KR102439429B1 - Annotation method for easy object extraction and a computer program recorded on a recording medium to execute the annotation method - Google Patents

Abstract

The present invention proposes an annotation method that can easily designate an object included in an image when annotating data for artificial intelligence (AI) learning. The method may include the steps of: setting, by an annotation device, a partial area in an image that is a target of annotation work for AI learning under the control of an operator as a bounding box; extracting, by the annotation device, an edge of an object from an inner region of the bounding box, and distinguishing an object from a background based on the extracted edge; and modifying, by the annotation device, a bounding box to an area of the object from which the background has been deleted.

Description

Annotation method for easy object extraction and a computer program recorded on a recording medium to execute the same

The present invention relates to data design for artificial intelligence (AI) learning. More specifically, in annotating data for artificial intelligence (AI) learning, it relates to an annotation method capable of easily designating an object included in an image, and a computer program recorded on a recording medium for executing the annotation method.

Artificial intelligence (AI) refers to a technology that artificially implements some or all of human learning ability, reasoning ability, and perception ability using computer programs. In relation to artificial intelligence (AI), machine learning refers to learning to optimize parameters with given data using a model composed of multiple parameters. Such machine learning is classified into supervised learning, unsupervised learning, and reinforcement learning according to the form of data for learning.

In general, the design of data for artificial intelligence (AI) learning proceeds in the steps of designing a data structure, collecting data, refining data, processing data, expanding data, and verifying data.

In more detail with respect to each step, the design of the data structure is made through the definition of an ontology, a definition of a classification system, and the like. Data is collected by direct shooting, web crawling, or by collecting data through associations/professional groups. Data purification is performed by removing redundant data from the collected data and de-identifying personal information. Data processing is performed by inputting metadata and performing annotations. Data expansion is performed by performing ontology mapping and supplementing or extending the ontology as necessary. And, the verification of the data is performed by verifying the validity according to the set target quality using various verification tools.

In general, the annotation of the data processing step is performed by processing a bounding box on an object included in an image, and inputting property information of the object that has been processed with the bounding box. Such annotations are also referred to as data labeling. And, the dataset corresponding to the work result of the annotation is calculated in the form of a JSON (Java Script Object Notation) file.

Specifically, in the process of processing the bounding box in the annotation, for each object to be recognized in the image, the operator manually processes the bounding box according to the outline of the object.

However, the bounding box processing must follow various pre-determined rules to ensure the AI learning effect (ie, the recognition rate of objects in the image). For example, the bounding box processing should follow rules such as a minimum size value of an area in which an object can be recognized, and a maximum number of pixels spaced apart from the outline of the object. Therefore, it is not easy for the operator to completely comply with all predetermined rules in the process of manually processing the bounding box for a large number of objects.

Korean Patent Application Laid-Open No. 10-2020-0042629, ‘Method for generating touch-based annotation and image of mobile device for artificial intelligence learning, and device therefor’, (published on April 24, 2020)

An object of the present invention is to provide an annotation method capable of facilitating extraction of an object included in an image when annotating data for artificial intelligence (AI) learning.

Another object of the present invention is to provide a computer program recorded on a recording medium for executing an annotation method that can facilitate extraction of an object included in an image.

The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the following description.

In order to achieve the technical problem as described above, the present invention proposes an annotation method that can easily designate an object included in an image. The method includes, by an annotation device, setting a partial region in an image to be subjected to an annotation operation for artificial intelligence (AI) learning as a bounding box under the control of an operator; extracting, by the annotation device, an edge of an object from an inner region of the bounding box, and classifying an object and a background based on the extracted edge; and modifying, by the annotation device, a bounding box to the area of the object from which the background has been deleted.

Specifically, in the setting of the bounding box, the bounding box may be set based on a rectangle having two coordinates input from the operator in the image as the coordinates of the upper left vertex and the coordinates of the lower right vertex.

In the setting of the bounding box, the bounding box may be set based on a rectangle connecting the four coordinates input by the operator in the image.

In the step of classifying the background, an RGB value of each pixel may be detected in an inner region of the bounding box, and an edge of an object may be extracted through the detected RGB value.

In the step of classifying the background, the RGB value for each pixel is detected in the inner region of the bounding box, and a point where the change amount of the RGB value with respect to an adjacent pixel is higher than a preset threshold value may be extracted as an edge. .

Separating the background may include: generating an RGB image by detecting RGB (red, green, blue) values for each pixel in an inner region of the bounding box; changing the generated RGB channel to an RGBA (red, green, blue, alpha) image; and extracting an edge of an object based on transparency from the RGBA image.

In the step of extracting the edge, a transparency channel is extracted from the RGBA image, an area having a transparency value lower than a preset transparency threshold value is removed, but the image is preprocessed by blurring the edge of the extracted transparency channel to do; and separating and extracting the object from the background in the RGBA image by deriving the probability that the important object belongs to each pixel as an intensity value using salient object detection (SOD). may include

In the step of extracting the edge, a transparency channel is extracted from the RGBA image, an area having a transparency value lower than a preset transparency threshold value is removed, but the image is preprocessed by blurring the edge of the extracted transparency channel to do; and separating and extracting the object from the background in the RGBA image based on a boundary value at which the object is not visible while imparting transparency to the preprocessed image.

In the extracting of the edge, a transparency channel may be extracted from the RGBA image, and a pixel area having a transparency lower than a preset threshold value may be distinguished as a background.

In order to achieve the technical problem as described above, the present invention proposes a computer program recorded on a recording medium for executing an annotation method. The computer program includes a memory; transceiver; And coupled with a computing device configured to include a processor (processor) for processing instructions resident in the memory, the processor, according to the control of the operator, artificial intelligence (Artificial Intelligence, AI) annotation for learning (annotation) operation setting a partial area in the image to be the target of as a bounding box; extracting, by the processor, an edge of an object from an inner region of the bounding box, and classifying an object and a background based on the extracted edge; and modifying, by the processor, the bounding box to the area of the object from which the background has been deleted, it may be recorded on the recording medium.

Specific details of other embodiments are included in the detailed description and drawings.

According to embodiments of the present invention, in performing annotation on each of a plurality of images for artificial intelligence (AI) learning, an edge of an object is extracted from an inner region of a bounding box, and based on the extracted edge By classifying an object and a background and modifying the bounding box to the area of the object from which the background has been deleted, it is possible to shorten the worker's work time and have a more uniform quality of the annotation work result.

1 is a block diagram of a security management system according to an embodiment of the present invention.
2 to 4 are block diagrams of an artificial intelligence learning system according to various embodiments of the present invention.
5 and 6 are logical configuration diagrams of an annotation apparatus according to an embodiment of the present invention.
7 is a hardware configuration diagram of an annotation apparatus according to an embodiment of the present invention.
8 to 12 are exemplary views for explaining a process of setting a bounding box according to an embodiment of the present invention.
13 is a flowchart illustrating an annotation method according to an embodiment of the present invention.
14 is a flowchart illustrating a method of setting a bounding box according to an embodiment of the present invention.
15 is a flowchart illustrating a method of setting a bounding box according to another embodiment of the present invention.
16 is a flowchart illustrating a method of setting a bounding box according to another embodiment of the present invention.
17 is a flowchart illustrating a security management method according to an embodiment of the present invention.

It should be noted that technical terms used herein are used only to describe specific embodiments, and are not intended to limit the present invention. In addition, the technical terms used in this specification should be interpreted in the meaning generally understood by those of ordinary skill in the art to which the present invention belongs, unless otherwise defined in this specification, and excessively inclusive. It should not be construed as a human meaning or in an excessively reduced meaning. In addition, when the technical terms used in this specification are incorrect technical terms that do not accurately express the spirit of the present invention, they should be understood by being replaced with technical terms that can be correctly understood by those skilled in the art. In addition, general terms used in the present invention should be interpreted as defined in advance or according to the context before and after, and should not be interpreted in an excessively reduced meaning.

Also, as used herein, the singular expression includes the plural expression unless the context clearly dictates otherwise. In the present application, terms such as "consisting of" or "having" should not be construed as necessarily including all of the various components or various steps described in the specification, and some components or some of the steps are included. It should be construed that it may not, or may further include additional components or steps.

Also, terms including ordinal numbers such as first, second, etc. used herein may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component.

When a component is referred to as being “connected” or “connected” to another component, it may be directly connected or connected to the other component, but another component may exist in between. On the other hand, when it is said that a certain element is "directly connected" or "directly connected" to another element, it should be understood that no other element is present in the middle.

Hereinafter, a preferred embodiment according to the present invention will be described in detail with reference to the accompanying drawings, but the same or similar components are given the same reference numerals regardless of the reference numerals, and the redundant description thereof will be omitted. In addition, in the description of the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, it should be noted that the accompanying drawings are only for easy understanding of the spirit of the present invention, and should not be construed as limiting the spirit of the present invention by the accompanying drawings. The spirit of the present invention should be construed as extending to all changes, equivalents, or substitutes other than the accompanying drawings.

On the other hand, the worker performing the annotation work may perform the annotation work by accessing the foreign artificial intelligence learning system as a foreigner, without performing the work in Korea as a Korean. Accordingly, there is a problem in that the personal information of the worker is leaked to the corresponding AI learning system, or the personal information included in the annotation target image collected in the AI learning system is leaked to the worker. That is, when performing annotation work, personal information of each country may be leaked abroad, and export of personal information to other countries may be against the laws of each country's personal information protection laws, etc.

Accordingly, in performing annotation work worldwide, it is necessary to establish a system that can efficiently manage data for learning in accordance with country-specific personal information-related laws.

In order to overcome these difficulties, the present invention intends to propose means for efficiently managing learning data according to country-specific personal information-related laws when performing annotation work worldwide.

1 is a block diagram of a security management system according to an embodiment of the present invention.

Referring to Figure 1, the security management system 600 according to an embodiment of the present invention is an artificial intelligence learning system (400-1, 400-2, 400-3, ..., 400-n; 400) and the security management device 500 may be included.

Here, the artificial intelligence learning system 400 may include an annotation device, a learning data design device, and an artificial intelligence learning device. Meanwhile, a detailed configuration and operation of the artificial intelligence learning system 400 will be described later with reference to FIGS. 2 to 4 .

Such an artificial intelligence learning system 400 may be provided for each country as shown. Here, the artificial intelligence learning system 400 is illustrated as provided one by one for each country, but is not limited thereto, and may be provided in plurality for each country.

In addition, the annotation devices included in the artificial intelligence learning system 400 may receive the task for the annotation work from the learning data design device provided in the same system, but may also access other systems to perform the annotation work.

That is, the worker performing the annotation work may perform the annotation work by accessing the artificial intelligence learning system 400 abroad as a foreigner without performing the work in Korea as a Korean.

In this case, when the worker performing the annotation work accesses the overseas artificial intelligence learning system 400 through the annotation device, the worker's personal information is leaked to the corresponding artificial intelligence learning system 400, or the artificial intelligence learning system ( 400), personal information included in the annotation object collected may be leaked to the annotation device.

For example, when the annotation device included in the Korean AI learning system 400-3 accesses the learning data design device included in the American AI learning system 400-1 for annotation work, Because privacy laws may be different, illegal situations may arise.

Accordingly, each of the learning data designing devices of each artificial intelligence learning system 400 may include a non-security area storing a public annotation work object set according to a security policy, and a security area storing a private annotation work object.

Here, the security area and the non-security area may be set in consideration of at least one of an attribute of an annotation operation, the number of annotation workers, and personal information related laws of a corresponding country. That is, in the artificial intelligence learning system 400, the annotation work is carried out in units of projects, and each artificial intelligence learning system 400 determines the properties of the project, the number of annotation workers performing the project, and the project. In consideration of the privacy-related laws and regulations of the relevant country of the system, it is possible to separate and store the object to be disclosed and the object to be annotated to be private.

With the following configuration, the security management device 500 may collect and manage security policies for each of the plurality of artificial intelligence learning systems 400 as described above.

Then, when the security management device 500 receives a request for access for annotation work from the annotation device to one of the learning data design devices of the plurality of artificial intelligence learning systems 400, the security policy of the artificial intelligence learning system 400 that has requested access can confirm.

And, the security management device 500 may allow access to the annotation device according to the security policy of the artificial intelligence learning system 400 that requires access. That is, the security management device 500 may provide the public annotation object stored in the non-security area of the learning data design device requiring access to perform the annotation operation.

In this case, the security management device 500 identifies the country of the annotation device that has requested access based on the identification information of the annotation device that has requested access, and allows access to the AI learning system that can be accessed from the identified country. .

In addition, the security management device 500 may identify the country of the annotation device requesting access based on the identification information of the annotation device requesting access, and allow access to the non-security area accessible from the identified country.

That is, the security management device 500 identifies the country of the annotation device that has requested access, and selectively allows access to the non-security area included in the artificial intelligence learning system 400 or the artificial intelligence learning system 400 by country. have.

Thereafter, the annotation device may perform an annotation operation on the public annotation object stored in the non-security area of the learning data design device to which access is allowed.

Here, the security management device 500 may identify the country of the annotation device requesting access by recognizing the location information or language of the annotation device requesting access.

In addition, the security management device 500 identifies worker information using the annotation device requesting access based on the identification information of the annotation device requesting access, and artificial intelligence learning system 400 that can be accessed from the identified worker information can allow access. Here, the worker information may be personal information of a worker performing an annotation operation.

That is, the security management device 500 may control access in consideration of not only the security policy of each country, but also the country of the annotation device attempting to access, and information of a worker performing annotation work through the annotation device.

In this way, the security management system 600 according to an embodiment of the present invention efficiently manages learning data according to the privacy-related laws of each country in performing the annotation work worldwide through the security management device 500 . can do.

As such, the security management system 600 according to an embodiment of the present invention may be any one of a fixed computing device such as a desktop, a workstation, or a server, but is not limited thereto. , smart phone, laptop, tablet, phablet, portable multimedia player (PMP), personal digital assistants (PDA) or e-book reader It may be any one of the portable computing devices such as (E-book reader).

2 to 4 are block diagrams of an artificial intelligence learning system according to various embodiments of the present invention.

As shown in Figure 2, the artificial intelligence learning system 400 according to an embodiment of the present invention is one or more annotation devices (100-1, 100-2, 100-3, ..., 100-n; 100) and It may be configured to include the artificial intelligence learning apparatus 300 .

As shown in FIG. 3 , the artificial intelligence learning system 400 according to another embodiment of the present invention additionally includes a learning data design apparatus 200 in addition to the annotation apparatus 100 and the artificial intelligence learning apparatus 300 . can be

In addition, as shown in FIG. 4 , in the artificial intelligence learning system 400 according to another embodiment of the present invention, the annotation apparatus 100 , the learning data design apparatus 200 and the artificial intelligence learning apparatus 300 are disclosed. They may be connected to each other through a public network. In this case, a part of the annotation device 100 may be a device for performing annotation by a clouding service.

As such, since the components of the AI learning system according to various embodiments are merely functionally distinct elements, two or more components are integrated with each other in the actual physical environment, or one component is the actual physical environment. may be implemented separately from each other.

Each component is described. The annotation apparatus 100 is a device that can be used to annotate an image provided from the learning data design apparatus 200 or the artificial intelligence learning apparatus 300 .

In particular, the annotation apparatus 100 according to the present invention has a feature of providing various conveniences to an operator in the process of setting a bounding box in relation to an annotation operation and inputting attribute information of an object.

As such, the annotation device 100 can transmit and receive data to and from the learning data design device 200 or the artificial intelligence learning device 300, and any device capable of performing an operation using the transmitted/received data may be permitted. have.

For example, the annotation device 100 may be any one of a fixed computing device such as a desktop, a workstation, or a server, but is not limited thereto, and a smart phone, such as laptops, tablets, phablets, Portable Multimedia Players (PMPs), Personal Digital Assistants (PDAs), or E-book readers. It may be any of the mobile computing devices.

A detailed configuration and operation of the annotation apparatus 100 as described above will be described later with reference to FIGS. 5 to 16 .

With the following configuration, the learning data design device 200 is a device that can be used to design and generate data for artificial intelligence (AI) learning. As described above, the learning data design apparatus 200 is basically a device distinct from the artificial intelligence learning apparatus 300 , but may be implemented by being integrated into the artificial intelligence learning apparatus 300 in an actual physical environment.

Specifically, the learning data design apparatus 200 may receive the properties of the project related to artificial intelligence (AI) learning from the artificial intelligence learning apparatus 300 . The learning data design device 200 performs design of a data structure for artificial intelligence (AI) learning, purification of collected data, processing of data, expansion of data, and verification of data, based on the user's control and the properties of the project. can be done

In particular, the learning data design apparatus 200 may transmit an image to be annotated to the annotation apparatus 100 for data processing for artificial intelligence (AI) learning. The learning data design apparatus 200 may receive an annotation work result from the annotation apparatus 100 . In this case, the annotation operation result may have a JSON (Java Script Object Notation) file format. Alternatively, the learning data design apparatus 200 may generate a JSON file based on the received result after receiving an annotation work result in a format different from JSON. Then, the learning data design apparatus 200 may inspect the received or generated JSON file, package it, and transmit it to the artificial intelligence learning apparatus 300 .

As such, the learning data design device 200 may be any device as long as it is a device capable of transmitting and receiving data to and from the annotation device 100 and the artificial intelligence learning device 300 and performing an operation using the transmitted/received data. have. For example, the learning data design apparatus 200 may be any one of a fixed computing device such as a desktop, a workstation, or a server, but is not limited thereto.

With the following configuration, the artificial intelligence learning device 300 is a device that can be used to perform machine learning of artificial intelligence (AI) based on data for artificial intelligence (AI) learning.

Specifically, the artificial intelligence learning apparatus 300 may receive the packaged JSON file directly from the annotation apparatus 100 or from the learning data design apparatus 200 . And, the artificial intelligence learning apparatus 300 may perform machine learning of artificial intelligence (AI) using the received JSON file.

As such, the artificial intelligence learning apparatus 300 can transmit and receive data to and from the annotation apparatus 100 or the learning data design apparatus 200, and any apparatus that can perform an operation using the transmitted and received data is allowed. have. For example, the artificial intelligence learning apparatus 300 may be any one of a fixed computing device such as a desktop, a workstation, or a server, but is not limited thereto.

As described above, at least one annotation device 100, learning data design device 200, and artificial intelligence learning device 300 are a combination of at least one of a secure line, a public wired communication network, or a mobile communication network directly connected between the devices. Data can be transmitted and received using the network.

For example, public wired networks include Ethernet, x Digital Subscriber Line (xDSL), Hybrid Fiber Coax (HFC), and Fiber To The Home (FTTH). However, it is not limited thereto. In addition, mobile communication networks include Code Division Multiple Access (CDMA), Wideband CDMA, WCDMA, High Speed Packet Access (HSPA), Long Term Evolution, LTE) and 5th generation mobile telecommunication may be included, but are not limited thereto.

Hereinafter, the configuration of the annotation apparatus 100 as described above will be described in more detail.

5 and 6 are logical configuration diagrams of an annotation apparatus according to an embodiment of the present invention.

5 and 6, the annotation apparatus 100 includes a communication unit 105, an input/output unit 110, a storage unit 115, a bounding box setting unit 120, an object property setting unit 125, and It may be configured to include a result generating unit 130 .

As such, the components of the annotation apparatus 100 merely represent functionally distinct elements, so that two or more components are integrated with each other in the actual physical environment, or one component is separated from each other in the actual physical environment. could be implemented.

When describing each component, the communication unit 105 may transmit/receive data to and from the learning data design apparatus 200 and the artificial intelligence learning apparatus 300 .

Specifically, the communication unit 105 may receive one or more images from the learning data design apparatus 200 or the artificial intelligence learning apparatus 300 .

Here, the image is an image to be subjected to annotation work for artificial intelligence (AI) learning. As such, according to the data processing plan designed by the learning data design device 200 or the artificial intelligence learning device 300 , the images are individually received, or a plurality of images are collectively received. can receive

The communication unit 105 may transmit the work result of the annotation to the learning data design apparatus 200 or the artificial intelligence learning apparatus 300 .

Here, the work result may include the coordinates of the bounding box set under the control of the operator and property information of the object. In addition, the work result may have a JSON file format, but is not limited thereto.

In addition, the communication unit 150 may receive a project property, an image property, or a worker property from the learning data design device 200 or the artificial intelligence learning device 300 .

Here, the properties of the project may include the learning purpose of the project related to the learning of artificial intelligence (AI), the learning period, the number of images required for learning, the properties of the object to be identified in the image, the bounding box setting rule, etc. However, the present invention is not limited thereto.

The properties of the image include the file name of the image, size (width, height), resolution, bit level, compression format, recording device name, exposure time, ISO sensitivity, focal length, aperture opening value, and location coordinates (GPS latitude, longitude) of the image. , shooting time, etc. may be included, but is not limited thereto.

The attributes of the worker may include, but are not limited to, the name of the worker, an identification number, the amount of work assigned, the cost according to the work, and evaluation of the work result.

In addition, the bounding box setting rule is a rule that the operator must follow in the process of setting the bounding box to the object in the image according to the properties of the project, the properties of the image, or the properties of the worker. Such a bounding box setting rule may include, but is not limited to, a minimum size value of a region, a maximum number of pixels spaced apart from an outline of an object, and the like.

With the following configuration, the input/output unit 110 may input a signal from an operator through a user interface (UI) or output the calculated result to the outside.

Here, the worker means a person who performs the annotation work. Such an operator may be referred to as a user, performer, labeler, or data labeler, but is not limited thereto.

Specifically, the input/output unit 110 may output an image to be annotated. The input/output unit 110 may receive a control signal for setting a bounding box from an operator. In addition, the input/output unit 110 may output a bounding box by overlaying it on the image.

Here, the bounding box is an area for specifying an object to be subjected to artificial intelligence (AI) learning among objects included in the image. As such, the bounding box may have a rectangular or polygonal shape, but is not limited thereto.

The input/output unit 110 may output a list of recommendation information that can be used as attribute information of an object. In addition, the input/output unit 110 may receive a control signal for setting object attribute information from an operator.

Here, the list of recommended information is a list of information that is composed of information that is highly likely to be included in the attribute information of the object and can be utilized as the attribute information of the object in the process of annotating by the operator.

In addition, the attribute information of the object is information for designating the attribute of the object to be the target of artificial intelligence (AI) learning. As such, the attribute information of the object includes information on the type of annotation, the class name, the classification item (tags), whether the object is truncated, the large classification, the small classification, or the upper level (instance upper). may, but is not limited thereto.

With the following configuration, the storage unit 115 may store data required for annotation work.

Specifically, the storage unit 115 may store the image received through the communication unit 105 . The storage unit 115 may store the properties of the project, the properties of the images, or the properties of the workers received through the communication unit 105 .

The storage unit 115 may temporarily store the position and size of the bounding box set according to the control signal input through the input/output unit 110 . The storage unit 115 may temporarily store the properties of the object input through the input/output unit 110 .

With the following configuration, the bounding box setting unit 120 may set a bounding box for specifying an object to be a target of artificial intelligence (AI) learning among objects included in the image.

The bounding box setting unit 120 may include a bounding box setting module 121 , an object identification module 123 , a bounding box correction module 125 , and an object mutual complementation module 127 .

The bounding box setting module 121 may set a partial region in an image to be annotated for learning, input through the input/output unit 110 , as a bounding box according to the control of the operator.

In addition, the bounding box setting module 121 sets, as the first bounding box, a partial area including one of a plurality of objects of an image to be annotated for artificial intelligence learning under the control of the operator, A partial region including an object included in the first bounding box and another object at least partially overlapping may be set as the second bounding box.

At this time, the bounding box setting module 121 receives two coordinates from the operator through the input/output unit 110, and a rectangle having the input two coordinates as the coordinates of the upper left vertex and the coordinates of the lower right vertex in the image. You can set the bounding box based on . In this case, the two coordinates may be set by the operator inputting one type of input signal twice (eg, clicking the mouse), or the operator entering two types of input signals once (eg, dragging the mouse) to set the two coordinates. However, the present invention is not limited thereto.

Alternatively, the bounding box setting module 121 may receive four coordinates from an operator through the input/output unit 110 and set the bounding box based on a line connecting the inputted four coordinates. In this case, the four coordinates may be set by the operator inputting one type of input signal four times (eg, clicking a mouse).

Also, the bounding box setting module 121 may receive a selection of a specific point in an image to be annotated for learning under the control of the operator. Here, the specific point may be one coordinate received from the operator through the input/output unit 11 . In addition, the bounding box setting module 121 may set a bounding box corresponding to a region including an object in the image based on a selected specific point. Specifically, the bounding box setting module 121 may form a circle having a preset radius value with a specific point as a center point, and move the circle under the control of the operator to set the selected area as the bounding box.

Here, the bounding box setting module 121 may expand the area around a specific point and set the corresponding area at the point in time when all the specific objects are included in the area as the bounding box. In this case, the bounding box setting module 121 may expand the area around a specific point according to the control of the operator, and set the specific area selected by the operator as the bounding box. Specifically, the bounding box setting module 121 forms a circle having a specific point as the center point, and sets the corresponding area at the point in time when all the specific objects are included in the circle as the bounding box while expanding the radius of the circle around the center point. can

Thereafter, the bounding box setting module 121 may select, within the first bounding box, a second specific point for specifying an object included in the first bounding box. That is, the bounding box setting module 121 may designate the second specific point again within the first bounding box area in order to more accurately recognize the object.

In addition, the bounding box setting module 212 may expand the area around the second specific point, and set the corresponding area at the point in time when all the specific objects are included in the area as the final bounding box. In this case, the second specific point may be set such that the background area excluding the object in the final bounding box is minimized.

Here, the bounding box setting module 121 may designate a specific point through a pre-trained machine learning model, and set a bounding box including an object around the designated specific point.

The object identification module 123 may identify an object that is completely included in the inner region of the bounding box set by the control of the operator or spans one or more line segments among a plurality of line segments constituting the bounding box.

Specifically, the object identification module 123 may identify all or a part of the object included in the area inside the bounding box set by the control of the operator. To this end, the object identification module 123 may perform image processing on an inner region of the bounding box.

To this end, the object identification module 123 may detect an RGB value for each pixel in the inner region of the bounding box, and extract an edge of the object through the detected RGB value. At this time, the object identification module 123 may detect the RGB value for each pixel in the inner region of the bounding box, and extract a point where the change amount of the RGB value with the adjacent pixel is higher than a preset threshold as an edge. .

For example, the object identification module 123 may divide the area inside the bounding box into three images according to RGB (Red, Green, Blue). The object identification module 123 may extract an edge of each image divided into three images (edge detection). In more detail, the object identification module 123 may use either a Laplacian of Gaussian (LoG) algorithm or a Difference of Gaussian (DoG) algorithm for edge extraction of each image.

When using the LoG algorithm, the object identification module 123 may use a Gaussian filter to remove noise existing in the image. The object identification module 123 may apply a Laplacian filter to the image from which the noise has been removed. Then, the object identification module 123 may extract an edge by detecting a zero crossing in the image to which the Laplacian filter is applied.

When using the DoG algorithm, the object identification module 123 generates two Gaussian masks having different variances from the image. The object identification module 123 subtracts another mask from the generated one mask. Then, the object identification module 123 may extract the edge by applying the subtracted mask to the image.

Here, the object identification module 123 may perform a background removal process of classifying an object and a background based on the extracted edge, deleting the background, recognizing only the object, and extracting only object information.

In addition, before extracting the object from the image, the object identification module 123 removes relatively transparent pixels to improve the boundary of objects existing in the image and then performs a process of smoothing the edge. To this end, a transparency channel is extracted from the target image, a region having a transparency value lower than a preset transparency threshold is removed, but the image can be pre-processed by blurring the edge of the object in the extracted transparency channel. . More specifically, the object identification module 123 extracts the alpha channel (transparency channel) from the RGBA image of the image, and displays the area above the transparency threshold as opaque, but makes the area lower than the transparency threshold completely transparent. can be converted Then, by blurring the edge of the object in the previously extracted alpha channel, it makes the subsequent object separation process easier. In terms of performance guarantee, the above process may be performed, for example, twice.

Thereafter, the object identification module 123 derives the probability that the important object belongs to each pixel by using salient object detection (SOD) as an intensity value, thereby extracting the object from the background in the target image. can be isolated and extracted. There are various algorithms for detecting important objects. Here, an example implemented using U2 net is introduced. First, the loss function for object extraction is expressed in Equation 1 below.

은 모델의 각 스테이지마다의 중요 맵 손실(saliency map loss)이고, (r, c)는 픽셀 좌표이고, (H, W)는 이미지의 높이와 너비이고, P_G는 픽셀의 ground truth 값이며, P_S는 중요 맵(saliency map)을 예측한 값을 나타낸다.here,

is the saliency map loss for each stage of the model, (r, c) is the pixel coordinate, (H, W) is the height and width of the image, P _G is the ground truth value of the pixel, P _S represents a predicted value of a saliency map.

In addition, the final loss function is expressed by Equation 2 below.

는 U2 net의 각 스테이지의 중요 맵을 업샘플링하고 1Х1 컨볼루션 및 시그모이드(sigmoid) 함수를 융합해 연산한 결과이고,

는 U2 net의 각 스테이지에서의 중요 맵 손실의 결과이고,

는 s_fuse의 손실(loss)이며,

, w_fuse는 각 스테이지에서의 가중치, 및 모든 스테이지를 융합한 가중치를 나타낸다.here,

is the result of upsampling the important map of each stage of U2 net and fusion of 1Х1 convolution and sigmoid function,

is the result of significant map loss in each stage of U2 net,

is the loss of s _fuse ,

, w _fuse represents a weight in each stage, and a weight obtained by merging all stages.

Meanwhile, the object identification module 123 may separate and extract the object from the background in the RGBA image based on the boundary value at which the object is not visible while imparting transparency to the preprocessed image.

In addition, the object identification module 123 may extract a transparency channel from the RGBA image and classify a pixel region having a transparency lower than a preset threshold value as a background.

The bounding box correction module 125 may correct the bounding box to the area of the object from which the background is deleted. That is, the bounding box correction module 125 may correct the bounding box according to the size of the recognized object in which the background is deleted from the image in which the object and the background are separated by the object identification module 123 .

Object mutual complementation module 127 when each bounding box is set in a plurality of objects overlapped by the bounding box setting module 121, extracts the object included in each bounding box, and extracts the object extracted from each bounding box can complement each other.

For example, the object mutual complementation module 127 may delete a part of the object extracted from the second bounding box from the object extracted from the first bounding box. Accordingly, the object mutual complementation module 127 may detect a plurality of closely related objects without duplication.

Also, the object mutual complementation module 127 may add a part of the object extracted from the second bounding box to the object extracted from the first bounding box. Accordingly, the object mutual complementation module 127 may add a part whose boundary is not closed or a missing part due to an erroneous creation of a bounding box from an object extracted from another bounding box.

With the following configuration, the object property setting unit 125 may set the property of the object specified by the bounding box. In particular, the object attribute setting unit 125 according to the present invention may provide a list of recommended information that can be preemptively utilized as attribute information of an object before an operator directly inputs attribute information.

First, the object property setting unit 125 may create a list of recommended information that can be used as property information of an object corresponding to the bounding box set by the bounding box setting unit 120 .

In one embodiment, the object property setting unit 125 selects information that is likely to be used as property information of an object according to a property of a project related to artificial intelligence (AI) learning set in advance, a property of an image, or a property of a worker. You can include it to create a list of recommended information.

In another embodiment, the object property setting unit 125 may calculate the average color, saturation, or brightness of pixels included in the bounding box set by the bounding box setting unit 120 . The object property setting unit 125 may select one recommended object type according to the calculated average color, saturation, or brightness based on a probability distribution diagram of color, saturation, or brightness configured for each type of object. In addition, the object property setting unit 125 may create a list of recommendation information by including information related to the type of the selected recommendation object.

In another embodiment, the object property setting unit 125 identifies the input frequency of the property information previously input by the worker from the worker's work log stored in the local storage of the annotation device 100 can do. In addition, the object attribute setting unit 125 may create a list of recommended information by including attribute information having a high input frequency based on the identified input frequency.

Next, the object property setting unit 125 may filter the list of pre-written recommendation information.

Specifically, the object property setting unit 125 is information related to the type of object that cannot be set by the work tool used in the process of setting the bounding box by the operator, among the information included in the list of pre-written recommendation information. (ie, information irrelevant to the object) can be removed.

In one embodiment, when the operator uses a tool capable of designating a rectangular area in the process of setting the bounding box, the object property setting unit 125 determines the type and Relevant information can be removed.

In another embodiment, when an operator uses a tool capable of designating a skeleton structure of an object in the process of setting a bounding box, the object property setting unit 125 may include an object having no skeleton in the list of pre-written recommendation information. It is also possible to remove information related to the type of

The object property setting unit 125 may sort information remaining after information irrelevant to an object is removed from the list of recommendation information.

In an embodiment, the object property setting unit 125 may sort information included in the list of recommendation information according to a preset alignment criterion based on a position or size occupied by the bounding box in the image.

In another embodiment, the object property setting unit 125 compares the basic shape provided in advance for each type of object and the shape of the object included in the bounding box, and is included in the list of recommended information based on the similarity of the two shapes. You can also sort information.

Next, the object property setting unit 125 may output the input/output unit 110 a list of the written and filtered recommendation information. In addition, the object property setting unit 125 may receive a control signal for setting object property information from an operator through the input/output unit 110 .

When one piece of information is selected from the list of recommended information under the control of the operator, the object property setting unit 125 may provide feedback according to the type of object corresponding to the selected information.

In one embodiment, the object property setting unit 125 reflects a color or transparency set differently depending on the type of object corresponding to the selected information, and a user interface related to an area inside a bounding box or an area inside the object. , UI) can be changed.

With the following configuration, the result generating unit 130 may generate a work result of the annotation and transmit it to the learning data design apparatus 200 or the artificial intelligence learning apparatus 300 .

Specifically, the result generating unit 130 determines the position and size of the enlarged or reduced bounding box under the control of the operator, and when the property information of the object is determined by the control of the operator based on the list of recommended information, the determined A work result of the annotation can be generated including coordinates and determined attribute information according to the location and size of the bounding box. Such a work result may have a JSON file format, but is not limited thereto.

In addition, the result generating unit 130 may transmit the generated annotation work result to the learning data design apparatus 200 or the artificial intelligence learning apparatus 300 through the communication unit 105 .

Hereinafter, hardware for implementing the logical components of the annotation apparatus 100 as described above will be described in more detail.

7 is a hardware configuration diagram of an annotation apparatus according to an embodiment of the present invention.

As shown in FIG. 7 , the annotation device 100 includes a processor 150 , a memory 155 , a transceiver 160 , an input/output device 165 , and a data bus Bus , 170) and storage (Storage, 175).

The processor 150 may implement the operations and functions of the annotation apparatus 100 based on an instruction according to the software 180a in which the annotation method is implemented, which resides in the memory 155 . The memory 155 may be loaded with software 180a in which the annotation method is implemented. The transceiver 160 may transmit and receive data to and from the learning data design apparatus 200 and the artificial intelligence learning apparatus 300 . The input/output device 165 may receive data necessary for the operation of the annotation device 100 , and may output an image, a bounding box, and a list of recommendation information. The data bus 170 is connected to the processor 150 , the memory 155 , the transceiver 160 , the input/output device 165 , and the storage 175 . can play a role.

The storage 175 may store an application programming interface (API), a library file, a resource file, and the like necessary for the execution of the software 180a in which the annotation method is implemented. The storage 175 may store the software 180b in which the annotation method is implemented. Also, the storage 175 may store information necessary for performing the annotation method. In particular, the storage 175 may include a database 185 that stores an image to be annotated, properties of a project, properties of an image, properties of workers, and work logs of workers.

According to an embodiment of the present invention, the software 180a, 180b for implementing the annotation method resident in the memory 155 or stored in the storage 175 is the processor 150 under the control of the operator, artificial intelligence ( Step of setting a partial area in the image to be subjected to annotation work for artificial intelligence (AI) learning as a bounding box, the processor 150 is the edge of the object in the inner area of the bounding box extracts, distinguishing an object and a background based on the extracted edge, and the processor corrects the bounding box to the area of the object from which the background has been deleted. It can be a computer program.

According to another embodiment of the present invention, the software 180a, 180b for implementing the annotation method resident in the memory 155 or stored in the storage 175 is the processor 150 under the control of the operator, artificial intelligence ( A step of setting a partial area including one of a plurality of objects in an image to be annotated for artificial intelligence (AI) learning as a first bounding box, the processor 150 , according to the control of the operator, setting a partial area including an object included in the first bounding box and another object at least partially overlapping with the second bounding box, and the processor 150, the first bounding box and the second It may be a computer program recorded on a recording medium to execute the steps of extracting each object included in the bounding box, and complementing each extracted object from the first bounding box and the second bounding box.

According to another embodiment of the present invention, the software 180a, 180b for implementing the annotation method resident in the memory 155 or stored in the storage 175 is performed by the processor 150 under the control of the operator, A step of selecting a specific point in an image to be subjected to an annotation operation for artificial intelligence (AI) learning and the processor 150 corresponds to an area including an object in the image centered on the selected specific point It may be a computer program recorded on a recording medium to execute the step of setting a bounding box.

More specifically, the processor 150 may include an application-specific integrated circuit (ASIC), another chipset, a logic circuit, and/or a data processing device. The memory 155 may include read-only memory (ROM), random access memory (RAM), flash memory, memory cards, storage media, and/or other storage devices. The transceiver 160 may include a baseband circuit for processing wired and wireless signals. The input/output device 165 includes an input device such as a keyboard, a mouse, and/or a joystick, and a liquid crystal display (LCD), an organic light emitting diode (OLED) and/or an input device such as a joystick. Alternatively, an image output device such as an active matrix OLED (AMOLED) may include a printing device such as a printer or a plotter.

When the embodiment included in this specification is implemented in software, the above-described method may be implemented as a module (process, function, etc.) that performs the above-described function. The module resides in the memory 155 and may be executed by the processor 150 . The memory 155 may be internal or external to the processor 150 , and may be connected to the processor 150 by various well-known means.

Each component shown in FIG. 7 may be implemented by various means, for example, hardware, firmware, software, or a combination thereof. In the case of implementation by hardware, an embodiment of the present invention provides one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), FPGAs ( Field Programmable Gate Arrays), a processor, a controller, a microcontroller, a microprocessor, etc. may be implemented.

In addition, in the case of implementation by firmware or software, an embodiment of the present invention is implemented in the form of a module, procedure, function, etc. that performs the functions or operations described above, and is stored in a recording medium readable through various computer means. can be recorded. Here, the recording medium may include a program command, a data file, a data structure, etc. alone or in combination. The program instructions recorded on the recording medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. For example, the recording medium includes a magnetic medium such as a hard disk, a floppy disk, and a magnetic tape, an optical recording medium such as a compact disk read only memory (CD-ROM), a digital video disk (DVD), and a floppy disk. magneto-optical media, such as a disk, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions may include high-level language codes that can be executed by a computer using an interpreter or the like as well as machine language codes such as those generated by a compiler. Such hardware devices may be configured to operate as one or more software to perform the operations of the present invention, and vice versa.

Hereinafter, the features of the artificial intelligence learning system according to various embodiments of the present invention as described above will be described in detail with reference to the drawings.

8 to 12 are exemplary views for explaining a process of setting a bounding box according to an embodiment of the present invention.

8 and 12 , the annotation apparatus 100 of the artificial intelligence learning system according to an embodiment of the present invention performs an annotation on each of a plurality of images for artificial intelligence (AI) learning, a bounding box By extracting the edge of the object from the inner region of It is possible to have a more uniform quality of the annotation work while reducing the work time.

For example, as shown in FIG. 8 , the annotation apparatus 100 may set an area including the object a2 as a bounding box by an operator. Thereafter, the annotation apparatus 100 may extract the edge of the object from the inner region of the bounding box, and distinguish the object a2 and the background a1 based on the extracted edge. And as shown in FIG. 9 , the annotation apparatus 100 may delete the background a1 from the bounding box and modify the bounding box to the area of the object a2 .

Also, referring to FIG. 10 , the annotation apparatus 100 of the artificial intelligence learning system according to another embodiment of the present invention performs annotation on each of a plurality of images for artificial intelligence (AI) learning. By setting each bounding box for a plurality of overlapping objects and complementing objects extracted from each bounding box, it is possible to minimize errors in extraction of a plurality of overlapping objects.

For example, as shown in FIG. 10 , the annotation apparatus 100 first creates a bounding box b1 including all overlapping objects, and after secondly creates a bounding box b2 for an object to be distinguished. , by deleting the object specified in the secondary bounding box b2 from the object specified in the primary bounding box b1, overlapping objects can be detected without duplication.

In addition, referring to FIGS. 11 and 12 , the annotation apparatus 100 of an artificial intelligence learning system according to another embodiment of the present invention performs annotation on each of a plurality of images for artificial intelligence (AI) learning, By setting a bounding box corresponding to an area including an object in an image with a focus on a specific point selected by the operator, the operator's work time can be reduced.

For example, as shown in FIG. 11 , the annotation apparatus 100 may receive a selection of a specific point p1 under the control of an operator. Thereafter, the annotation apparatus 100 may set a circular bounding box a1 including all objects based on a specific point. And, as shown in FIG. 12 , the annotation apparatus 100 specifies a specific point p2 within the firstly created bounding box a1 in order to clearly recognize the object, and secondly specifies a specific specific point p2 ), a bounding box a2 including all objects may be finally set.

13 is a flowchart illustrating an annotation method according to an embodiment of the present invention.

In describing the annotation method according to an embodiment of the present invention with reference to FIG. 13 , the same description as described with reference to FIGS. 8 to 12 will not be repeated.

Referring to FIG. 13 , the annotation apparatus 100 may receive coordinates for setting a partial region in an image to be annotated as a bounding box from an operator ( S100 ).

The annotation apparatus 100 may set a bounding box based on the coordinates input from the operator (S200). Meanwhile, specific embodiments of setting the bounding box will be described later.

The annotation apparatus 100 may input attribute information of the object specified by the bounding box (S300).

The annotation apparatus 100 determines the position and size of the enlarged or reduced bounding box by the control of the operator, and when the property information of the object is determined by the control of the operator based on the recommended information list, the position of the determined bounding box And it is possible to generate a work result of the annotation including the coordinates and the determined attribute information according to the size (S400). Such a work result may have a JSON file format, but is not limited thereto.

In addition, the annotation apparatus 100 may transmit a work result of the generated annotation to the learning data design apparatus 200 or the artificial intelligence learning apparatus 300 .

14 is a flowchart illustrating a method of setting a bounding box according to an embodiment of the present invention.

Referring to FIG. 14 , in the annotation apparatus 100 according to an embodiment of the present invention, a partial region in an image to be subjected to an annotation operation for artificial intelligence (AI) learning under the control of an operator. can be set as a bounding box (S211).

Next, the annotation apparatus 100 may extract an edge of the object from the inner region of the bounding box and classify the object and the background based on the extracted edge ( S213 ). In this case, the annotation apparatus 100 may detect an RGB value for each pixel in the inner region of the bounding box, and extract an edge of the object through the detected RGB value. At this time, the object identification module 123 may detect the RGB value for each pixel in the inner region of the bounding box, and extract a point where the change amount of the RGB value with the adjacent pixel is higher than a preset threshold as an edge. .

Also, the annotation apparatus 100 may extract a transparency channel from the RGBA image and classify a pixel area having a transparency lower than a preset threshold value as a background.

And the annotation apparatus 100 may correct the bounding box to the area of the object from which the background is deleted (S215).

15 is a flowchart illustrating a method of setting a bounding box according to another embodiment of the present invention.

Referring to FIG. 15 , the annotation apparatus 100 according to another embodiment of the present invention includes a plurality of images in an image to be subjected to an annotation operation for artificial intelligence (AI) learning under the control of an operator. A partial area including one of the objects may be set as a first bounding box (S221).

Next, the annotation apparatus 100 may set a partial region including an object included in the first bounding box and another object at least partially overlapping with the second bounding box under the control of the operator ( S223 ).

In addition, the annotation apparatus 100 may extract objects included in the first and second bounding boxes, respectively, and complement the objects respectively extracted from the first and second bounding boxes ( S225 ).

That is, the annotation apparatus 100 may delete a part of the object extracted from the second bounding box from the object extracted from the first bounding box. Accordingly, it is possible to detect a plurality of closely related objects without duplication.

Also, the annotation apparatus 100 may add a part of the object extracted from the second bounding box to the object extracted from the first bounding box. Accordingly, the boundary may not be closed, or a missing part due to misgeneration of the bounding box may be added from an object extracted from another bounding box.

16 is a flowchart illustrating a method of setting a bounding box according to another embodiment of the present invention.

Referring to FIG. 16 , the annotation apparatus 100 according to another embodiment of the present invention is a specific image in an image to be subjected to an annotation operation for artificial intelligence (AI) learning under the control of an operator. A point may be selected, and a bounding box corresponding to an area including an object in the image may be set around the selected specific point (S231).

In this case, the annotation apparatus 100 may form a circle having a specific point as a center point, and set a corresponding area at a viewpoint in which all specific objects are included in the circle as a bounding box while extending the radius of the circle around the center point.

Next, the annotation apparatus 100 selects, within the first bounding box, a second specific point for specifying an object included in the first bounding box ( S233 ).

In addition, the annotation apparatus 100 may expand the area around the second specific point, set the corresponding area at the point in time when all the specific objects are included in the area as the second bounding box, and set it as the final bounding box.

17 is a flowchart illustrating a security management method according to an embodiment of the present invention.

Referring to FIG. 17 , the security management device 500 collects security policies for annotation tasks of a plurality of AI learning systems ( S600 ). Here, each of the learning data design apparatus 200 may include a non-security area storing a public annotation work object set according to a security policy, and a security area storing a private annotation work object. In this case, the security area and the non-security area may be set in consideration of at least one of the properties of the annotation work, the number of annotation workers, and privacy-related laws of the country.

Next, the security management device 500 may receive a request for access for annotation work from the annotation device 100 to one of the learning data design devices 200 of the plurality of artificial intelligence learning systems 400 ( S700 ).

And, the security management device 500 may allow access to the annotation device 100 according to the security policy of the artificial intelligence learning system 400 that requires access (S800). In this case, the security management apparatus 500 may provide the public annotation object stored in the non-security area of the learning data design apparatus 200 that requires access to perform the annotation operation. In addition, the security management device 500 identifies the country of the annotation device that has requested access, based on the identification information of the annotation device 100 that has requested access, and accesses the artificial intelligence learning system 400 that can be accessed from the identified country. can allow

Thereafter, the annotation apparatus 100 may perform an annotation operation on the public annotation object stored in the non-security area of the learning data design apparatus 200 to which access is allowed.

As described above, although preferred embodiments of the present invention have been disclosed in the present specification and drawings, it is in the technical field to which the present invention pertains that other modifications based on the technical idea of the present invention are possible in addition to the embodiments disclosed herein. It is obvious to those with ordinary knowledge. In addition, although specific terms have been used in the present specification and drawings, these are only used in a general sense to easily describe the technical contents of the present invention and help the understanding of the present invention, and are not intended to limit the scope of the present invention. Accordingly, the foregoing detailed description should not be construed as restrictive in all respects but as exemplary. The scope of the present invention should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the present invention are included in the scope of the present invention.

Annotation devices: 100-1, 100-2, 100-3, … , 100-n; 100
Communication unit: 105 Input/output unit: 110
Storage unit: 115 Bounding box setting unit: 120
Object property setting unit: 125 Result generation unit: 130
Learning data design device: 200 Artificial intelligence learning device: 300
AI Learning System: 400 Security Management Unit: 500
Security Management System: 600

Claims (10)

setting, by an annotation device, a partial region in an image to be subjected to an annotation operation for artificial intelligence (AI) learning as a bounding box according to the control of the operator;
extracting, by the annotation device, an edge of an object from an inner region of the bounding box, and classifying an object and a background based on the extracted edge; and
Including, by the annotation device, modifying a bounding box to the area of the object from which the background has been deleted,
The step of classifying the background is
generating an RGB image by detecting RGB (red, green, blue) values for each pixel in an inner region of the bounding box;
changing the generated RGB channel to an RGBA (red, green, blue, alpha) image; and
Extracting an edge of an object based on transparency from the RGBA image,
The step of extracting the edge is
extracting a transparency channel from the RGBA image, removing a region having a transparency value lower than a preset transparency threshold value, and preprocessing the image by blurring an edge of the object from the extracted transparency channel; and
Separating and extracting the object from the background in the RGBA image by deriving the probability that the important object belongs to each pixel as an intensity value using salient object detection (SOD) Annotation method, characterized in that. The method of claim 1, wherein the setting of the bounding box comprises:
Annotation method, characterized in that the bounding box is set based on a rectangle having two coordinates input from the operator in the image as the coordinates of the upper-left vertex and the coordinates of the lower-right vertex. The method of claim 1, wherein the setting of the bounding box comprises:
Annotation method, characterized in that the bounding box is set based on a rectangle connecting the four coordinates input from the operator in the image. The method of claim 1, wherein the step of classifying the background comprises:
Annotation method, characterized in that the RGB value of each pixel is detected in the inner region of the bounding box, and the edge of the object is extracted through the detected RGB value. The method of claim 4, wherein the step of classifying the background comprises:
Annotation method, characterized in that the RGB value of each pixel is detected in the inner region of the bounding box, and a point where the change amount of the RGB value with respect to an adjacent pixel is higher than a preset threshold value is extracted as an edge. delete delete delete According to claim 1, wherein the step of extracting the edge,
Extracting a transparency channel from the RGBA image, and separating a pixel region having a transparency lower than a preset threshold value as a background, the annotation method. memory;
transceiver; and
In combination with a computing device configured to include a processor (processor) for processing instructions resident in the memory,
setting, by the processor, a partial region in an image to be subjected to an annotation operation for artificial intelligence (AI) learning as a bounding box under the control of the operator;
extracting, by the processor, an edge of an object from an inner region of the bounding box, and classifying an object and a background based on the extracted edge; and
Including, by the processor, modifying a bounding box to the area of the object from which the background has been deleted,
The step of classifying the background is
generating an RGB image by detecting RGB (red, green, blue) values for each pixel in an inner region of the bounding box;
changing the generated RGB channel to an RGBA (red, green, blue, alpha) image; and
Extracting an edge of an object based on transparency from the RGBA image,
The step of extracting the edge is
extracting a transparency channel from the RGBA image, removing a region having a transparency value lower than a preset transparency threshold, and preprocessing the image by blurring an edge of the object from the extracted transparency channel; and
Execute the step of separating and extracting the object from the background in the RGBA image by deriving the probability that the important object belongs to each pixel as an intensity value using salient object detection (SOD) In order to do so, a computer program recorded on a recording medium.

Images