WO2023051377A1 - 图像数据的脱敏方法和装置 - Google Patents

图像数据的脱敏方法和装置 Download PDF

Info

Publication number
WO2023051377A1
WO2023051377A1 PCT/CN2022/120560 CN2022120560W WO2023051377A1 WO 2023051377 A1 WO2023051377 A1 WO 2023051377A1 CN 2022120560 W CN2022120560 W CN 2022120560W WO 2023051377 A1 WO2023051377 A1 WO 2023051377A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
channel
target
data
original
Prior art date
Application number
PCT/CN2022/120560
Other languages
English (en)
French (fr)
Inventor
左翠莲
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 WO2023051377A1 publication Critical patent/WO2023051377A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60Protecting data
    • G06F21/62Protecting access to data via a platform, e.g. using keys or access control rules
    • G06F21/6218Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
    • G06F21/6245Protecting personal data, e.g. for financial or medical purposes
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/084Backpropagation, e.g. using gradient descent

Definitions

  • the present disclosure relates to computer vision technology, in particular to a method and device for desensitizing image data.
  • Embodiments of the present disclosure provide a method and device for desensitizing image data.
  • a desensitization method for image data including:
  • a desensitization device for image data including:
  • An image channel determination module configured to determine the target image channel that needs to be identified in the original image
  • a single-channel image acquisition module configured to restore the original data corresponding to the target image channel to obtain a single-channel image corresponding to the color component of the target image channel;
  • An image recognition module configured to recognize a target object from the single-channel image
  • a sensitive area determining module configured to determine a sensitive area in the native image based on the target object
  • a data desensitization module configured to perform data desensitization processing on the original data in the sensitive area.
  • a computer-readable storage medium stores a computer program, and the computer program is used to execute the image data desensitization method described in the first aspect above.
  • an electronic device includes:
  • the processor is configured to read the executable instructions from the memory, and execute the instructions to implement the image data desensitization method described in the first aspect above.
  • the image data desensitization method and device provided by the above-mentioned embodiments of the present disclosure, first determine the target image channel in the original image that requires target recognition, and then generate a single-channel image for the original data of the target image channel, and then generate a single-channel image on the single-channel image Identify the target object, and then determine the sensitive area in the original image based on the identified target object, and finally perform data desensitization on the original data in the sensitive area to ensure that the private image data cannot be saved and processed in the subsequent digital image signal processing flow Leakage, the privacy data is protected at the source of image data generation, which improves data security.
  • FIG. 1 is a schematic flow diagram of a desensitization method for image data according to an embodiment of the present disclosure
  • Fig. 2 is a schematic diagram from an original image to determining a sensitive area in the original image in an example of the present disclosure
  • FIG. 3 is a schematic flow diagram of step S1 in an embodiment of the present disclosure.
  • Fig. 4 is a schematic flow chart of step S1 in another embodiment of the present disclosure.
  • Fig. 5 is a schematic diagram of performing image interpolation processing on a native image to obtain a single-channel image in another example of the present disclosure
  • Fig. 6 is a schematic diagram of determining a sensitive area in an original image based on a single-channel image in an example corresponding to Fig. 5;
  • FIG. 7 is a structural block diagram of an image data desensitization device according to an embodiment of the present disclosure.
  • FIG. 8 is a structural block diagram of an image channel determination module 100 in an embodiment of the present disclosure.
  • FIG. 9 is a structural block diagram of an image channel determination module 100 in another embodiment of the present disclosure.
  • Fig. 10 is a structural diagram of an electronic device provided by an exemplary 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 may operate 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, 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, etc.
  • FIG. 1 is a schematic flowchart of a method for desensitizing image data according to an embodiment of the present disclosure. This embodiment can be applied to electronic equipment, as shown in Figure 1, including the following steps:
  • S1 Determine the target image channel that needs to be recognized in the original image.
  • the native image is collected by an image sensor.
  • the original image may be a raw image of the driver collected by the image sensor in the car and not restored to the image during assisted driving or automatic driving of the vehicle, or a raw image of the driver collected by the image sensor in the car, The raw image of the passenger without image restoration.
  • the vehicle control system or a terminal (such as a mobile phone or server) that can be connected to control the vehicle can determine the target image channel in the original image that needs to be recognized.
  • Fig. 2 is a schematic diagram from an original image to determining a sensitive area in the original image in an example of the present disclosure.
  • the resolution of the original image is 8x8, and the resolution of the single-channel image obtained after channel separation can be 4x4 (for example, the resolution of the R channel image is 4x4), in fact the original image
  • the resolution of the image sensor is determined by the resolution of the image sensor.
  • the specific size of the mask image for example, the original image can be an image with a resolution of 1920x1280, an image with a resolution of 1024x768, or an image of other sizes.
  • the target image channel can be one of the R channel, G channel and B channel, depending on user settings, or the target image channel can be determined according to the image quality of the R channel image, G channel image and B channel image.
  • S2 Perform image restoration on the original data corresponding to the target image channel to obtain a single-channel image corresponding to the color components of the target image channel.
  • the target image channel is the R channel
  • image restoration is performed on the original data corresponding to the R channel to obtain a single-channel image whose color component is R, that is, the R channel image.
  • the target object in the single-channel image is determined through image recognition technology.
  • the target object depends on the behavior recognition task.
  • the driver's side face area and hand area in the R channel image can be used as the target object.
  • the position of the target object in the single-channel image includes the middle 4-frame image area of the single-channel image. Then, based on the correspondence between the single-channel image and the original image, the sensitive area in the original image can be determined.
  • the sensitive area in the original image is the middle 16-frame image area of the original image.
  • S5 Perform data desensitization processing on the original data in the sensitive area.
  • the manner of data desensitization processing may include adjusting pixel values of pixel points in the sensitive area, or performing image blurring processing on the sensitive area, and so on.
  • first determine the target image channel that needs to be recognized in the original image then generate a single-channel image for the original data of the target image channel, and then identify the target object on the single-channel image, and then based on the identified target object Determine the sensitive areas in the original image, and finally perform data desensitization on the original data in the sensitive area, which can ensure that the private image data cannot be saved and leaked in the subsequent digital image signal processing process, and the source of image data is generated. Data is protected and data security is improved.
  • step S1 includes:
  • S1-A-1 Based on the array distribution information of the image sensor, determine multiple single-channel images corresponding to the original image.
  • the array distribution information includes the setting positions of each channel, that is, including the setting positions of the R channel, the G channel and the B channel. Based on the array distribution information of the original image, the R channel image, the G channel image and the B channel image corresponding to the original image can be obtained.
  • an R-channel image, a G-channel image, and a B-channel image corresponding to the original image can be obtained.
  • the resolution of the R-channel image is 4x4, and the resolution of the native image is 8x8.
  • the image size of the R-channel image is smaller than that of the native image.
  • S1-A-2 Determine a target image channel based on multiple single-channel images.
  • one image channel can be selected as the target image channel, for example, the R channel is selected as the target image channel.
  • the advantage of performing target recognition directly based on the single-channel image data of the original image is that the full-resolution image is interpolated without an interpolation step, thereby greatly reducing the complexity of algorithm implementation.
  • step S1 includes:
  • S1-B-1 Perform image interpolation processing on the original image, and determine multiple single-channel images corresponding to the original image. Wherein, the image size of the multiple single-channel images is the same as that of the original image.
  • Fig. 5 is a schematic diagram of performing image interpolation processing on an original image to obtain a single-channel image in another example of the present disclosure.
  • the resolution of the original image is 8x8, and the resolution of multiple single-channel images after image interpolation processing is also 8x8.
  • the resolution of the R channel image is 8x8, the resolution of the G channel image is 8x8, the resolution of the B channel image is 8x8, and the resolution of the native image is also 8x8.
  • the image size of the R-channel image, G-channel image, and B-channel image is equal to the image size of the original image.
  • the 8x8 resolution shown in Figure 5 is only used as a schematic illustration.
  • the original image and the single-channel image after image interpolation processing can use an image with a resolution of 1920x1280, an image with a resolution of 1024x768, or other size of the image.
  • image interpolation processing methods include nearest neighbor interpolation processing, bilinear interpolation processing, bicubic interpolation processing, adaptive image interpolation processing based on image edge gradient information, and the like.
  • the nearest neighbor interpolation process is also called zero-order interpolation, which makes the gray value of the transformed pixel equal to the gray value of the nearest input pixel.
  • Bilinear interpolation processing is to perform a linear interpolation in two directions, for example, first perform a linear interpolation in the horizontal direction, then perform a linear interpolation in the vertical direction, and finally obtain a certain position through two linear interpolations pixel value.
  • the bicubic interpolation process can be obtained by weighting and averaging sixteen sampling points in the 4x4 neighborhood of the pixel to be interpolated, and two polynomial interpolation cubic functions need to be used, one for each direction.
  • the adaptive image interpolation method based on image edge gradient information is to consider the weight information related to the edge gradient according to the gradient of the image in addition to the weight information related to the distance when performing upsampling interpolation.
  • the gradient value of the image is small; perpendicular to the edge direction of the image, the gradient value of the image is larger, that is, the pixel point interpolation weight along the edge direction is larger, and the pixel point interpolation weight value perpendicular to the edge direction is smaller, which can ensure the edge of the image Interpolation of position performs better.
  • S1-B-2 Determine a target image channel based on multiple single-channel images.
  • the target image channel may be one of R channel, G channel and B channel, which is determined according to user settings, or determined according to the image quality of the R channel image, G channel image and B channel image.
  • Fig. 6 is a schematic diagram of determining a sensitive area in an original image based on a single-channel image in an example corresponding to Fig. 5 .
  • the position of the target object in the single-channel image includes the middle 16-grid image area of the single-channel image, and the sensitive area in the native image is also the original There are 16 image areas in the middle of the image, and the positions of the two image areas correspond to each other.
  • the full resolution image of the three channels of RGB is obtained by interpolation of the original image.
  • the resolution of the single-channel image obtained by the image interpolation process as shown in Figure 5 is 8x8
  • the resolution of the single-channel image obtained by channel separation as shown in Figure 2 is 4x4
  • the single-channel image obtained by image interpolation has a higher resolution than the single-channel image obtained by channel separation of the original image
  • step S3 specifically includes: recognizing the target object from the single-channel image by using a pre-trained recognition model.
  • the recognition model is trained in the following manner.
  • Downsampling is performed from a sample single-channel image through a fully convolutional network based on the inception model.
  • upsampling is performed after downsampling based on the initial model to restore the same image size as the sample single-channel image.
  • the sample single-channel image is predicted pixel by pixel, and the parameters of the initial model are updated by backpropagation based on the difference between the prediction result and the classification label of the sample single-channel image until the stop iteration condition is satisfied, and the final recognition model.
  • the single-channel image that needs to be recognized is used as the input of the recognition model, and the target object in the single-channel image can be recognized through the recognition model.
  • the target object in the single-channel image can be quickly and accurately identified, which is convenient for subsequent steps to determine the sensitive area in the original image and perform data desensitization on the original data in the sensitive area deal with.
  • step S5 includes: setting the pixel points in the sensitive area as target pixel values.
  • the difference between the target pixel value and the pixel boundary value is within a preset difference range, for example, the preset difference range may be [0, 5].
  • the target pixel value can be 0, 1, 2, 3 at this time , 4, 251, 252, 253, 254 or 255.
  • step S5 includes: performing image blur processing on the sensitive area, for example, using Gaussian blur processing.
  • user privacy can be effectively protected by performing image blurring processing on sensitive areas in the original image.
  • the preset convolution kernel is used to blur the sensitive area of the original image.
  • the size of the preset convolution kernel and the weight distribution of the convolution kernel can be determined based on the classification type of the target object corresponding to the sensitive area. If the target object The confidentiality type logo of the target object (such as the driver's eye area in the original image) needs to be highly confidential, so the resolution of the convolution kernel can be set larger, and the weight of the convolution kernel needs to be set to have a higher blur Intensity, for example, the size of the preset convolution kernel can be 21x21, so as to ensure that the target in the sensitive area will not be restored later; if the security type of the target indicates the target (such as the driver's forehead area in the original image) If the density level is low, the resolution of the convolution kernel can be set smaller, and the weight of the convolution kernel needs to be set to have a relatively weak fuzzy strength. For example, the preset convolution kernel size can be 5x5, and the smaller The size of the convolution kernel performs image
  • user privacy can be effectively protected by performing image blurring processing on the sensitive area of the original image through a preset convolution kernel.
  • Any image data desensitization method provided in the embodiments of the present disclosure may be executed by any appropriate device with data processing capabilities, including but not limited to: terminal devices, servers, and the like.
  • any image data desensitization method provided in the embodiments of the present disclosure may be executed by a processor, for example, the processor executes any image data desensitization mentioned in the embodiments of the present disclosure by calling the corresponding instructions stored in the memory method. I won't go into details below.
  • Fig. 7 is a structural block diagram of an image data desensitization device according to an embodiment of the disclosure.
  • the image data desensitization device of the embodiment of the present disclosure includes: an image channel determination module 100 , a single-channel image acquisition module 200 , an image recognition module 300 , a sensitive area determination module 400 and a data desensitization module 500 .
  • the image channel determination module 100 is used to determine the target image channel that needs to be recognized in the original image; the single-channel image acquisition module 200 is used to restore the original data corresponding to the target image channel to obtain the target image channel
  • the sensitive module 500 is used to desensitize the original data in the sensitive area.
  • Fig. 8 is a structural block diagram of the image channel determination module 100 in an embodiment of the present disclosure. As shown in Figure 8, the image channel determination module 100 includes:
  • the first determining unit 101 is configured to determine a plurality of single-channel images corresponding to the original image based on the array distribution information of the image sensor, wherein the image sizes of the plurality of single-channel images are all smaller than those of the original image size;
  • the second determining unit 102 is configured to determine the target image channel based on the multiple single-channel images.
  • Fig. 9 is a structural block diagram of an image channel determination module 100 in another embodiment of the present disclosure. As shown in Figure 9, the image channel determination module 100 includes:
  • the third determining unit 103 is configured to perform image interpolation processing on the original image, and determine a plurality of single-channel images corresponding to the original image, wherein the image sizes of the plurality of single-channel images are all the same as the original image of the same image size;
  • the fourth determining unit 104 is configured to determine the target image channel based on the multiple single-channel images.
  • the image recognition module 300 recognizes the target object from the single-channel image through a pre-trained recognition model.
  • the data desensitization module 500 is configured to set the pixel points in the sensitive area as target pixel values, wherein the difference between the target pixel value and the pixel boundary value is within a preset within the difference range.
  • the data desensitization module 500 is used to perform image blurring on the sensitive area.
  • the data desensitization module 500 is configured to perform image blurring processing on the sensitive area of the original image through a preset convolution kernel.
  • image data desensitization device in the embodiment of the present disclosure is similar to the specific implementation of the image data desensitization method in the embodiment of the present disclosure.
  • image data desensitization method Redundant, do not repeat them.
  • the electronic device includes one or more processors 110 and memory 120 .
  • Processor 110 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 to perform desired functions.
  • CPU central processing unit
  • Processor 110 may control other components in the electronic device to perform desired functions.
  • Memory 120 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 may include, for example, random access memory (RAM) and/or cache memory (cache).
  • the non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, and the like.
  • One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 110 may execute the program instructions to implement the image data desensitization method and / or other desired functionality.
  • Various contents such as input signals, signal components, noise components, etc. may also be stored in the computer-readable storage medium.
  • the electronic device may further include: an input device 130 and an output device 140, and these components are interconnected through a bus system and/or other forms of connection mechanisms (not shown).
  • the input device 130 may be, for example, a keyboard, a mouse, and the like.
  • the output device 140 may include, for example, a display, a speaker, a printer, a communication network and its connected remote output devices, and the like.
  • the electronic device may also include any other suitable components according to specific applications.
  • 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.
  • the steps in the image data desensitization method according to various embodiments of the present disclosure are described in the section.
  • the computer program product can be written in any combination of one or more programming languages to execute the program codes for performing 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.
  • 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.
  • 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 include: electrical connection with one or more conductors, portable disk, hard disk, random access memory (RAM), read only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
  • each component or each step can be decomposed and/or reassembled. These decompositions and/or recombinations should be considered equivalents of the present disclosure.

Abstract

本公开实施例公开了一种图像数据的脱敏方法和装置,其中,该脱敏方法包括:确定原生图像中需要进行目标识别的目标图像通道;对所述目标图像通道对应的原生数据进行图像复原,得到所述目标图像通道的颜色分量对应的单通道图像;从所述单通道图像上识别目标对象;基于所述目标对象确定所述原生图像中的敏感区域;对所述敏感区域中的原生数据进行数据脱敏处理。本公开实施例可以保证隐私图像数据在后续的数字图像信号处理流程中无法被保存和泄露,在图像数据产生的源头对隐私数据进行了保护,提升了数据安全性。

Description

图像数据的脱敏方法和装置
本公开要求在2021年9月30日提交的、申请号为202111161187.X、发明名称为“图像数据的脱敏方法和装置”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
本公开涉及计算机视觉技术,尤其是一种图像数据的脱敏方法和装置。
背景技术
随着智能识别技术的发展,数据隐私问题逐渐引起广泛的关注。如何保证采集图像中隐私数据的安全性,是一个亟待解决的问题。
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图像数据的脱敏方法和装置。
根据本公开实施例的第一方面,提供了一种图像数据的脱敏方法,包括:
确定原生图像中需要进行目标识别的目标图像通道;
对所述目标图像通道对应的原生数据进行图像复原,得到所述目标图像通道的颜色分量对应的单通道图像;
从所述单通道图像上识别目标对象;
基于所述目标对象确定所述原生图像中的敏感区域;
对所述敏感区域中的原生数据进行数据脱敏处理。
根据本公开实施例的第二方面,提供了一种图像数据的脱敏装置,包括:
图像通道确定模块,用于确定原生图像中需要进行目标识别的目标图像通道;
单通道图像获取模块,用于对所述目标图像通道对应的原生数据进行图像复原,得到所述目标图像通道的颜色分量对应的单通道图像;
图像识别模块,用于从所述单通道图像上识别目标对象;
敏感区域确定模块,用于基于所述目标对象确定所述原生图像中的敏感区域;
数据脱敏模块,用于对所述敏感区域中的原生数据进行数据脱敏处理。
根据本公开实施例的第三方面,提供了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第一方面所述的图像数据的脱敏方法。
根据本公开实施例的第四方面,提供了一种电子设备,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,用于从所述存储器中读取所述可执行指令,并执行所述指令以实现上述第一方面所述的图像数据的脱敏方法。
基于本公开上述实施例提供的图像数据的脱敏方法和装置,首先确定原生图像中需要进行目标识别的目标图像通道,然后针对目标图像通道的原生数据生成单通道图像,接着在单通道图像上识别目标对象,进而基于识别出的目标对象确定原生图像中的敏感区域,最后对敏感区域中的原生数据进行数据脱敏处理,保证隐私图像数据在后续的数字图像信号处理流程中无法被保存和泄露,在图像数据产生的源头对隐私数 据进行了保护,提升了数据安全性。
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、特征和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是本公开实施例的图像数据的脱敏方法的流程示意图;
图2是本公开一个示例中从原生图像到确定原生图像中敏感区域的示意图;
图3是本公开一个实施例中步骤S1的流程示意图;
图4是本公开另一个实施例中步骤S1的流程示意图;
图5是本公开另一个示例中对原生图像进行图像插值处理得到单通道图像的示意图;
图6是与图5对应示例中基于单通道图像确局部定原生图像中敏感区域的示意图;
图7是本公开实施例的图像数据的脱敏装置的结构框图;
图8是本公开一个实施例中图像通道确定模块100的结构框图;
图9是本公开另一个实施例中图像通道确定模块100的结构框图;
图10是本公开一示例性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”可以指两个或两个以上,“至少一个”可以指一个、两个或两个以上。
本公开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、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。
示例性方法
图1是本公开实施例的图像数据的脱敏方法的流程示意图。本实施例可应用在电子设备上,如图1所示,包括如下步骤:
S1:确定原生图像中需要进行目标识别的目标图像通道。
在本公开实施例中,原生图像是通过图像传感器采集的。示例性地,原生图像可以是在车辆辅助驾驶或自动驾驶时,通过车内的图像传感器采集到的、且未经图像复原的驾驶员的raw图像,或通过车内的图像传感器采集到的、且未经图像复原的乘客的raw图像。
在获取到原生图像之后,可以由本车控制系统或可以连接控制本车的终端(例如手机或服务器),确定原生图像中需要进行目标识别的目标图像通道。
图2是本公开一个示例中从原生图像到确定原生图像中敏感区域的示意图。如图2所示,在本示例中,原生图像的分辨率为8x8,通过通道分离之后得到的单通道图像的分辨率可以为4x4(例如R通道图像的分辨率为4x4),实际上原生图像的分辨率由图像传感器的分辨率来确定,图2所示的8x8的分辨率和4x4的分辨率仅作为示意性说明,其不能形成对本公开的限制,本公开实施例并不限定原生图像和掩码图像的具体尺寸,例如原生图像可以采用1920x1280分辨率的图像,1024x768分辨率的图像,或者其他尺寸的图像。
目标图像通道可以是R通道、G通道和B通道中的其中之一,具体根据用户设置而定,或者根据R通道图像、G通道图像和B通道图像的图像质量确定目标图像通道。
需要说明的是,以下实施例将以本车控制系统为例进行图像数据的脱敏,但本领域技术人员可以知晓,采用可以连接控制本车的终端同样可以进行图像数据的脱敏。
S2:对目标图像通道对应的原生数据进行图像复原,得到目标图像通道的颜色分量对应的单通道图像。例如当目标图像通道为R通道时,则对R通道对应的原生数据进行图像复原,得到颜色分量为R的单通道图像,即R通道图像。
S3:从单通道图像上识别目标对象。
具体地,通过图像识别技术,确定单通道图像中的目标对象。其中,目标对象根据行为识别任务而定。示例性地,当原生图像为车内驾驶员的raw图像时,且行为识别任务为驾驶员接打电话识别时,可以将R通道图像中驾驶员侧脸区域和手部区域作为目标对象。
S4:基于目标对象确定原生图像中的敏感区域。
具体地,首先确定目标对象在单通道图像中的位置,例如在图2所示的示例中,目标对象在单通道图像中的位置包括单通道图像的中间4格图像区域。接着基于单通道图像与原生图像之间的对应关系,可以确定原生图像中的敏感区域,例在如图2所示的示例中,原生图像中的敏感区域为原生图像的中间16格图像区域。
S5:对敏感区域中的原生数据进行数据脱敏处理。其中,数据脱敏处理的方式可以包括调整敏感区域内像素点的像素值,或者对敏感区域进行图像模糊处理,等等。
在本实施例中,首先确定原生图像中需要进行目标识别的目标图像通道,然后针对目标图像通道的原生数据生成单通道图像,接着在单通道图像上识别目标对象,进而基于识别出的目标对象确定原生图像中的敏感区域,最后对敏感区域中的原生数据进行数据脱敏处理,可以保证隐私图像数据在后续的数字图像信号处理流程中无法被保存和泄露,在图像数据产生的源头对隐私数据进行了保护,提升了数据安全性。
图3是本公开一个实施例中步骤S1的流程示意图。如图3所示,步骤S1包括:
S1-A-1:基于图像传感器的阵列分布信息,确定与原生图像对应的多个单通道图像。该阵列分布信息包括各个通道的设置位置,即包括了R通道、G通道和B通道的设置位置。基于原生图像的阵列分布信息,可以得到原生图像对应的R通道图像、G通道图像和B通道图像。
请再次参考图2中的原生图像,原生图像中体现了图像传感器的阵列分布信息。基于该阵列分布信息,可以得到原生图像对应的R通道图像、G通道图像和B通道图像。例如R通道图像的分辨率为4x4,原生图像的分辨率为8x8,本示例中R通道图像的图像尺寸小于原生图像的图像尺寸。
S1-A-2:基于多个单通道图像,确定目标图像通道。
具体地,基于系统设置,或者基于R通道图像、G通道图像和B通道图像的图像质量,可以选 择一个图像通道作为目标图像通道,例如选择R通道作为目标图像通道。
在本实施例中,直接基于原生图像的单通道图像数据进行目标识别优点是无需经过插值步骤将全分辨率图像插值出来,从而大大降低算法实现上的复杂度。
图4是本公开另一个实施例中步骤S1的流程示意图。如图4所示,步骤S1包括:
S1-B-1:对原生图像进行图像插值处理,确定与原生图像对应的多个单通道图像。其中,多个单通道图像的图像尺寸均与原生图像的图像尺寸相同。
图5是本公开另一个示例中对原生图像进行图像插值处理得到单通道图像的示意图。如图5所示,在本示例中,原生图像的分辨率为8x8,经过图像插值处理后的多个单通道图像的分辨率也为8x8。例如在本示例中,R通道图像的分辨率为8x8,G通道图像的分辨率为8x8,B通道图像的分辨率为8x8,原生图像的分辨率也为8x8。本示例中R通道图像、G通道图像和B通道图像的图像尺寸均等于原生图像的图像尺寸。
需要说明的是,图5所示的8x8的分辨率仅作为示意性说明,实际上原生图像和进行图像插值处理后的单通道图像可以采用1920x1280分辨率的图像,1024x768分辨率的图像,或者其他尺寸的图像。
在本公开实施例中,图像插值处理的方式包括最近邻插值处理、双线性插值处理、双三次插值处理和基于图像边缘梯度信息的自适应图像插值处理,等等。其中,最近邻插值处理也称作零阶插值,令变换后像素的灰度值等于距它最近的输入像素的灰度值。双线性插值处理是在两个方向分别进行一次线性插值,例如先在水平方向上进行一次线性插值,然后在竖直方向上再进行一次线性插值,通过两次线性插值最终可以得到某个位置的像素值。双三次插值处理可以通过对待插值像素点的4x4邻域内的十六个采样点进行加权平均得到,需要使用两个多项式插值三次函数,每个方向使用一个。基于图像边缘梯度信息的自适应图像插值方式是在进行上采样插值时,除了考虑和距离相关的权重信息,同时需根据图像的梯度,考虑和边缘梯度相关的权重信息,沿着图像边缘方向,图像的梯度值较小;垂直于图像边缘方向,图像的梯度值较大,即沿着边缘方向的像素点插值权重更大,垂直于边缘方向的像素点插值权重值更小,可保证图像边缘位置的插值表现更好。
S1-B-2:基于多个单通道图像,确定目标图像通道。
具体地,目标图像通道可以是R通道、G通道和B通道中的其中之一,具体根据用户设置而定,或者根据R通道图像、G通道图像和B通道图像的图像质量而定。
图6是与图5对应示例中基于单通道图像确定原生图像中敏感区域的示意图。图6所示,当单通道图像的图像尺寸与原生图像的图像尺寸相同时,目标对象在单通道图像中的位置包括单通道图像的中间16格图像区域,原生图像中的敏感区域也为原生图像的中间16格图像区域,两者的图像区域位置相对应。
在本实施例中,利用原生图像插值得到RGB三个通道的完整分辨率图像,例如针对分辨率为8x8的原生图像,通过如图5所示的图像插值处理得到单通道图像的分辨率为8x8,通过如图2所示通道分离得到单通道图像的分辨率为4x4,因此通过图像插值处理得到的单通道图像相对于对原生图像进行通道分离得到的单通道图像而言,分辨率更高,包含的信息更加全面,更有利于后续步骤进行敏感信息的识别。
在本公开的一个实施例中,步骤S3具体包括:通过预训练的识别模型从单通道图像上识别目标对象。
在本实施例中,通过以下方式训练识别模型。
基于初始模型从样本单通道图像通过全卷积网络进行下采样。
接着,基于初始模型在下采样之后进行上采样恢复到样本单通道图像相同的图像尺寸。
然后,基于初始模型对该样本单通道图像逐像素进行预测,基于预测结果和样本单通道图像的分类标签之间的差异进行反向传播更新初始模型的参数,直至满足停止迭代条件后,得到最终的识别模型。
以需要进行目标识别单通道图像作为识别模型的输入,通过识别模型可以识别出单通道图像中的目标对象。
在本实施例中,通过预训练的识别模型,可以快速、准确地识别出单通道图像中的目标对象,便于后续步骤确定原生图像中的敏感区域和对敏感区域中的原生数据进行数据脱敏处理。
在本公开的一个实施例中,步骤S5包括:对敏感区域内的像素点设置为目标像素值。其中,目标像素值与像素边界值之间的差值在预设差值范围内,例如预设差值范围可以取[0,5]。
在本公开的一个示例中,当原生图像中的像素取值范围为0至255时,预设差值范围取[0,5],则此时目标像素值可以是0、1、2、3、4、251、252、253、254或255。
在本实施例中,通过设定原生图像中的目标敏感图像区域内的像素点设置为目标像素值,可以防止真实的用户隐私数据被反向恢复,避免在通过将原生图像进行图像复原之后形成RGB真彩图像之后,用户的隐私数据也被复原,从而有效保护用户隐私。
在本公开的另一个实施例中,步骤S5包括:对敏感区域进行图像模糊处理,例如采用高斯模糊处理。
在本实施例中,通过对原生图像中的敏感区域进行图像模糊处理,可以有效保护用户隐私。
进一步地,通过预设卷积核对原生图像的敏感区域进行模糊处理,预设卷积核的尺寸大小与卷积核的权重分布可以基于敏感区域对应的目标物的密级类型来确定,若目标物的密级类型标识表示目标物(例如原生图像中驾驶员的眼部区域)需要高度保密,则卷积核的分辨率可以设置的较大,同时卷积核的权重需设置的具有更高的模糊强度,例如,预设卷积核的尺寸可以是21x21,从而确保敏感区域的目标物在后续避免被复原;若目标物的密级类型标识表示目标物(例如原生图像中驾驶员的额头区域)的密级较低,则卷积核的分辨率可以设置的较小,同时卷积核的权重需设置的具有相对较弱的模糊强度,例如,预设卷积核的尺寸可以是5x5,通过较小的卷积核的尺寸进行图像模糊,可以降低图像模糊处理的复杂度,同时大大提高图像模糊处理的效率。
在本实施例中,通过对原生图像的敏感区域通过预设卷积核进行图像模糊处理,可以有效保护用户隐私。
本公开实施例提供的任一种图像数据的脱敏方法可以由任意适当的具有数据处理能力的设备执行,包括但不限于:终端设备和服务器等。或者,本公开实施例提供的任一种图像数据的脱敏方法可以由处理器执行,如处理器通过调用存储器存储的相应指令来执行本公开实施例提及的任一种图像数据的脱敏方法。下文不再赘述。
示例性装置
图7是本公开实施例的图像数据的脱敏装置的结构框图。如图7所示,本公开实施例的图像数据的脱敏装置,包括:图像通道确定模块100、单通道图像获取模块200、图像识别模块300、敏感区域确定模块400和数据脱敏模块500。
其中,图像通道确定模块100用于确定原生图像中需要进行目标识别的目标图像通道;单通道图像获取模块200用于对所述目标图像通道对应的原生数据进行图像复原,得到所述目标图像通道的颜色分量对应的单通道图像;图像识别模块300用于从所述单通道图像上识别目标对象;敏感区 域确定模块400用于基于所述目标对象确定所述原生图像中的敏感区域;数据脱敏模块500用于对所述敏感区域中的原生数据进行数据脱敏处理。
图8是本公开一个实施例中图像通道确定模块100的结构框图。如图8所示,图像通道确定模块100包括:
第一确定单元101,用于基于图像传感器的阵列分布信息,确定与所述原生图像对应的多个单通道图像,其中,所述多个单通道图像的图像尺寸均小于所述原生图像的图像尺寸;
第二确定单元102,用于基于所述多个单通道图像,确定所述目标图像通道。
图9是本公开另一个实施例中图像通道确定模块100的结构框图。如图9所示,图像通道确定模块100包括:
第三确定单元103,用于对所述原生图像进行图像插值处理,确定与所述原生图像对应的多个单通道图像,其中,所述多个单通道图像的图像尺寸均与所述原生图像的图像尺寸相同;
第四确定单元104,用于基于所述多个单通道图像,确定所述目标图像通道。
在本公开的一个实施例中,图像识别模块300通过预训练的识别模型从所述单通道图像上识别目标对象。
在本公开的一个实施例中,数据脱敏模块500用于对所述敏感区域内的像素点设置为目标像素值,其中,所述目标像素值与像素边界值之间的差值在预设差值范围内。
在本公开的一个实施例中,数据脱敏模块500用于对所述敏感区域进行图像模糊处理。
在本公开的一个实施例中,数据脱敏模块500用于通过预设卷积核对所述原生图像的敏感区域进行图像模糊处理。
需要说明的是,本公开实施例的图像数据的脱敏装置的具体实施方式与本公开实施例的图像数据的脱敏方法的具体实施方式类似,具体参见图像数据的脱敏方法部分,为了减少冗余,不作赘述。
示例性电子设备
下面,参考图10来描述根据本公开实施例的电子设备。如图10所示,电子设备包括一个或多个处理器110和存储器120。
处理器110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中的其他组件以执行期望的功能。
存储器120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、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110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图像数据的脱敏方法以及/或者其他期望的功能。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,电子设备还可以包括:输入装置130和输出装置140,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。输入装置130可以例如键盘、鼠标等等。输出装置140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10中仅示出了该电子设备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图像数据的脱敏方法中的步骤。
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图像数据的脱敏方法中的步骤。
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(非穷举的列表)包括:具有一个或多个导线的电连接、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、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、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。
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。尽管以上已经讨论了多个示例方面和实施例,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 (10)

  1. 一种图像数据的脱敏方法,包括:
    确定原生图像中需要进行目标识别的目标图像通道;
    对所述目标图像通道对应的原生数据进行图像复原,得到所述目标图像通道的颜色分量对应的单通道图像;
    从所述单通道图像上识别目标对象;
    基于所述目标对象确定所述原生图像中的敏感区域;
    对所述敏感区域中的原生数据进行数据脱敏处理。
  2. 根据权利要求1所述的图像数据的脱敏方法,其中,所述确定原生图像中需要进行目标识别的目标图像通道,包括:
    基于图像传感器的阵列分布信息,确定与所述原生图像对应的多个单通道图像,其中,所述多个单通道图像的图像尺寸均小于所述原生图像的图像尺寸;
    基于所述多个单通道图像,确定所述目标图像通道。
  3. 根据权利要求1所述的图像数据的脱敏方法,其中,所述确定原生图像中需要进行目标识别的目标图像通道,包括:
    对所述原生图像进行图像插值处理,确定与所述原生图像对应的多个单通道图像,其中,所述多个单通道图像的图像尺寸均与所述原生图像的图像尺寸相同;
    基于所述多个单通道图像,确定所述目标图像通道。
  4. 根据权利要求1所述的图像数据的脱敏方法,其中,所述从所述单通道图像上识别目标对象,包括:
    通过预训练的识别模型从所述单通道图像上识别目标对象。
  5. 根据权利要求1所述的图像数据的脱敏方法,其中,所述对所述敏感区域中的原生数据进行数据脱敏处理,包括:
    对所述敏感区域内的像素点设置为目标像素值,其中,所述目标像素值与像素边界值之间的差值在预设差值范围内。
  6. 根据权利要求1所述的图像数据的脱敏方法,其中,所述对所述敏感区域中的原生数据进行数据脱敏处理,包括:
    对所述敏感区域进行图像模糊处理。
  7. 根据权利要求1所述的图像数据的脱敏方法,其中,所述对所述敏感区域进行图像模糊处理,包括:
    通过预设卷积核对所述原生图像的敏感区域进行图像模糊处理。
  8. 一种图像数据的脱敏装置,包括:
    图像通道确定模块,用于确定原生图像中需要进行目标识别的目标图像通道;
    单通道图像获取模块,用于对所述目标图像通道对应的原生数据进行图像复原,得到所述目标图像通道的颜色分量对应的单通道图像;
    图像识别模块,用于从所述单通道图像上识别目标对象;
    敏感区域确定模块,用于基于所述目标对象确定所述原生图像中的敏感区域;
    数据脱敏模块,用于对所述敏感区域中的原生数据进行数据脱敏处理。
  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任一所述的图像数据的脱敏方法。
  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,用于从所述存储器中读取所述可执行指令,并执行所述指令以实现上述权利要求1-7任一所述的图像数据的脱敏方法。
PCT/CN2022/120560 2021-09-30 2022-09-22 图像数据的脱敏方法和装置 WO2023051377A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111161187.X 2021-09-30
CN202111161187.XA CN113837970B (zh) 2021-09-30 2021-09-30 图像数据的脱敏方法和装置

Publications (1)

Publication Number Publication Date
WO2023051377A1 true WO2023051377A1 (zh) 2023-04-06

Family

ID=78967803

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/120560 WO2023051377A1 (zh) 2021-09-30 2022-09-22 图像数据的脱敏方法和装置

Country Status (2)

Country Link
CN (1) CN113837970B (zh)
WO (1) WO2023051377A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117496276A (zh) * 2023-12-29 2024-02-02 广州锟元方青医疗科技有限公司 肺癌细胞形态学分析、识别方法及计算机可读存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113837970B (zh) * 2021-09-30 2024-04-26 北京地平线信息技术有限公司 图像数据的脱敏方法和装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111310734A (zh) * 2020-03-19 2020-06-19 支付宝(杭州)信息技术有限公司 保护用户隐私的人脸识别方法和装置
CN111402146A (zh) * 2020-02-21 2020-07-10 华为技术有限公司 图像处理方法以及图像处理装置
CN111652352A (zh) * 2020-05-13 2020-09-11 北京航天自动控制研究所 一种针对迁移学习的神经网络模型输入通道整合方法
CN112633230A (zh) * 2020-12-30 2021-04-09 深圳云天励飞技术股份有限公司 一种人脸加密方法、装置、电子设备及存储介质
CN113838070A (zh) * 2021-09-28 2021-12-24 北京地平线信息技术有限公司 数据脱敏方法和装置
CN113837970A (zh) * 2021-09-30 2021-12-24 北京地平线信息技术有限公司 图像数据的脱敏方法和装置

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111008561B (zh) * 2019-10-31 2023-07-21 重庆小雨点小额贷款有限公司 一种牲畜的数量确定方法、终端及计算机存储介质
CN111814678A (zh) * 2020-07-08 2020-10-23 江苏三恒科技股份有限公司 一种基于视频监控的运输皮带中煤流识别方法及系统
CN111899755A (zh) * 2020-08-11 2020-11-06 华院数据技术(上海)有限公司 一种说话人语音分离方法及相关设备
CN111783803B (zh) * 2020-08-14 2022-06-28 支付宝(杭州)信息技术有限公司 实现隐私保护的图像处理方法及装置
CN112115811A (zh) * 2020-08-31 2020-12-22 支付宝(杭州)信息技术有限公司 基于隐私保护的图像处理方法、装置和电子设备
CN111814194B (zh) * 2020-09-04 2020-12-25 支付宝(杭州)信息技术有限公司 基于隐私保护的图像处理方法、装置和电子设备
CN111783146B (zh) * 2020-09-04 2021-02-12 支付宝(杭州)信息技术有限公司 基于隐私保护的图像处理方法、装置和电子设备
CN112200134B (zh) * 2020-10-28 2023-05-30 支付宝(杭州)信息技术有限公司 基于用户隐私保护的图像处理方法及装置
CN112381104A (zh) * 2020-11-16 2021-02-19 腾讯科技(深圳)有限公司 一种图像识别方法、装置、计算机设备及存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111402146A (zh) * 2020-02-21 2020-07-10 华为技术有限公司 图像处理方法以及图像处理装置
CN111310734A (zh) * 2020-03-19 2020-06-19 支付宝(杭州)信息技术有限公司 保护用户隐私的人脸识别方法和装置
CN111652352A (zh) * 2020-05-13 2020-09-11 北京航天自动控制研究所 一种针对迁移学习的神经网络模型输入通道整合方法
CN112633230A (zh) * 2020-12-30 2021-04-09 深圳云天励飞技术股份有限公司 一种人脸加密方法、装置、电子设备及存储介质
CN113838070A (zh) * 2021-09-28 2021-12-24 北京地平线信息技术有限公司 数据脱敏方法和装置
CN113837970A (zh) * 2021-09-30 2021-12-24 北京地平线信息技术有限公司 图像数据的脱敏方法和装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117496276A (zh) * 2023-12-29 2024-02-02 广州锟元方青医疗科技有限公司 肺癌细胞形态学分析、识别方法及计算机可读存储介质
CN117496276B (zh) * 2023-12-29 2024-04-19 广州锟元方青医疗科技有限公司 肺癌细胞形态学分析、识别方法及计算机可读存储介质

Also Published As

Publication number Publication date
CN113837970B (zh) 2024-04-26
CN113837970A (zh) 2021-12-24

Similar Documents

Publication Publication Date Title
WO2023051377A1 (zh) 图像数据的脱敏方法和装置
WO2023051143A1 (zh) 数据脱敏方法和装置
JP2020509488A (ja) 物体検出方法、ニューラルネットワークの訓練方法、装置および電子機器
EP4071707A1 (en) Method and apparatus for correcting face distortion, electronic device, and storage medium
US11361587B2 (en) Age recognition method, storage medium and electronic device
JP6731529B1 (ja) 単一画素攻撃サンプルの生成方法、装置、設備及び記憶媒体
KR20220011207A (ko) 이미지 처리 방법 및 장치, 전자 기기 및 저장 매체
KR102287407B1 (ko) 이미지 생성을 위한 학습 장치 및 방법과 이미지 생성 장치 및 방법
KR20210012012A (ko) 물체 추적 방법들 및 장치들, 전자 디바이스들 및 저장 매체
CN112602088A (zh) 提高弱光图像的质量的方法、系统和计算机可读介质
CN113989156A (zh) 脱敏方法的可靠性验证的方法、装置、介质、设备和程序
JP2022544635A (ja) 危険運転行動認識方法、装置、電子機器および記憶媒体
CN111145202B (zh) 模型生成方法、图像处理方法、装置、设备及存储介质
CN113327193A (zh) 图像处理方法、装置、电子设备和介质
CN111062922B (zh) 一种翻拍图像的判别方法、系统及电子设备
WO2024045421A1 (zh) 图像保护方法及相关设备
WO2023231435A1 (zh) 视觉感知方法、装置、存储介质和电子设备
CN114626090A (zh) 图像数据的处理方法、装置和车辆
KR20220155882A (ko) 뉴럴 네트워크를 이용하는 데이터 처리 방법 및 장치
CN114764839A (zh) 动态视频生成方法、装置、可读存储介质及终端设备
CN113139527A (zh) 视频隐私保护方法、装置、设备及存储介质
CN113887379A (zh) 唇语识别方法和装置
EP4064215A2 (en) Method and apparatus for face anti-spoofing
US10699381B2 (en) Text enhancement using a binary image generated with a grid-based grayscale-conversion filter
CN111626283B (zh) 文字提取方法、装置和电子设备

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: 22874763

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2022874763

Country of ref document: EP

Effective date: 20240430