private key sk. During the verification, the client computer 100 checks whether the digital signature from the transaction matches the expected value when verifying against the public key.

[0231] Alternatively to camera 10 another electronic device may perform the function of camera 10. Also, instead of the camera-specific signature, other types of signatures, for example for signing an audio signal (see Fig. 9) may be used.

[0232] Fig. 12 shows an exemplifying communication flow related to the system of Fig. 10.

[0233] At 307, a hash 4d (image ID 4, Figs. 2, 6 and 8) of the signed image 3 of Fig. 1 is generated by the camera 10. Also, signed image 3 is stored off-chain in a database (see e.g., Fig. 2).

[0234] At 308, the public key pk (e.g., generated at 301, Fig. 11) of the camera 10 and the image hash 4d are published on the blockchain pk. The transaction is signed by private key sk of the camera 10 (e.g., generated at 301, Fig. 11). The blockchain BC provides an immutable and transparent ledger, ensuring that the camera-specific signature 2 and image hash 4d are securely stored and publicly accessible. Camera 10 uses the private key sk to sign the transaction. This cryptographic signature provides proof that the transaction was created by the owner of camera 10, i.e., by the owner of the private key sk.Sony Group Corporation et al.

[0235] Step 303 corresponds to step 303 of Fig. 11, wherein the camera-specific signature 2 corresponding to new image 7 is determined by the client computer 100.

[0236] At step 304, the client computer 100 searches the blockchain for the camera-specific signature 2. This search involves querying the blockchain BC to locate the transaction that recorded the camera-specific signature 2. Alternatively, the client computer 100 may search the blockchain BC for the image hash 4d.

[0237] At 309, the client computer 100 uses blockchain APIs or other blockchain access methods to retrieve the camera-specific signature 2 and image hash 4d transaction which also includes the timestamp of the transaction and the public key pk.

[0238] At 310, the client computer 100 retrieves the signed image 3 from the database and uses image hash 4d to verify that the image hash 4d corresponds to the signed image 3. During the verification, the client computer 100 checks whether the image hash computed from the signed image 3 matches the image hash 4d. Furthermore, by performing a cryptographic verification process based on the published public key pk and the retrieved transaction which was signed by the private key sk ownership of the signed image 3 is verified. During the verification, the client computer 100 checks whether the digital signature from the transaction matches the expected value when verifying against the public key.

[0239] Furthermore, for example prior to process 307, the process 301 of Fig. 1 and / or the process 302 of Fig. 11 may be performed by camera 10.

[0240] Alternatively to camera 10 another electronic device may perform the function of camera 10. Also, instead of the camera-specific signature 2, other types of signatures, for example for signing an audio signal (see Fig. 9) may be used. In the same vein instead of image hash 4d another hash, e.g., of an audio signal may be used.

[0241] Fig. 13 shows a block diagram of a camera configured to perform a signing process of a captured image.

[0242] The camera 10 includes a CPU 101 as processor. Additionally, or alternatively, other computation hardware, such as GPU, TPU, DSP etc. may be used.

[0243] The camera 10 includes an imaging unit 106, microphone(s) 107 and loudspeaker(s) 108 that are connected to the processor 101.

[0244] The processor 101 may for example implement the image capture of captured image 1 of Fig. 1 and the signing process of Fig. 1. Furthermore, the processor 101 may for example implementSony Group Corporation et al.

[0245] the process of Fig. 2. The processor 101 may also implement the process 301, 302 of Fig. 11 and / or the process of 307, 308 of Fig. 12.

[0246] The microphone 107 may be configured to receive any kind of audio signal. Loudspeaker 108 may include one or more loudspeakers.

[0247] The camera 10 may include one or more cameras, such as an RGB camera, a ToF camera, for example, an iToF or dTof or the like.

[0248] The camera 10 further includes a user interface 109 that is connected to the processor 101. This user interface 109 acts as a man-machine interface and enables a dialogue between a user and the camera 10.

[0249] The camera 10 further includes a Bluetooth interface 104, and a WLAN interface 105. These units 104, 105 act as I / O interfaces for data communication with external devices. For example, additional loudspeakers, microphones, and cameras with WLAN or Bluetooth connection may be coupled to the processor 101 via these interfaces 104 and 105.

[0250] The camera 10 further includes a data storage 102 and a data memory 103 (here a RAM). The data memory 103 is arranged to temporarily store or cache data or computer instructions for processing by the processor 101.

[0251] The data storage 102 is arranged as a long-term storage, e.g., the captured image 1 or signed image 3 of Fig. 1 may be stored.

[0252] Furthermore, the camera 10 includes a neural network processor 110. The neural network processor 110 may include a graphics processing unit (GPU) and / or a tensor processing unit 20 (TPU). The neural network processor 110 may be configured to perform the signing process 14 of Fig. 1.

[0253] Fig. 14 shows a block diagram of an electronic device configured to perform any one of the processes of Figs. 1, 2, 4 to 9.

[0254] The electronic device 200 includes a CPU 201 as processor. Additionally, or alternatively, other computation hardware, such as GPU, TPU, DSP etc. may be used.

[0255] The electronic device 200 includes camera(s) 206, microphone(s) 207 and loudspeaker(s) 208 that are connected to the processor 201.

[0256] The processor 201 may for example implement the process of Fig. 4, the extraction process 8 of Figs. 2, the matching process of Figs. 6 or 7 and / or the process of Fig. 8. The processor may for example implement the database generation process 30 and / or the content identification processSony Group Corporation et al.

[0257] 40 of Fig. 9. The processor 201 may also implement the signing process of Fig. 1 or a corresponding signing process of other type of data, such as audio data, and / or the process of Fig. 2 or a corresponding process for other type of data, such as audio data. The processor may implement any one or more of the processes 301 to 306 of Fig. 11 and / or 307 to 310 of Fig. 12. An electronic device 200 may be used instead of camera 10 in the system of Fig. 10.

[0258] The microphone 107 may be configured to receive any kind of audio signal. Loudspeaker 108 may be headphones, e.g., on-ear, in-ear, over-ear, wireless headphones and the like, or may consist of one or more loudspeakers that are distributed over a predefined space and are configured to render any kind of audio, such as 3D audio.

[0259] The camera 106 may include one or more cameras, such as an RGB camera, a ToF camera, for example, an iToF or dTof or the like.

[0260] The electronic device 200 further includes a user interface 109 that is connected to the processor 101. This user interface 109 acts as a man-machine interface and enables a dialogue between a user and the electronic device 200. The user interface may be used to notify a user, for example of a flagged new image 7 or similar data of different type which was determined to be generated based on property of the user.

[0261] The electronic device 200 further includes a Bluetooth interface 104, and a WLAN interface 105. These units 104, 105 act as I / O interfaces for data communication with external devices. For example, additional loudspeakers, microphones, and cameras with WLAN or Bluetooth connection may be coupled to the processor 101 via these interfaces 104 and 105. Also computation platform 50 of Fig. 10 may be connected to the electronic device 200 based on units 104 and / or 105.

[0262] The electronic device 200 further includes a data storage 102 and a data memory 103 (here a RAM). The data memory 103 is arranged to temporarily store or cache data or computer instructions for processing by the processor 101. The data storage 102 is arranged as a long-term storage.

[0263] The connection between the processor 101 and the camera 106 may include a camera serial interface (CSI). The CSI is an interface between a camera 106 and a host processor 101. Thus, control signals and data from the processor 101 to the camera 106 as well as from the camera 106 to the processor 101 may be sent.

[0264] Furthermore, the electronic device 200 includes a neural network processor 210. The neural network processor 210 may include a graphics processing unit (GPU) and / or a tensor processingSony Group Corporation et al.

[0265] unit 20 (TPU). The neural network processor 210 may be configured to perform the signing process 14 of Fig. 1 and / or the extraction process 8 of Figs. 5 or 8.

[0266] The electronic device 200 may be a stationary electronic device, such as a laptop computer, a personal computer, or a mobile device of any other kind of portable or wearable device, for example, a smartphone, a tablet computer, smart glasses, head mounted displays (HMDs), earphones or other types of smart wearable devices, or the like. The electronic device 200 may correspond to client computer 100 of Figs. 10 to 12.

[0267] It should be recognized that the embodiments describe methods with an exemplary ordering of method steps. The specific ordering of method steps is however given for illustrative purposes only and should not be construed as binding. For example, the ordering of 23 and 24 in the embodiment of Fig. 4 may be exchanged. Other changes of the ordering of method steps may be apparent to the skilled person.

[0268] Please note that the division of the CPU 101 and 201 into units is only made for illustration purposes and that the present disclosure is not limited to any specific division of functions in specific units. For instance, the CPU 101, 201 could be implemented by a respective programmed processor, field programmable gate array (FPGA) and the like.

[0269] A method for controlling a camera 10, such as the camera discussed above, is described under reference of Figs. 1, 2 as well as Figs. 11 and 12. The method can also be implemented as a computer program causing a processor, such as processor 101 discussed above, to perform the method, when being carried out on the processor 201. A method for controlling an electronic device 200, such as the device discussed above, is described under reference of Figs. 1, 2 and 4 to 12. The method can also be implemented as a computer program causing a computer and / or a processor, such as processor 201 discussed above, to perform the method, when being carried out on the computer and / or processor 201. In some embodiments, also a non-transitory computer-readable recording medium is provided that stores therein a computer program product, which, when executed by a processor, such as the processor described above, causes the method described to be performed.

[0270] All units and entities described in this specification and claimed in the appended claims can, if not stated otherwise, be implemented as integrated circuit logic, for example on a chip, and functionality provided by such units and entities can, if not stated otherwise, be implemented by software.Sony Group Corporation et al.

[0271] In so far as the embodiments of the disclosure described above are implemented, at least in part, using software-controlled data processing apparatus, it will be appreciated that a computer program providing such software control and a transmission, storage or other medium by which such a computer program is provided are envisaged as aspects of the present disclosure.

[0272] Note that the present technology can also be configured as described below.

[0273] [1] A camera (10) configured to:

[0274] capture an image (1); and

[0275] sign (14) the captured image (1) with a signature (2) based on changing the image pixels to generate a signed image (3), wherein the difference between the signed image (3) and the captured image (1) is imperceptible to human visual perception.

[0276] [2] The camera ( 10) of

[0001] ,

[0277] wherein signing (14) the captured image (1) is based on a neural network.

[0278] [3] The camera (10) of [1] or [2],

[0279] wherein signing (14) the captured image (1) with the signature (2) includes adding a hidden perturbation to the image (1) as the signature (2).

[0280] [4] The camera (10) of any one of [1] to [3],

[0281] wherein signing (14) the captured image (1) with the signature (2) includes adding the signature (2) to the captured image (1) such that it can be extracted from the signed image (3).

[0282] [5] The camera (10) of any one of [1] to [4],

[0283] wherein the signature (2) within the signed image (3) is configured to be resilient to image modification.

[0284] [6] The camera (10) of any one of [1] to [5],

[0285] wherein the signature (2) within the signed image (3) is configured to be resilient, such that the signature (2) can be extracted from an image (7) generated by artificial intelligence assisted image generation based on the signed image (3) with statistical confidence.

[0286] [7] The camera (10) of any one of [1] to [6],

[0287] wherein the signature (2) is a camera-specific signature associated with a cryptographic key (sk, pk).

[0288] [8] The camera (10) of any one of [1] to [7] further configured to:Sony Group Corporation et al.

[0289] provide (15) an image identification (4) assigned to the signed image (3);

[0290] save (16, 302, 308) the image identification (4) and the signature (2) associated with each other to a distributed ledger (BC, 50).

[0291] [9] An electronic device (100) configured to extract (8) a signature (9) from a target image (7) generated based on the signed image (3) output by the camera (10) of any one of [1] to [8].

[0292]

[0010] The electronic device (100) of [9], configured to match (13, 13a) the extracted signature (9) of the target image (7) to the signature (2) of the signed image (3).

[0293]

[0011] The electronic device (100) of [9] or

[0010] , configured to verify (306, 310) the ownership of the extracted signature (9) matched to the signature (2) of the signed image (3).

[0294]

[0012] The electronic device (100) of

[0011] , wherein verifying (306, 310) the ownership includes retrieving the signature (2) of the signed image (3) from a distributed ledger (BC, 50).

[0295]

[0013] An electronic signing device (10, 200) configured to:

[0296] sign (14) a first signal (1) with a signature (2, 2a-2c) to generate a signed signal (3, 3a-3c), such that the difference between the signed signal (3, 3a-3c) and the first signal (1) is imperceptible to human perception;

[0297] provide (15) the signed signal (3, 3a-3c) with a signal identification (4, 4a-4c); and save (16, 302, 308) the signature (2, 2a-2c) and the signal identification (4, 4a-4c) in a distributed ledger (BC, 50).

[0298]

[0014] The electronic signing device (10, 200) of

[0013] ,

[0299] wherein signing (14) the first signal (1) is based on a neural network.

[0300]

[0015] The electronic signing device (10, 200) of

[0013] or

[0014] ,

[0301] wherein signing (14) the first signal (1) with the signature (2, 2a-2c) includes adding a hidden perturbation to the first signal (1) as the signature (2, 2a-2c).

[0302]

[0016] The electronic signing device (10, 200) of any one of

[0013] to

[0015] ,

[0303] wherein signing (14) the first signal (1) with the signature (2, 2a-2c) includes adding the signature (2, 2a-2c) to the first signal (1) such that it can be extracted from the signed signal (3, 3a-3c).

[0304]

[0017] The electronic signing device (10, 200) of any one of

[0013] to

[0016] ,Sony Group Corporation et al.

[0305] wherein the signature (2, 2a-2c) within the signed signal (3, 3a-3c) is configured to be resilient to signal modification.

[0306]

[0018] The electronic signing device (10, 200) of any one of

[0013] to

[0017] ,

[0307] wherein the signature (2, 2a-2c) within the signed signal (3, 3a-3c) is configured to be resilient, such that the signature (2, 2a-2c) can be extracted from data (7, 7a) generated by artificial intelligence assisted data generation based on the signed signal (3, 3a-3c) with statistical confidence.

[0308]

[0019] An electronic extraction device (100, 200) configured to retrieve (305, 309) the signature (2, 2a-2c) stored by the electronic signing device (10, 200) of any one of

[0013] to

[0018] on the distributed ledger (BC, 50) from the distributed ledger (BC, 50), and to extract (8) a signature (9, 9a) from data (7, 7a) generated based on the signed signal (3, 3a-3c) output by the electronic signing device (10, 200) of any one of

[0013] to

[0018] ,

[0309]

[0020] The electronic extraction device (100, 200) of

[0019] , configured to match (13, 13a, 13b) the extracted signature (9, 9a) of the data (7, 7a) to the signature (2, 2a-2c) of the signed signal (3, 3a-3c).

[0310]

[0021] The electronic extraction device (100, 200) of

[0019] or

[0020] , configured to verify (306, 310) the ownership of the extracted signature (9, 9a) based on the retrieved signature (2, 2a-2c).

[0311]

[0022] A system comprising a distributed ledger (BC, 50) configured to store a signature (2, 2a-2c), the signature (2, 2a-2c) enabling signing (14) of a first signal (1) to generate a signed signal (3, 3a-3c), such that the difference between the signed signal (3, 3a-3c) and the first signal (1) is imperceptible to human perception.

[0312]

[0023] The system of

[0022] , wherein signing is performed by the electronic signing device (10, 100) of any one of

[0013] to

[0018] or the camera of any one of [1] to [8],

[0313]

[0024] The system of

[0022] or

[0023] , wherein the system further includes the electronic signing device (10, 100) of any one of

[0013] to

[0018] or the camera of any one of [1] to [8],

[0314]

[0025] The system of any one of

[0022] to

[0024] , wherein the system further includes the electronic device (100) of any one of [9] to

[0012] and / or the electronic extraction device (100, 200) of any one of

[0019] to

[0021] ,

[0315]

[0026] A method for controlling the camera (10) of any one of [1] to [8] comprising the steps of capturing an image (1); andSony Group Corporation et al.

[0316] signing (14) the captured image (1) with a signature (2) based on changing the image pixels to generate a signed image (3), wherein the difference between the signed image (3) and the captured image (1) is imperceptible to human visual perception.

[0317]

[0027] The method of

[0026] ,

[0318] wherein signing (14) the captured image (1) is based on a neural network.

[0319]

[0028] The method of

[0026] or

[0027] ,

[0320] wherein signing (14) the captured image (1) with the signature (2) includes adding a hidden perturbation to the image (1) as the signature (2).

[0321]

[0029] The method of any one of

[0026] to

[0028] ,

[0322] wherein signing (14) the captured image (1) with the signature (2) includes adding the signature (2) to the captured image (1) such that it can be extracted from the signed image (3).

[0323]

[0030] The method of any one of

[0026] to

[0029] ,

[0324] wherein the signature (2) within the signed image (3) is configured to be resilient to image modification.

[0325]

[0031] The method of any one of

[0026] to

[0030] ,

[0326] wherein the signature (2) within the signed image (3) is configured to be resilient, such that the signature (2) can be extracted from an image (7) generated by artificial intelligence assisted image generation based on the signed image (3) with statistical confidence.

[0327]

[0032] The method of any one of

[0026] to

[0031] ,

[0328] wherein the signature (2) is a camera-specific signature associated with a cryptographic key (sk, pk).

[0329]

[0033] The method of any one of

[0026] to

[0032] further including the steps of:

[0330] providing (15) an image identification (4) assigned to the signed image (3); saving (16, 302, 308) the image identification (4) and the signature (2) associated with each other to a distributed ledger (BC, 50).

[0331]

[0034] A method comprising the step of extracting (8) a signature (9) from a target image (7) generated based on the signed image (3) output by the camera (10) of any one of [1] to [8],

[0035] The method of

[0034] , further including the step of matching (13, 13a) the extracted signature (2) of the target image (7) to the signature (2) of the signed image (3).Sony Group Corporation et al.

[0332]

[0036] The method of

[0034] or

[0035] , further including the step of verifying (306, 310) the ownership of the extracted signature (9) matched to the signature (2) of the signed image (3).

[0333]

[0037] The method of any one of

[0034] to

[0036] , wherein verifying (306, 310) the ownership includes retrieving (305, 309) the signature (2) of the signed image (3) from a distributed ledger (BC, 50).

[0334]

[0038] A method comprising the steps of

[0335] signing (14) a first signal (1) with a signature (2, 2a-2c) to generate a signed signal (3, 3a-3c), such that the difference between the signed signal (3, 3a-3c) and the first signal (1) is imperceptible to human perception;

[0336] providing (15) the signed signal (1) with a signal identification (4, 4a-4c); and

[0337] saving (16, 302, 308) the signature (2) and the signal identification (4, 4a-4c) in a distributed ledger (50, BC).

[0338]

[0039] The method of

[0039] ,

[0339] wherein signing (14) the first signal (1) is based on a neural network.

[0340]

[0040] The method of

[0038] or

[0039] ,

[0341] wherein signing (14) the first signal (1) with the signature (2, 2a-2c) includes adding a hidden perturbation to the first signal (1) as the signature (2, 2a-2c).

[0342]

[0041] The method of any one of

[0038] to

[0040] ,

[0343] wherein signing (14) the first signal (1) with the signature (2, 2a-2c) includes adding the signature (2, 2a-2c) to the first signal (1) such that it can be extracted from the signed signal (3, 3a-3c).

[0344]

[0042] The method of any one of

[0038] to

[0041] ,

[0345] wherein the signature (2, 2a-2c) within the signed signal (3, 3a-3c) is configured to be resilient to signal modification.

[0346]

[0043] The method of any one of

[0038] to

[0042] ,

[0347] wherein the signature (2, 2a-2c) within the signed signal (3, 3a-3c) is configured to be resilient, such that the signature (2, 2a-2c) can be extracted from data (7, 7a) generated by artificial intelligence assisted data generation based on the signed signal (3, 3a-3c) with statistical confidence.Sony Group Corporation et al.

[0348]

[0044] A method comprising the steps of

[0349] retrieving (305, 309) the signature (2), stored by the electronic signing device of any one of

[0013] to

[0017] on the distributed ledger (BC, 50) from the distributed ledger (BC, 50), and extracting (8) a signature (9, 9a) from data (7, 7a) generated based on the signed signal (3, 3a-3c) output by the electronic signing device of any one of

[0013] to

[0017] ,

[0350]

[0045] The method of

[0044] further including the step of matching (13, 13a, 13b) the extracted signature (9, 9a) of the data (7, 7a) to the signature (2, 2a-2c) used for signing (14) the signed signal (3, 3a-3c).

[0351]

[0046] The method of

[0044] or

[0045] further including the step of verifying (306, 310) the ownership of the extracted signature (9, 9a) based on the retrieved signature (2, 2a-2c).

[0352]

[0047] A method comprising the steps of

[0353] storing (302, 308) a signature (2, 2a-2c) of a signed signal (3, 3a-3c) on a distributed ledger (BC, 50);

[0354] retrieving (305, 309) the signature (2, 2a-2c) of the signed signal (3, 3a-3c) from the distributed ledger (BC, 50); and

[0355] verifying (306, 310) ownership of data (7, 7a) based on the retrieved signature (2, 2a-2c).

[0356]

[0048] A computer program comprising program code causing a computer to perform the method according to anyone of

[0026] to

[0046] , when being carried out on a computer.

[0357]

[0049] A non-transitory computer-readable recording medium that stores therein a computer program product, which, when executed by a processor, causes the method according to anyone of

[0026] to

[0046] to be performed.

Claims

Sony Group Corporation et al.CLAIMS1. A camera configured to:capture an image; andsign the captured image with a signature based on changing the image pixels to generate a signed image, wherein the difference between the signed image and the captured image is imperceptible to human visual perception.

2. The camera of claim 1,wherein signing the captured image is based on a neural network.

3. The camera of claim 1,wherein signing the captured image with the signature includes adding a hidden perturbation to the image as the signature.

4. The camera of claim 1,wherein signing the captured image with the signature includes adding the signature to the captured image such that it can be extracted from the signed image.

5. The camera of claim 1,wherein the signature within the signed image is configured to be resilient to image modification.

6. The camera of claim 5,wherein the signature within the signed image is configured to be resilient, such that the signature can be extracted from an image generated by artificial intelligence assisted image generation based on the signed image with statistical confidence.

7. The camera of claim 1,wherein the signature is a camera-specific signature associated with a cryptographic key.

8. The camera of claim 1 further configured to:provide an image identification assigned to the signed image;save the image identification and the signature associated with each other to a distributed ledger.Sony Group Corporation et al.

9. An electronic device configured to extract a signature from a target image generated based on the signed image output by the camera of claim 1.

10. The electronic device of claim 9, configured to match the extracted signature of the target image to the signature of the signed image.

11. The electronic device of claim 10, configured to verify the ownership of the extracted signature matched to the signature of the signed image.

12. The electronic device of claim 11, wherein verifying the ownership includes retrieving the signature of the signed image from a distributed ledger.

13. An electronic signing device configured to:sign a first signal with a signature to generate a signed signal, such that the difference between the signed signal and the first signal is imperceptible to human perception;provide the signed signal with a signal identification; andsave the signature and the signal identification in a distributed ledger.

14. The electronic signing device of claim 13,wherein signing the first signal is based on a neural network.

15. The electronic signing device of claim 13,wherein signing the first signal with the signature includes adding a hidden perturbation to the first signal as the signature.

16. The electronic signing device of claim 13,wherein signing the first signal with the signature includes adding the signature to the first signal such that it can be extracted from the signed signal.

17. The electronic signing device of claim 13,wherein the signature within the signed signal is configured to be resilient to signal modification.

18. The electronic signing device of claim 13,wherein the signature within the signed signal is configured to be resilient, such that the signature can be extracted from data generated by artificial intelligence assisted data generation based on the signed signal with statistical confidence.Sony Group Corporation et al.

19. An electronic extraction device configured to retrieve the signature stored by the electronic signing device of claim 13 on the distributed ledger from the distributed ledger, and to extract a signature from data generated based on the signed signal output by the electronic signing device of claim 13.

20. The electronic extraction device of claim 19, configured to match the extracted signature of the data to the signature of the signed signal.

21. The electronic extraction device of claim 19, configured to verify the ownership of the extracted signature based on the retrieved signature.

22. A system comprising a distributed ledger configured to store a signature, the signature enabling signing of a first signal to generate a signed signal, such that the difference between the signed signal and the first signal is imperceptible to human perception.

23. A method for controlling the camera of claim 1 comprising the steps ofcapturing an image; andsigning the captured image with a signature based on changing the image pixels to generate a signed image, wherein the difference between the signed image and the captured image is imperceptible to human visual perception.

24. A method comprising the step of extracting a signature from a target image generated based on the signed image output by the camera of claim 1.

25. The method of claim 24, further including the step of matching the extracted signature of the target image to the signature of the signed image.

26. The method of claim 24, further including the step of verifying the ownership of the extracted signature matched to the signature of the signed image.

27. The method of claim 26, wherein verifying the ownership includes retrieving the signature of the signed image from a distributed ledger.

28. A method comprising the steps ofsigning a first signal with a signature to generate a signed signal, such that the difference between the signed signal and the first signal is imperceptible to human perception;providing the signed signal with a signal identification; andsaving the signature and the signal identification in a distributed ledger.Sony Group Corporation et al.

29. A method comprising the steps ofretrieving the signature stored by the electronic signing device of claim 13 on the distributed ledger from the distributed ledger, and extracting a signature from data generated based on the signed signal output by the electronic signing device of claim 13.

30. The method of claim 29 further including the step of matching the extracted signature of the data to the signature used for signing the signed signal.

31. The method of claim 29 further including the step of verifying the ownership of the extracted signature based on the retrieved signature.

Images