A method and electronic device for generating shielded image against ai manipulation

Abstract

The present disclosure relates to a method electronic devicefor generating shielded image against artificial intelligence manipulation. The method includes generating a plurality of compressed images with different compression ratio from an input image; generating an intermediate image by adding a predefined noise to input image; selecting reference image from plurality of predefined reference images based on an attribute of each of the plurality of predefined reference images; calculating an embedding for each of the generated plurality of compressed images, the reference image, and the generated intermediate image; and updating the intermediate image to generate the shielded image by iteratively estimating noise based on the calculated embedding of the generated plurality of compressed images, the reference image, and generated intermediate image; and adding the estimated noise to the intermediate image.

Description

A METHOD AND ELECTRONIC DEVICE FOR GENERATING SHIELDED IMAGE AGAINST AI MANIPULATION

[0001] The present disclosure generally relates to image processing. Particularly, but not exclusively, the present disclosure relates to a method and an electronic device for generating shielded image against manipulation.

[0002] In recent times, image editing is performed using simple prompt, and any third-party user may transform an existing image of a user where the user may be swapped with another user or object which leads to creation of deepfakes. In other words, the ease of image editing / generation has opened doors for creativity, but it also raises concerns about their potential misuse, e.g., creation of deepfakes. Deepfakes may be defined as creating entirely original content where the user is represented doing that may not be performed by him. The greatest danger posed by deepfakes is their ability to spread false information that appears to come from trusted sources. While deepfakes pose serious threats, they also have legitimate uses, such as video game audio and entertainment, and customer support and caller response applications, such as call forwarding and receptionist services.

[0003] To counter the above-mentioned threat, there are many approaches like deepfake detection and Artificial Intelligence (AI) watermarking. However, these approaches are remedial measures that do not prevent harm. The existing technology discloses the detection of fake images, but current solutions do not provide protection against all Generative AI models that are capable of editing image with simple prompts. One of the existing technologies describes protecting an image from manipulation by adding noise to the image. However, when the third-party user tries to modify the image, the image may be initially compressed into multiple compression ratios where the added noise may be lost. Further, the image may be easily modified and misused.

[0004] Therefore, there is a need for a mechanism that may protect the image from AI manipulation even after the image may be compressed into multiple compression ratios and protecting the image from AI manipulation by adding necessary amount of noise to the image.

[0005] The information disclosed in this background of the disclosure section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.

[0006] One or more shortcomings discussed above are overcome, and additional advantages and features are provided by the present disclosure. Other embodiments and aspects of the disclosure are described in detail herein and are considered a part of the disclosure.

[0007] In a non-limiting embodiment of the present disclosure, a method for generating shielded image against artificial intelligence manipulation is disclosed. The method includes generating a plurality of compressed images with a different compression ratio from an input image. The method comprises generating an intermediate image by adding a predefined noise to the input image. The method comprises selecting a reference image from a plurality of predefined reference images based on an attribute of the plurality of predefined reference images that differs from the attribute of the input image. The method includes calculating an embedding for each of the generated plurality of compressed images, the reference image, and the generated intermediate image. The method comprises updating the intermediate image to generate the shielded image by iteratively (ⅰ) estimating a noise based on the calculated embedding of the generated plurality of compressed images, the reference image, and the generated intermediate image, and (ⅱ) adding the estimated noise to the intermediate image.

[0008] In a non-limiting embodiment of the present disclosure, the method further comprises dividing an image into a plurality of sub-images. A shielded image may be generated for each of the plurality of sub-images. The method comprises combining the shielded image generated for each of the plurality of sub-images to generate a combined shielded image having a resolution same as the resolution of the image

[0009] In a non-limiting embodiment of the present disclosure, to select the reference image, the method discloses determining the attribute of the input image and the attribute of each of the plurality of predefined reference images. Further, the determined attribute of the input image is compared with the attribute of the plurality of predefined reference images. Then the reference image from the plurality of predefined reference images may be selected that has highest attribute difference from the attribute of the input image, based on the comparison.

[0010] In a non-limiting embodiment of the present disclosure, a predefined noise is added to the input image by predicting the predefined noise for the input image using a pre-trained noise prediction machine learning (ML) model. Further, the predicted predefined noise is added to the input image to generate the intermediate image.

[0011] In a non-limiting embodiment of the present disclosure, to calculate the embeddings, edges are detected within each of the plurality of compressed images, the intermediate image, and the reference image. Further, the embeddings are calculated based on the detected edges for each of the plurality of compressed images, the reference image, and the intermediate image.

[0012] In a non-limiting embodiment of the present disclosure, to add the estimated noise to the intermediate image, at least one Region of Interest (ROI) in the input image is identified. The estimated noise may be filtered based on the at least one ROI. The filtered noise may be added to the intermediate image.

[0013] In a non-limiting embodiment of the present disclosure, estimating the noise comprises calculating a first loss value by comparing the calculated embedding of the generated plurality of compressed images and the reference image. Further, calculating a second loss value by comparing the calculated embedding of the input image and the reference image. Finally, based on the calculated first loss value and the second loss value, the noise is estimated.

[0014] In a non-limiting embodiment of the present disclosure, to update the intermediate image, a difference between edges present in the updated intermediate image with the edges present in the input image may be determined. The intermediate image may be updated till the difference between the edges present in the updated intermediate image and the edges present in the input image is greater than a predefined threshold.

[0015] In a non-limiting embodiment of the present disclosure, a computer-readable storage medium storing one or more instructions is disclosed. The one or more instructions, when executed by at least one processor individually or collectively, cause the at least one processor to perform any method described herein or any combination of steps, operations, and / or functionalities described herein.

[0016] In a non-limiting embodiment of the present disclosure, an electronic device to generate shielded image against AI manipulation is disclosed. The electronic device comprises at least one processor comprising processing circuitry and memory coupled to the processor. The memory may comprise one or more storage media storing one or more instructions. When executed by the at least one processor individually or collectively, the one or more instructions cause the electronic device to generate a plurality of compressed images with a different compression ratio from an input image. The one or more instructions, when executed by the at least one processor individually or collectively, cause the electronic device to generate an intermediate image by adding a predefined noise to the input image and select a reference image from a plurality of predefined reference images based on an attribute of each of the plurality of predefined reference images. The one or more instructions, when executed by the at least one processor individually or collectively, cause the electronic device to calculate an embedding for each of the generated plurality of compressed images, the reference image, and the generated intermediate image. The one or more instructions, when executed by the at least one processor individually or collectively, cause the electronic device to update the intermediate image to generate the shielded image by iteratively (ⅰ) estimating a noise based on the calculated embedding of the generated plurality of compressed images, the reference image, and the generated intermediate image, and (ⅱ) adding the estimated noise to the intermediate image.

[0017] The foregoing summary is illustrative only and is not intended to be in any way limiting. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features will become apparent by reference to the drawings and the following detailed description.

[0018] The embodiments of the disclosure itself, as well as a preferred mode of use, further objectives and advantages thereof, will best be understood by reference to the following detailed description of an illustrative embodiment when read in conjunction with the accompanying drawings. One or more embodiments are now described, by way of example only, with reference to the accompanying drawings in which:

[0019] Fig. 1illustrates an electronic device for generating shielded image against artificial intelligence (AI) manipulation, in accordance with various embodiments of the present disclosure.

[0020] Fig. 2illustrates a block diagram of an electronic device to generate shielded image against AI manipulation, in accordance with various embodiments of the present disclosure.

[0021] Fig. 3A illustrates a diagram to generate a plurality of compressed images with a different compression ratio, in accordance with various embodiments of the present disclosure.

[0022] Fig. 3B illustrates a diagram to generate the shielded image from the generated plurality of compressed images, in accordance with various embodiments of the present disclosure.

[0023] Fig. 3Cillustrates a diagram to generate an intermediate image by adding a predefined noise, in accordance with various embodiments of the present disclosure.

[0024] Fig. 3Dillustrates a diagram to select a reference image from a plurality of predefined reference images, in accordance with various embodiments of the present disclosure.

[0025] Fig. 3Eillustrates a diagram to select of reference image from reference images databank, in accordance with various embodiments of the present disclosure.

[0026] Fig. 3Fillustrates a diagram to calculate an embedding, in accordance with various embodiments of the present disclosure.

[0027] Fig. 3Gillustrates an embedding of an image and the embedding indicating disturbed edges of the image, in accordance with various embodiments of the present disclosure.

[0028] Fig. 3Hillustrates a diagram to estimate noise, in accordance with various embodiments of the present disclosure.

[0029] Fig. 3Iillustrates a diagram to identify Region of Interest (ROI) and adding noise to the input image, in accordance with various embodiments of the present disclosure.

[0030] Figs. 3J-Lshow a use-case scenario indicating adding of the estimated noise to ROI in an image, in accordance with various embodiments of the present disclosure.

[0031] Fig. 3Millustrates a diagram to estimate noise, in accordance with various embodiments of the present disclosure.

[0032] Fig. 4illustrates a flowchart of a method for generating shielded image against AI manipulation, in accordance with various embodiments of the present disclosure.

[0033] Fig. 5 illustrates a computer electronic device in accordance with various embodiments of the present disclosure.

[0034] The figures depict embodiments of the disclosure for purposes of illustration only. One skilled in the art will readily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.

[0035] In the present document, the word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any embodiment or implementation of the present subject matter described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

[0036] While the disclosure is susceptible to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the drawings and will be described in detail below. It should be understood, however, that it is not intended to limit the disclosure to the particular forms disclosed, but on the contrary, the disclosure is to cover all modifications, equivalents, and alternative falling within the scope of the disclosure.

[0037] The terms "comprises," "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a setup, device, or method that comprises a list of components or steps does not include only those components or steps but may include other components or steps not expressly listed or inherent to such setup or device or method. In other words, one or more elements in a electronic device or apparatus proceeded by "comprises a" does not, without more constraints, preclude the existence of other elements or additional elements in the electronic device or apparatus.

[0038] In the following detailed description of the embodiments of the disclosure, reference is made to the accompanying drawings that form a part hereof, and in which are shown by way of illustration specific embodiments in which the description may be practiced. These embodiments are described in sufficient detail to enable those skilled in art to practice the disclosure, and it is to be understood that other embodiments may be utilized and that changes may be made without departing from the scope of the present disclosure. The following description is, therefore, not to be taken in a limiting sense.

[0039] Deepfake technology is creating highly realistic fake videos or pictures that may be difficult to distinguish from real ones. The rise of social media platforms and online platforms has raised the challenges and concerns of detecting misinformation and malicious content. Any third-party platforms or the user may easily modify the video or images posted on the social media platforms using simple prompts in form of text or voice. Thus, the present disclosure describes a method and electronic device for generating shielded image against AI manipulation.

[0040] Fig. 1illustrates an electronic deviceelectronic device 100 for generating shielded image against AI manipulation, in accordance with various embodiments of the present disclosure.

[0041] The electronic device 100 may be implemented as a user device (or a user terminal or user equipment), e.g., a smartphone or a mobile device, for generating shielded image against AI manipulation. However, this should not be construed as a limitation. The electronic device 100 for generating shielded image against AI manipulation can be implemented as any of electronic device such as laptop, personal computer (PC), tablet PC, server, wearable device, home appliance and the like.

[0042] As per the present disclosure, the electronic device 100 for generating the shielded image against AI manipulation comprises a processor 103, an Input / Output (I / O) interface 105 and a memory 107. In some embodiments, the electronic device 100 may exclude at least one of these components or may add at least one other component.

[0043] The processor 103 includes one or more processing devices or processing circuitry, such as one or more microprocessors, microcontrollers, digital signal processors (DSPs), application specific integrated circuits (ASICs), or field programmable gate arrays (FPGAs). In some embodiments, the processor 103 includes one or more of a central processing unit (CPU), an application processor (AP), a communication processor (CP), or a graphics processor unit (GPU). The processor 103 is able to perform control on at least one of the other components of the electronic device 100 and / or perform an operation or data processing relating to image processing or other functions. As described in more detail below, the processor 103 may perform various operations related to generating a shield image from a given image.

[0044] According to an embodiment of the present disclosure, the electronic device 100 may comprise a plurality of processors, including the processor 103. The plurality of processors may execute one or more commands, instructions, and / or program code, thereby causing the electronic device 100 to perform any combination of operations described herein.

[0045] The I / O interface 105 serves as an interface that may, for example, transfer commands or data input from a user or other external devices to other component(s) of the electronic device 100. The I / O interface 105 may also output commands or data received from other component(s) of the electronic device 100 to the user or the other external device. In an embodiment, the I / O interface 105 may comprise one or more input devices such as a keyboard, a touch screen, a sensor, and / or a microphone. The I / O interface 105 may comprise one or more output devices such as a display device and / or a speaker.

[0046] The memory 107 may include a volatile and / or non-volatile memory. For example, the memory 107 may store commands, instructions, program codes, and / or data related to at least one other component of the electronic device 100. According to embodiments of this disclosure, the memory 107 may store software and / or a program. The program may include, for example, a kernel, middleware, an application programming interface (API), and / or an application program (or an application).

[0047] According to an embodiment of the present disclosure, the memory 107 may include one or more storage media. The storage medium may store one or more instructions. The one or more instructions may be executed by the processor 103 and / or other processor(s). When the one or more instructions are executed by the processor 103 and / or other processor(s) individually or collectively, the one or more instructions may cause the electronic device 100 to perform any combination of operations described herein.

[0048] In some embodiments, the components of the electronic device may communicate with each other via a bus included in the electronic device 100. The bus may include a circuit for connecting the components included in the electronic device 100 with one another and for transferring communications (such as control messages and / or data) between the components.

[0049] In recent times, images may be easily edited using different online AI tools with the help of simple commands. However, this leads to serious concerns about potential misuse, e.g., creation of deepfakes. Deepfakes may include images, videos, or audio but not limited thereto which may be edited or generated using artificial intelligence tools that depicts real or non-existent people. Thus, protecting the images, video and audio from AI manipulation plays a vital role. Therefore, the present disclosure describes generation of a shielded image that may not be modified using any of the AI tools. To generate the shielded image, devices such as smartphone, tablets, personal computer, laptop may generate the shielded image which may not be edited or manipulated using the AI tools.

[0050] For instance, the user may capture the plurality of images, and the captured images may be stored in the memory 107 of the electronic device 100. When the user wishes to shield the image, the processor 103 of the electronic device 100 may receive the user selected image from the memory of the smartphone and may be processed to generate the shielded image. In an embodiment, the electronic device 100 may be a server which receives the image from another device like client device. The processor 103 may process the image to generate the shielded image.

[0051] In an embodiment, the processor 103 of the electronic device may initially divide the received image into a plurality of blocks. A plurality of compressed images with a different compression ratio may be generated for each of plurality of blocks. An intermediate image may be generated by adding a predefined noise to the input image and a reference image may be selected that differs from the attribute of the input image. Based on the generated plurality of compressed images, the intermediate image and the reference image, the processor 103 may calculate the embeddings which may be further utilized to estimate the noise. The processor 103 may update the intermediate image by iteratively estimating and adding the estimated noise to the intermediate image to generate the shielded image.

[0052] The present disclosure may protect the image from AI manipulation. For example, in the present disclosure, the image with plurality of compression ratios may be shielded against AI manipulation. As the noise is calculated and added iteratively to the intermediate image which may be further verified with a predefined threshold based on which the final shielded image may be generated which may not be modified using AI tools. However, if the third-party user tries to modify the final shielded image, a distorted image may be obtained which may not reflect the features of the input image associated with the user.

[0053] Fig. 2illustrates a block diagram of the electronic device 100 to generate shielded image against AI manipulation, in accordance with various embodiments of the present disclosure.

[0054] In some implementations, the electronic device 100 comprises the processor 103, the I / O Interface 105, and the memory 107. The electronic device 100 may further include data 209 and modules 211. As an example, the data 209 may be stored in the memory 107 configured in the electronic device 100 as shown in the figure 2.

[0055] In one embodiment, the data 209 may include embedding data 213, estimated noise 215 and other data 217. In some embodiments, the data 209 may be stored in the memory 107 in form of various data structures. Additionally, the data 209 can be organized using data models, such as relational or hierarchical data models. The other data 217 may store data, including temporary data and temporary files, generated by the modules for performing the various functions of the electronic device 100.

[0056] In some embodiments, the data 209 stored in the memory 107 may be processed by the modules 211 of the electronic device 100. In an example, the modules 211 communicatively coupled to the processor 103 configured in the electronic device 100, may also be present outside the memory 107 as shown inFig.2 and implemented as hardware. As used herein, the term modules refer to an application specific integrated circuit (ASIC), an electronic circuit, a processor 203 (shared, dedicated, or group) and memory that execute one or more software or firmware programs, a combinational logic circuit, and / or other suitable components that provide the described functionality.

[0057] In an embodiment, any combinations of functions of modules 211 described herein may be implemented as one or more commands, instructions, applications and / or computer programs which can be loaded in the memory 107 and executed by the processor 103. For example, one or more commands, instructions, applications and / or computer programs may be, when loaded in memory 107 and / or executed by the processor 103, cause the electronic device 100 to perform any combination of operations and / or functions performed by any one of modules 211, in accordance with the present disclosure.

[0058] In some embodiments, the modules 211 may include, for example an image generation module 219, an embedding calculation module 221, a noise estimation module 223, a reference image selection module 227, a loss function calculation module 229, and other modules 225. The other modules225 may be usedto perform various miscellaneous functionalities of the electronic device 100. It will be appreciated that such aforementioned modules may be represented as a single module or a combination of different modules. The operations of each module included in the modules 221 may be described below.

[0059] Fig. 3A illustrates a diagram to generate a plurality of compressed images with a different compression ratio, in accordance with various embodiments of the present disclosure.

[0060] To protect images or videos from AI manipulation, a user may select the image for the electronic device 100 and may generate a shielded image for the selected image. In an embodiment, when the image is selected, the image may be divided into a plurality of blocks. For instance, the input image may be divided into 6 tiles or blocks. Each of the tiles is considered as an input image and processed independently to generate the shielded image. Finally, once the shielded image is generated for each of the 6 tiles as per the above example, then the shielded image may be combined to generate the combined shielded image having a resolution same as the resolution of the image.

[0061] In an embodiment, when the user selects an image, the image is divided into plurality of sub-images and each of the sub-images may act as the input image. For example, the electronic device 100 may generate respective shielded image for each sub-image. In an embodiment, the image may be retrieved from the user device (e.g., from the memory 107 in the electronic device 100). Once the input image may be received, the image generation module 219 may be configured to pre-process the received image. For example, the image may be re-sized and / or split, or the dimension of the image may be changed. The image generation module 219 may process the image to remove the unwanted artifacts or random variations in the image and / or may adjust the image to improve visibility of objects or features. The step of pre-processing may include more or less functions as described and as not limited to the above description. Once the pre-processing is performed, the image generation module 219 may generate a plurality of compressed images with different compression ratio.

[0062] In an embodiment, when the user selects the input image, the image generation module 219 may be configured to generate the plurality of compressed images with a different compression ratio as shown in Fig. 3A. Image compression may be defined as a technique that reduces the amount of data required to represent the image by removing the redundant data. For instance, image compression may be performed by reducing the size of the image by retaining the necessary information. The present disclosure may use different compressed images and each of the compressed images may be processed independently. As a result, when any third-party user tries to edit the compressed image, a distorted image may be obtained instead of the original image. For instance, consider the user may select one image from a plurality of images. The image generation module 219 may generate plurality of compressed images with a different compression ratio. For example, the input image may be compressed to 50%, 25%, and the like.

[0063] Fig. 3B illustrates a diagram to generate the shielded image from the generated plurality of compressed images, in accordance with various embodiments of the present disclosure.

[0064] Once the plurality of compressed images with a different compression ratio is generated, an appropriate reference image may be selected from the plurality of predefined reference images based on the attribute of plurality of predefined reference images that differs from an attribute of the compressed image. Further, the embedding calculation module 221 (e.g., using an edge detection model included the embedding calculation module 221) may calculate the embeddings for each of the compressed images and the selected reference image and add the suitable noise to each of the compressed images based on the calculated embeddings. Loss function calculation module 229 may calculate an estimated noise based on the calculated embeddings. For example, to provide protection even if the input image is compressed, the loss function calculation module 229. The loss function may be updated based on the embedding of the input image (e.g., 'Emb(Image_100)'), the embeddings of the compressed images (e.g., 'Emb(Image_50)' and 'Emb(Image_25)') and the embedding of the reference (e.g., 'Emb(Ref)'). The loss function may be calculated by calculating the sum of the absolute differences between each embedding associated with the input image and the embedding of the reference image. The updated loss function takes into account not only the embedding of the input image but also the embeddings of the compress image as well. The updated loss function protects shielded images even after arbitrary compression as shown in Fig. 3B. The calculated loss function may be used to backpropagate over the edge detection model (e.g., the embedding calculation) or to calculated the noise to be added to the images. The estimated noise may be calculated based on the calculated loss, and added to the input image and the compressed images. In an embodiment, the calculation of embeddings and / or the loss function may be performed for each iteration.

[0065] Fig. 3Cillustrates a diagram to generate an intermediate image by adding a predefined noise, in accordance with various embodiments of the present disclosure.

[0066] In an embodiment, the image generation module 219 may be configured to generate an intermediate image for input image by adding a predefined noise to the input image using a pre-trained noise model 302. The pre-trained noise model 302 may predict the noise to be added to the input image to provide shielding. For instance, the noise referred herein may be random variations in brightness or color information across the image. Once, the noise is added to the input image, the pre-trained noise model 302 may check if the added noise meets the predefined threshold. When the noise added is greater than the predefined threshold, then there is a significant difference between the edges present in the updated intermediate image and the edges present in the input image and the intermediate image with the added noise may be considered as the shielded image. However, if the added noise to the input image is not the best noise which may be less than the threshold criteria, then based on the added noise to the input image, the image generation module 219 may generate the intermediate image as shown in Fig. 3C. For example, the intermediate image may be generated by adding the predicted noise (or the estimaged noise) to the input image. In some embodiments, the threshold criteria may be set by a domain expert, and the reference image may be selected by the electronic device 100 to meet the threshold criteria.

[0067] In an embodiment, the pre-trained noise model 302 may be stored in the memory 107, and executed by the processor 103 or other module(s) of the modules 221. The pre-trained noise model 302 is pre-trained to predict an adversarial noise for a given image (e.g., the input image). The predicted noise from the pre-trained noise model 302 may be added to the input image. In an embodiment, the pre-trained noise model may be trained based on a photo immunization.

[0068] Fig. 3Dillustrates a diagram to select a reference image from a plurality of predefined reference images, in accordance with various embodiments of the present disclosure.

[0069] The reference image selection module 227 is configured to select the reference image from a plurality of predefined reference images based on an attribute of the plurality of predefined reference images that differs from the attribute of the input image. For example, consider that the input image may represent "a person in-front of Taj Mahal", the reference image selection module 227 which is in association with the plurality of the reference image dataset may select the reference image such that attributes like style, type, and embedding but not limited thereto of the reference image differs from that of the input image as shown in Fig. 3D.

[0070] Fig. 3Eillustrates a diagram to select of reference image from reference images databank, in accordance with various embodiments of the present disclosure.

[0071] The reference image selection module 227 may select the refence image to be a "flower vase" which represent different embedding, style in comparison with the input image. For example, the attribute (e.g., the calculated embedding) of the input image and the attribute of each of the plurality of predefined reference images may be determined. The plurality of predefined reference images may be stored in a reference image databank 304. The reference image databank 304 may be included in the memory 107. Each of the attributes of the plurality of reference images may be compared with the determined attribute of the input image. Based on comparison, the reference image may be selected from the plurality of predefined reference images that have the highest attribute difference from the attribute of the input image as shown in Fig. 3E.

[0072] Fig. 3Fillustrates a diagram to calculate an embedding, in accordance with various embodiments of the present disclosure.

[0073] Once the image generation module 219 generates the intermediate image, a plurality of images with different compression ratios and selects the reference image for the input image, the embedding calculation module 221 may be configured to calculate the embedding for each of the generated plurality of compressed images, the reference image, and the generated intermediate image as shown in Fig. 3F. In an embodiment, the embedding calculation may be performed by using a predefined technique such as stable diffusion technique.

[0074] Fig. 3Gillustrates an embedding of an image and the embedding indicating disturbed edges of the image, in accordance with various embodiments of the present disclosure.

[0075] The stable diffusion technique is used in image processing to provide smoother images by removing noise in the image (e.g., generated plurality of compressed images, the reference image, and the generated intermediate image], that may improve the quality and enhance edges. The stable diffusion technique improves the visual quality of images by removing unwanted details and highlighting edges. The embedding calculation module 221 may be configured to identify edges defined as curves in which the image brightness changes sharply or, more formally, containing discontinuities in each of the generated plurality of compressed images, the reference image, and the generated intermediate image. The embedding calculation module 221 may detect the set of connected curves that indicate the boundaries of objects, the boundaries of surface markings as well as curves that correspond to discontinuities in surface orientation. For example, the embedding calculation module 221 may be configured to detect the edges in plurality of compressed images, the reference image and the intermediate image with which high-level features are formed and the further downstream task may be performed. The edges are identified so that to disrupt the edges so that computer vision models or any of the AI tools may not successfully modify the image as shown in Fig. 3G. For example, the embedding calculation module 221 may generate an embedding 314 from an image 310 by using an edge detection model 312.

[0076] Fig. 3Hillustrates a diagram to estimate noise, in accordance with various embodiments of the present disclosure.

[0077] When the embedding for each of the generated plurality of compressed images, the reference image, and the generated intermediate image are calculated, the noise estimation module 223 may be configured to estimate a noise based on the calculated embedding of the generated plurality of compressed images, the reference image, and the generated intermediate image. For example, the noise estimation module 223 calculates the first loss for comparing the calculated embedding of the generated plurality of compressed images and the reference image. The noise estimation module 223 may be configured to calculate a second loss value by comparing the calculated embedding of the input image and the reference image. Once the first loss value and second loss value are calculated, the noise estimation module 223 may estimate the noise to be added to the intermediate image. To estimate the noise, the noise estimation module 223 may calculate first loss and second loss so that any third-party user may not be successful in editing the compressed image or using the original image as the edges detected are disturbed and then the noise may be added to the images. For example, the noise estimation module 223 may estimate the noise that is to be added to the intermediate image e.g., to each pixel in the intermediate image. As a result, even if the third-party user or AI may try to compress the image and then manipulate, the AI tool may not modify the image as the loss function is updated which takes in input comprising of compressed input images embeddings. The calculated loss is used again to backpropagate over the embedding calculation module 221 and calculate noise to be added to the intermediate image.

[0078] Fig. 3Iillustrates a diagram to identify Region of Interest (ROI) and adding noise to the input image, in accordance with various embodiments of the present disclosure.

[0079] When the noise is calculated, the noise estimation module 223 may identify at least one Region of Interest (ROI) in the input image to add the estimated noise. For instance, the input image may contain sensitive information such as logo, personal information, personal documents, personal preferences, and / or personally identification information (PII) regions like the face, palms, eyes. The PII detector 314 may detect and mark the bounding box around the PII region present in the image. The user may provide his / her inputs regarding the ROI present in the image (block 316). The noise estimation module 223 may consider the above inputs and then estimate the noise and add to the intermediate shielded image as shown in Fig. 3I. For example, the noise estimation module 223 may filter the estimated noise based on the at least one ROI (block 318). The filtered noise may correspond to the at least one ROI from the estimated noise. For example, noise estimation module 223 may filter noise only for the region(s) containing the PII and / or the selected regions so that only noises corresponding to the PII and / or the selected regions remain to be added to the image. The noise estimation module 223 may add the filtered noise to the intermediate image. At each instance the noise is added to the intermediate image, a difference between edges present in the updated intermediate image with the edges present in the input image may be determined. The intermediate image is updated by iteratively estimating and adding the estimated noise to the intermediate image till the difference between the edges present in the updated intermediate image and the edges present in the input image is greater than a predefined threshold. As a result, accurate amount of noise may be added to the intermediate image which may not be visible to the naked eye of the user. In an embodiment, the purpose of this filtering the noise is to add noise only to PII regions as malicious users may try to only manipulate those images. This reduces also the noise added to other regions making it less visible.

[0080] Figs. 3J-Lshows a use-case scenario indicating adding of the estimated noise to ROI in an image, in accordance with various embodiments of the present disclosure.

[0081] For instance, consider the user may click multiple images and post it on any of the social media platforms. Any third-party user may easily access the picture associated with the user and may edit or modify the image which may look realistic. In Fig. 3J, the user may post multiple images 320 of social media. The images posted in the social media platforms may be shielded images 322 associated with the user. For example, the image may contain sensitive information such as logo, personal information, PII regions like the face, palms, eyes. The PII detector may detect and mark the bounding box around the PII region present in the image and then the user may provide his / her inputs regarding the ROI present in the image. The noise estimation module 223 may consider the above inputs and then estimate the noise and add to the image (not shown in Fig. 3J) which is now the shielded image that is posted on social media platforms (block 324). When the third-party user access the image and tries to modify the image using any of a generative AI (GenAI) model 326, the third-party may obtain the distorted image 328 rather than the modified realistic image as shown in Fig. 3J.

[0082] In an embodiment, consider that the user has posted an image 330 indicating his palm. A shielded image 332 may be generated from the image 330. For example, the shielded image 332 may comprise a noise added to the ROI (e.g., the palm region) in the image 330. The shielded image 323 may be posted on the social media (block 334). Now, if any of the other user tries to manipulate the image indicating the user palm using any of the GenAI tools 336, then the other user may obtain distorted image 338 which is easily detected (or considered) as manipulated image instead of realistic image as shown in Fig. 3K.

[0083] In an embodiment, consider that the user hopes to post an image 340 that includes plurality of faces associated with the user on the social media platform. As the image 340 includes plurality of faces, the electronic device 100 according to the present disclosure may add the estimated to the PII regions like the face, palms, eyes to generated a shielded image 342, and post the shielded image 342 on the social media (block 344). When the third-party user tries to manipulate the image indicating face regions using any of the GenAI tools 346, then the third-party user may obtain distorted image 348 which is easily detected as manipulated image instead of realistic image as shown in Fig. 3L.

[0084] Fig. 3Millustrates a diagram to estimate noise, in accordance with various embodiments of the present disclosure.

[0085] Once the noise is added to each of plurality of compressed images with different compression ratios, the shielded image may be generated by combining the shielded image generated for each of the plurality of tiles of the input image having a resolution same as the resolution of the image. As shown in Fig. 3M, when the compressed image 350 is fed to the AI tool for manipulation, the distorted image 352 may be obtained instead of the original image which may be taken from any of the social media platform of the user.

[0086] FIG 4.shows a flowchart illustrating a method for generating shielded image against AI manipulation, in accordance with some embodiments of the present disclosure.

[0087] As illustrated in FIG.4, the method 400 includes one or more blocks (or steps or operations) illustrating a method for generating shielded image against AI manipulation. The method 400 may be described in the general context of computer executable instructions. Generally, computer executable instructions can include routines, programs, objects, components, data structures, procedures, modules, and functions, which perform functions or implement abstract data types. In an embodiment, the method 400 may be performed by the electronic device 100, but not limited thereto.

[0088] The order in which the method 400 is described is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method 400. Additionally, individual blocks may be deleted from the methods without departing from the spirit and scope of the subject matter described herein. Furthermore, the method 400can be implemented with any suitable hardware, software, firmware, or combination thereof.

[0089] At block 401, the electronic device 100may generate a plurality of compressed images with a different compression ratio from an input image.

[0090] At block 403, the electronic device 100may generate an intermediate image by adding a predefined noise to the input image. For example, the predefined noise may be predicted for the input image using a pre-trained noise prediction machine learning (ML) model. The ML models are trained to protect the images by introducing trivial perturbation, so that the protected images are immune to the viruses in that the tampered contents can be auto recovered. The predicted predefined noise is added to the input image to generate the intermediate image.

[0091] At block 405, the electronic device 100may select a reference image from a plurality of predefined reference images based on an attribute of each of the plurality of predefined reference images. The attribute of the input image and the attribute of each of the plurality of predefined reference images are determined by the electronic device 100. The attributes of input image and the plurality of reference image may be compared. The electronic device 100 may select the reference image from the plurality of predefined reference images that has the highest attribute difference from the attribute of the input image.

[0092] At block 407, the electronic device 100may calculate an embedding for each of the generated plurality of compressed images, the reference image, and the generated intermediate image. The edges are detected (or identified) within each of the plurality of compressed images, the intermediate image, and the reference image. The embeddings are calculated corresponding to the identified edges for each of the plurality of compressed images, the reference image, and the intermediate image.

[0093] At block 409, the electronic device 100 may update the intermediate image to generate the shielded image by iteratively (ⅰ) estimating a noise based on the calculated embedding of the generated plurality of compressed images, the reference image, and the generated intermediate image; and (ⅱ) adding the estimated noise to the intermediate image. For example, a first loss value may be calculated by comparing the calculated embedding of the generated plurality of compressed images and the reference image. The electronic device 100 may calculate a second loss value by comparing the calculated embedding of the input image and the reference image. The noise may be estimated based on the calculated first loss value and the second loss value.

[0094] In an embodiment, the electronic device 100 may identify at least one Region of Interest (ROI) in the input image. The electronic device 100 may filter the estimated noise based on the at least one ROI. The filtered noise may be added to the intermediate image.

[0095] In an embodiment, the estimated noise is added to the intermediate image to generate the updated intermediate image. Further, difference between edges present in the updated intermediate image with the edges present in the input image may be determined. Based on the above determination, the intermediate image may be updated by iteratively estimating and adding the estimated noise to the intermediate image till the difference between the edges present in the updated intermediate image and the edges present in the input image is greater than a predefined threshold.

[0096] In an embodiment, an image may be divided into a plurality of sub-images. The electronic device 100 may generate a shielded image for each of the plurality of sub-images. The electronic device 100 may combine the shielded images generated for each of the plurality of sub-images to generate a combined shielded image having a resolution same as the resolution of the image.

[0097] According to an embodiment, the computer electronic device may implement embodiments consistent with the present disclosure as shown in Fig. 5.

[0098] In some embodiments, FIG.5illustrates a block diagram of an exemplary computing system 500for implementing embodiments consistent with the present invention. In some embodiments, the computing system 500 for generating shielded image. The computer system500 may include a central processing unit ("CPU" or "processor)502. The processor 502 may include at least one data processor 502 for executing program components for executing user or electronic device-generated business processes. The processor 502may include specialized processing units such as integrated system (bus) controllers, memory management control units, floating point units, graphics processing units, digital signal processing units, etc.

[0099] The processor 502 may be disposed in communication with input devices 511and output devices 512 via I / O interface 501. The I / O interface 501may employ communication protocols / methods such as, without limitation, audio, analog, digital, stereo, IEEE-1394, serial bus, Universal Serial Bus (USB), infrared, PS / 2, BNC, coaxial, component, composite, Digital Visual Interface (DVI), high-definition multimedia interface (HDMI), Radio Frequency (RF) antennas, S-Video, Video Graphics Array (VGA), IEEE 802.n / b / g / n / x, Bluetooth, cellular (e.g., Code-Division Multiple Access (CDMA), High-Speed Packet Access (HSPA+), Global System For Mobile Communications (GSM), Long-Term Evolution (LTE), WiMax, or the like), etc. Using the I / O interface501, computer system(or a UE)may communicate with input devices 511and output devices 512.

[0100] In some embodiments, the processor502 may be disposed in communication with a communication network 509 via a network interface 503. The network interface 503 may communicate with the communication network 509. The network interface 503may employ connection protocols including, without limitation, direct connect, Ethernet (e.g., twisted pair 10 / 100 / 1000 Base T), Transmission Control Protocol / Internet Protocol (TCP / IP), token ring, IEEE 802.11a / b / g / n / x, etc. The communication network 509 can be implemented as one of the different types of networks, such as intranet or Local Area Network (LAN), Closed Area Network (CAN) and such within the vehicle. The communication network 509 may either be a dedicated network or a shared network, which represents an association of the different types of networks that use a variety of protocols, for example, Hypertext Transfer Protocol (HTTP), CAN Protocol, Transmission Control Protocol / Internet Protocol (TCP / IP), Wireless Application Protocol (WAP), etc., to communicate with each other. Further, the communication network 509may include a variety of network devices, including routers, bridges, servers, computing devices, storage devices, etc. The one or more computing devicesmay include, but not limited to, a mobile phone, a tablet phone, a laptop and the like. In some embodiments, the processor 502 may be disposed in communication with a memory 505(e.g., RAM, ROM, etc. not shown in FIG. 5) via a storage interface 504. The storage interface 504may connect to memory 505 including, without limitation, memory drives, removable disc drives, etc., employing connection protocols such as Serial Advanced Technology Attachment (SATA), Integrated Drive Electronics (IDE), IEEE-1394, Universal Serial Bus (USB), fibre channel, Small Computer Electronic devices Interface (SCSI), etc. The memory drives may further include a drum, magnetic disc drive, magneto-optical drive, optical drive, Redundant Array of Independent Discs (RAID), solid-state memory devices, solid-state drives, etc. The memory 505 may store a collection of program or database components, including, without limitation, a user interface 506, an operating electronic device 507, a web browser 508 etc. In some embodiments, the computer electronic device500 may store user / application data, such as the data, variables, records, etc. as described in this invention. Such databases may be implemented as fault-tolerant, relational, scalable, secure databases such as Oracle or Sybase.

[0101] The operating electronic device 507 may facilitate resource management and operation of the computer electronic device500. Examples of operating electronic devices include, without limitation, APPLE®MACINTOSH®OS X®, UNIX®, UNIX-like electronic device distributions (E.G., BERKELEY SOFTWARE DISTRIBUTION®(BSD), FREEBSD®, NETBSD®, OPENBSD, etc.), LINUX®DISTRIBUTIONS (E.G., RED HAT®, UBUNTU®, KUBUNTU®, etc.), IBM®OS / 2®, MICROSOFT®WINDOWS®(XP®, VISTA® / 7 / 8, 10 etc.), APPLE®IOS®, GOOGLETMANDROIDTM, BLACKBERRY®OS, or the like. The User interface 506 may facilitate display, execution, interaction, manipulation, or operation of program components through textual or graphical facilities. For example, user interfaces may provide computer interaction interface elements on a display system operatively connected to the computer system500, such as cursors, icons, check boxes, menus, scrollers, windows, widgets, etc. Graphical User Interfaces (GUIs) may be employed, including, without limitation, Apple®Macintosh®operating systems' Aqua®, IBM®OS / 2®, Microsoft®Windows®(e.g., Aero, Metro, etc.), web interface libraries (e.g., ActiveX®, Java®,　Javascript®, AJAX, HTML, Adobe®Flash®, etc.), or the like.

[0102] In some embodiments, the computer system500 may implement the web browser 508stored program components. The web browser 508may be a hypertext viewing application, such as MICROSOFT®INTERNET EXPLORER®, GOOGLETMCHROMETM, MOZILLA®FIREFOX®, APPLE®SAFARI®, etc. Secure web browsing may be provided using Secure Hypertext Transport Protocol (HTTPS), Secure Sockets Layer (SSL), Transport Layer Security (TLS), etc. Web browsers 508 may utilize facilities such as AJAX, DHTML, ADOBE®FLASH®, JAVASCRIPT®, JAVA®, Application Programming Interfaces (APIs), etc. In some embodiments, the computer system500 may implement a mail server stored program component. The mail server may be an Internet mail server such as Microsoft Exchange, or the like. The mail server may utilize facilities such as Active Server Pages (ASP), ACTIVEX®, ANSI®C++ / C#, MICROSOFT®, .NET, CGI SCRIPTS, JAVA®, JAVASCRIPT®, PERL®, PHP, PYTHON®, WEBOBJECTS®, etc. The mail server may utilize communication protocols such as Internet Message Access Protocol (IMAP), Messaging Application Programming Interface (MAPI), MICROSOFT®exchange, Post Office Protocol (POP), Simple Mail Transfer Protocol (SMTP), or the like. In some embodiments, the computer system 500 may implement a mail client stored program component. The mail client may be a mail viewing application, such as APPLE®MAIL, MICROSOFT®ENTOURAGE®, MICROSOFT®OUTLOOK®, MOZILLA®THUNDERBIRD®, etc.

[0103] Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present invention. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor 502 may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processor 402, including instructions for causing the processor 502to perform steps or stages consistent with the embodiments described herein. The term "computer-readable medium" should be understood to include tangible items and exclude carrier waves and transient signals, i.e., non-transitory. Examples include Random Access Memory (RAM), Read-Only Memory (ROM), volatile memory, non-volatile memory, hard drives, Compact Disc (CD) ROMs, Digital Video Disc (DVDs), flash drives, disks, and any other known physical storage media.

[0104] The terms "an embodiment", "embodiment", "embodiments", "the embodiment", "the embodiments", "one or more embodiments", "some embodiments", and "one embodiment" mean "one or more (but not all) embodiments of the invention(s)" unless expressly specified otherwise.

[0105] The terms "including", "comprising", "having" and variations thereof mean "including but not limited to", unless expressly specified otherwise. The enumerated listing of items does not imply that any or all the items are mutually exclusive, unless expressly specified otherwise.

[0106] The terms "a", "an" and "the" mean "one or more", unless expressly specified otherwise. A description of an embodiment with several components in communication with each other does not imply that all such components are required. On the contrary, a variety of optional components are described to illustrate the wide variety of possible embodiments of the invention.

[0107] When a single device or article is described herein, it will be clear that more than one device / article (whether they cooperate) may be used in place of a single device / article. Similarly, where more than one device or article is described herein (whether they cooperate), it will be clear that a single device / article may be used in place of the more than one device or article or a different number of devices / articles　may be used instead of the shown number of devices or programs. The functionality and / or the features of a device may be alternatively embodied by one or more other devices which are not explicitly described as having such functionality / features. Thus, other embodiments of the invention need not include the device itself.

[0108] Finally, the language used in the specification has been principally selected for readability and instructional purposes, and it may not have been selected to delineate or circumscribe the inventive subject matter. It is therefore intended that the scope of the invention be limited not by this detailed description, but rather by any claims that issue on an application based here on. Accordingly, the embodiments of the present invention are intended to be illustrative, but not limiting, of the scope of the invention, which is set forth in the following claims.

[0109] While various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are for purposes of illustration and are not intended to be limiting, with the true scope and spirit being indicated by the following claims.

Claims

A method for generating a shielded image against artificial intelligence (AI) manipulation, the method comprising:generating a plurality of compressed images with a different compression ratio from an input image;generating an intermediate image by adding a predefined noise to the input image;selecting a reference image from a plurality of predefined reference images based on an attribute of each of the plurality of predefined reference images;calculating an embedding for each of the generated plurality of compressed images, the reference image, and the generated intermediate image; andupdating the intermediate image to generate the shielded image by iteratively (ⅰ) estimating a noise based on the calculated embedding of the generated plurality of compressed images, the reference image, and the generated intermediate image; and (ⅱ) adding the estimated noise to the intermediate image.The method as claimed in claim 1, further comprises:dividing an image into a plurality of sub-images;generating a shielded image for each of the plurality of sub-images; andcombining the shielded images generated for each of the plurality of sub-images to generate a combined shielded image having a resolution same as the resolution of the image.The method as claimed in claim 1 or claim 2, wherein selecting the reference image further comprises:determining an attribute of the input image and the attribute of each of the plurality of predefined reference images;comparing the determined attribute of the input image with the attribute of the plurality of predefined reference images; andselecting the reference image from the plurality of predefined reference images that has the highest attribute difference from the attribute of the input image, based on the comparison.The method as claimed in any one of claims 1 to 3, wherein generating the intermediate image comprises:predicting the predefined noise for the input image using a pre-trained noise prediction machine learning (ML) model; andadding the predicted predefined noise to the input image to generate the intermediate image.The method as claimed in any one of claims 1 to 4, wherein calculating the embedding further comprises:detecting edges within each of the plurality of compressed images, the intermediate image, and the reference image; andcalculating the embedding based on the detected edges for each of the plurality of compressed images, the reference image, and the intermediate image.The method as claimed in any one of claims 1 to 5, wherein adding the estimated noise to the intermediate image comprises:identifying at least one Region of Interest (ROI) in the input image;filtering the estimated noise based on the at least one ROI; andadding the filtered noise to the intermediate image.The method as claimed in any one of claims 1 to 6, wherein estimating the noise comprises:calculating a first loss value by comparing the calculated embeddings of the generated plurality of compressed images and the reference image;calculating a second loss value by comparing the calculated embeddings of the input image and the reference image; andestimating the noise based on the calculated first loss value and the second loss value.The method as claimed in any one of claims 1 to 7, wherein updating the intermediate image comprises:determining a difference between edges present in the updated intermediate image with the edges present in the input image; andupdating the intermediate image till the difference between the edges present in the updated intermediate image and the edges present in the input image is greater than a predefined threshold.A computer-readable storage medium storing one or more instructions, wherein the one or more instructions, when executed by at least one processor individually or collectively, cause the at least one processor to perform the method of any one of claims 1 to 8.An electronic device electronic deviceto generate shielded image against artificial intelligence (AI) manipulation, the electronic device comprises:at least one processor comprising processing circuitry; andmemory coupled to the at least one processor, the memory comprising one or more storage media storing one or more instructions,wherein, when executed by the at least one processor individually or collectively, the one or more instructions cause the electronic device to:generate a plurality of compressed images with a different compression ratio from an input image;generate an intermediate image by adding a predefined noise to the input image;select a reference image from a plurality of predefined reference images based on an attribute of each of the plurality of predefined reference images;calculate an embedding for each of the generated plurality of compressed images, the reference image, and the generated intermediate image; andupdate the intermediate image to generate the shielded image by iteratively (ⅰ) estimating a noise based on the calculated embedding of the generated plurality of compressed images, the reference image, and the generated intermediate image; and (ⅱ) adding the estimated noise to the intermediate image.The electronic device as claimed in claim 10, wherein, when executed by the at least processor individually or collectively, the one or more instructions further cause the electronic device to:predict the predefined noise for the input image using a pre-trained noise prediction machine learning (ML) model; andadd the predicted predefined noise to the input image to generate the intermediate image.The electronic device as claimed in claim 10 or 11, wherein, when executed by the at least processor individually or collectively, the one or more instructions further cause the electronic device to:detect one or more edges within each of the plurality of compressed images, the intermediate image, and the reference image; andcalculate the embedding based on the detected edges for each of the plurality of compressed images, the reference image and the intermediate image.The electronic device as claimed in any one of claims 10 to 12, wherein, when executed by the at least processor individually or collectively, the one or more instructions further cause the electronic device to:identify at least one Region of Interest (ROI) in the input image;filter the estimated noise based on the at least one ROI; andadd the filtered noise to the intermediate image.The electronic device as claimed in any one of claims 10 to 13, wherein, when executed by the at least processor individually or collectively, the one or more instructions further cause the electronic device to:calculate a first loss value by comparing the calculated embedding of the generated plurality of compressed images and the reference image;calculate a second loss value by comparing the calculated embedding of the input image and the reference image; andestimate the noise based on the calculated first loss value and the second loss value.The electronic device as claimed in any one of claims 10 to 14, wherein, when executed by the at least processor individually or collectively, the one or more instructions further cause the electronic device to:determine a difference between edges present in the updated intermediate image with the edges present in the input image; andupdate the intermediate image till the difference between the edges present in the updated intermediate image and the edges present in the input image is greater than a predefined threshold.

Images