Intelligent photographing service method and system for preventing leakage of portrait information

Abstract

This application discloses a smart photo-taking service method and system for preventing the leakage of facial information, belonging to the field of image privacy processing technology. The method includes: before shooting, the camera previews the image, performs real-time detection and positioning of all facial images in the image, categorizes the detected images into authorized and unauthorized subjects, and generates a role intent tag for each image. After shooting, the device locally crops and extracts the areas of unauthorized subjects in the captured image according to the determined protection level and protection strategy, while simultaneously performing corresponding privacy desensitization processing to generate a desensitized image of the facial area and fuse it back to the original image position. When the authorized party requests to restore the original image, the identity and permissions of the authorized party are verified by multiple parties. After successful verification, the corresponding decryption key is released, restoring the original facial information before desensitization. This method can preserve the original expression of the photo to the greatest extent and avoid the privacy leakage risk caused by incomplete desensitization.

Description

Technical Field

[0001] This application belongs to the field of image privacy processing technology, specifically, it relates to a smart photography service method and system for preventing the leakage of facial information. Background Technology

[0002] With the rapid development of artificial intelligence and camera technology, photo services have been widely used in various scenarios such as daily life, security monitoring, and commercial marketing, bringing great convenience to people's lives and social operations. However, as a unique and unchangeable biometric feature, the risk of leakage of facial information is becoming increasingly prominent. Unprotected photo-taking may lead to the illegal collection and misuse of facial information by unauthorized personnel, which may cause a series of security problems and even threaten personal property safety and public safety.

[0003] While existing portrait privacy protection technologies used in photo-taking services can reduce the risk of information leakage to some extent, most adopt a one-size-fits-all approach. They either desensitize all portraits, affecting the shooting effect and the user experience of authorized users, or simply blur some portraits, failing to accurately protect the privacy of unauthorized users. Moreover, they often fail to differentiate processing based on the shooting intent and scene characteristics, potentially leading to the risk that desensitized images can be reverse-engineered into identifiable original images. This makes it impossible to completely eliminate privacy leaks. Furthermore, they lack traceability and flexible recovery capabilities, making auditing and traceability impossible. Either the original image cannot be recovered, or the recovery process lacks security verification, which can easily lead to the illegal acquisition of original privacy information. They fail to balance the needs of privacy protection and the reuse of original data. Summary of the Invention

[0004] To address the aforementioned problems and technical deficiencies, this application adopts the following technical solution: a smart photography service method for preventing the leakage of facial information, comprising the following steps:

[0005] Before shooting, the camera previews the image, detects and locates all the people in the image in real time, divides the detected people into authorized subjects and unauthorized subjects, and generates a role intent label for each person.

[0006] After shooting, the device locally crops and extracts the portrait area of ​​the unauthorized subject in the captured image according to the determined protection level and protection strategy, and performs corresponding privacy desensitization processing to generate a desensitized portrait area image and merge it back into the original image position.

[0007] When the authorizing party requests to restore the original image, the identity and permissions of the authorizing party are verified by multiple parties. After the verification is successful, the corresponding decryption key is released to restore the original portrait information before desensitization.

[0008] Preferably, the image preview is performed by the camera before the camera shutter is pressed or a shooting command is issued, and the camera performs image preview and scene perception to extract image features, scene features and behavioral features.

[0009] Scene awareness includes:

[0010] Based on ambient lighting, scene categories are determined, and image and scene features are extracted.

[0011] The user behavior context is determined based on the application type, historical photo shooting mode, and geographical location. Then, behavioral features are extracted from the user behavior context based on the principle of minimum necessity and privacy compliance policy. The extraction of behavioral features is completed locally on the device and is not associated with user identity.

[0012] Furthermore, the generation of the character intent tag includes:

[0013] Multi-scale detection is performed on all portraits in the captured image to locate and extract the structured data of each portrait. Subject determination is made based on the focus position and the proportion of the image, the bounding boxes of all portrait regions are determined, and all portrait regions are extracted and globally sorted.

[0014] The system uses a pre-defined authorized facial feature database to match the extracted facial regions for identity verification, and categorizes the detected facial images into authorized subjects and unauthorized subjects.

[0015] Then, based on the statistical analysis of users' historical photo-taking features, an intent-aware classification model is constructed, which combines image features, scene features, and behavioral features to determine the category of photo-taking intent;

[0016] The intent category is combined with whether the portrait is an authorized subject, converted into a corresponding tag, and then bound to the portrait.

[0017] Furthermore, the determination of the intent category involves identifying the subject of the image based on image features, then determining the intent based on scene features and behavioral features, and finally combining the subject of the image and the intent to determine the intent category, including:

[0018] If the main subject of the image is a portrait and the intention is to take a selfie, engage in social interaction, or live a lifestyle-related activity, the intention of taking the photo is classified as a selfie.

[0019] If the subject of the image is not the photographer, and the intention is to take a portrait or a group photo, the photographic intent is classified as "photographing another person's subject."

[0020] If the main subject of the image is a person, not the subject of the photograph, and the intention is to record scenery, architecture, or events, the photographic intention category is determined to be scene recording.

[0021] If the subject of the image is a business-related photograph, and the intent is to collect documents, conduct surveys, or perform business-related tasks, and the intent is to create business-related images, then the photographic intent category is determined to be task-oriented.

[0022] If the subject of the image is captured by a surveillance camera and the intention is related to recording surveillance footage, the category of the intention to take the photo is determined to be security monitoring.

[0023] Furthermore, the privacy desensitization process includes:

[0024] Based on image and scene features, semantic recognition of the current shooting environment is performed to determine the scene privacy level;

[0025] The corresponding protection level is determined based on the screen proportions and character intent tags between each unauthorized and authorized entity.

[0026] Combining scene privacy level and protection level, the privacy protection strategy corresponding to each portrait is matched from the preset scene and role mapping strategy table, and the privacy protection strategy is quantified.

[0027] The identity attribute decoupling network is used to decompose the facial features of each unauthorized person into identity feature vector and attribute feature vector. Based on the identity feature vector and attribute feature vector, the quantized privacy protection level and protection strategy are feature-encoded to generate a portrait privacy processing matrix.

[0028] A facial privacy processing matrix is ​​used to perform differentiated privacy desensitization processing on the facial image area of ​​each unauthorized subject.

[0029] Furthermore, after the privacy desensitization process, each processed image will undergo reverse risk verification and sensitivity checks.

[0030] Reverse risk verification verifies whether the image after privacy desensitization can be reversed to be restored to the original identity-recognizable image. If it can, then privacy desensitization is performed again.

[0031] Sensitivity check detects whether there is sensitive information in the image other than human figures. If so, the sensitive area is first cropped out, then the non-sensitive, non-human figures area around the sensitive area within a preset range is stretched to the original sensitive area, and finally the sensitive area is Gaussian blurred to maintain visual consistency.

[0032] Reverse risk verification and sensitivity checks are both completed before image encoding and storage, ultimately generating unauthorized photos whose original information is unrecognizable.

[0033] Furthermore, each photo-taking operation will generate a privacy processing log, which will be hashed and then uploaded to a distributed evidence storage network for auditing.

[0034] The privacy processing log includes: photo capture time, extracted features, intent category, portrait attributes, type of processing operation performed on each portrait, and encryption key;

[0035] The encryption key is generated by combining extracted features, timestamps, device fingerprints, privacy levels, and protection levels to create an encryption key that is associated with the scenario.

[0036] After the privacy desensitization process, an encryption key is used to encrypt the original human image region data and sensitive regions extracted before desensitization. The encrypted original human image region data and sensitive regions are then embedded in the privacy desensitized image file in a steganographic or encrypted metadata manner.

[0037] A smart photo-taking service system for preventing the leakage of facial information includes:

[0038] The portrait preprocessing module performs a preview of the image before shooting, detects and locates all portraits in the image in real time, divides the detected portraits into authorized subjects and unauthorized subjects, and generates a role intent label for each portrait.

[0039] The privacy encryption module, after shooting, crops and extracts the portrait area of ​​the unauthorized subject in the shooting image according to the determined protection level and protection policy on the device local, and performs corresponding privacy desensitization processing to generate a desensitized portrait area image and merge it back into the original image position;

[0040] The privacy recovery module performs multi-party security verification of the authorizing party's identity and permissions when the authorizing party requests the restoration of the original image. After successful verification, it releases the corresponding decryption key and restores the original portrait information before desensitization.

[0041] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the content of a smart photography service method for preventing the leakage of facial information as described above.

[0042] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the content of a smart photography service method for preventing the leakage of facial information as described above.

[0043] Compared to existing technologies, the beneficial effects of this application are as follows:

[0044] (1) By introducing role intent tags and scene semantic recognition in the preview stage, this application can dynamically determine the protection level and corresponding privacy desensitization strategy of each unauthorized subject, distinguish between unintentionally intruding passers-by and unauthorized objects of interest, and take corresponding desensitization processing in combination with scene privacy, so as to effectively protect privacy while preserving the original expression of the photo to the greatest extent.

[0045] (2) This application determines the privacy level of the scene through scene semantic recognition, assigns the protection level by combining the role intent label of the unauthorized subject, matches the corresponding privacy protection strategy, and uses the identity attribute decoupling network to generate the portrait privacy processing matrix to achieve differentiated privacy desensitization of the unauthorized subject. At the same time, after desensitization, reverse risk verification is set to ensure that the processed image cannot be reversed to be restored to the original identity-identifiable image, thus avoiding the privacy leakage risk caused by incomplete desensitization.

[0046] (3) Each photo-taking operation generates a privacy processing log, which is then hashed and uploaded to a distributed evidence storage network for auditing. This ensures that the privacy processing process is traceable and verifiable, facilitating subsequent auditing and supervision. At the same time, by generating a composite encryption key associated with the scene, the original portrait area data and sensitive areas before desensitization are encrypted and embedded in the processed image file in the form of steganography or encrypted metadata. Combined with the multi-party security verification mechanism of the licensor, the original image can be securely restored. This not only protects privacy but also enables the reuse of the original data when needed. Attached Figure Description

[0047] In the attached diagram:

[0048] Figure 1 This is a schematic diagram of the method steps in an embodiment of this application.

[0049] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments. Generally, the components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0051] Example 1, such as Figure 1 As shown, a smart photo-taking service method to prevent the leakage of facial information includes the following steps:

[0052] Before shooting, the camera previews the image, detects and locates all the people in the image in real time, divides the detected people into authorized subjects and unauthorized subjects, and generates a role intent label for each person.

[0053] Image preview is the process by which the camera previews the image and perceives the scene before the camera shutter is pressed or the shooting command is issued, extracting image features, scene features, and behavioral features.

[0054] Scene awareness includes:

[0055] Based on ambient lighting, scene categories are determined, and image and scene features are extracted.

[0056] The user behavior context is determined based on the application type, historical photo shooting mode, and geographical location. Then, behavioral features are extracted from the user behavior context based on the principle of minimum necessity and privacy compliance policy. The extraction of behavioral features is completed locally on the device and is not associated with user identity.

[0057] The generation of character intent tags includes:

[0058] Multi-scale detection is performed on all portraits in the captured image to locate and extract the structured data of each portrait. Subject determination is made based on the focus position and the proportion of the image, the bounding boxes of all portrait regions are determined, and all portrait regions are extracted and globally sorted.

[0059] Structured metadata includes: the size, location, sharpness, and degree of occlusion of each person's image;

[0060] Subject determination includes: focus position and screen occupancy.

[0061] The system uses a pre-defined authorized facial feature database to match the extracted facial regions for identity verification, and categorizes the detected facial images into authorized subjects and unauthorized subjects.

[0062] Then, based on the statistical analysis of users' historical photo-taking features, an intent-aware classification model is constructed, which combines image features, scene features, and behavioral features to determine the category of photo-taking intent;

[0063] The intent category is combined with whether the portrait is an authorized subject, converted into a corresponding tag, and then bound to the portrait.

[0064] Determining the intent category involves identifying the subject of the image based on its features, then determining the intent based on scene and behavioral characteristics, and finally combining the subject and intent to determine the intent category, which includes:

[0065] If the main subject of the image is a portrait and the intention is to take a selfie, engage in social interaction, or live a lifestyle-related activity, the intention of taking the photo is classified as a selfie.

[0066] If the subject of the image is not the photographer, and the intention is to take a portrait or a group photo, the photographic intent is classified as "photographing another person's subject."

[0067] If the main subject of the image is a person, not the subject of the photograph, and the intention is to record scenery, architecture, or events, the photographic intention category is determined to be scene recording.

[0068] If the subject of the image is a business-related photograph, and the intent is to collect documents, conduct surveys, or perform business-related tasks, and the intent is to create business-related images, then the photographic intent category is determined to be task-oriented.

[0069] If the subject of the image is captured by a surveillance camera and the intention is related to recording surveillance footage, the category of the intention to take the photo is determined to be security monitoring.

[0070] After shooting, the device locally crops and extracts the portrait area of ​​the unauthorized subject in the captured image according to the determined protection level and protection strategy, and performs corresponding privacy desensitization processing to generate a desensitized portrait area image and merge it back into the original image position.

[0071] Privacy desensitization processing includes:

[0072] Based on image and scene features, semantic recognition of the current shooting environment is performed to determine the scene privacy level;

[0073] The corresponding protection level is determined based on the screen proportions and character intent tags between each unauthorized and authorized entity.

[0074] Combining scene privacy level and protection level, the privacy protection strategy corresponding to each portrait is matched from the preset scene and role mapping strategy table, and the privacy protection strategy is quantified.

[0075] The identity attribute decoupling network is used to decompose the facial features of each unauthorized person into identity feature vector and attribute feature vector. Based on the identity feature vector and attribute feature vector, the quantized privacy protection level and protection strategy are feature-encoded to generate a portrait privacy processing matrix.

[0076] A facial privacy processing matrix is ​​used to perform differentiated privacy desensitization processing on the facial image area of ​​each unauthorized subject.

[0077] After privacy desensitization, each processed image will undergo reverse risk verification and sensitivity checks.

[0078] Reverse risk verification verifies whether the image after privacy desensitization can be reversed to be restored to the original identity-recognizable image. If it can, then privacy desensitization is performed again.

[0079] Sensitivity checks detect whether there is sensitive information in an image other than human figures, such as ID cards, bank cards, license plate numbers, or classified information. If so, the sensitive area is first cropped out, then the non-sensitive, non-human figures area around the sensitive area is stretched to the original sensitive area within a preset range, and finally the sensitive area is Gaussian blurred to maintain visual consistency.

[0080] Reverse risk verification and sensitivity checks are both completed before image encoding and storage, ultimately generating unauthorized photos whose original information is unrecognizable.

[0081] Each photo-taking operation generates a privacy processing log, which is then hashed and uploaded to a distributed evidence storage network for auditing.

[0082] The privacy processing log includes: photo capture time, extracted features, intent category, portrait attributes, type of processing operation performed on each portrait, and encryption key;

[0083] The encryption key is generated by combining extracted features, timestamps, device fingerprints, privacy levels, and protection levels to create an encryption key that is associated with the scenario.

[0084] After the privacy desensitization process, an encryption key is used to encrypt the original human image region data and sensitive regions extracted before desensitization. The encrypted original human image region data and sensitive regions are then embedded in the privacy desensitized image file in a steganographic or encrypted metadata manner.

[0085] When the authorizing party requests to restore the original image, the identity and permissions of the authorizing party are verified by multiple parties. After the verification is successful, the corresponding decryption key is released to restore the original portrait information before desensitization.

[0086] Example 2: A smart photo-taking service method to prevent the leakage of facial information, comprising the following steps:

[0087] Before the camera app receives the user's command to take a picture, the camera starts a real-time preview and simultaneously performs scene perception and feature extraction to extract image features. Scene characteristics and behavioral characteristics .

[0088] Image features include: color histogram, texture features, light intensity, and contrast;

[0089] Scene characteristics include: indoor signage, outdoor signage, scene category, and scene density;

[0090] Behavioral characteristics include: posture, movement sequence, gaze direction, and changes in distance from the camera.

[0091] Multi-scale detection is performed on all portraits in the preview image to locate and extract the structured data of each portrait. Subject determination is made based on focus position and screen proportion, and the bounding boxes of all portrait regions are determined. ,in , This represents the total number of human images detected.

[0092] The extracted facial region features are matched against a pre-defined authorized facial feature database to determine identity and calculate the similarity score.

[0093]

[0094] in, To match similarity, For the first Feature vector of a personal portrait The feature vector is from the authorized face feature library.

[0095] like If the image matches the preset identity threshold, the person is determined to be an authorized subject; otherwise, they are determined to be an unauthorized subject.

[0096] Construct an intent-aware classification model based on statistical analysis of users' historical photo features. ,enter The output is the category of the intention to take the photo. categorize intent Combined with the person's identity attributes, a unique character intent tag is generated. :

[0097]

[0098] Based on scene features Perform semantic recognition on the current shooting environment to determine the scene privacy level. The privacy levels are categorized as follows: 1 represents low privacy, such as open spaces like squares or scenic spots; 2 represents medium privacy, such as public places like shopping malls or restaurants; and 3 represents high privacy, such as private places like hospitals, banks, and government offices.

[0099] Then, based on the role intent tags of each unauthorized subject and the proportion of the authorized subject's image Calculate the protection level of the i-th portrait. :

[0100]

[0101] in, The weighting coefficient for the character's intention tag. As a weighting adjustment factor, .

[0102] Finally, combining and Match the corresponding privacy protection policy from the preset scenario and role mapping policy table. The strategies include: desensitizing identity features, face replacement, edge feathering, and region cropping.

[0103] Constructing an identity attribute decoupled network , the feature vector of each portrait Decomposed into identity feature vectors and attribute feature vector :

[0104]

[0105] in, It contains information that uniquely identifies an individual, such as facial features and proportions. It includes attribute information unrelated to identity, such as facial expressions, posture, lighting, and makeup.

[0106] According to protection level and protection strategies For identity feature vectors Perform feature encoding to generate a facial privacy processing matrix. ,in and The height and width of the human portrait area.

[0107] use Perform pixel-by-pixel processing on the portrait areas of unauthorized subjects:

[0108]

[0109] in, The original human portrait region image pixels, The background image pixels surrounding the portrait area.

[0110] Only identifiable information is removed, while non-identifiable attributes such as facial expressions and postures are fully preserved to ensure the visual coherence and naturalness of the image.

[0111] Before image encoding and storage, perform reverse risk verification and sensitive information checks;

[0112] Reverse risk verification includes: using a pre-trained face reconstruction model to analyze the processed human face region. Recovering Identity Feature Vectors Calculate the restored similarity:

[0113]

[0114] like If the value exceeds the preset threshold, it is determined that there is a risk of reverse recovery, and a more intensive desensitization process is performed again.

[0115] Sensitive information checks include: using a target detection model to detect whether there is sensitive information such as ID cards, bank cards, license plate numbers, or classified information in the image. If so, the sensitive area is first cropped out, and then a non-sensitive, non-human image area of ​​the same size within a preset range around the sensitive area is copied and stretched to the original sensitive area. The original sensitive area is then subjected to Gaussian blurring to maintain visual consistency with the surrounding environment.

[0116] A privacy processing log is generated after each photo-taking operation, including: photo timestamp. The extracted feature vector hash value, intent category, attribute information of all portraits, processing operation type and parameters performed on each portrait, and encryption key.

[0117] The logs are hashed to generate log hash values, which are then uploaded to a blockchain distributed evidence storage network for permanent storage, ensuring that the processing is tamper-proof and auditable.

[0118] Use encryption key The original human image area data and sensitive area data extracted before desensitization are hashed and encrypted to generate encrypted data.

[0119] Will By embedding the original data and the desensitized data into the reserved fields of the EXIF ​​metadata of the privacy-desensitized image file in a steganographic manner, the security risks caused by separate storage are avoided.

[0120] When an authorized party requests to restore the original image, it must provide identity credentials, device fingerprint, and request timestamp. The system verifies the authorized party's identity and permissions through a multi-party secure computation protocol.

[0121] After successful verification, the system generates and releases the corresponding decryption key. It extracts encrypted data from the EXIF ​​metadata of the image file and decrypts it, then merges the original portrait area and sensitive areas back into the original image location to restore the complete original image.

[0122] Example 3, as Figure 2 As shown, a smart photo-taking service system for preventing the leakage of facial information includes:

[0123] The portrait preprocessing module performs a preview of the image before shooting, detects and locates all portraits in the image in real time, divides the detected portraits into authorized subjects and unauthorized subjects, and generates a role intent label for each portrait.

[0124] The privacy encryption module, after shooting, crops and extracts the portrait area of ​​the unauthorized subject in the shooting image according to the determined protection level and protection policy on the device local, and performs corresponding privacy desensitization processing to generate a desensitized portrait area image and merge it back into the original image position;

[0125] The privacy recovery module performs multi-party security verification of the authorizing party's identity and permissions when the authorizing party requests the restoration of the original image. After successful verification, it releases the corresponding decryption key and restores the original portrait information before desensitization.

[0126] Example 4, from a hardware perspective, this application provides an embodiment of an electronic device containing all or part of a smart photography service method for preventing the leakage of facial information. The electronic device includes a service processor and a distributed memory. The service processor is connected to the memory. The distributed memory stores a service self-management program configured to store machine-readable instructions. The service processor executes the service self-management program. When the instructions are executed by the processor, a smart photography service method for preventing the leakage of facial information as described above can be implemented.

[0127] Example 5: This application also provides a computer-readable storage medium capable of implementing a smart photography service method for preventing the leakage of facial information, where the execution subject is a server or client as described in the above embodiments. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements all the contents of the smart photography service method for preventing the leakage of facial information, where the execution subject is a server or client as described in the above embodiments.

[0128] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of this application, and these all fall within the protection scope of this application.

Claims

1. A smart photography service method to prevent the leakage of facial information, characterized in that, Includes the following steps: Before shooting, the camera previews the image, detects and locates all the people in the image in real time, divides the detected people into authorized subjects and unauthorized subjects, and generates a role intent label for each person. After shooting, the device locally crops and extracts the portrait area of ​​the unauthorized subject in the captured image according to the determined protection level and protection strategy, and performs corresponding privacy desensitization processing to generate a desensitized portrait area image and merge it back into the original image position. When the authorizing party requests to restore the original image, the identity and permissions of the authorizing party are verified by multiple parties. After the verification is successful, the corresponding decryption key is released to restore the original portrait information before desensitization. 2.The intelligent photographing service method for preventing leakage of portrait information according to claim 1, wherein, The image preview refers to the camera previewing the image and perceiving the scene before the camera shutter is pressed or the shooting command is issued, and extracting image features, scene features and behavioral features. Scene awareness includes: Based on ambient lighting, scene categories are determined, and image and scene features are extracted. The user behavior context is determined based on the application type, historical photo shooting mode, and geographical location. Then, behavioral features are extracted from the user behavior context based on the principle of minimum necessity and privacy compliance policy. The extraction of behavioral features is completed locally on the device and is not associated with user identity. 3.The intelligent photographing service method for preventing leakage of portrait information according to claim 2, wherein, The generation of the role intent tag includes: Multi-scale detection is performed on all portraits in the captured image to locate and extract the structured data of each portrait. Subject determination is made based on the focus position and the proportion of the image, the bounding boxes of all portrait regions are determined, and all portrait regions are extracted and globally sorted. The system uses a pre-defined authorized facial feature database to match the extracted facial regions for identity verification, and categorizes the detected facial images into authorized subjects and unauthorized subjects. Then, based on the statistical analysis of users' historical photo-taking features, an intent-aware classification model is constructed, which combines image features, scene features, and behavioral features to determine the category of photo-taking intent; The intent category is combined with whether the portrait is an authorized subject, converted into a corresponding tag, and then bound to the portrait. 4.The intelligent photographing service method for preventing leakage of portrait information according to claim 3, wherein, The determination of the intent category is based on identifying the subject of the image based on image features, then determining the intent based on scene features and behavioral features, and finally combining the subject of the image and the intent to determine the intent category, including: If the main subject of the image is a portrait and the intention is to take a selfie, engage in social interaction, or live a lifestyle-related activity, the intention of taking the photo is classified as a selfie. If the subject of the image is not the photographer, and the intention is to take a portrait or a group photo, the category of intention to take the photo is classified as "photographing another person's subject". If the main subject of the image is a person, not the subject of the photograph, and the intention is to record scenery, architecture, or events, the photographic intention category is determined to be scene recording. If the subject of the image is a business-related photograph, and the intent is to collect documents, conduct surveys, or perform business-related tasks, and the intent is to create business-related images, then the photographic intent category is determined to be task-oriented. If the subject of the image is captured by a surveillance camera and the intention is related to recording surveillance footage, the category of the intention to take the photo is determined to be security monitoring. 5.The intelligent photographing service method for preventing leakage of portrait information according to claim 3, wherein, The privacy desensitization process includes: Based on image and scene features, semantic recognition is performed on the current shooting environment to determine the scene privacy level; The corresponding protection level is determined based on the screen proportion allocation between each unauthorized subject and authorized subject, and the role intent tag. Combining scene privacy level and protection level, the privacy protection strategy corresponding to each portrait is matched from the preset scene and role mapping strategy table, and the privacy protection strategy is quantified. The identity attribute decoupling network is used to decompose the facial features of each unauthorized person into identity feature vector and attribute feature vector. Based on the identity feature vector and attribute feature vector, the quantized privacy protection level and protection strategy are feature-encoded to generate a portrait privacy processing matrix. A facial privacy processing matrix is ​​used to perform differentiated privacy desensitization processing on the facial image area of ​​each unauthorized subject. 6.The intelligent photographing service method for preventing leakage of portrait information according to claim 5, wherein, After the privacy desensitization process, each processed image will undergo reverse risk verification and sensitivity checks. Reverse risk verification verifies whether the image after privacy desensitization can be reversed to be restored to the original identity-recognizable image. If it can, then privacy desensitization is performed again. Sensitivity check detects whether there is sensitive information in the image other than human figures. If so, the sensitive area is first cropped out, then the non-sensitive, non-human figures area around the sensitive area within a preset range is stretched to the original sensitive area, and finally the sensitive area is Gaussian blurred to maintain visual consistency. Reverse risk verification and sensitivity checks are both completed before image encoding and storage, ultimately generating unauthorized photos whose original information is unrecognizable. 7.The intelligent photographing service method for preventing portrait information leakage according to claim 6, wherein, Each photo-taking operation will generate a privacy processing log, which will be hashed and then uploaded to a distributed evidence storage network for auditing. The privacy processing log includes: photo capture time, extracted features, intent category, portrait attributes, type of processing operation performed on each portrait, and encryption key; The encryption key is generated by combining extracted features, timestamps, device fingerprints, privacy levels, and protection levels to create an encryption key that is associated with the scenario. After the privacy desensitization process, an encryption key is used to encrypt the original human image region data and sensitive regions extracted before desensitization. The encrypted original human image region data and sensitive regions are then embedded in the privacy desensitized image file in a steganographic or encrypted metadata manner.

8. A smart photo-taking service system for preventing the leakage of facial information, characterized in that, include: The portrait preprocessing module performs a preview of the image before shooting, detects and locates all portraits in the image in real time, divides the detected portraits into authorized subjects and unauthorized subjects, and generates a role intent label for each portrait. The privacy encryption module, after shooting, crops and extracts the portrait area of ​​the unauthorized subject in the shooting image according to the determined protection level and protection policy on the device local, and performs corresponding privacy desensitization processing to generate a desensitized portrait area image and merge it back into the original image position; The privacy recovery module performs multi-party security verification of the authorizing party's identity and permissions when the authorizing party requests the restoration of the original image. After successful verification, it releases the corresponding decryption key and restores the original portrait information before desensitization.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the content of the intelligent photography service method for preventing the leakage of facial information as described in claim 1.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the content of the intelligent photography service method for preventing the leakage of facial information as described in claim 1.

Images