WO2020067615A1 - Procédé de commande d'un dispositif d'anonymisation vidéo permettant d'améliorer les performances d'anonymisation, et dispositif associé - Google Patents

Procédé de commande d'un dispositif d'anonymisation vidéo permettant d'améliorer les performances d'anonymisation, et dispositif associé Download PDF

Info

Publication number
WO2020067615A1
WO2020067615A1 PCT/KR2019/002197 KR2019002197W WO2020067615A1 WO 2020067615 A1 WO2020067615 A1 WO 2020067615A1 KR 2019002197 W KR2019002197 W KR 2019002197W WO 2020067615 A1 WO2020067615 A1 WO 2020067615A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
subject
video
unit
recognition
Prior art date
Application number
PCT/KR2019/002197
Other languages
English (en)
Korean (ko)
Inventor
유상원
이용재
렌종젱
Original Assignee
주식회사 이고비드
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 주식회사 이고비드 filed Critical 주식회사 이고비드
Publication of WO2020067615A1 publication Critical patent/WO2020067615A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
    • G06F21/6218—Protecting 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
    • G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
    • G06F21/6254—Protecting personal data, e.g. for financial or medical purposes by anonymising data, e.g. decorrelating personal data from the owner's identification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/60—Editing figures and text; Combining figures or text
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/431—Generation of visual interfaces for content selection or interaction; Content or additional data rendering
    • H04N21/4318—Generation of visual interfaces for content selection or interaction; Content or additional data rendering by altering the content in the rendering process, e.g. blanking, blurring or masking an image region

Definitions

  • the present invention relates to an apparatus and method for anonymizing an image, and more specifically, to an image anonymizing apparatus capable of producing improved performance and results, and a method for further improving anonymization performance by controlling the apparatus. will be.
  • Anonymization of video refers to the process of performing a predetermined graphic processing so that other people cannot easily recognize information that can identify an individual in a video or still image, and the processed video is called anonymized video or anonymized video. It is called.
  • the present invention has been devised in view of the above problems, and can solve the above-mentioned salpin technical problems, as well as provide additional technical elements that can not be easily invented by a person skilled in the art. Became.
  • An object of the present invention is to provide an apparatus for realizing anonymization of an image, but the apparatus is intended to perform image processing so that a subject of an action cannot be identified while preserving the behavior information in the image so that it can be identified.
  • the present invention aims to enable the device as described above to improve its own image anonymization performance by using an adversarial learning method in which an image conversion unit and a subject recognition unit in the device are implemented in order to automatically enable image anonymization. Is done.
  • an object of the present invention is to improve the recognition rate through learning the behavior recognition unit in order to improve the performance of recognizing the behavior information in the image.
  • control method of the video anonymization apparatus for improving the video anonymization performance comprises the steps of: (a) extracting a subject-type region from an image by an image editing unit; (b) an image converting unit converting the extracted subject-chain region to generate a transformed image; (c) the image editing unit generating a converted image in which the main chain in the image is replaced with the converted image; (d) determining, by the subject recognition unit, an action subject from the converted image; It includes.
  • the main chain recognition area may be characterized as a face area.
  • the image conversion unit may change at least one or more conversion parameters in order to reduce the recognition rate of the behavioral subject of the subject recognition unit.
  • the subject recognition unit may change at least one subject recognition parameter in order to increase the behavior subject recognition rate.
  • the control method of the image anonymization device is characterized by repeating steps (a) to (d), but altering either one of the conversion parameter of the image conversion unit or the subject recognition parameter of the subject recognition unit alternately for each iteration. Can be done with
  • control method of the image anonymization device may further include (e) after the step (c), the behavior recognition unit determines the behavior information from the converted image.
  • the image conversion unit may change at least one or more transformation parameters to increase the behavior information recognition rate of the behavior recognition unit, and the behavior recognition unit may change at least one behavior recognition parameter to increase the behavior information recognition rate.
  • the control method of the image anonymization device repeats steps (a) to (e), but for each iteration, the conversion parameter of the image conversion unit, the subject recognition parameter of the subject recognition unit, or the behavior recognition parameter of the behavior recognition unit It can be characterized by changing only one.
  • the behavioral subject recognition rate may be determined by a behavioral subject recognition loss value.
  • M image conversion unit
  • A behavior information recognition unit
  • V first image
  • v zero 1 Any image constituting the image
  • L A loss function
  • v ' transformed image
  • b i
  • the image conversion unit calculating the image conversion loss value that defines the size of the difference between the main recognition region and the converted image; may further include.
  • control method of the video anonymization apparatus comprises: (a) the image conversion unit converts the learning data for the subject, and generates a first converted image; (b) a subject recognition unit determining an action subject from the first transformed image; (c) the image editing unit extracting the main chain-type region from the image; (d) generating a second transformed image by converting the extracted subject-chain region by an image converter; (e) an image editing unit generating a transformed image in which the main chain in the image is replaced with the second transformed image; (d) the behavior recognition unit determines behavior information from the converted image.
  • the image anonymization apparatus includes an image editing unit for extracting a subject-chained region from an image, and replacing the subject-chained region with a transformed image generated by an image converter to generate a transformed image; An image conversion unit that converts a subject-chain-type region to generate a converted image; A subject recognition unit that determines a subject from the converted image; A behavior recognition unit for determining behavior information from the converted image; And a control unit controlling the image editing unit, the image conversion unit, the subject recognition unit, and the behavior recognition unit.
  • the image conversion unit generates a converted image after changing at least one conversion parameter in order to reduce the recognition rate of the behavioral subject of the subject recognition unit or increase the recognition rate of the behavioral information of the behavior recognition unit. You can.
  • the subject recognition unit may determine a subject from the converted image after changing at least one subject recognition parameter in order to increase a behavior subject recognition rate.
  • the behavior recognition unit may determine behavior information from the converted image after changing at least one behavior recognition parameter to increase a behavior information recognition rate.
  • the anonymization of the video is implemented, but the action information in the video is preserved while the subject of the action cannot be specified.
  • 1 is a conceptual diagram for understanding image anonymization.
  • FIG. 3 shows a detailed configuration of the video anonymization device according to the present invention.
  • FIG. 4 illustrates a method for controlling an image anonymization device according to a first embodiment of the present invention
  • FIG. 5 illustrates a method of controlling an image anonymization device according to a second embodiment of the present invention.
  • a terminal is a mobile station, a mobile terminal, a subscriber station, a portable subscriber station, a user equipment, an access terminal, or a personal computer.
  • Personal computer refers to all types of devices capable of data communication with external devices, such as a telephone, and may include all or some functions of the device.
  • Video anonymization refers to the process of unidentifiably processing information that can identify an individual in a moving image (video) or still image (image), and the processed image is referred to as an anonymized image.
  • FIG. 1 briefly illustrates one embodiment for understanding the concept of image anonymization, according to which an image region (face region) capable of identifying an individual from the original image is determined when the original image is present, This includes converting the image area of a certain area including the corresponding image area using various graphic conversion methods and outputting the converted image to the user.
  • image anonymization refers to a process of processing an image so that it is difficult to identify information capable of identifying the subject of an action as much as possible, and information that can identify an action or action is preserved as much as possible.
  • the level of image anonymization is further processed from making the image unidentifiable by simply blurring or masking the image area capable of identifying an individual, as well as making the graphic unrecognizable.
  • the behavioral information on what the person is doing is required to the level of graphic processing so that it can be accurately determined.
  • 'object behavior information' which is one of the important information that can be extracted from the image while the number of objects that can be photographed gradually increases and the image data requiring anonymization increases.
  • the demand for technology that can accurately identify is gradually increasing.
  • the present invention applies appropriate graphic processing to information capable of identifying an individual in the image to make it difficult to identify, but the information on the behavior of the individual in the image is maintained as much as possible without damage.
  • a video anonymization device that can do video anonymization while maintaining it.
  • a video anonymization device and a method of controlling the video anonymization device that can effectively improve the video anonymization performance by allowing the video anonymization device to hostile itself. do.
  • the image anonymization device may include an image editing unit 110, an image conversion unit 120, a subject recognition unit 130, and a behavior recognition unit 140, and further control the above components as a whole
  • the control unit 150 and a storage unit 160 in which data and instructions can be stored will be described.
  • the image editing unit 110 basically functions to extract a specific image region from the image or to generate a new image by replacing the extracted image region with another image region. Specifically, when the image anonymization device receives the original image from the outside, the image editing unit 110 distinguishes and extracts the main chain-type region, that is, the face region, from the received original image, and also performs graphic processing on the main chain-type region in the future. When applied and converted, a new image (hereinafter referred to as a 'converted image') is generated by inserting the converted image in place of the original subject-identifying region in the original image.
  • the subject-chain domain can be understood to mean an image domain including information that can identify who the subject of the action is in the video. For example, the face region of individual people who can identify an individual It can be regarded as a representative subject-chain domain.
  • the image conversion unit 120 functions to convert a specific image region extracted from the image, that is, a subject-type region, into another type. Specifically, the image conversion unit 120 applies a predetermined graphic processing method to the image of the subject-chain area, that is, the face area, previously extracted by the image editing unit 110, so that the face area is different from that of the individual in the original image. This is the configuration for making conversion.
  • the image conversion unit 120 is competitively learning (adversarial learning) with the subject recognition unit 130 in the learning process for improving the ability to transform the subject-chain domain, specifically, image conversion.
  • the unit 120 repeats the learning process with the aim of preventing the subject recognition unit 130 from properly recognizing the action subject, that is, reducing the recognition rate of the action subject through competition with the subject recognition unit 130. From the standpoint of the subject recognition unit 130, the learning process is repeated with the goal of determining who the converted image is as high as possible.
  • the image conversion unit 120 may repeat the learning process with the goal of increasing the recognition rate of the behavior information of the behavior recognition unit 140.
  • the image conversion unit 120 may be preferably implemented as a convolutional neural network (CNN). Since many related papers are known in the CNN, detailed description will be omitted in this detailed description.
  • CNN convolutional neural network
  • the subject recognition unit 130 functions to determine who the corresponding converted image is based on an image (hereinafter referred to as a 'converted image') of the subject-chain area converted by the preceding image conversion unit 120. .
  • the subject recognition unit 130 compares the face images of the previously learned data set or the data set stored in the storage unit 160 to the converted image to identify who the converted image is, and at this time, the subject recognition unit ( 130) can also be implemented with CNN.
  • the subject recognition unit 130 aims to increase the recognition rate of the subject of action through a hostile learning process with the image conversion unit 120.
  • the behavior recognition unit 140 functions to determine what action any subject in the converted image is based on, based on the converted image, that is, the image in which the main subject recognition area has been replaced with the converted image.
  • the behavior recognition unit 140 compares the transformation image with various behavior information images of a data set previously learned about the behavior information or a data set stored in the storage 160 to identify the behavior in the transformation image,
  • the behavior recognition unit 140 may also be implemented as a CNN.
  • the behavior recognition unit 140 unlike the subject recognition unit 130 described above, through a learning process of raising each other in a relationship with the image conversion unit 120, that is, the image conversion unit 120 is a behavior recognition unit 140 The goal is to ultimately increase the recognition rate of the behavior information through the process of learning to increase the recognition rate of the behavior information and the behavior recognition unit 140 to increase the recognition rate of the behavior information.
  • the image anonymization apparatus further includes a control unit 150 and a storage unit 160.
  • the control unit 150 may control the image editing unit 110, the image conversion unit 120, the subject recognition unit 130, and the behavior recognition unit 140 described above, wherein the control unit 150 is a controller ), A microcontroller, a microprocessor, and a microcomputer.
  • the controller 150 may be implemented by hardware or firmware, software, or a combination thereof. When implemented using hardware, an application specific integrated circuit (ASIC) or digital signal processor (DSP) , DSPD (digital signal processing device), PLD (programmable logic device), FPGA (field programmable gate array), etc.
  • ASIC application specific integrated circuit
  • DSP digital signal processor
  • DSPD digital signal processing device
  • PLD programmable logic device
  • FPGA field programmable gate array
  • program instructions and contents necessary to provide the search service may be stored in a memory provided with the controller 150, wherein the memory includes read only memory (ROM), random access memory (RAM), and EPROM ( Erasable Programmable Read Only Memory (EEPROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash (flash) memory, static RAM (SRAM), hard disk drive (HDD), solid state drive (SSD), or the like may be implemented.
  • ROM read only memory
  • RAM random access memory
  • EPROM Erasable Programmable Read Only Memory
  • EEPROM Electrically Erasable Programmable Read-Only Memory
  • flash memory flash memory
  • SRAM static RAM
  • HDD hard disk drive
  • SSD solid state drive
  • Adversarial learning is a learning method in which modules having two different functions enhance their original functions through competition with each other.
  • the image conversion unit 120 and the subject recognition unit 130 To lead hostile learning.
  • learning referred to in the present invention refers to a video image anonymization device indicating a learning image and a label given to a training image (description of the corresponding training image), which frame type of a certain image represents specific behavior information, and It refers to the process of accumulating data about which type of image represents which specific person.
  • the meaning of the image conversion unit 120 and the subject recognition unit 130 performing mutually antagonistic learning is that the image conversion unit 120 is configured to prevent the subject recognition unit 130 from properly identifying individuals in the converted image. While learning to widen the width of the transformation, the subject recognition unit 130 is a methodology for recognizing the subject or a parameter for recognizing the subject even if the image conversion unit 120 widens the transformation to convert the first face region image It means that you can learn to recognize who the face of the transformed image is better by adjusting them in various ways.
  • FIG. 4 shows a first embodiment of a method for controlling an image anonymization device according to the present invention.
  • the first embodiment starts from the step S410 in which the image editing unit 110 extracts the subject-type region from the image.
  • Video refers to a set of image frames that contain information that can identify a person and the actions of a person.
  • video is a video or photo taken by a camera or a video taken by a wearable device.
  • the image may be largely divided into a main chain-type area and an area excluding the main chain-type area, which means an image area including personally identifiable information, for example, a face area, and excludes the main chain-type area.
  • the area may be regarded as an image area containing information that can identify the person's behavior.
  • the face area is shown as a representative example of the subject-chain area, but this is only an example, and an image containing information that can identify who the subject of the action is, although not necessarily the face area. All of my domain should be understood as the main chain.
  • the image conversion unit 120 converts the subject-chain area extracted by the image editing unit 110 to generate a converted image.
  • the converted image may be a color changed by changing a pixel in the main chain area in the original image, or may be replaced by a pixel of the face of another person in the main chain, or else the individual in the original image may be replaced. Any method of changing the pixel information in the subject identification area with the goal of making it unidentifiable will be irrelevant.
  • various kinds of transformation parameters may be defined and adjusted, for example, by specifying the main chain-type region to a certain extent. Parameters such as whether to convert, how much to change the pixels within a specified range, and which graphic processing algorithm to use for image conversion, etc. can be defined and adjusted. These conversion parameters may derive an optimization value of each parameter by continuously adjusting the image converter 120 to improve performance as the learning process progresses.
  • the image editing unit 110 After the step S420, the image editing unit 110 generates a transformed image in which the subject identification area in the original image is replaced with the transformed image generated in the step S420.
  • the converted image preferably, may be in a state in which it is not easy to identify who is acting, and conversely, the original action information is preserved so that it is easy to identify what the person is doing. Can be
  • the subject recognition unit 130 may determine who the subject of the action is from the transform image previously generated in step S420.
  • the meaning that the subject recognition unit 130 determines the subject of the action may be defined in two ways, one of which is a corresponding converted image from a labeled learning image or a labeled learning image. It can be defined in two ways: recognizing who is the other, and recognizing who the corresponding converted image is through image comparison from unlabeled data. The former is understood as the definition in the process in which the subject recognition unit 130 performs learning through the learning data, and the latter in the process of performing actual image anonymization after learning is completed.
  • various types of subject-chain parameters may be defined and adjusted. For example, to what extent pixel information is used for the subject-chain expression, the individual The subject-chain parameters, such as how many pixels to sample for identification, can be defined and adjusted.
  • the subject recognition unit 130 may further perform a process of calculating a behavior subject recognition rate.
  • the behavioral subject recognition rate is a value for determining how accurately the subject recognition unit 130 has identified an individual from an image, and this number is the host recognition unit 130 or the image conversion unit 120 in the future It is also the value that is the basis of the goal in the learning process. That is, the image conversion unit 120 adjusts the conversion parameters so that the subject recognition unit 130 recognizes at a lower value than the previous action subject recognition rate in the next cycle when steps S410 to S440 are cycled, and then cycles. In the subject recognition unit 130 adjusts the recognition parameters to further improve the behavioral subject recognition rate compared to the previous cycle.
  • the recognition rate of the behavioral subject may be calculated based on a loss value generated in the recognition process, that is, a behavioral subject recognition loss value.
  • the behavioral subject recognition loss value L adv may be defined by, for example, Equation 1 below, and the higher the value, the higher the recognition rate.
  • M is the image conversion unit 120
  • D is the subject recognition unit 130
  • F is the subject-chain dataset
  • L D is the loss function
  • f is the data of the subject-chain domain
  • I is the behavioral subject label set
  • i f is the actor label.
  • control method of the image anonymization apparatus may further include a learning process for identifying behavior information in the transformed image in addition to the subject recognition learning process from steps S410 to S440 (or step S441). That is, when referring back to FIG. 4, the control method of the image anonymizing device further includes a step of determining, by the behavior recognition unit 140, personal behavior information from the converted image generated in step S430. (S450) As in the previous description of the subject recognition unit 130 in step S440, the behavior information discrimination of the behavior recognition unit 140 may also be understood in two meanings, one of which is learning with a label.
  • each Behavior recognition parameters such as to which range of pixel information in a frame is used for behavior information recognition, may be defined and adjusted.
  • the behavior recognition unit 140 may further perform a process of calculating a behavior information recognition rate.
  • the behavior information recognition rate is a value for determining how accurately the behavior recognition unit 140 has identified the behavior information from the converted image, and this value is also the future behavior recognition unit 140 or the image conversion unit 120 A can be the standard of goal in the process of learning.
  • the subject recognition unit 130 Adjusts the conversion parameters so that the recognition rate is lower than the previous behavioral subject recognition rate, and in the next cycle, the subject recognition unit 130 adjusts the subject-chain parameters to further improve the behavioral subject recognition rate compared to the previous cycle.
  • the behavior recognition unit 140 may adjust the behavior recognition parameters to further improve the behavior information recognition rate compared to the previous cycle.
  • the video conversion unit 120, the subject recognition unit 130, and the behavior recognition unit 140 adjust the parameters in order to learn in each cycle.
  • the other parameters are fixed when any one parameter is adjusted in each cycle.
  • the behavior information recognition rate may be calculated based on a loss value that occurs in the recognition process, that is, a behavior information recognition loss value.
  • the behavior information recognition loss value L det may be defined by, for example, Equation 2 below, and the lower the value, the higher the recognition rate.
  • M is an image conversion unit 120
  • A is a behavior recognition unit 140
  • V is an original image
  • v is an arbitrary image frame constituting the original image
  • L A is a loss function
  • v ' is a converted image
  • the frame b i is ground-truth action bounding boxes
  • t i is the action information category label.
  • the video anonymization device can calculate behavioral subject recognition loss and behavioral information recognition loss to improve performance, such as subject recognition and behavioral recognition, respectively, and further define an image conversion loss value (L l1 ). By doing so, it is possible to prevent the difference between the main chain-type region image and the converted image in the original image from being excessively different. (S461) If the purpose of the video anonymization device is to process the part or information that can identify the individual in the original video, but the difference between the main subject area of the original video and the converted image is too large, as a user, Since it can be difficult to understand information from the converted image, to prevent this, the difference between the original subject identification area and the converted image is limited so as not to fall outside a certain range.
  • the above image conversion loss value L l1 may be defined by Equation 3 below.
  • M is the image conversion unit 120
  • F is the main chain-type data set
  • f is the main chain-type data
  • M (f) is the converted main chain-type area.
  • the behavioral subject recognition loss value, behavioral information recognition loss value, and image conversion loss value described above have been described as being calculated by the subject recognition unit 130, the behavior recognition unit 140, and the image conversion unit 120, respectively.
  • the above values are not necessarily calculated by each functional unit, and may be implemented to be calculated by other components constituting a video anonymization device according to the developer's intention, for example, by the control unit 150.
  • the first embodiment described above is based on the premise that there is only one person who needs to be identified in the video, but when a plurality of persons are included in the video, the same process can be performed for each person.
  • the second embodiment of the image anonymization device control method is similar to the first embodiment described above, except that the data set (image) for subject recognition and the data set (image) for subject recognition are independent sources from each other. have.
  • the data set (image) for subject recognition and the data set (image) for subject recognition are independent sources from each other. have.
  • the action label and the face label of the same image are always required at the same time. Because it is not, there is a limit to learning the video anonymization device.
  • the second embodiment according to FIG. 5 is proposed to overcome the above disadvantages, and in the second embodiment, the subject recognition unit 130 does not use a converted image in which the subject-identified region in the original image is converted, and is separate. It is possible to individually train using a data set, that is, a training data set.
  • steps S510, S521, S530, S550, and S551 in which the behavior recognition unit 140 identifies behavior information from the converted image is large as corresponding to steps S410, S420, S430, S450, and S451 of FIG.
  • the subject recognition unit 130 differs from FIG. 4 before steps S522, S540, and S541 in which the subject is identified from the converted image.
  • the image conversion unit 120 receives and converts data for subject identification separate from the image from the outside, and then converts the first converted image (clear distinction)
  • the image in which the subject chain recognition area extracted from the image is converted is referred to as a second transform image
  • the recognition rate of the behavioral subject calculated in step S541 will also be calculated according to the subject identification process from the first transformed image, and the image transformation loss value L 1 that limits the difference between the original image and the transformed image is also received separately. It will be calculated between the main chain type data and the first converted image.
  • FIGS. 4 and 5 The method of controlling the device to improve the performance of the video anonymization device according to the present invention has been described with reference to FIGS. 4 and 5.
  • the embodiment of FIGS. 4 and 5 has been described as a representative example of the face area as information capable of recognizing the subject of the action, but the present invention is limited to the face area and controls the image anonymization device. It should be understood that it is intended to anonymize by processing information that can recognize the subject of the action in the video in a comprehensive sense (which may exist across all areas of the video).
  • FIGS. 4 and 5 may be understood as a process of learning image anonymization or a process of anonymizing an actual image depending on whether the cyclic process is repeated.
  • it may be understood as a process of learning video anonymization, and when there is no cyclic process It can be understood as a process of receiving an image from the outside and performing anonymization after all learning has already been done.
  • the video anonymization device that has undergone the learning process of FIG. 4 or 5 may be applied in various fields, for example, behavior information of actors in an image captured by implementing a camera fixed at a certain position, such as a CCTV camera. Can be used to convert only the face area while preserving it, or implemented with a mobile camera such as a smart glass or a camera equipped with a mobile robot to convert only the face area while preserving the behavioral information of the actor in the captured image. You may.
  • the video anonymization device that has undergone the learning process of FIG. 4 or 5 may be used to perform privacy protection behavior recognition.
  • the video anonymization device may It is possible to recognize the user's behavior without accessing a private high-resolution face image.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Health & Medical Sciences (AREA)
  • Bioethics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Computer Security & Cryptography (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Medical Informatics (AREA)
  • Signal Processing (AREA)
  • Image Analysis (AREA)

Abstract

La présente invention concerne un dispositif et un procédé d'anonymisation d'une vidéo et, plus particulièrement, un dispositif d'anonymisation vidéo qui peut présenter une performance et une sortie améliorées par rapport à l'état de la technique classique, et un procédé permettant d'améliorer davantage la performance d'anonymisation en commandant ledit dispositif.
PCT/KR2019/002197 2018-09-28 2019-02-22 Procédé de commande d'un dispositif d'anonymisation vidéo permettant d'améliorer les performances d'anonymisation, et dispositif associé WO2020067615A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR10-2018-0116426 2018-09-28
KR1020180116426A KR102083192B1 (ko) 2018-09-28 2018-09-28 익명화 성능을 향상시키기 위한 영상 익명화 장치의 제어 방법 및 이를 위한 장치

Publications (1)

Publication Number Publication Date
WO2020067615A1 true WO2020067615A1 (fr) 2020-04-02

Family

ID=69805401

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2019/002197 WO2020067615A1 (fr) 2018-09-28 2019-02-22 Procédé de commande d'un dispositif d'anonymisation vidéo permettant d'améliorer les performances d'anonymisation, et dispositif associé

Country Status (2)

Country Link
KR (1) KR102083192B1 (fr)
WO (1) WO2020067615A1 (fr)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102542016B1 (ko) * 2020-08-26 2023-06-09 (주)어셈블써클 의료 영상의 가명화 방법 및 장치
KR102531368B1 (ko) * 2020-11-26 2023-05-12 주식회사 딥핑소스 원본 의료 이미지의 영역별로 서로 다른 컨실링 프로세스가 수행되도록 하여 원본 의료 이미지를 컨실링 처리하는 변조 네트워크를 학습하는 방법 및 테스트하는 방법, 그리고 이를 이용한 학습 장치 및 테스트 장치
KR102473801B1 (ko) * 2021-07-07 2022-12-05 주식회사 딥핑소스 공간 내에서 오브젝트들을 익명화하여 추적하는 방법 및 이를 이용한 익명화 추적 장치

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101458136B1 (ko) * 2014-02-28 2014-11-05 김호 영상 처리 방법, 이를 실행하는 영상 처리 서버, 감시 서버, 영상 처리 시스템 및 이를 저장한 기록 매체
KR20140131188A (ko) * 2013-05-03 2014-11-12 딕스비전 주식회사 저비용으로 사생활 보호 및 관심 강조 기능을 구비하는 통합 관제 시스템 및 방법
KR101563297B1 (ko) * 2014-04-23 2015-10-26 한양대학교 산학협력단 영상에서 행동을 인식하는 방법 및 장치
JP2015222460A (ja) * 2014-05-22 2015-12-10 日本電信電話株式会社 公開画像編集方法及び公開画像編集装置

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105844128B (zh) 2015-01-15 2021-03-02 北京三星通信技术研究有限公司 身份识别方法和装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20140131188A (ko) * 2013-05-03 2014-11-12 딕스비전 주식회사 저비용으로 사생활 보호 및 관심 강조 기능을 구비하는 통합 관제 시스템 및 방법
KR101458136B1 (ko) * 2014-02-28 2014-11-05 김호 영상 처리 방법, 이를 실행하는 영상 처리 서버, 감시 서버, 영상 처리 시스템 및 이를 저장한 기록 매체
KR101563297B1 (ko) * 2014-04-23 2015-10-26 한양대학교 산학협력단 영상에서 행동을 인식하는 방법 및 장치
JP2015222460A (ja) * 2014-05-22 2015-12-10 日本電信電話株式会社 公開画像編集方法及び公開画像編集装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
REN, ZHONGZHENG ET AL.: "Learning to Anonymize Faces for Privacy Preserving Action Detection", ARXIV:1803.11556V2, 26 July 2018 (2018-07-26), XP047488312, Retrieved from the Internet <URL:https://arxiv.org/abs/1803.11556v2> *

Also Published As

Publication number Publication date
KR102083192B1 (ko) 2020-03-02

Similar Documents

Publication Publication Date Title
WO2020067615A1 (fr) Procédé de commande d&#39;un dispositif d&#39;anonymisation vidéo permettant d&#39;améliorer les performances d&#39;anonymisation, et dispositif associé
WO2020130309A1 (fr) Dispositif de masquage d&#39;image et procédé de masquage d&#39;image
WO2014189232A1 (fr) Capteur électronique et méthode de commande de celui-ci
KR101911900B1 (ko) 사생활 보호 카메라와 그 시스템 및 얼굴 인식 기반 실시간 자동 영상 익명화 방법
WO2017034220A1 (fr) Procédé de mise au point automatique sur une région d&#39;intérêt par un dispositif électronique
WO2020122403A1 (fr) Procédé de service lié à un animal à base de chaîne de blocs, et dispositif pour réaliser le procédé
WO2018097389A1 (fr) Dispositif de recherche d&#39;image, procédé de stockage de données et dispositif de stockage de données
WO2017090892A1 (fr) Caméra de génération d&#39;informations d&#39;affichage à l&#39;écran, terminal de synthèse d&#39;informations d&#39;affichage à l&#39;écran (20) et système de partage d&#39;informations d&#39;affichage à l&#39;écran le comprenant
WO2021118039A1 (fr) Procédé fondé sur l&#39;apprentissage profond permettant de filtrer des images similaires, et appareil faisant appel audit procédé
WO2021100919A1 (fr) Procédé, programme et système pour déterminer si un comportement anormal se produit, sur la base d&#39;une séquence de comportement
WO2010041836A2 (fr) Procédé de détection d&#39;une zone de couleur peau à l&#39;aide d&#39;un modèle de couleur de peau variable
EP3172683A1 (fr) Procédé d&#39;extraction d&#39;image et dispositif électronique associé
WO2021153971A1 (fr) Procédé d&#39;apprentissage de réseau neuronal utilisant une image anonymisée et serveur mettant en œuvre celui-ci
WO2023282520A1 (fr) Procédé et système permettant de construire une base de données d&#39;entraînement au moyen d&#39;une technologie de protection des informations personnelles vocales
WO2019017720A1 (fr) Système de caméra permettant la protection de la confidentialité et procédé correspondant
WO2022086147A1 (fr) Procédé permettant d&#39;entraîner et de tester un réseau d&#39;apprentissage utilisateur à utiliser pour reconnaître des données brouillées créées par brouillage de données originales pour protéger des informations personnelles et dispositif d&#39;apprentissage utilisateur et dispositif de test faisant appel à celui-ci
EP3459009A2 (fr) Procédé de quantification adaptative pour codage d&#39;image d&#39;iris
WO2021049855A1 (fr) Procédé et dispositif électronique pour capturer une région d&#39;intérêt (roi)
WO2012137994A1 (fr) Dispositif de reconnaissance d&#39;image et son procédé de surveillance d&#39;image
CN111898448B (zh) 一种基于深度学习的行人属性识别方法和系统
Asghar et al. Deep learning based effective identification of eu-gdpr compliant privacy safeguards in surveillance videos
WO2014148784A1 (fr) Base de données de modèles linguistiques pour la reconnaissance linguistique, dispositif et procédé et système de reconnaissance linguistique
WO2020050550A1 (fr) Procédés et systèmes de réalisation d&#39;opérations de modification sur un support
WO2023080667A1 (fr) Traitement d&#39;image wdr de caméra de surveillance par reconnaissance d&#39;objets basée sur l&#39;ia
WO2018056488A1 (fr) Système et procédé d&#39;extraction de valeurs de caractéristiques d&#39;une vidéo numérique, et système et procédé de détermination de similitude entre des vidéos numériques à l&#39;aide desdites valeurs de caractéristiques

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: 19865879

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 08.07.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19865879

Country of ref document: EP

Kind code of ref document: A1