WO2023071563A1 - 脱敏方法的可靠性验证的方法、装置、介质、设备和程序 - Google Patents

脱敏方法的可靠性验证的方法、装置、介质、设备和程序 Download PDF

Info

Publication number
WO2023071563A1
WO2023071563A1 PCT/CN2022/118401 CN2022118401W WO2023071563A1 WO 2023071563 A1 WO2023071563 A1 WO 2023071563A1 CN 2022118401 W CN2022118401 W CN 2022118401W WO 2023071563 A1 WO2023071563 A1 WO 2023071563A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
desensitized
probability value
desensitization
evaluation information
Prior art date
Application number
PCT/CN2022/118401
Other languages
English (en)
French (fr)
Inventor
徐宁
王国利
张骞
黄畅
Original Assignee
北京地平线信息技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京地平线信息技术有限公司 filed Critical 北京地平线信息技术有限公司
Publication of WO2023071563A1 publication Critical patent/WO2023071563A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning

Definitions

  • the present disclosure relates to the technical field of image processing, and in particular to a method, device, storage medium, electronic equipment and computer program for reliability verification of a desensitization method.
  • desensitization processing is required to prevent leakage of sensitive information.
  • desensitization processing (such as coding) is performed on the position of sensitive information in the image to obtain a desensitized image.
  • the desensitized image it can be restored by anti-desensitization technology (for example, by image repair or restoration) to obtain sensitive information in the original image.
  • anti-desensitization technology for example, by image repair or restoration
  • the desensitized images generated by different data desensitization processing methods perform differently in the face of anti-desensitization technology, that is, the reliability of different data desensitization processing methods is different.
  • Embodiments of the present disclosure provide a method, device, storage medium, electronic equipment, and computer program for verifying the reliability of a desensitization method.
  • a method for verifying the reliability of a desensitization method including: desensitizing the first original image through a preset desensitization method to obtain a first desensitized image; Perform image restoration processing on the first desensitized image to obtain a first image corresponding to the first desensitized image; estimate a first probability value that the first image is a real image; determine evaluation information based on the first probability value, and use the evaluation information To evaluate the reliability of preset desensitization methods.
  • a device for verifying the reliability of a desensitization method including: an image desensitization unit configured to desensitize the first original image by a preset desensitization method, The first desensitized image is obtained; the image restoration module is configured to perform image restoration processing on the first desensitized image to obtain a first image corresponding to the first desensitized image; the probability prediction unit is configured to estimate the first The image is a first probability value of the real image; the information generation unit is configured to determine evaluation information based on the first probability value, and the evaluation information is used to evaluate the reliability of the preset desensitization method.
  • a computer-readable storage medium stores a computer program, and the computer program is used to execute the verification of the reliability of the desensitization method in the above-mentioned embodiments.
  • an electronic device includes: a processor; a memory for storing instructions executable by the processor; method of verification.
  • a computer program product including computer programs/instructions, characterized in that, when the computer program/instructions are executed by a processor, the reliability verification of the desensitization method in the above-mentioned embodiments is implemented Methods.
  • the first original image is desensitized by the preset desensitization method, and the first desensitized image is obtained.
  • image and then perform restoration processing on the first desensitized image to obtain the first image after restoration processing, and estimate the first probability value that the first image is a real image, and then generate evaluation information based on the first probability value, and evaluate The information was used to evaluate the reliability of the preset desensitization method.
  • An evaluation of the reliability of image data desensitization methods is achieved.
  • FIG. 1 is a schematic diagram of a scene applicable to the present disclosure
  • Fig. 2 is the flowchart of an embodiment of the method for the reliability verification of the desensitization method of the present disclosure
  • FIG. 3 is a flow chart of generating the first desensitized image in one embodiment of the method for reliability verification of the desensitization method of the present disclosure
  • FIG. 4 is a flow chart of generating a first desensitized image in yet another embodiment of the method for reliability verification of the desensitization method of the present disclosure
  • FIG. 5 is a flow chart of another embodiment of the method for reliability verification of the desensitization method of the present disclosure
  • FIG. 6 is a flow chart of generating evaluation information in an embodiment of the method for reliability verification of the desensitization method of the present disclosure
  • FIG. 7 is a schematic structural diagram of an embodiment of a device for verifying the reliability of the desensitization method of the present disclosure
  • Fig. 8 is a schematic structural diagram of the probability prediction unit in an embodiment of the device for reliability verification of the desensitization method of the present disclosure
  • FIG. 9 is a schematic structural diagram of an image desensitization unit in an embodiment of the device for verifying the reliability of the desensitization method of the present disclosure
  • FIG. 10 is a schematic structural diagram of an image desensitization unit in another embodiment of the device for verifying the reliability of the desensitization method of the present disclosure
  • FIG. 11 is a schematic structural diagram of an information generating unit in an embodiment of a device for verifying the reliability of the desensitization method of the present disclosure
  • FIG. 12 is a schematic structural diagram of an information generation module in an embodiment of a device for verifying the reliability of the desensitization method of the present disclosure
  • Fig. 13 is a structural diagram of an electronic device provided by an embodiment of the present disclosure.
  • plural may refer to two or more than two, and “at least one” may refer to one, two or more than two.
  • Embodiments of the present disclosure may be applied to electronic devices such as terminal devices, computer systems, servers, etc., which are operable with numerous other general purpose or special purpose computing system environments or configurations.
  • Examples of well-known terminal devices, computing systems, environments and/or configurations suitable for use with electronic devices such as terminal devices, computer systems, or servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick client Computers, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, networked personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the foregoing, among others.
  • Electronic devices such as terminal devices, computer systems, servers, etc. may be described in the general context of computer system-executable instructions, such as program modules, being executed by the computer system.
  • program modules may include routines, programs, objects, components, logic, data structures, etc., that perform particular tasks or implement particular abstract data types.
  • Computer systems/servers can be implemented in a distributed cloud computing environment. In a distributed cloud computing environment, tasks can be performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computing system storage media including storage devices.
  • This disclosure can use the preset desensitization method to be evaluated to desensitize the original image to obtain the desensitized image; then restore the desensitized image to obtain the first image corresponding to the desensitized image; after that, estimate Find out the first probability value that the first image is a real image, and determine the evaluation information of the preset desensitization method based on the first probability value, where the evaluation information is used to evaluate the reliability of the preset desensitization method.
  • the method for verifying the reliability of the desensitization method of the present disclosure will be exemplarily described below in conjunction with FIG. 1 , which is a schematic diagram of a scene to which the present disclosure is applicable.
  • the electronic device 100 is the subject of execution of the reliability verification method of the desensitization method of the present disclosure, and the electronic device 100 can be, for example, a terminal computer or a server, on which a preset desensitization method to be evaluated is loaded.
  • the execution subject can desensitize the first original image 110 through a preset desensitization method, for example, the area where the sensitive information in the image is located can be coded to obtain the first desensitized image 120; after that, use the preset
  • the image restoration algorithm performs restoration processing on the first desensitized image 120, for example, it may be a deep neural network for image generation, to obtain the first image 130 corresponding to the first desensitized image 120.
  • the evaluation information 150 is generated based on the first probability value 140.
  • the evaluation information 150 can characterize the reliability of the preset desensitization method. For example, the higher the value of the first probability value 140, the closer the first image 130 is to the first original image. 110, indicating that the higher the probability that the desensitized image obtained by the preset desensitization method can obtain sensitive information through image restoration, the lower the reliability of the preset desensitization method.
  • FIG. 2 is a flowchart of an embodiment of a method for verifying the reliability of the desensitization method of the present disclosure. As shown in Figure 2, the method includes the following steps:
  • Step 200 Desensitize the first original image by using a preset desensitization method to obtain a first desensitized image.
  • the preset desensitization method characterizes the desensitization method to be evaluated.
  • the first original image represents an image without desensitization processing, for example, it may be a native image (image data in raw format) directly output by a camera sensor, or it may be preprocessed by the above-mentioned native image (for example RGB image obtained after color interpolation).
  • the first desensitized image represents an image obtained by desensitizing the region where the sensitive information in the first original image is located by using a preset desensitization method.
  • Step 210 Perform image restoration processing on the first desensitized image to obtain a first image corresponding to the first desensitized image.
  • the first image represents the image obtained after performing image restoration processing on the first desensitized image
  • the image restoration processing method may include one or more image anti-desensitization processing algorithms, the purpose of which is to reproduce Sensitivity information of a specific region (such as the first target region or the second target region) in the image after the first desensitization.
  • the execution subject may input the first desensitized image obtained in step 200 into a generator in a pre-trained conditional adversarial generation network.
  • the first desensitized image is used as a conditional label to learn the image data distribution law in the first desensitized image, and then the sensitive information in the first desensitized image is simulated according to the distribution law through random noise, and the first image is obtained .
  • Step 220 estimating a first probability value that the first image is a real image.
  • the real image represents the first original image without desensitization processing.
  • the first probability value represents the degree of similarity between the first image and the first original image. The larger the first probability value, the more similar the first image is to the first original image, and the closer the restored sensitive information in the first image is to the first image. A real sensitive information in the original image.
  • the execution subject can use the image recognition model to estimate the first probability value that the first image is a real image, which may specifically include the following steps: first extract features related to image restoration processing in the first image, such as abnormal pixel values The number, position, and pixel value of pixels, etc. Afterwards, a first probability value is estimated based on the extracted features.
  • the execution subject can respectively restore the first desensitized image through various image restoration algorithms to obtain the first image, and then respectively estimate that each first image is real The probability value of the image, and the mean or weighted mean of the plurality of probability values is determined as the first probability value.
  • Step 230 Determine evaluation information based on the first probability value.
  • the evaluation information is used to evaluate the reliability of the preset desensitization method, and may be in the form of text description, numerical value or image data presentation.
  • the execution subject may pre-establish the correspondence between the numerical range of the first probability value and the evaluation level, for example, when the first probability value is in [0.8, 1.0], it means that the first image is highly similar to the first original image, Indicates that the reliability of the preset desensitization method is extremely poor.
  • the evaluation level corresponding to this interval can be determined as "extremely poor”.
  • the execution subject can determine the corresponding evaluation level according to the interval of the first probability value obtained in step 220, and obtain the evaluation information of the preset desensitization method.
  • the executive body can set different colors for different numerical ranges, and use the colors to represent the reliability of the preset desensitization method.
  • the first original image is desensitized by a preset desensitization method to obtain the first desensitized image, and then the first desensitized image is restored. , obtain the restored first image, and estimate the first probability value that the first image is a real image, then generate evaluation information based on the first probability value, and the evaluation information is used to evaluate the reliability of the preset desensitization method, An evaluation of the reliability of image data desensitization methods is achieved.
  • the method can also obtain the first image in the following manner: input the first desensitized image into the generator in the pre-trained confrontational generation network to obtain the first desensitized The latter image corresponds to the first image.
  • the generator in the adversarial generative network can learn an image restoration processing strategy through training, so as to realize the restoration processing of the first desensitized image.
  • the method can estimate the first probability value in the following way, input the first image into the discriminator in the confrontational generation network, and obtain the confidence degree of the first image; based on the confidence degree of the first image, determine the first image is the first probability value of the real image.
  • the discriminator in the adversarial generative network can learn an image recognition strategy through training, so as to judge the probability that the first image output by the generator is a real image.
  • the adversarial generative network is trained through the following steps: first, construct a sample set, the sample set includes a first sample image marked as 0 and a second sample image marked as 1, the first The sample image is the image generated by the generator of the pre-built initial adversarial generation network, and the second sample image is an image without desensitization; then, the parameters of the generator are fixed, the discriminator is trained for the first time, and the The image is input into the discriminator of the initial adversarial generation network, and the label of the image is used as the expected output, and the discriminator of the initial adversarial generation network is trained to obtain the discriminator after the initial training; after that, the sample image pair is constructed, and the sample image pair is obtained from the first Composed of three sample images and their sample labels, where the sample label is an undesensitized image, and the third sample image is a desensitized image obtained after the sample label is desensitized by data; then, the parameters of the discriminator are fixed , perform initial
  • the termination condition can be, for example, is the preset number of iterations or the confidence value output by the discriminator.
  • the parameters of the generator are adjusted through the output of the discriminator, so that the generator can generate more realistic images, so as to improve the generation ability of the generator;
  • the parameters are adjusted so that the discriminator can recognize the image more accurately, so as to improve the discriminative ability of the discriminator.
  • the collaborative training and game between the generator and the discriminator in the adversarial generative network can be used to improve the image restoration ability of the generator and the image recognition ability of the discriminator, thereby improving the desensitization method of the present disclosure.
  • the pertinence and accuracy of the reliability verification method can be used to improve the image restoration ability of the generator and the image recognition ability of the discriminator, thereby improving the desensitization method of the present disclosure.
  • FIG. 3 is a flow chart of generating a first desensitized image in an embodiment of the method for verifying the reliability of the desensitization method of the present disclosure.
  • step 200 may further include the following steps:
  • Step 300 perform demosaic processing on the first original image, and convert the first original image into a three-channel image.
  • the first original image may be a native image.
  • the execution subject can input a native image into an ISP (Image Signal Processing, Image Signal Processor), and perform demosaic processing on the native image through a color restoration module preset in the ISP to obtain a three-channel image corresponding to the native image (for example, can be an RGB image).
  • ISP Image Signal Processing, Image Signal Processor
  • Step 310 identifying the first target area where the sensitive information in the three-channel image is located.
  • sensitive information may include privacy information, portrait information, security information, and other types of information
  • the execution subject may input the three-channel image obtained in step 300 into a pre-trained image recognition model, such as a convolutional neural network
  • the network model identifies the first target area where the sensitive information is located from the three-channel image, for example, the outline of the image area where the sensitive information is located can be marked by a detection frame.
  • Step 320 Adjust the pixel values of the first target area to obtain a first desensitized image.
  • the execution subject can set the pixel value of the first target area to 0 (that is, the values of the three colors of RGB are all 0) or other values, so that each pixel in the first target area is black, so as to realize The sensitive information is blanked to obtain the first desensitized image.
  • the execution subject may also input the image marked with the first target area into the ISP, and adjust the pixel values of the first target area in the desensitization module of the ISP to obtain the first desensitized image.
  • the execution subject may first convert the first original image into a three-channel image, and then identify the first target area where the sensitive information is located from the three-channel image, and Desensitization is performed on the first target area to obtain a first desensitized image of three channels.
  • FIG. 4 is a flow chart of generating a first desensitized image in yet another embodiment of the method for verifying the reliability of the desensitization method of the present disclosure.
  • step 200 may also adopt the following process:
  • Step 400 identifying a second target area where sensitive information in the first original image is located.
  • the first original image represents an original image without desensitization processing.
  • the execution subject may input the original image into a pre-built image recognition model to identify the second target area from the first original image, and the image recognition model represents the correspondence between the original image and the second target area.
  • Step 410 adjusting the pixel values of the second target area to obtain the first desensitized image.
  • the execution subject can directly adjust the pixel value of the original image, for example, the brightness value of each pixel point in the second target area can be adjusted to the minimum to obtain the first desensitized image, so as to realize the Escapement of sensitive information in native images.
  • the execution subject can directly perform identification and desensitization processing on the first original image, and the type of the obtained first desensitized image is an original image.
  • FIG. 5 is a flow chart of another embodiment of the method for verifying the reliability of the desensitization method of the present disclosure. As shown in Figure 5, the process includes the following steps:
  • Step 500 Desensitize the first original image by using a preset desensitization method to obtain a first desensitized image.
  • Step 510 Perform image restoration processing on the first desensitized image to obtain a first image corresponding to the first desensitized image.
  • Step 520 estimating a first probability value that the first image is a real image.
  • steps 500 to 520 correspond to the aforementioned steps 200 to 220 respectively, and will not be repeated here.
  • Step 530 Desensitize at least one second original image by using a preset desensitization method to obtain at least one second desensitized image.
  • At least one second original image is an image different from the first original image.
  • at least one second original image may include one image, or may include multiple different images.
  • Step 540 Perform image restoration processing on at least one second desensitized image to obtain second images corresponding to each of the at least one desensitized image.
  • Step 550 Determine the probability value that the second image corresponding to each of the at least one second desensitized image is a real image, and obtain at least one second probability value.
  • the process of desensitizing, restoring and estimating the first probability value of the first original image corresponds to the process of desensitizing, restoring and estimating the second probability of at least one second original image, where No longer.
  • Step 560 Determine evaluation information based on the first probability value and at least one second probability value.
  • the execution subject may first determine the mean value of the first probability value and at least one second probability value, and then determine the evaluation level according to the value range of the mean value, so as to obtain the evaluation information of the preset desensitization method.
  • the embodiment shown in Fig. 5 embodies: based on the probability values that the multiple images obtained by the preset desensitization method are real images, the evaluation information is determined .
  • the overall performance of the preset desensitization method can be more accurately described through multiple images, and thus the accuracy of reliability verification of the preset desensitization method can be improved.
  • FIG. 6 is a flow chart of generating evaluation information in an embodiment of the method for reliability verification of the desensitization method of the present disclosure. As shown in FIG. 6, in some optional implementations of this embodiment, step 560 may further include the following steps:
  • Step 600 Determine a first weight coefficient of the first desensitized image in determining evaluation information.
  • the first weight coefficient represents the importance of the first desensitized image to the evaluation result.
  • Step 610 Determine respective second weight coefficients of at least one second desensitized image in determining evaluation information.
  • the second weight coefficient represents the importance of at least one second desensitized image to the evaluation result.
  • Step 620 Weight the first probability value and at least one second probability value based on the first weight coefficient and the corresponding second weight coefficient to determine evaluation information.
  • mapping relationship between the weighted sum or weighted average value and the evaluation level can be constructed in advance, and then the executive body can determine the weighted sum or weighted average of the first probability value and at least one second probability value, and compare the value with the evaluation level Carry out mapping, determine the evaluation level of the preset desensitization method, and obtain evaluation information.
  • the importance of the first desensitized image in the evaluation result is represented by the first weight coefficient
  • the importance of the second desensitized image in the evaluation result is represented by the second weight coefficient
  • FIG. 7 is a schematic structural diagram of an embodiment of a device for verifying the reliability of the desensitization method of the present disclosure.
  • the device of this embodiment can be used to implement the corresponding method embodiment of the present disclosure.
  • the device shown in FIG. 7 includes: an image desensitization unit 710 configured to desensitize the first original image by a preset desensitization method to obtain a first desensitized image; an image restoration module 720 configured to Perform image restoration processing on the first desensitized image to obtain a first image corresponding to the first desensitized image; the probability prediction unit 730 is configured to estimate the first probability value that the first image is a real image; the information generation unit 740 , configured to determine evaluation information based on the first probability value, where the evaluation information is used to evaluate the reliability of the preset desensitization method.
  • the image desensitization unit 710 is further configured to: input the first desensitized image into the generator in the pre-trained adversarial generation network to obtain the first image corresponding to the first desensitized image.
  • Fig. 8 is a schematic structural diagram of a probability prediction unit in an embodiment of the device for verifying the reliability of the desensitization method of the present disclosure.
  • the probability prediction unit 730 further includes: a prediction module 731 configured to input the first image into the discriminator in the confrontational generation network to obtain the confidence of the first image; the determination module 732, configured to determine a probability value that the first image is a real image based on the confidence of the first image.
  • FIG. 9 is a schematic structural diagram of an image desensitization unit in an embodiment of the device for verifying the reliability of the desensitization method of the present disclosure.
  • the image desensitization unit 710 further includes: an image conversion module 711 configured to perform demosaic processing on the first original image, and convert the first original image into a three-channel image; An identification module 712 configured to identify the first target area where the sensitive information in the three-channel image is located; a first desensitization module 713 configured to adjust the pixel values of the first target area to obtain the first desensitized after image.
  • FIG. 10 is a schematic structural diagram of an image desensitization unit in another embodiment of the device for verifying the reliability of the desensitization method of the present disclosure.
  • the image desensitization unit 710 further includes: a second identification module 714 configured to identify the second target area where the sensitive information in the first original image is located; the second desensitization module 714 Module 715, configured to adjust the pixel values of the second target area to obtain the first desensitized image.
  • Fig. 11 is a schematic structural diagram of an information generation unit in an embodiment of the device for verifying the reliability of the desensitization method of the present disclosure.
  • the information generation unit 740 further includes: a third desensitization module 741 configured to determine to desensitize at least one second original image by a preset desensitization method to obtain at least A second desensitized image, at least one second original image is an image different from the first original image;
  • the image restoration module 742 is configured to perform image restoration processing on at least one second desensitized image to obtain The second image corresponding to each of the at least one desensitized image;
  • the probability prediction module 743 is configured to determine the probability value that the second image corresponding to the at least one second desensitized image is a real image, and obtain at least one second Probability value;
  • information generation module 744 configured to determine evaluation information based on the first probability value and at least one second probability value.
  • Fig. 12 is a schematic structural diagram of an information generation module in an embodiment of the device for verifying the reliability of the desensitization method of the present disclosure.
  • the information generation module 744 further includes: a first weight sub-module 7441 configured to determine the first weight coefficient of the first desensitized image in determining the evaluation information; the second weight The sub-module 7442 is configured to determine the respective second weight coefficients of at least one second desensitized image in determining the evaluation information; the weighting sub-module 7443 is configured to determine based on the first weight coefficients and the corresponding second weight coefficients The evaluation information is determined by weighting the first probability value and the at least one second probability value.
  • FIG. 13 is a structural diagram of an electronic device provided by an embodiment of the present disclosure. As shown in FIG. 13 , electronic device 1300 includes one or more processors 1310 and memory 1320 .
  • the processor 1310 may be a central processing unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device 1300 to perform desired functions.
  • CPU central processing unit
  • the processor 1310 may control other components in the electronic device 1300 to perform desired functions.
  • Memory 1320 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and/or nonvolatile memory.
  • the volatile memory for example, may include: a random access memory (RAM) and/or a cache memory (cache), and the like.
  • the non-volatile memory for example, may include: a read-only memory (ROM), a hard disk, and a flash memory.
  • One or more computer program instructions can be stored on the computer-readable storage medium, and the processor 1310 can execute the program instructions to realize the reliability verification of the desensitization method of the various embodiments of the present disclosure described above methods and other functions.
  • Various contents such as input signal, signal component, noise component, etc. may also be stored in the computer-readable storage medium.
  • the electronic device 1300 may further include: an input device 1330, an output device 1340, etc., and these components are interconnected through a bus system and/or other forms of connection mechanisms (not shown).
  • the input device 1330 may also include, for example, a keyboard, a mouse, and the like.
  • the output device 1340 can output various information to the outside.
  • the output device 1340 may include, for example, a display, a speaker, a printer, a communication network and remote output devices connected thereto, and the like.
  • the electronic device 1300 may further include any other appropriate components.
  • embodiments of the present disclosure may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the above-mentioned "exemplary method" of this specification. Steps in the method for reliability verification of the desensitization method according to various embodiments of the present disclosure described in the section.
  • the computer program product can be written in any combination of one or more programming languages for executing the program codes for the operations of the embodiments of the present disclosure, and the programming languages include object-oriented programming languages, such as Java, C++, etc. , also includes conventional procedural programming languages, such as the "C" language or similar programming languages.
  • the program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server to execute.
  • the embodiments of the present disclosure may also be a computer-readable storage medium, on which computer program instructions are stored, and the computer program instructions, when executed by a processor, cause the processor to perform the above-mentioned "Exemplary Method" section of this specification. Steps in the method for training a language model or the method for predicting the occurrence probability of a word based on a language model according to various embodiments of the present disclosure described in .
  • the computer readable storage medium may employ any combination of one or more readable media.
  • the readable medium may be a readable signal medium or a readable storage medium.
  • the readable storage medium may include, but not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media may include: an electrical connection with one or more conductors, a portable disk, a hard disk, random access memory (RAM), read only memory (ROM), computer Erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • the methods and apparatus of the present disclosure may be implemented in many ways.
  • the methods and apparatuses of the present disclosure may be implemented by software, hardware, firmware or any combination of software, hardware, and firmware.
  • the above sequence of steps for the method is for illustration only, and the steps of the method of the present disclosure are not limited to the sequence specifically described above unless specifically stated otherwise.
  • the present disclosure can also be implemented as programs recorded in recording media, the programs including machine-readable instructions for realizing the method according to the present disclosure.
  • the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.
  • each component or each step can be decomposed and/or reassembled. These decompositions and/or recombinations should be considered equivalents of the present disclosure.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

本公开实施例公开了一种脱敏方法的可靠性验证的方法、装置、存储介质、设备和计算机程序，其中，方法包括：通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像；对第一脱敏后图像进行图像复原处理，得到第一脱敏后图像对应的第一图像；估计第一图像为真实图像的第一概率值；基于第一概率值，确定评估信息，评估信息用于评价预设脱敏方法的可靠性，实现了对图像数据脱敏方法的可靠性的评估。

Description

脱敏方法的可靠性验证的方法、装置、介质、设备和程序
本公开要求在2021年11月01日提交的、申请号为202111283508.3、发明名称为“脱敏方法的可靠性验证的方法、装置、介质、设备和程序”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本公开涉及图像处理技术领域，尤其涉及一种脱敏方法的可靠性验证的方法、装置、存储介质、电子设备和计算机程序。
背景技术
实践中，对于包含敏感信息图像，需要进行脱敏处理，以防止敏感信息的泄漏，通常是对图像中的敏感信息所在的位置进行脱敏处理(例如打码)，得到脱敏后的图像。
对于脱敏后的图像，可以通过反脱敏技术(例如通过图像修复或复原)进行复原，以获取原图像中的敏感信息。不同的数据脱敏处理方法所生成的脱敏后的图像面对反脱敏技术时的表现也不同，即不同的数据脱敏处理方法的可靠性存在差异。
相关技术中尚不存在对图像数据脱敏方法的可靠性进行验证的方法。
发明内容
为了解决上述技术问题，提出了本公开。本公开的实施例提供了一种脱敏方法的可靠性验证的方法、装置、存储介质、电子设备和计算机程序。
根据本公开实施例的一个方面，提供了一种脱敏方法的可靠性验证的方法，包括：通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像；对第一脱敏后图像进行图像复原处理，得到第一脱敏后图像对应的第一图像；估计第一图像为真实图像的第一概率值；基于第一概率值，确定评估信息，评估信息用于评价预设脱敏方法的可靠性。
根据本公开实施例的又一个方面，提供了一种脱敏方法的可靠性验证的装置，包括：图像脱敏单元，被配置成通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像；图像复原模块，被配置成对第一脱敏后图像进行图像复原处理，得到第一脱敏后图像对应的第一图像；概率预测单元，被配置成估计第一图像为真实图像的第一概率值；信息生成单元，被配置成基于第一概率值，确定评估信息，评估信息用于评价预设脱敏方法的可靠性。
根据本公开实施例的又一个方面，提供了一种计算机可读存储介质，存储介质存储有计算机程序，计算机程序用于执行上述实施例中脱敏方法的可靠性验证的。
根据本公开实施例的又一个方面，提供了一种电子设备，电子设备包括：处理器；用于存储处理器可执行指令的存储器；处理器，用于执行上述实施例中脱敏方法的可靠性验证的方法。
根据本公开实施例的又一个方面，提供了一种计算机程序产品，包括计算机程序/指令，其特征在于，该计算机程序/指令被处理器执行时实现上述实施例中脱敏方法的可靠性验证的方法。
基于本公开上述实施例提供的一种脱敏方法的可靠性验证的方法、装置、存储介质以及电子设备，通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像，然后对第一脱敏后图像进行复原处理，得到经过复原处理后的第一图像，并估计出第一图像为真实图像的第一概率值，之后基于第一概率值生成评估信息，评估信息用于评价预设脱敏方法的可靠性。实现了对图像数据脱敏方法的可靠性的评估。
下面通过附图和实施例，对本公开的技术方案做进一步的详细描述。
附图说明
通过结合附图对本公开实施例进行更详细的描述，本公开的上述以及其他目的、特征以及优势将变得更加明显。附图用来提供对本公开实施例的进一步的理解，并且构成说明书的一部分，与本公开实施例一起用于解释本公开，并不构成对本公开的限制。在附图中，相同的参考标号通常代表相同部件或步骤。
图1为本公开所适用的场景示意图；
图2为本公开的脱敏方法的可靠性验证的方法的一个实施例的流程图；
图3为本公开的脱敏方法的可靠性验证的方法的一个实施例中生成第一脱敏后图像的流程图；
图4为本公开的脱敏方法的可靠性验证的方法的又一个实施例中生成第一脱敏后图像的流程图；
图5为本公开的脱敏方法的可靠性验证的方法的又一个实施例的流程图；
图6为本公开的脱敏方法的可靠性验证的方法的一个实施例中生成评估信息的流程图；
图7为本公开的脱敏方法的可靠性验证的装置的一个实施例的结构示意图；
图8为本公开的脱敏方法的可靠性验证的装置的一个实施例中概率预测单元的结构示意图；
图9为本公开的脱敏方法的可靠性验证的装置的一个实施例中图像脱敏单元的结构示意图；
图10为本公开的脱敏方法的可靠性验证的装置的又一个实施例中图像脱敏单元的结构示意图；
图11为本公开的脱敏方法的可靠性验证的装置的一个实施例中信息生成单元的结构示意图；
图12为本公开的脱敏方法的可靠性验证的装置的一个实施例中信息生成模块的结构示意图；
图13为本公开的一个实施例提供的电子设备的结构图。
具体实施方式
下面将参考附图详细地描述根据本公开的示例实施例。显然，所描述的实施例仅仅是本公开的一部分实施例，而不是本公开的全部实施例，应理解，本公开不受这里描述的示例实施例的限制。
应注意到：除非另外具体说明，否则在这些实施例中阐述的部件和步骤的相对布置、数字表达式和数值不限制本公开的范围。
本领域技术人员可以理解，本公开实施例中的“第一”、“第二”等术语仅用于区别不同步骤、设备或模块等，既不代表任何特定技术含义，也不表示它们之间的必然逻辑顺序。
还应理解，在本公开实施例中，“多个”可以指两个或者两个以上，“至少一个”可以指一个、两个或两个以上。
还应理解，对于本公开实施例中提及的任一部件、数据或结构，在没有明确限定或者在前后文给出相反启示的情况下，一般可以理解为一个或多个。
另外，本公开中术语“和/或”，仅是一种描述关联对象的关联关系，表示可以存在三种关系，如A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本公开中字符“/”，一般表示前后关联对象是一种“或”的关系。
还应理解，本公开对各个实施例的描述着重强调各个实施例之间的不同之处，其相同或相似之处可以相互参考，为了简洁，不再一一赘述。
同时，应当明白，为了便于描述，附图中所示出的各个部分的尺寸并不是按照实际的比例关系绘制的。
以下对至少一个示例性实施例的描述实际上仅仅是说明性的，决不作为对本公开及其应用或使用的任何限制。
对于相关领域普通技术人员已知的技术、方法和设备可能不作详细讨论，但在适当情况下，所述技术、方法和设备应当被视为说明书的一部分。
应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步讨论。
本公开的实施例可以应用于终端设备、计算机系统、服务器等电子设备，其可与众多其它通用或者专用计算系统环境或配置一起操作。适于与终端设备、计算机系统或者服务器等电子设备一起使用的众所周知的终端设备、计算系统、环境和/或配置的例子包括但不限于：个人计算机系统、服务器计算机系统、瘦客户机、厚客户机、手持或膝上设备、基于微处理器的系统、机顶盒、可编程消费电子产品、网络个人电脑、小型计算机系统、大型计算机系统和包括上述任何系统的分布式云计算技术环境等等。
终端设备、计算机系统、服务器等电子设备可以在由计算机系统执行的计算机系统可执行指令(诸如程序模块)的一般语境下描述。通常，程序模块可以包括例程、程序、目标程序、组件、逻辑、数据结构等等，它们执行特定的任务或者实现特定的抽象数据类型。计算机系统/服务器可以在分布式云计算环境中实施。在分布式云计算环境中，任务可以是由通过通信网络链接的远程处理设备执行的。在分布式云计算环境中，程序模块可以位于包括存储设备的本地或远程计算系统存储介质上。
示例性概述
本公开可以利用待评估的预设脱敏方法对原始图像进行脱敏处理，得到脱敏后图像；然后对脱敏后图像进行复原处理，得到脱敏后图像对应的第一图像；之后，估计出第一图像为真实图像的第一概率值，并基于第一概率值确定预设脱敏方法的评估信息，该评估信息用于评价预设脱敏方法的可靠性。下面结合图1对本公开的脱敏方法的可靠性验证的方法进行示例性说明，图1为本公开所适用的场景示意图。
在图1所示的场景中，电子设备100为本公开的脱敏方法的可靠性验证的方法的执行主体，电子设备100例如可以是终端电脑或服务器，其上装载有待评估的预设脱敏方法所对应的计算机软件或代码以及图象复原算法所对应的计算机软件或代码。执行主体可以通过预设脱敏方法对第一原始图像110进行脱敏处理，例如可以将图像中的敏感信息所在的区域进行打码处理，得到第一脱敏后图像120；之后，利用预设的图像复原算法对第一脱敏后图像120进行复原处理，例如可以是用于图像生成的深度神经网络，得到第 一脱敏后图像120对应的第一图像130。再然后，估计出第一图像130为真实图像的第一概率值140，例如可以将第一图像130输入预先训练的用于图像识别的机器学习模型或深度网络模型。最后，基于第一概率值140生成评估信息150，评估信息150可以表征预设脱敏方法的可靠性，例如，第一概率值140的数值越高，表示第一图像130越贴近第一原始图像110，说明通过预设脱敏方法得到的脱敏后图像可以通过图象复原以获取敏感信息的概率越大，则预设脱敏方法的可靠性就越低。
示例性方法
图2为本公开的脱敏方法的可靠性验证的方法的一个实施例的流程图。如图2所示，该方法包括以下步骤：
步骤200、通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像。
在本实施例中，预设脱敏方法表征待评估的脱敏方法。第一原始图像表征未经脱敏处理的图像，例如可以是由相机传感器(camera sensor)直接输出的原生图像(raw格式的图像数据)，或者，也可以是由上述原生图像经预处理(例如色彩插值)后得到的RGB图像。第一脱敏后图像表示通过预设脱敏方法对第一原始图像中的敏感信息所在的区域进行脱敏处理后得到的图像。
步骤210、对第一脱敏后图像进行图像复原处理，得到第一脱敏后图像对应的第一图像。
在本实施例中，第一图像表示对第一脱敏后图像进行图像复原处理后得到的图像，图像复原处理的方法可以包括一种或多种图像反脱敏处理算法，其目的是重现第一脱敏后图像中特定区域(例如第一目标区域或第二目标区域)的敏感信息。
在一个可选地示例中，执行主体可以将步骤200中得到的第一脱敏后图像输入预先训练的条件对抗式生成网络中的生成器。将第一脱敏后图像作为条件标签，学习第一脱敏后图像中的图像数据分布规律，然后通过随机噪声按照该分布规律模拟出第一脱敏后图像中的敏感信息，得到第一图像。
步骤220、估计第一图像为真实图像的第一概率值。
在本实施例中，真实图像表示未经脱敏处理的第一原始图像。第一概率值表征第一图像与第一原始图像的相似程度，第一概率值越大，表示第一图像与第一原始图像越相似，则第一图像中复原出的敏感信息越接近于第一原始图像中的真实敏感信息。
作为示例，执行主体可以通过图像识别模型估计第一图像为真实图像的第一概率值，具体可以包括以下步骤：首先提取第一图像中与图像复原处理相关的特征，例如可以是像素值异常的像素点的数量、位置和像素值等。之后，基于提取出的特征估计出第一概率值。
在本实施例的一些可选的实现方式中，执行主体可以通过多种图象复原算法分别对第一脱敏后图像进行复原，得到第一图像，然后分别估计出每个第一图像为真实图像的概率值，并将多个概率值的均值或加权平均值确定为第一概率值。
步骤230、基于第一概率值，确定评估信息。
在本实施例中，评估信息用于评价预设脱敏方法的可靠性，可以采用文字描述、数值或图像的数据呈现形式。
作为示例，执行主体可以预先建立第一概率值的数值区间与评价等级的对应关系，例如第一概率值处于[0.8，1.0]时，表示第一图像与第一原始图像的相似程度极高，表示预设脱敏方法的可靠性极差，此时可以将该区间对应的评价等级确定为“极差”。第一概率值处于[0，0.3]时，表示第一图像与原始图像的相似程度极低，表示预设脱敏方法的可靠性极高，此时可以将该区间对应的评价等级确定为“优秀”。之后，执行主体可以根据步骤220中得到的第一概率值所处的区间，确定对应的评价等级，即可得到预设脱敏方法的评估信息。
再例如，执行主体可以为不同的数值区间设置不同的颜色，以颜色表征预设脱敏方法的可靠性。
本实施例提供的脱敏方法的可靠性验证的方法，通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像，然后对第一脱敏后图像进行复原处理，得到经过复原处理后的第一图像，并估计出第一图像为真实图像的第一概率值，之后基于第一概率值生成评估信息，评估信息用于评价预设脱敏方法的可靠性，实现了对图像数据脱敏方法的可靠性的评估。
在本实施例的一些可选的实现方式中，该方法还可以采用如下方式得到第一图像：将第一脱敏后图像输入预先训练的对抗式生成网络中的生成器，得到第一脱敏后图像对应的第一图像。
在本实现方式中，可以通过训练使对抗式生成网络中的生成器学习图象复原处理策略，以此实现对第一脱敏后图像的复原处理。
进一步地，该方法可以采用如下方式估计出第一概率值，将第一图像输入对抗式生成网络中的判别器，得到第一图像的置信度；基于第一图像的置信度，确定第一图像为真实图像的第一概率值。
在本实现方式中，可以通过训练使得对抗式生成网络中的判别器学习图像识别策略，以此判断生成器输出的第一图像为真实图像的概率。
在本实现方式的一个具体示例中，对抗式生成网络经由如下步骤训练得到：首先，构建样本集，样本集包括标记为0的第一样本图像和标记为1的第二样本图像，第一样本图像为预先构建的初始对抗式生成网络的生成器生成的图像，第二样本图像为未经脱敏的图像；然后，固定生成器的参数，对判别器进行初次训练，将样本集中的图像输入初始对抗式生成网络的判别器，将该图像的标记作为期望输出，训练初始对抗式生成网络的判别器，得到初次训练后的判别器；之后，构建样本图像对，样本图像对由第三样本图像及其样本标签组成，其中，样本标签为未经脱敏的图像，第三样本图像为样本标签经数据脱敏处理后得到的脱敏后的图像；再然后，固定判别器的参数，对生成器进行初次训练，将样本图像对中的第三样本图像输入初始对抗式生成网络的生成器，以该样本图像对应的样本标签为期望输出，训练初始对抗式生成网络的生成器，得到初次训练后的生成器。
之后，将初次训练后的生成器和判别器串联，交替固定生成器和判别器的参数，对二者进行交替迭代，直至满足训练终止条件，得到训练后的对抗式生成网络，终止条件例如可以是预设的迭代次数或者判别器输出的置信度数值。在交替迭代的过程中，通过判别器输出的结果调整生成器的参数，使得生成器可以生成更真实的图像，以提升生成器的生成能力；基于提升后的生成器输出的图像对判别器的参数进行调整，使得判别器可以更准确地识别图像，以提升判别器的判别能力。通过生成器与判别器之间的对抗博弈，交替提升二者的性能。
在本实现方式中，可以借助对抗式生成网络中生成器与判别器之间的协同训练和博弈，提高生成器的图像复原能力以及判别器对图像的识别能力，进而提高本公开的脱敏方法的可靠性验证的方法的针对性和准确度。
接着参考图3，图3为本公开的脱敏方法的可靠性验证的方法的一个实施例中生成第一脱敏后图像的流程图。如图3所示，在本实施例的一些可选的实现方式中，步骤200可以进一步包括如下步骤：
步骤300、对第一原始图像进行去马赛克处理，将第一原始图像转换为三通道图像。
在本实现方式中，第一原始图像可以是原生图像。
作为示例，执行主体可以将原生图像输入ISP(Image Signal Processing，图像信号处理器)，通过ISP中预设的色彩复原模块对原生图像进行去马赛克处理，得到该原生图像对应的三通道图像(例如可以是RGB图像)。
步骤310、识别出三通道图像中的敏感信息所在的第一目标区域。
在本实现方式中，敏感信息可以包括隐私信息、肖像信息、安全信息等类型的信息，执行主体可以将步骤300中得到的三通道图像输入预先训练的图像识别模型中，例如可以是卷积神经网络模型，从三通道图像中识别出敏感信息所在的第一目标区域，例如可以通过检测框标记出敏感信息所在的图像区域的轮廓。
步骤320、对第一目标区域的像素值进行调整，得到第一脱敏后图像。
作为示例，执行主体可以将第一目标区域的像素值设置为0(即RGB三种色彩的值均为0)或其他数值，使得第一目标区域中的各像素点均呈现黑色，以此实现对敏感信息的消隐，得到第一脱敏后图像。
在一个示例中，执行主体还可以将标记出第一目标区域的图像输入ISP中，在ISP中的脱敏模块中实现对第一目标区域的像素值的调整，得到第一脱敏后图像。
从图3可以看出，在图3所示的实现方式中，执行主体可以首先将第一原始图像转换为三通道图像，然后从三通道图像中识别出敏感信息所在的第一目标区域，并针对第一目标区域进行脱敏处理，得到三通道的第一脱敏后图像。
接着参考图4，图4为本公开的脱敏方法的可靠性验证的方法的又一个实施例中生成第一脱敏后图像的流程图。如图4所示，在本实施例的另一些可选的实现方式中，步骤200还可以采用如下流程：
步骤400、识别出第一原始图像中的敏感信息所在的第二目标区域。
在本实现方式中，第一原始图像表示未经脱敏处理的原生图像。作为示例，执行主体可以将原生图像输入预先构建的图像识别模型中，以从第一原始图像中识别出第二目标区域，该图像识别模型表征原生图像与第二目标区域的对应关系。
步骤410、对第二目标区域的像素值进行调整，得到第一脱敏后图像。
在本实现方式中，执行主体可以直接对原生图像的像素值进行调整，例如可以将第二目标区域中每个像素点的亮度值调整至最低，得到第一脱敏后图像，以此实现对原生图像中敏感信息的消隐。
从图4可以看出，在图4所示的实现方式中，执行主体可以直接对第一原始图像进行识别及脱敏处理，得到第一脱敏后图像的类型为原生图像。
接着参考图5，图5为本公开的脱敏方法的可靠性验证的方法的又一个实施例的流程图。如图5所示，该流程包括以下步骤：
步骤500、通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像。
步骤510、对第一脱敏后图像进行图像复原处理，得到第一脱敏后图像对应的第一图像。
步骤520、估计第一图像为真实图像的第一概率值。
在实施例中，步骤500至步骤520分别与前述步骤200至步骤220相对应，此处不再赘述。
步骤530、通过预设脱敏方法分别对至少一幅第二原始图像进行脱敏，得到至少一幅第二脱敏后图像。
在本实施例中，至少一幅第二原始图像为不同于第一原始图像的图像。例如，至少一幅第二原始图像可以包括一幅图像，也可以包括多幅不同图像。
步骤540、对至少一幅第二脱敏后图像进行图像复原处理，得到至少一幅脱敏后图像各自对应的第二图像。
步骤550、确定至少一幅第二脱敏后图像各自对应的第二图像为真实图像的概率值，得到至少一个第二概率值。
在本实施中，第一原始图像的脱敏、复原以及估计第一概率值的处理过程与至少一幅第二原始图像的脱敏、复原以及估计第二概率之的处理过程相对应，此处不再赘述。
步骤560、基于第一概率值和至少一个第二概率值，确定评估信息。
作为示例，执行主体可以首先确定第一概率值和至少一个第二概率值的均值，然后根据该均值所处的数值区间确定评价等级，得到预设脱敏方法的评估信息。
从图5可以看出，与图2所示的实施例相比，图5所示的实施例体现了：基于预设脱敏方法得到的多张图像分别为真实图像的概率值，确定评估信息。通过多张图像可以更准确地刻画预设脱敏方法的整体性能，因而可以提高对预设脱敏方法的可靠性验证的准确度。
接着参考图6，图6为本公开的脱敏方法的可靠性验证的方法的一个实施例中生成评估信息的流程图。如图6所示，在本实施例的一些可选的实现方式中，步骤560可以进一步包括如下步骤：
步骤600、确定第一脱敏后图像在确定评估信息中的第一权重系数。
在本实现方式中，第一权重系数表征第一脱敏后图像对评估结果的重要程度。
步骤610、确定至少一幅第二脱敏后图像在确定评估信息中各自的第二权重系数。
在本实现方式中，第二权重系数表征至少一幅第二脱敏后图像对评估结果的重要程度。
步骤620、基于第一权重系数和各自对应的第二权重系数对第一概率值和至少一个第二概率值进行加权，确定评估信息。
作为示例，可以预先构建加权和或者加权平均的数值与评价等级的映射关系，然后执行主体可以确定第一概率值和至少一个第二概率值的加权和或者加权平均，并将该数值与评价等级进行映射，确定预设脱敏方法的评价等级，得到评估信息。
在本实现方式中，通过第一权重系数表征第一脱敏后图像在评估结果中的重要程度，通过第二权重系数表征第二脱敏后图像在评估结果中的重要程度，并基于第一概率值和至少一个第二概率值加权结果确定预设脱敏方法的评估信息，可以更准确地评价预设脱敏方法的可靠性。
示例性装置
图7为本公开的脱敏方法的可靠性验证的装置一个实施例的结构示意图。该实施例的装置可用于实现本公开相应的方法实施例。如图7所示的装置包括：图像脱敏单元710，被配置成通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像；图像复原模块720，被配置成对第一脱敏后图像进行图像复原处理，得到第一脱敏后图像对应的第一图像；概率预测单元730，被配置成估计第一图像为真实图像的第一概率值；信息生成单元740，被配置成基于第一概率值，确定评估信息，评估信息用于评价预设脱敏方法的可靠性。
在本实施例中，图像脱敏单元710被进一步配置成：将第一脱敏后图像输入预先训练的对抗式生成网络中的生成器，得到第一脱敏后图像对应的第一图像。
图8为本公开的脱敏方法的可靠性验证的装置的一个实施例中概率预测单元的结构示意图。如图8所示，在本实施例中，概率预测单元730进一步包括：预测模块731，被配置成将第一图像输入对抗式生成网络中的判别器，得到第一图像的置信度；确定模块732，被配置成基于第一图像的置信度确定第一图像为真实图像的概率值。
图9为本公开的脱敏方法的可靠性验证的装置的一个实施例中图像脱敏单元的结构示意图。如图9所示，在本实施例中，图像脱敏单元710进一步包括：图像转换模块711，被配置成对第一原始图像进行去马赛克处理，将第一原始图像转换为三通道图像；第一识别模块712，被配置成识别出三通道图像中的敏感信息所在的第一目标区域；第一脱敏模块713，被配置成对第一目标区域的像素值进行调整，得到第一脱敏后图像。
图10为本公开的脱敏方法的可靠性验证的装置的又一个实施例中图像脱敏单元的结构示意图。如图10所示，在本实施例中，图像脱敏单元710进一步包括：第二识别模块714，被配置成识别出第一原始图像中的敏感信息所在的第二目标区域；第二脱敏模块715，被配置成对第二目标区域的像素值进行调整，得到第一脱敏后图像。
图11为本公开的脱敏方法的可靠性验证的装置的一个实施例中信息生成单元的结构示意图。如图11所示，在本实施例中，信息生成单元740进一步包括：第三脱敏模块741，被配置成确定通过预设脱敏方法分别对至少一幅第二原始图像进行脱敏得到至少一幅第二脱敏后图像，至少一幅第二原始图像为不同于第一原始图像的图像；图像复原模块742，被配置成对至少一幅第二脱敏后图像进行图像复原处理，得到至少一幅脱敏后图像各自对应的第二图像；概率预测模块743，被配置成确定至少一幅第二脱敏后图像各自对应的第二图像为真实图像的概率值，得到至少一个第二概率值；信息生成模块744，被配置成基于第一概率值和至少一个第二概率值，确定评估信息。
图12为本公开的脱敏方法的可靠性验证的装置的一个实施例中信息生成模块的结构示意图。如图12所示，在本实施例中，信息生成模块744进一步包括：第一权重子模块7441，被配置成确定第一脱敏后图像在确定评估信息中的第一权重系数；第二权重子模块7442，被配置成确定至少一幅第二脱敏后图像在确定评估信息中各自的第二权重系数；加权子模块7443，被配置成基于第一权重系数和各自对应的第二权重系数对第一概率值和至少一个第二概率值进行加权，确定评估信息。
示例性电子设备
下面参考图13来描述根据本公开实施例的电子设备。图13为本公开的一个实施例提供的电子设备的结构图。如图13所示，电子设备1300包括一个或多个处理器1310和存储器1320。
处理器1310可以是中央处理单元(CPU)或者具有数据处理能力和/或指令执行能力的其他形式的处理单元，并且可以控制电子设备1300中的其他组件以执行期望的功能。
存储器1320可以包括一个或多个计算机程序产品，所述计算机程序产品可以包括各种形式的计算机可读存储介质，例如易失性存储器和/或非易失性存储器。所述易失性存储器，例如，可以包括：随机存取存储器(RAM)和/或高速缓冲存储器(cache)等。所述非易失性存储器，例如，可以包括：只读存储器(ROM)、硬盘以及闪存等。在所述计算机可读存储介质上可以存储一个或多个计算机程序指令，处理器1310可以运行所述程序指令，以实现上文所述的本公开的各个实施例的脱敏方法的可靠性验证的方法等功能。在所述计算机可读存储介质中还可以存储诸如输入信号、信号分量、噪声分量等各种内容。
在一个示例中，电子设备1300还可以包括：输入装置1330以及输出装置1340等，这些组件通过总线系统和/或其他形式的连接机构(未示出)互连。此外，该输入装置1330还可以包括例如键盘、鼠标等等。该输出装置1340可以向外部输出各种信息。该输出装置1340可以包括例如显示器、扬声器、打印机、以及通信网络及其所连接的远程输出设备等等。
当然，为了简化，图13中仅示出了该电子设备1300中与本公开有关的组件中的一些，省略了诸如总线、输入/输出接口等等的组件。除此之外，根据具体应用情况，电子设备1300还可以包括任何其他适当的组件。
示例性计算机程序产品和计算机可读存储介质
除了上述方法和设备以外，本公开的实施例还可以是计算机程序产品，其包括计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行本说明书上述“示例性方法”部分中描述的根据本公开各种实施例的脱敏方法的可靠性验证的方法中的步骤。
所述计算机程序产品可以以一种或多种程序设计语言的任意组合来编写用于执行本公开实施例操作的程序代码，所述程序设计语言包括面向对象的程序设计语言，诸如Java、C++等，还包括常规的过程式程序设计语言，诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算设备上执行、部分地在用户设备上执行、作为一个独立的软件包执行、部分在用户计算设备上部分在远程计算设备上执行、或者完全在远程计算设备或服务器上执行。
此外，本公开的实施例还可以是计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行本说明书上述“示例性方法”部分中描述的根据本公开各种实施例的语言模型的训练方法或者基于语言模型预测词的出现概率的方法中的步骤。
所述计算机可读存储介质可以采用一个或多个可读介质的任意组合。可读介质可以是可读信号介质或者可读存储介质。可读存储介质例如可以包括但不限于电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。可读存储介质的更具体的例子(非穷举的列举)可以包括：具有一个或者多个导线的电连接、便携式盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。
以上结合具体实施例描述了本公开的基本原理，但是，需要指出的是，在本公开中提及的优点、优势、效果等仅是示例而非限制，不能认为这些优点、优势以及效果等是本公开的各个实施例必须具备的。另外，上述公开的具体细节仅是为了示例的作用和便于理解的作用，而非限制，上述细节并不限制本公开为必须 采用上述具体的细节来实现。
本说明书中各个实施例均采用递进的方式描述，每个实施例重点说明的都是与其它实施例的不同之处，各个实施例之间相同或相似的部分相互参见即可。对于系统实施例而言，由于其与方法实施例基本对应，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
本公开中涉及的器件、装置、设备、系统的方框图仅作为例示性的例子并且不意图要求或暗示必须按照方框图示出的方式进行连接、布置、配置。如本领域技术人员将认识到的，可以按任意方式连接、布置、配置这些器件、装置、设备以及系统。诸如“包括”、“包含、“具有”等等的词语是开放性词汇，指“包括但不限于”，且可与其互换使用。这里所使用的词汇“或”和“和”指词汇“和/或”，且可与其互换使用，除非上下文明确指示不是如此。这里所使用的词汇“诸如”指词组“诸如但不限于”，且可与其互换使用。
可能以许多方式来实现本公开的方法和装置。例如，可通过软件、硬件、固件或者软件、硬件、固件的任何组合来实现本公开的方法和装置。用于所述方法的步骤的上述顺序仅是为了进行说明，本公开的方法的步骤不限于以上具体描述的顺序，除非以其它方式特别说明。此外，在一些实施例中，还可将本公开实施为记录在记录介质中的程序，这些程序包括用于实现根据本公开的方法的机器可读指令。因而，本公开还覆盖存储用于执行根据本公开的方法的程序的记录介质。
还需要指出的是，在本公开的装置、设备和方法中，各部件或各步骤是可以分解和/或重新组合的。这些分解和/或重新组合应视为本公开的等效方案。
提供所公开的方面的以上描述，以使本领域的任何技术人员能够做出或者使用本公开。对这些方面的各种修改等对于本领域技术人员而言，是非常显而易见的，并且在此定义的一般原理可以应用于其他方面，而不脱离本公开的范围。因此，本公开不意图被限制到在此示出的方面，而是按照与在此公开的原理和新颖的特征一致的最宽范围。
为了例示和描述的目的已经给出了以上描述。此外，此描述不意图将本公开的实施例限制到在此公开的形式中。尽管以上已经讨论了多个示例方面以及实施例，但是本领域技术人员将认识到其某些变型、修改、改变、添加和子组合。

Claims (11)

  1. 一种脱敏方法的可靠性验证的方法，包括：
    通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像；
    对所述第一脱敏后图像进行图像复原处理，得到所述第一脱敏后图像对应的第一图像；
    估计所述第一图像为真实图像的第一概率值；
    基于所述第一概率值，确定评估信息，所述评估信息用于评价所述预设脱敏方法的可靠性。
  2. 根据权利要求1所述的方法，其中，所述对所述第一脱敏后图像进行图像复原处理，得到所述第一脱敏后图像对应的第一图像，包括：
    将所述第一脱敏后图像输入预先训练的对抗式生成网络中的生成器，得到所述第一脱敏后图像对应的第一图像。
  3. 根据权利要求2所述的方法，其中，所述估计所述第一图像为真实图像的概率值，包括：
    将所述第一图像输入所述对抗式生成网络中的判别器，得到所述第一图像的置信度；
    基于所述第一图像的置信度确定所述第一图像为真实图像的第一概率值。
  4. 根据权利要求1所述的方法，其中，所述通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像，包括：
    对所述第一原始图像进行去马赛克处理，将所述第一原始图像转换为三通道图像；
    识别出所述三通道图像中的敏感信息所在的第一目标区域；
    对所述第一目标区域的像素值进行调整，得到所述第一脱敏后图像。
  5. 根据权利要求1所述的方法，其中，所述通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像，包括：
    识别出所述第一原始图像中的敏感信息所在的第二目标区域；
    对所述第二目标区域的像素值进行调整，得到所述第一脱敏后图像。
  6. 根据权利要求1至5之一所述的方法，其中，所述基于所述第一概率值，确定评估信息，包括：
    通过所述预设脱敏方法分别对至少一幅第二原始图像进行脱敏，得到至少一幅第二脱敏后图像，所述至少一幅第二原始图像为不同于所述第一原始图像的图像；
    对所述至少一幅第二脱敏后图像进行图像复原处理，得到所述至少一幅脱敏后图像各自对应的第二图像；
    确定所述至少一幅第二脱敏后图像各自对应的所述第二图像为真实图像的概率值，得到至少一个第二概率值；
    基于所述第一概率值和所述至少一个第二概率值，确定所述评估信息。
  7. 根据权利要求6所述的方法，其中，所述基于所述第一概率值和所述至少一个第二概率值，确定评估信息，包括：
    确定所述第一脱敏后图像在确定所述评估信息中的第一权重系数；
    确定所述至少一幅第二脱敏后图像在确定所述评估信息中各自的第二权重系数；
    基于所述第一权重系数和所述各自对应的所述第二权重系数对所述第一概率值和所述至少一个第二概率值进行加权，确定所述评估信息。
  8. 一种脱敏方法的可靠性验证的装置，包括：
    图像脱敏单元，被配置成通过预设脱敏方法对第一原始图像进行脱敏处理，得到第一脱敏后图像；
    图像复原模块，被配置成对所述第一脱敏后图像进行图像复原处理，得到所述第一脱敏后图像对应的第一图像；
    概率预测单元，被配置成估计所述第一图像为真实图像的第一概率值；
    信息生成单元，被配置成基于所述第一概率值，确定评估信息，所述评估信息用于评价所述预设脱敏方法的可靠性。
  9. 一种计算机可读存储介质，所述存储介质存储有计算机程序，所述计算机程序用于执行上述权利要求1-7任一项所述的方法。
  10. 一种电子设备，所述电子设备包括：
    处理器；
    用于存储所述处理器可执行指令的存储器；
    所述处理器，用于执行上述权利要求1-7任一项所述的方法。
  11. 一种计算机程序产品，包括计算机程序/指令，其中，该计算机程序/指令被处理器执行时实 现上述权利要求1-7任一所述的方法。
PCT/CN2022/118401 2021-11-01 2022-09-13 脱敏方法的可靠性验证的方法、装置、介质、设备和程序 WO2023071563A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111283508.3 2021-11-01
CN202111283508.3A CN113989156A (zh) 2021-11-01 2021-11-01 脱敏方法的可靠性验证的方法、装置、介质、设备和程序

Publications (1)

Publication Number Publication Date
WO2023071563A1 true WO2023071563A1 (zh) 2023-05-04

Family

ID=79745385

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/118401 WO2023071563A1 (zh) 2021-11-01 2022-09-13 脱敏方法的可靠性验证的方法、装置、介质、设备和程序

Country Status (2)

Country Link
CN (1) CN113989156A (zh)
WO (1) WO2023071563A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117892349A (zh) * 2024-01-16 2024-04-16 唐山启奥科技股份有限公司 图像脱敏测试方法、设备及计算机可读存储介质

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113989156A (zh) * 2021-11-01 2022-01-28 北京地平线信息技术有限公司 脱敏方法的可靠性验证的方法、装置、介质、设备和程序
CN117455751B (zh) * 2023-12-22 2024-03-26 新华三网络信息安全软件有限公司 路段图像的处理系统及方法
CN117892358B (zh) * 2024-03-18 2024-07-05 北方健康医疗大数据科技有限公司 一种受限数据脱敏方法验证方法及系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110135193A (zh) * 2019-05-15 2019-08-16 广东工业大学 一种数据脱敏方法、装置、设备及计算机可读存储介质
CN111737750A (zh) * 2020-06-30 2020-10-02 绿盟科技集团股份有限公司 一种数据处理方法、装置、电子设备及存储介质
WO2020207079A1 (zh) * 2019-04-10 2020-10-15 创新先进技术有限公司 基于图像识别的脱敏处理方法以及装置
CN112200132A (zh) * 2020-10-28 2021-01-08 支付宝(杭州)信息技术有限公司 一种基于隐私保护的数据处理方法、装置及设备
CN113379036A (zh) * 2021-06-18 2021-09-10 西安石油大学 一种基于上下文编码器的油气图像脱敏方法
CN113989156A (zh) * 2021-11-01 2022-01-28 北京地平线信息技术有限公司 脱敏方法的可靠性验证的方法、装置、介质、设备和程序

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020207079A1 (zh) * 2019-04-10 2020-10-15 创新先进技术有限公司 基于图像识别的脱敏处理方法以及装置
CN110135193A (zh) * 2019-05-15 2019-08-16 广东工业大学 一种数据脱敏方法、装置、设备及计算机可读存储介质
CN111737750A (zh) * 2020-06-30 2020-10-02 绿盟科技集团股份有限公司 一种数据处理方法、装置、电子设备及存储介质
CN112200132A (zh) * 2020-10-28 2021-01-08 支付宝(杭州)信息技术有限公司 一种基于隐私保护的数据处理方法、装置及设备
CN113379036A (zh) * 2021-06-18 2021-09-10 西安石油大学 一种基于上下文编码器的油气图像脱敏方法
CN113989156A (zh) * 2021-11-01 2022-01-28 北京地平线信息技术有限公司 脱敏方法的可靠性验证的方法、装置、介质、设备和程序

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117892349A (zh) * 2024-01-16 2024-04-16 唐山启奥科技股份有限公司 图像脱敏测试方法、设备及计算机可读存储介质

Also Published As

Publication number Publication date
CN113989156A (zh) 2022-01-28

Similar Documents

Publication Publication Date Title
WO2023071563A1 (zh) 脱敏方法的可靠性验证的方法、装置、介质、设备和程序
US20230041233A1 (en) Image recognition method and apparatus, computing device, and computer-readable storage medium
WO2019024808A1 (zh) 语义分割模型的训练方法和装置、电子设备、存储介质
CN108229287B (zh) 图像识别方法和装置、电子设备和计算机存储介质
US20230154167A1 (en) Source-free cross domain detection method with strong data augmentation and self-trained mean teacher modeling
US11315037B2 (en) Systems and methods for generating and applying a secure statistical classifier
CN108229313B (zh) 人脸识别方法和装置、电子设备和计算机程序及存储介质
WO2019029459A1 (zh) 用于识别面部年龄的方法、装置和电子设备
JP2023549070A (ja) 意味特徴の学習を介したＵｎｓｅｅｎドメインからの顔認識
US11637858B2 (en) Detecting malware with deep generative models
US20220189083A1 (en) Training method for character generation model, character generation method, apparatus, and medium
CN112200380A (zh) 优化风险检测模型的方法及装置
CN114187483A (zh) 生成对抗样本的方法、检测器的训练方法及相关设备
US10885593B2 (en) Hybrid classification system
CN114612688B (zh) 对抗样本生成方法、模型训练方法、处理方法及电子设备
US20220180043A1 (en) Training method for character generation model, character generation method, apparatus and storage medium
CN114821066A (zh) 模型训练方法、装置、电子设备及计算机可读存储介质
WO2021000407A1 (zh) 字符验证方法、装置、计算机设备及存储介质
CN113222480A (zh) 对抗样本生成模型的训练方法及装置
CN116740421A (zh) 基于无监督公共子空间学习模型的遥感图像分类方法和装置
KR101854804B1 (ko) 텍스트와 연관된 개체명을 결정하기 위한 사용자 인증 서비스 제공 및 학습 데이터 구축 방법
CN112199978B (zh) 视频物体检测方法和装置、存储介质和电子设备
US20230274133A1 (en) Learning method, learning apparatus and program
CN114186039A (zh) 一种视觉问答方法、装置及电子设备
CN114742170A (zh) 对抗样本生成方法、模型训练方法、图像识别方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22885450

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE