Anonymization of personally identifiable information in images and other spatial signals

Abstract

A system for anonymization of personally identifiable information is provided. The system stores an encoder-decoder neural network that includes an encoder and a decoder. The system receives a first image comprising an object with personally identifiable information and extracts a block of feature maps as output of an intermediate layer of the encoder based on application of the encoder on the first image. The system identifies the object based on the extracted block of feature maps and selects a size label for the identified object from a set of size labels. The system further selects an anonymization operation from a set of anonymization operations based on the selected size label and reconstructs a second image based on application of the selected anonymization operation on the block of feature maps or the first image. The reconstructed second image includes an anonymized version of the personally identifiable information of the object.

Description

ANONYMIZATION OF PERSONALLY IDENTIFIABLE INFORMATION IN IMAGES AND OTHER SPATIAL SIGNALSCROSS-REFERENCE TO RELATED APPLICATIONS / INCORPORATION BY REFERENCE

[0001] This Application also makes reference to Indian Provisional Application No. 202411032498, which was filed on April 24, 2024. The above stated Patent Applications are hereby incorporated herein by reference in their entirety.FIELD

[0002] Various embodiments of the disclosure relate to data privacy and protection. More specifically, various embodiments of the disclosure relate to anonymization of personally identifiable information in images and other spatial signals.BACKGROUND

[0003] The increasing digitization of personal data across various sectors, including healthcare, finance, and social media has heightened the need for robust data privacy measures. As organizations collect and process vast amounts of personally identifiable information, the risk of data breaches and unauthorized access has grown significantly. High-profile incidents of data theft and misuse have underscored the vulnerabilities inherent in handling sensitive information. Additionally, stringent data protection regulations in various regions mandate strict controls over the use and sharing of personal data. These regulations require organizations to implement measures that protect individuals’ privacy and ensure data security. Therefore, there is a need for a technique to enhance data security and data privacy which may not only help in compliance with legal requirements but also foster trust and confidence among consumers and stakeholders in the digital age.

[0004] Limitations and disadvantages of conventional and traditional approaches will become apparent to one of skill in the art, through comparison of described systems withsome aspects of the present disclosure, as set forth in the remainder of the present application and with reference to the drawings.SUMMARY

[0005] A system and method for anonymization of personally identifiable information in images and other spatial signals is provided substantially as shown in, and / or described in connection with, at least one of the figures, as set forth more completely in the claims.

[0006] These and other features and advantages of the present disclosure may be appreciated from a review of the following detailed description of the present disclosure, along with the accompanying figures in which like reference numerals refer to like parts throughout.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] FIG. 1 is a block diagram that illustrates an exemplary network environment for anonymization of personally identifiable information in images and other spatial signals, in accordance with an embodiment of the disclosure.

[0008] FIG. 2 is a block diagram that illustrates an exemplary system of FIG. 1 , in accordance with an embodiment of the disclosure.

[0009] FIG. 3A and FIG. 3B are flow diagrams that collectively illustrate various operations of the system of FIG. 1 , in accordance with an embodiment of the disclosure.

[0010] FIG. 4 is a block diagram illustrating operations executed by a system for anonymization of personally identifiable information of multiple objects present in an image, in accordance with an embodiment of the disclosure.

[0011] FIG. 5 is a flowchart that illustrates exemplary operations of a method for anonymization of personally identifiable information in images and other spatial signals, in accordance with an embodiment of the disclosure.DETAILED DESCRIPTION

[0012] The following described implementation may be found in a system and method for anonymization of personally identifiable information in images and other spatial signals. Exemplary aspects of the disclosure may provide a system, which may include a memory configured to store an encoder-decoder neural network including an encoder and a decoder. The encoder-decoder neural network may be a U-Net based diffusion model. The system may further include a circuitry configured to receive a first image including an object with personally identifiable information. The object may be a human face, a vehicle, or a segment or a region of an audio spectrogram. The circuitry may further extract a block of feature maps as an output of an intermediate layer of the encoder based on application of the encoder on the received first image. The circuitry may further identify the object based on the extracted block of feature maps. The circuitry may further select a size label for the identified object from a set of size labels. Additionally, the circuitry may select an anonymization operation from a set of anonymization operations based on the selected size label. Subsequently, the circuitry may reconstruct a second image based on application of the selected anonymization operation on the extracted block of feature maps or the first image. The reconstructed second image may include an anonymized version of the personally identifiable information of the object.

[0013] Data digitization refers to the process of converting information into a digital format. In the digital format, data is organized into bits and bytes that computers may process, store, and transmit. This transformation from analog to digital data may allow for more efficient storage, retrieval, and manipulation of information. In this age, data has been rapidly digitized across multiple sectors, such as healthcare, finance, and social media. In order to prevent any misuse of the digitized personal data, stringent data privacy measures may be required. Along with increased collection and processing of vast amounts of personally identifiable information by several organizations, the risk for databreaches and unauthorized access has also escalated. High-profile incidents of data theft and misuse have highlighted the inherent vulnerabilities in managing sensitive information. These incidents may not only compromise individual privacy but may also damage the reputation and trustworthiness of the organizations involved. Moreover, stringent data protection regulations in various regions have mandated strict controls over the use and sharing of personal data. Such regulations require organizations to implement comprehensive measures to protect individuals’ privacy and ensure data security. Compliance with these regulations may not be optional; failure to adhere may result in severe penalties and legal repercussions. Therefore, organizations have to prioritize data security and privacy to meet legal requirements and avoid potential fines. Beyond legal compliance, enhancing data security and privacy measures may be crucial to foster trust and confidence among consumers and stakeholders in the digital age.

[0014] The system may detect personally identifiable information of an object present in images. The system may provide a single platform for handling of the images, which results in enhanced efficiency of the system and more convenience to a user. Further, the system may convert an image associated into an anonymized image to provide data security to the user. The system may substitute detected personally identifiable information with transformed data for anonymization. This approach may not only protect individual privacy but also preserve the functional and aesthetic quality of the original data. Moreover, the anonymization of the personally identifiable information may give the user the authority to decide what and how much information the user would like to disclose to public or private information processing systems.

[0015] The system may utilize an encoder of the encoder-decoder neural network to automatically use feature activations across different layers to adaptively and robustly identify regions or sections of the image including personally identifiable information of the object visible in the image. The system may select an anonymization operation to replacepersonally identifiable information of the object with visually coherent substitutes, maintaining content's original context and aesthetics while ensuring anonymity. Further, the system may be optimized for minimal resource use, as the selection of the anonymization operation from a set of anonymization operations based on size of the object in the image, reduces computational load, hence innovatively using object’s sizebased personally identifiable information segregation for effective anonymization. The system may also dynamically adjust the selection of the anonymization operation and functioning of the anonymization operation in real-time based on resource and network availability, ensuring optimal performance and faster anonymization of the image under varying conditions.

[0016] FIG. 1 is a block diagram that illustrates an exemplary network environment for anonymization of personally identifiable information in images and other spatial signals, in accordance with an embodiment of the disclosure. With reference to FIG. 1 , there is shown a network environment 100. The network environment 100 may include a system 102, an encoder-decoder neural network 104, a classification network 106, a server 108, a database 110, a communication network 112, and a user device 114.

[0017] The system 102 may include suitable logic, circuitry, interfaces, and / or code that may be configured to receive an image 116 including an object 116-1 with corresponding personally identifiable information 116-2. In an embodiment, the object 116-1 may be a human face, and the corresponding personally identifiable information 116-2 may include facial features of the human face. In another embodiment, the object 116-1 may be a segment or a region of an audio spectrogram, and the corresponding personally identifiable information 116-2 may include acoustic features or speech features represented by the segment or the region. In another embodiment, the object 116-1 may be a vehicle, and the corresponding personally identifiable information 116-2 may include a license plate number of the vehicle. The system 102 may further utilize the encoder-decoder neural network 104 to reconstruct an anonymized version of the personally identifiable information 116-2 of the object 116-1. Examples of the system 102 may include, but are not limited to, a digital media player (DMP), a micro-console, a TV tuner, a digital media streamer, a media extender / regulator, a digital media hub, a computer workstation, a mainframe computer, a handheld computer, a smart appliance, a plug-in device, and / or any other computing device with content streaming functionality.

[0018] The system 102 may store the encoder-decoder neural network 104 or may be remotely connected to another system (such as the server 108) that hosts the encoderdecoder neural network 104. When hosted on another system, the system 102 may send instructions to control training or inference of the encoder-decoder neural network 104 via remote calls (e.g., API calls).

[0019] The encoder-decoder neural network 104 may be configured to anonymize personally identifiable information (for instance, personally identifiable information 116-2) of objects (for instance, the object 116-1 ) present in images and other spatial signals. The encoder-decoder neural network 104 may be a hybrid network, which may include multiple neural networks including an encoder 104-1 and a decoder 104-2. Output from the encoder 104-1 may be connected to an input layer of the encoder-decoder neural network 104, and output of the encoder-decoder neural network 104 may be connected to input of the decoder 104-2.

[0020] In an example embodiment, the encoder-decoder neural network 104 may be a ll-Net based diffusion network for image-to-image generation. The U-Net based diffusion network may leverage the U-Net architecture for a denoising process in diffusion models. The encoder of the U-Net based diffusion network may include multiple convolutional layers that progressively reduce the spatial dimensions of an input image (e.g., image 116) while increasing a number of feature channels. Each convolutional block may be followed by a down sampling operation, such as max pooling, to reduce the spatial resolution. Theencoder may capture hierarchical features from the input image, ranging from low-level details to high-level abstract features. The decoder may mirror the encoder but may perform up sampling operations, like transposed convolutions, to increase the spatial resolution of the feature maps. Skip connections from the encoder to the decoder may help to retain spatial information by concatenating corresponding feature maps from the encoder to the decoder. The decoder may reconstruct the image from the encoded features, progressively refining the image details. In the context of a diffusion network, the ll-Net architecture may be used iteratively to predict and remove noise from the input image at each denoising step. The encoder captures the noisy input's features, while the decoder reconstructs the denoised image, leveraging the skip connections to maintain high-frequency details. This combination of encoder-decoder architecture with skip connections makes the ll-Net based diffusion network highly effective for tasks like image denoising and generation.

[0021] The encoder 104-1 may include suitable logic, circuitry, interfaces, and / or code that may be configured to process the received image. The encoder 104-1 may further extract a block of feature maps. The encoder 104-1 may transform the received image into a compact, high-dimensional feature representation through a series of processing steps. For example, the image may be passed through convolutional layers that may apply filters to extract a block of feature maps as an output of an intermediate layer of the encoder 104-1. Initial layers of the encoder 104-1 may include corresponding blocks of feature maps that may capture low-level features like edges and textures. As the image progresses through deeper layers of the encoder 104-1 , the encoder 104-1 may extract corresponding blocks of feature maps that may capture higher-level features and abstractions. The encoder 104-1 may also include pooling layers that may reduce spatial dimensions of the received image, retaining the most significant information while discarding redundant data from the image. The output may be a block of feature maps thatencapsulate the essential characteristics of the image, making it suitable for the task of anonymization. This encoded representation is efficient for further processing and analysis by machine learning models.

[0022] The decoder 104-2 may include suitable logic, circuitry, interfaces, and / or code that may be configured to reconstruct an anonymized image based on output from the encoder 104-1. The decoder 104-2 may receive encoded feature maps from the encoder 104-1 and may perform a series of up sampling operations to generate the anonymized image. The decoder 104-2 may use transposed convolutions or other up sampling techniques to increase the spatial resolution of the feature maps. At each up-sampling stage, the decoder 104-2 may combine these up sampled features with corresponding feature maps from the encoder 104-1 through skip connections, which help retain spatial details and improve reconstruction accuracy.

[0023] The classification network 106 may be a type of artificial neural network designed to process a block of feature maps and categorize the block into predefined classes or categories. The classification network 106 may learn from labeled training data associated with the block of feature maps and may accurately predict class labels (for instance, a set of size labels 118-3) for new and unseen data.

[0024] Each neural network (i.e. , classification network 106, encoder 104-1 , or decoder 104-2) may be referred to as a computational network or a system of artificial neurons, arranged in a plurality of layers, as nodes. The plurality of layers of the neural network may include an input layer, one or more hidden layers, and an output layer. Each layer of the plurality of layers may include one or more nodes (or artificial neurons). Outputs of all nodes in the input layer may be coupled to at least one node of hidden layer(s). Similarly, inputs of each hidden layer may be coupled to outputs of at least one node in other layers of the neural network. Outputs of each hidden layer may be coupled to inputs of at least one node in other layers of the neural network. Node(s) in the final layer may receive inputsfrom at least one hidden layer to output a result. The number of layers and the number of nodes in each layer may be determined from hyper-parameters of the neural network. Such hyper-parameters may be set before or after training the neural network on the training dataset.

[0025] Each neural network of the neural networks may include electronic data, which may be implemented as, for example, a software component of an application executable on the system 102. Each of the neural networks may rely on libraries, external scripts, or other logic / instructions for execution by a processing device, such as the system 102. Further, each of the neural networks may rely on code and routines to enable a computing device, such as the system 102 to perform one or more operations, such as automatic audio dubbing. In some embodiments, each of the neural networks may be implemented using hardware including a processor, a microprocessor (e.g., to perform or control performance of one or more operations), a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). Alternatively, in some embodiments, each of the neural networks may be implemented using a combination of hardware and software.

[0026] The server 108 may include suitable logic, circuitry, and interfaces, and / or code that may be configured to receive the image 116 from a user interface. In an embodiment, the server 108 may store the image 116. In some instances, the server 108 may also host the encoder-decoder neural network 104 and the classification network 106. The server 108 may be implemented as a cloud server and may execute operations through web applications, cloud applications, HTTP requests, repository operations, file transfer, and the like. Other example implementations of the server 108 may include, but are not limited to, a database server, a file server, a web server, a media server, an application server, a mainframe server, a machine learning server (enabled with or hosting, for example, a computing resource, a memory resource, and a networking resource), or a cloud computing server.

[0027] In at least one embodiment, the server 108 may be implemented as a plurality of distributed cloud-based resources by use of several technologies that are well known to those ordinarily skilled in the art. A person with ordinary skill in the art will understand that the scope of the disclosure may not be limited to the implementation of the server 108 and the system 102, as two separate entities. In certain embodiments, the functionalities of the server 108 can be incorporated in its entirety or at least partially in the system 102 without a departure from the scope of the disclosure. In certain embodiments, the server 108 may host the database 110. Alternatively, the server 108 may be separate from the database 110 and may be communicatively coupled to the database 110.

[0028] The database 110 may include suitable logic, interfaces, and / or code that may be configured to store a dataset 118, which may include information related to object 116- 1 , personally identifiable information 116-2, and set of size labels 118-3. The database 110 may also include information about a multitude of objects, and the object 116-1 may be a part of the multitude of objects. The database 110 may be derived from data off a relational or non-relational database, or a set of comma-separated values (csv) files in conventional or big-data storage. The database 110 may be stored or cached on a device, such as a server (e.g., the server 108) or the system 102. The device storing the database 110 may be configured to receive commands or instructions from the system 102 or the server 108. In response, the device of the database 110 may be configured to retrieve and provide information about the object 116-1 , the personally identifiable information 116-2, or the set of size labels 118-3.

[0029] In some embodiments, the database 110 may be hosted on a plurality of servers stored at the same or different locations. The operations of the database 110 may be executed using hardware including a processor, a microprocessor (e.g., to perform or control performance of one or more operations), a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). In some other instances, the database110 may be implemented using software.

[0030] The communication network 112 may include a communication medium through which the system 102 and the server 108 may communicate with one another. The communication network 112 may be one of a wired connection or a wireless connection. Examples of the communication network 112 may include, but are not limited to, the Internet, a cloud network, Cellular or Wireless Mobile Network (such as Long-Term Evolution and 5thGeneration (5G) New Radio (NR)), satellite communication system (using, for example, low earth orbit satellites), a Wireless Fidelity (Wi-Fi) network, a Personal Area Network (PAN), a Local Area Network (LAN), or a Metropolitan Area Network (MAN). Various devices in the network environment 100 may be configured to connect to the communication network 112 in accordance with various wired and wireless communication protocols. Examples of such wired and wireless communication protocols may include, but are not limited to, at least one of a Transmission Control Protocol and Internet Protocol (TIP / IP), User Datagram Protocol (UDP), Hypertext Transfer Protocol (HTTP), File Transfer Protocol (FTP), Zig Bee, EDGE, IEEE 802.11 , light fidelity (Li-Fi), 802.16, IEEE 802.11 s, IEEE 802.11g, multi-hop communication, wireless access point (AP), device to device communication, cellular communication protocols, and Bluetooth (BT) communication protocols.

[0031] The user device 114 may include a user-interface through which a user may interact with the system 102, send queries, feed commands and instructions, provide the image 116. The user may be a human entity associated with the image 116 or a user of the system 102. The user device 114 may be fixed at a place or may be portable. Examples of the user device 114 may include, but not limited to, a smartphone, a touchpad, a GUI interface, a personal computer, a microphone, or a display device.

[0032] In operation, the system 102 may be configured to receive the image 116. The image 116 may be of the object 116-1 with personally identifiable information 116-2. Theimage 116 may include personally identifiable information 116-2 in the form of facial features, addresses or contact details, license numbers, or even speech data (in the form of a spectrogram). For example, if the object 116-1 is a human face, the corresponding personally identifiable information 116-2 may include facial features of the human face. Similarly, if the object 116-1 is a vehicle, the corresponding personally identifiable information 116-2 may include the license plate number of the vehicle. In another embodiment, the image 116 may represent an audio spectrogram. The object 116-1 may be a segment or a region of the audio spectrogram, and the corresponding personally identifiable information 116-2 may include acoustic features or speech features represented by the segment or region of the audio spectrogram. FIG. 3A provides further details related to the reception of the image 116.

[0033] The system 102 may further be configured to extract a block of feature maps as an output of an intermediate layer of the encoder 104-1 based on the application of the encoder 104-1 on the received image. In an embodiment, the encoder 104-1 may include multiple layers, where each layer may be associated with a block of feature maps that capture a specific level of features associated with the received image. For instance, initial layers of the encoder 104-1 may include corresponding blocks of feature maps that capture low-level features like edges and textures. On the other hand, deeper layers of the encoder 104-1 may include corresponding blocks of feature maps that capture higher-level features and abstractions. The layers of the encoder 104-1 may apply filters to generate a suitable block of feature maps as an output of an intermediate layer based on the received image or inputs from previous layer(s) of the encoder 104-1. FIG. 3A provides further details related to the extraction of the block of feature maps, at 304.

[0034] The system 102 may further be configured to identify the object 116-1 based on the extracted block of feature maps. The system 102 may apply the classification network 106 to generate a classification result. Further, the object 116-1 may be identified basedon the classification result. In an example embodiment, the object 116-1 may be identified as a human face, a vehicle, or a segment or region of an audio spectrogram. FIG. 3A provides further details related to the identification of the object, at 306.

[0035] The system 102 may further be configured to select a size label for the identified object 116-1 from the set of size labels. The set of size labels may include a first size, a second size, and a third size. The first size may be smaller than the second size and the third size, and the second size may be greater than the first size but smaller than the third size (i.e. , third size > second size > first size).

[0036] In an embodiment, the classification result may include a size label from the set of size labels for the identified object 116-1. In another embodiment, the size label for the identified object 116-1 may be selected based on the classification result and the position of the intermediate layer with respect to the positions of other layers of the encoder 104- 1. In yet another embodiment, the classification result may include an object label for the object 116-1 and bounding box coordinates to localize the object 116-1 in the image 116. The size label for the identified object 116-1 may be selected based on the bounding box coordinates. For instance, the system 102 may compute the size of the identified object 116-1 in pixels based on the classification result (i.e., the bounding box coordinates) and may select the size label for the identified object 116-1 from the set of size labels based on a comparison of the computed size with a threshold size range. FIG. 3A provides further details related to the selection of the size label.

[0037] The system 102 may further be configured to select an anonymization operation from a set of anonymization operations based on the selected size label. In an embodiment, the system 102 may be configured to select a blurring operation (for instance, blurring operation 310-1 , shown in FIG. 3B) as the anonymization operation based on the selection of the size label as the first size (also referred to as small size). Additionally, or alternatively, the system 102 may be configured to select a 3D morphing operation (forinstance, 3D morphing operation 310-2 shown in FIG. 3B) as the anonymization operation based on the selection of the size label as the second size (also referred to as moderate or medium size). Additionally, or alternatively, the system 102 may be configured to select a variational auto-encoder (for instance, variational auto-encoder 310-3 shown in FIG. 3B) as the anonymization operation based on the selection of the size label as the third size (also referred to as large size). FIG. 3B provides further details related to the selection of the anonymization operation.

[0038] The system 102 may further be configured to reconstruct a second image based on the application of the selected anonymization operation on the extracted block of feature maps. The reconstructed image may include an anonymized version of the personally identifiable information 116-2 of the object 116-1.

[0039] In an embodiment, the system 102 may be configured to generate a transformed block of feature maps based on the application of the selected anonymization operation on the extracted block of feature maps 304-1 . Further, the system 102 may be configured to generate an encoder output of the encoder 104-1 based on the transformed block of feature maps. Upon generation of the encoder output, the system 102 may be configured to reconstruct the second image based on the application of the decoder 104-2 on the encoder output.

[0040] In some instances, the image 116 may have multiple objects (e.g., faces) of different size labels, where each object may correspond to the same or a different block of feature maps with respective size labels (as shown in FIG. 4, for example). The generated image 116 in such a case may include the same or different transformation applied to a respective block of feature maps due to a particular size-specific anonymization operation. For instance, small faces in background of the image 116 may be blurred but larger faces in foreground may be modified using a variational auto-encoder (as shown in FIG. 3). FIG. 3B provides further details related to the reconstruction of thesecond image.

[0041] FIG. 2 is a block diagram that illustrates an exemplary system of FIG. 1 , in accordance with an embodiment of the disclosure. FIG. 2 is explained in conjunction with elements from FIG. 1 . With reference to FIG. 2, there is shown a block diagram 200 of the system 102. The system 102 may include circuitry 202, a memory 204, a network interface 206, and an input / output (I / O) device 208. The I / O device 208 may include a display device 208-A. The memory 204 may include the encoder-decoder neural network 104, the classification network 106, and data 210. The network interface 206 may connect the system 102 with the server 108, via the communication network 112.

[0042] The circuitry 202 may include suitable logic, circuitry, and / or interfaces that may be configured to execute program instructions associated with different operations to be executed by the system 102. The operations may include, for instance, image reception, block of feature maps extraction, object identification, size label selection, anonymization operation selection, anonymized image reconstruction, and the like. The circuitry 202 may include one or more processing units, which may be implemented as a separate processor. In an embodiment, the one or more processing units may be implemented as an integrated processor or a cluster of processors that perform the functions of the one or more specialized processing units, collectively. The circuitry 202 may be implemented based on a number of processor technologies known in the art. Examples of implementations of the circuitry 202 may be an X86-based processor, a Graphics Processing Unit (GPU), a Reduced Instruction Set Computing (RISC) processor, an Application-Specific Integrated Circuit (ASIC) processor, a Complex Instruction Set Computing (CISC) processor, a microcontroller, a central processing unit (CPU), and / or a combination thereof.

[0043] The memory 204 may include suitable logic, circuitry, interfaces, and / or code that may be configured to store one or more instructions to be executed by the circuitry 202. The one or more instructions stored in the memory 204 may be configured to execute thedifferent operations of the circuitry 202 (and / or the system 102). The memory 204 may be further configured to store the encoder-decoder neural network 104, the classification network 106, and the data 210. The data 210 may include a plurality of objects and corresponding personally identifiable information along with the dataset 118, where the object 116-1 and the personally identifiable information 116-2 may be a part of the plurality of objects and the corresponding personally identifiable information. Examples of implementation of the memory 204 may include, but are not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read- Only Memory (EEPROM), Hard Disk Drive (HDD), a Solid-State Drive (SSD), a CPU cache, and / or a Secure Digital (SD) card.

[0044] The network interface 206 may include suitable logic, circuitry, interfaces, and / or code that may be configured to facilitate communication between the system 102 and the server 108, via the communication network 112. The network interface 206 may be implemented by use of various known technologies to support wired or wireless communication of the system 102 with the communication network 112. The network interface 206 may include, but is not limited to, an antenna, a radio frequency (RF) transceiver, one or more amplifiers, a tuner, one or more oscillators, a digital signal processor, a coder-decoder (CODEC) chipset, a subscriber identity module (SIM) card, or a local buffer circuitry.

[0045] The network interface 206 may be configured to communicate via wireless communication with networks, such as the Internet, an Intranet, a wireless network, a cellular telephone network, a wireless local area network (LAN), or a metropolitan area network (MAN). The wireless communication may be configured to use one or more of a plurality of communication standards, protocols and technologies, such as Global System for Mobile Communications (GSM), Enhanced Data GSM Environment (EDGE), wideband code division multiple access (W-CDMA), Long Term Evolution (LTE), 5thGeneration (5G)New Radio (NR), code division multiple access (CDMA), time division multiple access (TDMA), Bluetooth, Wireless Fidelity (Wi-Fi) (such as IEEE 802.11a, IEEE 802.11 b, IEEE 802.11 g or IEEE 802.11 n), voice over Internet Protocol (VoIP), light fidelity (Li-Fi), Worldwide Interoperability for Microwave Access (Wi-MAX), a protocol for email, instant messaging, and a Short Message Service (SMS).

[0046] The I / O device 208 may include suitable logic, circuitry, interfaces, and / or code that may be configured to receive an image (for instance, image 116) and provide an anonymized image based on the received image. For example, the I / O device 208 may receive the image 116 including the object 116-1 and the personally identifiable information 116-2. The I / O device 208 may be further configured to render the reconstructed anonymized image on the user interface, for instance, the user device 114.

[0047] Examples of the I / O device 208 may include, but are not limited to, a display (e.g., a touch screen), a keyboard, a mouse, a joystick, a microphone, or a speaker. Examples of the I / O device 208 may further include braille I / O devices, such as, braille keyboards and braille readers.

[0048] The display device 208-A may include suitable logic, circuitry, and interfaces that may be configured to display or render the received image 116 and the anonymized image. In some embodiments, the display device 208-A may be a touch screen which may enable a user to provide a user-input via the display device 208-A. The display device 208-A may be realized through several known technologies such as, but not limited to, at least one of a Liquid Crystal Display (LCD) display, a Light Emitting Diode (LED) display, a plasma display, or an Organic LED (OLED) display technology, or other display devices. In accordance with an embodiment, the display device 208-A may refer to a display screen of a head mounted device (HMD), a smart-glass device, a see-through display, a projection-based display, an electro-chromic display, or a transparent display.

[0049] Various operations of the circuitry 202 are described further, for example, in FIG.3A and FIG. 3B.

[0050] FIG. 3A and FIG. 3B are flow diagrams that collectively illustrate various operations of the system of FIG. 1 , in accordance with an embodiment of the disclosure. FIG. 3A and FIG. 3B are explained in conjunction with elements from FIG. 1 and FIG. 2. With reference to FIG. 3A and FIG. 3B, there is shown an exemplary flow diagram 300 that illustrates a sequence of operations from 302 to 312 that may be executed by the circuitry 202 of FIG. 2 or the system 102 of FIG. 1. The sequence of operations may be executed for anonymization of personally identifiable information associated with objects in the image 116. The sequence of operations that may start at 302 and may terminate at 312.

[0051] At 302, an image including an object with corresponding personally identifiable information may be received. As shown, for instance, an image 302-1 may be received, including a human face 306-1 as object 302-2 and facial features of the human face 306- 1 as corresponding personally identifiable information. Although not shown, the received image may include a vehicle as an object, and the corresponding personally identifiable information may include a license plate number of the vehicle. In another embodiment, the received image may include segment or region of an audio spectrogram associated with the image as the object, and the corresponding personally identifiable information may include acoustic features or speech features represented by the segment or the region of the audio spectrogram.

[0052] In some instances, the image 302-1 may be a single frame sampled from a video. The image 302-1 may be transmitted to the system 102 via the user device 114. In some other instances, the image 302-1 may be retrieved from the server 108 or a local storage and stored into the memory 204. Alternatively, the image 302-1 may be acquired directly from an imaging device in near real-time.

[0053] At 304, a block of feature maps may be extracted based on application of encoder104-1 on the received image 302-1 . The circuitry 202 may extract, via the encoder 104-1 , a block of feature maps from a plurality of blocks of feature maps as an output of an intermediate layer of the encoder 104-1 (based on application of the encoder 104-1 on the received image 302-1 ). As shown, for example, the plurality of blocks of feature maps may include N blocks of feature maps 304-1 , 304-2... 304-N. A block of feature maps 304-1 may be extracted from the blocks of feature maps 304-1 , 304-2... 304-N. The extraction may be part of a layer-wise analysis of blocks and may be performed iteratively across all the layers (except input and final layers) of the encoder 104-1 to identify the intermediate layer(s) with target object(s) for anonymization.

[0054] The encoder 104-1 may include multiple layers, where each layer may be associated with a block of feature maps (of the plurality of blocks of feature maps) that may capture a specific level of features associated with object(s) in the received image 302-1. For instance, initial layers of the encoder 104-1 may output blocks of feature maps (such as block of feature maps 304-1) that may capture low-level features like edges and textures associated with the human face in the image 302-1 . Deeper layers of the encoder 104-1 may output corresponding blocks of feature maps (such as block of feature maps 304-N) that may capture higher-level features and abstractions of the human face in the image 302-1. Each layer of the encoder 104-1 may apply certain filters or operations to output a suitable block of feature maps (such as block of feature maps 304-1 ) as output of a respective layer of the encoder 104-1 based on input(s) from previously connected layer(s). These filters may be matrices or kernels that slide over the image 302-1 and perform element-wise multiplication and summation to produce the block of feature maps 304-1 as output.

[0055] It should be noted that the extraction may be part of a layer-wise analysis of blocks and may be performed iteratively across all the layers (except input and final layers) of the encoder 104-1 to identify the intermediate layer(s) with target object(s) foranonymization.

[0056] At 306, object(s) in the image 302-1 may be identified. The circuitry 202 may identify the object 302-2 such as a human face 306-1 in the image 302-1 based on the extracted block of feature maps 304-1. Specifically, the circuitry 202 may apply the classification network 106 on the block of feature maps 304-1 to generate a classification result. The classification result may include at least one of a class label (or an object class) and a size label for the identified object 302-2. The object 302-2 may be identified based on the classification result. The classification network 106 may generate the classification result by passing extracted block of feature maps 304-1 through a series of layers such as convolutional layers, pooling layers, and activation functions to output the classification result.

[0057] At 308, a size label may be selected for the object 302-2. The circuitry 202 may select a size label for the identified object 302-2 from the set of size labels (includes a first size, a second size, and a third size). The first size may be smaller than the second size and the third size, and the second size may be greater than the first size but smaller than the third size (i.e. , third size > second size > first size).

[0058] In an embodiment, the classification result may include a size label for the identified object 302-2. In another embodiment, the size label for the identified object 302- 2 may be selected based on the classification result and a position of the intermediate layer with respect to positions of other layers of the encoder 104-1. In yet another embodiment, the classification result may include an object label for the object 302-2 and bounding box coordinates to localize the object 302-2 in the image 302-1 . The object label may define the object 302-2, i.e., whether the object 302-2 is associated with a human entity (for instance, a human face) or a non-human entity (for instance, a vehicle). Based on the object label, the circuitry 202 may determine whether to take the object 302-2 into consideration. For instance, the image 302-1 may include various objects such as, ahuman face, house, garden, track, animal face, and the like. The objects may be determined based on the respective object labels, and the objects such as animals, streets signs, or people with no visible face in the image 302-1 may not be processed by the circuitry 202 for anonymization. In other words, the object label may be used to filter out any unwanted objects for anonymization, which may simplify the reconstruction process. Moreover, the bounding box coordinates may play a crucial role in the localization and size classification of various objects within the image 302-1. The bounding box coordinates may also define the precise location and dimensions of the object 302-1 .

[0059] In another embodiment, the circuitry 202 may compute a size of the identified object 302-2 in pixels based on the classification result and may select the size label for the identified object 302-1 from the set of size labels based on a comparison of the computed size with a threshold size range. The threshold size range may be predefined. For instance, the threshold size range may be defined as 200x200 pixels to 300x300 pixels. For an object having size less than 200*200 pixels, the size label may be selected as small-sized (at 308-1 ). For an object having size between 200x200 pixels and 300x300 pixels, the size label may be selected as medium-sized (at 308-2). For another object having size more than 300x300 pixels, the size label may be selected as large-sized (308- 3).

[0060] At 310, an anonymization operation may be selected based on the selected size label. The circuitry 202 may select a suitable anonymization operation from a set of anonymization operations based on the selected size label.

[0061] At 308-1 , it may be determined whether the size label is equal to the first size (e.g., small-sized object). In case the size label is equal to the first size, the circuitry 202 may select a blurring operation 310-1 as the anonymization operation. As shown, for example, the application of the blurring operation 310-1 on the block of feature maps 304- 1 associated with the object 302-2 (i. e. , human face 306-1 ) at the encoder side may resultin decoder-side reconstruction of an anonymized image 312-1 with blur around the human face 306-1 . In case the size label is not equal to the first size, the control may pass to 308- 2.

[0062] At 308-2, it may be determined whether the size label is equal to the second size (e.g., medium-sized object). In case the size label is equal to the second size, the circuitry 202 may select a 3D morphing operation 310-2 as the anonymization operation. As shown, for example, the application of the 3D morphing operation 310-2 on the block of feature maps 304-1 at the encoder side may result in decoder-side generation of an anonymized image 312-2 with anonymized version (i.e., a different hairstyle with a thinner face structure) of the human face 306-1 in the image 302-1 .

[0063] For the human face 306-1. the 3D morphing operation 310-2 may include transformation of various facial features to achieve different effects. For instance, the overall shape of the human face 306-1 may be adjusted to make the human face 306-1 rounder or more angular. The eyes may be resized or repositioned to appear larger, smaller, closer together, or farther apart. The nose may be reshaped to be more refined or broader, while the mouth may be altered to have fuller or thinner lips. Cheekbones may be enhanced or reduced to create a more defined or softer look, and the jawline may be adjusted to be more pronounced or subtle. Additionally, the skin texture may be smoothed to reduce wrinkles, or new features like freckles may be added. Even the hairline may be modified to change the hairstyle or forehead shape. These transformations may range from subtle enhancements to dramatic changes, allowing for a wide variety of creative possibilities.

[0064] In accordance with an embodiment, the 3D morphing operation 310-2 may involve detection of facial landmark information such as eyes, nose, chin, and the like associated with the human face 306-1 in the image 302-1. Further, a 3D morphable face model may be aligned to a specific pose. The aligned 3D morphable face model may beprojected onto a 2D plane to obtain a 2D face projection. Furthermore, the circuitry 202 may iteratively optimize parameters of the aligned 3D morphable version based on a fitment of landmarks in the 2D face projection to the facial landmark information of the human face 306-1 detected in the image 302-1 . Thereafter, the circuitry 202 may alter the optimized parameters based on an anonymization rule to obtain an anonymized 3D face version of the human face 306-1. The anonymization rule may be selected from a set of defined anonymization rules associated with the 3D morphing operation. Thereafter, the circuitry 202 may transfer a texture of the human face 306-1 from the image 302-1 to the anonymized 3D face version of the human face. After the transfer, the circuitry 202 may obtain the anonymized image 312-2 with a 2D view of the anonymized 3D face version as the anonymized image.

[0065] At 310-3, it may be determined whether the size label is equal to the third size (e.g., large-sized object). In case the size label is equal to the third size, the circuitry 202 may select a variational auto-encoder 310-3 as the anonymization operation. The variational auto-encoder 310-3 may be a type of generative model that may learn to encode input data, such as the block of feature maps 304-1 associated with the image 302-1 , into encoded data in a latent space. The encoded data may be decoded (at the decoder 104-2) back to the original input (i.e. , an image).

[0066] In some embodiments, the variational auto-encoder 310-3 may be used to anonymize facial features by altering specific aspects of the human face 306-1 while retaining overall structure of the human face 306-1 . For example, the variational autoencoder 310-3 may modify the shape and size of facial features such as the eyes, nose, and mouth to make such facial features unrecognizable while keeping the general face shape intact. The variational auto-encoder 310-3 may also change the skin texture and tone, add or remove facial hair, and adjust the hairstyle. Additionally, or alternatively, the variational auto-encoder 310-3 may blend features from multiple faces to create acomposite face that does not resemble any individual person. This may ensure that the anonymized face looks realistic and natural, making it difficult to identify the original human face 306-1 .

[0067] In these or other embodiments, the variational auto-encoder 310-3 may apply other operations such as a masking operation where the circuitry 202 may overlay a solid color or pattern onto the region of interest to completely obscure details and to ensure no identifiable information is visible. The circuitry 202 may use the variational auto-encoder 310-3 to replace the personally identifiable information with a different, non-identifiable image or pattern, such as a generic face or random characters, through a swapping operation. Geometric transformations, such as warping or stretching, may be applied by the circuitry 202 to distort a region of interest so as to alter the shape and appearance of the personally identifiable information. Noise addition is another operation where the circuitry 202 may use the variational auto-encoder 310-3 to introduce random noise to the region of interest. Alternatively, the circuitry 202 may use the variational auto-encoder 310- 3 to perform an inpainting operation which fills the region of interest with content that blends seamlessly with the surrounding area. This may remove the personally identifiable information and replace the personally identifiable information with plausible background content.

[0068] The variational auto-encoder 310-3 may be applied to the block of feature maps 304-1 to achieve more granular and localized transformations. For example, by application of the variational auto-encoder 310-3 to the block of feature maps 304-1 , specific regions or aspects of the image 302-1 may be modified without altering the entire image. This approach may allow for targeted anonymization or enhancement of particular features, such as a new hair style while preserving the overall context of the image 302-1. As shown, for example, the application of the variational auto-encoder 310-3 on the block of feature maps 304-1 may result in addition of features such as moustache, beard, moles, and othersuch features to the human face 306-1 in an anonymized image 312-3 (at the side of the decoder 104-2). The addition may result in transformation of the personally identifiable information of the human face 306-1 .

[0069] At 312, a second image may be reconstructed. The circuitry 202 may reconstruct a second image based on application of the selected anonymization operation on the extracted block of feature maps 304-1 or the received image 302-1 . The reconstructed image may include an anonymized version of the personally identifiable information of the object. For instance, the system 102 may be configured to generate a transformed block of feature maps based on application of the selected anonymization operation on the extracted block of feature maps. The transformed block of feature maps may include transformed version of personally identifiable information. Specifically, the system 102 may be configured to generate an encoder output of the encoder 104-1 based on the transformed block of feature maps. Upon generation of the encoder output, the system 102 may be configured to reconstruct the second image based on application of the decoder 104-2 on the encoder output.

[0070] In an embodiment, the blurring operation 310-1 may be applied to the extracted block of feature maps 304-1 (associated with the first size label) to generate a transformed block of feature maps. The transformed block of feature maps may replace the original output (i.e. , extracted block of feature maps 304-1 ) of the intermediate layer of the encoder 104-1 . The encoder 104-1 may generate the encoded data based on the transformed block of feature maps. The decoder 104-2 may process the encoded data to generate the anonymized image 312-1 (also referred to as second image). The application of the blurring operation 310-1 on the block of feature maps 304-1 may result in the transformation of the personally identifiable information of the human face 306-1 in the anonymized image 312-1.

[0071] In another embodiment, the 3D morphing operation 310-2 may be applied to theblock of feature maps 304-1 (associated with the second size label) to generate an anonymized image 312-2 (at the decoder side). The application of the 3D morphing operation 310-2 on the block of feature maps 304-1 may result in a localized transformation of the personally identifiable information such as a thinning of the face structure with a change in hairstyle of the human face 306-1 , as illustrated in the anonymized image 312- 2.

[0072] In another embodiment, the variational auto-encoder 310-3 may be applied to the block of feature maps 304-1 (associated with the third size) to generate the anonymized image 312-3 (as output of the decoder 104-2). The application of the variational auto-encoder 310-3 may result in transformation of the personally identifiable information of the human face 306-1 at the decoder side with addition of features such as moustaches and beard on the human face 306-1 , as illustrated in the anonymized image 312-1.

[0073] FIG. 4 is a block diagram that illustrates operations executed by a system for anonymization of personally identifiable information of multiple objects present in an image, in accordance with an embodiment of the disclosure. FIG. 4 is described in conjunction with elements from FIG. 1 , FIG. 2, and FIG. 3. With reference to FIG. 4, there is shown a diagram 400. The encoder 104-1 may receive an image 402 and may further traverse through the blocks of feature maps 304-1 to 304-N to extract a suitable block of feature maps (for instance, block of feature maps 304-1 ) as an output of an intermediate layer of the encoder 104-1 based on the received image 402. Further, the classification network 106 may identify objects (for instance, human faces) in the image 402 of the same and different size labels based on the respective blocks of feature maps. Each size label for a human face may be selected from a set of size labels. Furthermore, the circuitry 202 may select a suitable anonymization operation from the set of anonymization operations based on the selected size label. The set of anonymization operations may include blurringoperation 310-1 , 3D morphing operation 310-2, and variational auto-encoder 310-3.

[0074] As shown, for example, the blurring operation 310-1 may be selected for block of feature maps corresponding to small-sized faces 404-1 , 404-2... 402-N (associated with first size label) in the image 402, the 3D morphing operation 310-2 may be selected for block of feature maps corresponding to medium-sized faces 406-1 , 406-2 ... 406-N (associated with second size label) in the image 402, and the variational auto-encoder 310-3 may be selected for block of feature maps corresponding to large-sized faces 408- 1 , 408-2 ... 408-N (associated with third size label) in the image 402. The selected anonymization operation(s) may be applied on respective blocks of feature maps (extracted from respective layers of the encoder 104-1 , which may result in generation of anonymized data. The anonymized data may be fed to the decoder 104-2, which may process the data to reconstruct the second image with anonymized faces.

[0075] FIG. 5 is a flowchart that illustrates exemplary operations of a method for anonymization of personally identifiable information in images and other spatial signals, in accordance with an embodiment of the disclosure. FIG. 5 is described in conjunction with elements from FIG. 1 , FIG. 2, FIG. 3, and FIG. 4. With reference to FIG. 5, there is shown a flowchart 500. The flowchart 500 may include operations from 502 to 518 and may be implemented by the system 102 of FIG. 1 or by the circuitry 202 of FIG. 2. The flowchart 500 may start at 502 and proceed to 504.

[0076] At 504, an image including an object with corresponding personally identifiable information 116-2 may be received. The circuitry 202 may be configured to receive an image 116, which includes object 116-1 with corresponding personally identifiable information 116-2. The object 116-1 may be a human face, and the corresponding personally identifiable information 116-2 may include facial features of the human face. Alternatively, the object 116-1 may also be a vehicle, and the corresponding personally identifiable information 116-2 may include a license plate number of the vehicle.Alternatively, the object 116-1 may be a segment or a region of an audio spectrogram associated with the image 116, and the corresponding personally identifiable information 116-2 may include acoustic features or speech features represented by the segment or the region of the audio spectrogram. Details related to the receipt of the image are further described, for example, at 302 in FIG. 3.

[0077] At 506, a block of feature maps may be extracted based on application of encoder on the received image 116. The circuitry 202 may be configured to extract a block of feature maps (for instance, block of feature maps 304-1 ) of a plurality of blocks of feature maps 304-1 as an output of an intermediate layer of the encoder 104-1 based on application of the encoder 104-1 on the received image 116. Details related to the extraction of the block of feature maps are further described, for example, at 304 in FIG. 3.

[0078] At 508, object in the image may be identified. The circuitry 202 may be configured identify object 116-1 , for instance, the human face 306-1 in the image 116 based on the extracted block of feature maps 304-1. The circuitry 202 may apply the classification network 106 to generate a classification result. Further, the object 116-1 may be identified based on the classification result. Details related to the identification of the object in the image are further described, for example, at 306 in FIG. 3.

[0079] At 510, a size label may be selected for the object. The circuitry 202 may be configured to select a size label for the identified object 116-1 from the set of size labels 118-3. In an exemplary embodiment, the set of size labels may include a first size, a second size, and a third size. The first size may be smaller than the second size and the third size. The second size may be greater than the first size but smaller than the third size. Details related to the generation of the first speaker embedding are further described, for example, at 308 in FIG. 3.

[0080] At 512, an anonymization operation may be selected based on the selected sizelabel. The circuitry 202 may be configured to select a suitable anonymization operation from a set of anonymization operations based on the selected size label. In one embodiment, the circuitry 202 may select the blurring operation 310-1 as the anonymization operation based on the selection of the size label as the first size. In another embodiment, the circuitry 202 may select the 3D morphing operation 310-2 as the anonymization operation based on the selection of the size label as the second size. In yet another embodiment, the circuitry 202 may select the variational auto-encoder 310-3 as the anonymization operation based on the selection of the size label as the third size. Details related to the generation of the lip features are further described, for example, at 310 in FIG. 3.

[0081] At 514, a second image may be reconstructed. The circuitry 202 may be configured to reconstruct a second image based on application of the selected anonymization operation on the extracted block of feature maps304-1. The reconstructed image includes an anonymized version of the personally identifiable information 116-2 of the object 116-1. Details related to the reconstruction of the second image are further described, for example, at 312 in FIG. 3.

[0082] Although the flowchart 500 is illustrated as discrete operations, such as, 502, 504, 506, 508, 510, 512, and 514, the disclosure is not so limited. Accordingly, in certain embodiments, such discrete operations may be further divided into additional operations, combined into fewer operations, or eliminated, depending on the implementation without detracting from the essence of the disclosed embodiments.

[0083] Various embodiments of the disclosure may provide a non-transitory computer- readable medium and / or storage medium having stored thereon, computer-executable instructions executable by a machine and / or a computer to operate a system (for example, the system 102 of FIG. 1 ). Such instructions may cause the system 102 to perform operations that may include receipt of a first image (for example, the image 116 of FIG. 1 )including an object (for example, the object 116-1 of FIG. 1 ) with corresponding personally identifiable information (for example, the personally identifiable information 116-2 of FIG. 1 ). The operations may further include extraction of a block of feature maps as an output of an intermediate layer of an encoder (for example, the encoder 104-1 of FIG. 1 ) of an encoder-decoder neural network (for example, the encoder-decoder neural network 104 of FIG. 1 ) based on application of the encoder 104-1 on the received first image 116. The operations may further include identification of the object 116-1 based on the extracted block of feature maps. The operations may further include selection of a size label for the identified object 116-1 from a set of size labels (for example, the set of size labels of FIG. 1 ). The operations may further include selection of an anonymization operation from a set of anonymization operations based on the selected size label. The operations may further include reconstruction of a second image based on application of the anonymization operation on the extracted block of feature maps. The reconstructed second image includes an anonymized version of the personally identifiable information 116-2 of the object 116-1.

[0084] Exemplary aspects of the disclosure may provide a system (such as, the system 102 of FIG. 1 ) that includes circuitry (such as, the circuitry 202 of FIG. 2). The circuitry 202 may be configured to receive a first image for example, the image 116 of FIG. 1 ) including an object (for example, the object 116-1 of FIG. 1 ) with corresponding personally identifiable information (for example, the personally identifiable information 116-2 of FIG. 1 ). The circuitry 202 may be configured to extract a block of feature maps as an output of an intermediate layer of an encoder (for example, the encoder 104-1 of FIG. 1 ) of an encoder-decoder neural network (for example, the encoder-decoder neural network 104 of FIG. 1 ) based on application of the encoder 104-1 on the received first image 116. The circuitry 202 may be configured to identify the object 116-1 based on the extracted block of feature maps. The circuitry 202 may be configured to select a size label for the identifiedobject 116-1 from a set of size labels. The circuitry 202 may be configured to select an anonymization operation from a set of anonymization operations based on the selected size label. The circuitry 202 may be configured to reconstruct a second image based on application of the anonymization operation on the extracted block of feature maps. The reconstructed second image includes an anonymized version of the personally identifiable information 116-2 of the object 116-1.

[0085] In an embodiment, the encoder-decoder neural network 104 may be a U-Net based diffusion model.

[0086] In an embodiment, the object 116-1 may be a human face, and the personally identifiable information 116-2 may include facial features of the human face.

[0087] In an embodiment, the first image 116 may be an audio spectrogram, the object 116-1 may be a segment or a region of the audio spectrogram, and the personally identifiable information 116-2 may correspond to acoustic features or speech features represented by the segment or the region.

[0088] In an embodiment, the object 116-1 may be a vehicle, and the personally identifiable information 116-2 may include a license plate number of the vehicle.

[0089] In an embodiment, the circuitry 202 may further be configured to apply a pretrained classification network (for example, the classification network 106 of FIG. 1 ) on the block of feature maps to generate a classification result. The object 116-1 may be identified based on the classification result.

[0090] In an embodiment, the classification result may include the size label for the identified object 116-1.

[0091] In an embodiment, the size label for the identified object 116-1 may be selected based on the classification result and a position of the intermediate layer with respect to positions of other layers of the encoder 104-1 .

[0092] In an embodiment, the classification result may include an object label for theobject 116-1 and bounding box coordinates to localize the object 116-1 in the first image 116. The size label for the identified object is selected based on the bounding box coordinates.

[0093] In an embodiment, the circuitry 202 may further be configured to compute a size of the identified object 116-1 in pixels based on the classification result. The circuitry 202 may further be configured to select the size label for the identified object 116-1 from the set of size labels based on a comparison of the computed size with a threshold size range.

[0094] In an embodiment, the circuitry 202 may further be configured to generate a transformed block of feature maps based on application of the selected anonymization operation on the extracted block of feature maps. The circuitry 202 may further be configured to generate an encoder output of the encoder 104-1 based on the transformed block of feature maps. The circuitry 202 may further be configured to reconstruct the second image based on application of the decoder 104-2 on the encoder output.

[0095] The present disclosure may be realized in hardware, or a combination of hardware and software. The present disclosure may be realized in a centralized fashion, in at least one computer system, or in a distributed fashion, where different elements may be spread across several interconnected computer systems. A computer system or other apparatus adapted to carry out the methods described herein may be suited. A combination of hardware and software may be a general-purpose computer system with a computer program that, when loaded and executed, may control the computer system such that it carries out the methods described herein. The present disclosure may be realized in hardware that comprises a portion of an integrated circuit that also performs other functions.

[0096] The present disclosure may also be embedded in a computer program product, which comprises all the features that enable the implementation of the methods described herein, and which when loaded in a computer system is able to carry out these methods.Computer program, in the present context, means any expression, in any language, code or notation, of a set of instructions intended to cause a system with information processing capability to perform a particular function either directly, or after either or both of the following: a) conversion to another language, code or notation; b) reproduction in a different material form.

[0097] While the present disclosure is described with reference to certain embodiments, it will be understood by those skilled in the art that various changes may be made, and equivalents may be substituted without departure from the scope of the present disclosure. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the present disclosure without departure from its scope. Therefore, it is intended that the present disclosure is not limited to the embodiment disclosed, but that the present disclosure will include all embodiments that fall within the scope of the appended claims.

Claims

CLAIMSWhat is claimed is:1 . A system, comprising: a memory configured to a store an encoder-decoder neural network comprising an encoder and a decoder; and circuitry configured to: receive a first image comprising an object with personally identifiable information; extract a block of feature maps as an output of an intermediate layer of the encoder based on application of the encoder on the received first image; identify the object based on the extracted block of feature maps; select a size label for the identified object from a set of size labels; select an anonymization operation from a set of anonymization operations based on the selected size label; and reconstruct a second image based on application of the selected anonymization operation on the extracted block of feature maps or the first image, wherein the reconstructed second image includes an anonymized version of the personally identifiable information of the object.

2. The system according to claim 1 , wherein the encoder-decoder neural network is a U- Net based diffusion model.

3. The system according to claim 1 , wherein the object is a human face, and the personally identifiable information includes facial features of the human face.

4. The system according to claim 1 , wherein the first image is an audio spectrogram, the object is a segment or a region of the audio spectrogram, and the personally identifiable information corresponds to acoustic features or speech features represented by the segment or the region.

5. The system according to claim 1 , wherein the object is a vehicle, and the personally identifiable information includes a license plate number of the vehicle.

6. The system according to claim 1 , wherein the circuitry is further configured to apply a pre-trained classification network on the block of feature maps to generate a classification result, and the object is identified based on the classification result.

7. The system according to claim 6, wherein the classification result comprises the size label for the identified object.

8. The system according to claim 6, wherein the size label for the identified object is selected based on the classification result and a position of the intermediate layer with respect to positions of other layers of the encoder.

9. The system according to claim 6, wherein the classification result comprises an object label for the object and bounding box coordinates to localize the object in the first image, and the size label for the identified object is selected based on the bounding box coordinates.

10. The system according to claim 6, wherein the circuitry is further configured to:compute a size of the identified object in pixels based on the classification result; and select the size label for the identified object from the set of size labels based on a comparison of the computed size with a threshold size range.11 . The system according to claim 1 , wherein the circuitry is further configured to: generate a transformed block of feature maps based on application of the selected anonymization operation on the extracted block of feature maps; generate an encoder output of the encoder based on the transformed block of feature maps; and reconstruct the second image based on application of the decoder on the encoder output.

12. The system according to claim 1 , wherein the size label for the object is selected as one of a first size, a second size, or a third size, the first size is smaller than the second size and the third size, and the second size is greater than the first size but smaller than the third size.

13. The system according to claim 12, wherein the circuitry is further configured to select a variational auto-encoder as the anonymization operation based on the selection of the size label as the third size.

14. The system according to claim 12, wherein the circuitry is further configured to select a blurring operation as the anonymization operation based on the selection of the size label as the first size.

15. The system according to claim 12, wherein the circuitry is further configured to determine a 3D morphing operation as the anonymization operation based on the selection of the size label as the second size.

16. The system according to claim 15, wherein the application of the anonymization operation as the 3D morphing operation comprises: detect facial landmark information associated with the object in the first image; align a 3D morphable face model to a specific pose; project the aligned 3D morphable face model onto a 2D plane to obtain a 2D face projection; iteratively optimize parameters of the aligned 3D morphable model based on a fitment of landmarks in the 2D face projection to the facial landmark information; alter the optimized parameters based on an anonymization rule to obtain an anonymized 3D face model of the object; transfer a texture of the object from the first image to the anonymized 3D face model of the object; and obtain, after the transfer, a 2D view of the anonymized 3D face model as the second image.

17. A method, comprising: in a system that stores an encoder-decoder neural network comprising an encoder and a decoder: receiving a first image comprising an object with personally identifiable information; extracting a block of feature maps as an output of an intermediate layer of the encoder based on application of the encoder on the received first image;identifying the object based on the extracted block of feature maps; selecting a size label for the identified object from a set of size labels; selecting an anonymization operation from a set of anonymization operations based on the selected size label; and reconstructing a second image based on application of the anonymization operation on the extracted block of feature maps or the first image, wherein the reconstructed second image includes an anonymized version of the personally identifiable information of the object.

18. The method according to claim 17, further comprising: generating a transformed block of feature maps based on application of the selected anonymization operation on the extracted block of feature maps; generating an encoder output of the encoder based on the transformed block of feature maps; and reconstructing the second image based on application of the decoder on the encoder output.

19. A non-transitory computer-readable medium having stored thereon, computerexecutable instructions that when executed by a system, causes the system to execute operations, the operations comprising: receiving a first image comprising an object with personally identifiable information; extracting a block of feature maps as an output of an intermediate layer of an encoder of an encoder-decoder neural network based on application of the encoder on the received first image; identifying the object based on the extracted block of feature maps;selecting a size label for the identified object from a set of size labels; selecting an anonymization operation from a set of anonymization operations based on the selected size label; and reconstructing a second image based on application of the anonymization operation on the extracted block of feature maps or the first image, wherein the reconstructed second image includes an anonymized version of the personally identifiable information of the object.

20. The non-transitory computer-readable medium according to claim 19, wherein the operations comprise: generating a transformed block of feature maps based on application of the selected anonymization operation on the extracted block of feature maps; generating an encoder output of the encoder based on the transformed block of feature maps; and reconstructing the second image based on application of a decoder of the encoder-decoder neural network on the encoder output.

Images