KR102509343B1 - Method and system for analyzing layout of image - Google Patents

Abstract

An image layout analysis system according to an aspect according to the technical idea of the present disclosure includes an image acquisition module for acquiring an image including at least one class, and mask data for acquiring mask data designating a region to be analyzed among the acquired images. An acquisition module and a layout analysis module for predicting a class of a region designated by the mask data and a layout of the class among the acquired images, wherein the layout analysis module receives the image and the mask data as input, and a layout analysis network trained to output a prediction result of predicting the class and the layout based on the input image and the mask data.

Description

Layout analysis method and system of image {METHOD AND SYSTEM FOR ANALYZING LAYOUT OF IMAGE}

The technical idea of this disclosure relates to a method and system for analyzing the layout of an image.

When analyzing or processing an image (eg, document image, etc.) in which information of various classes such as text, table, picture, and photo is mixed, the analysis method or processing method may be different for each class in the image, and each class It may be important to accurately distinguish between . For example, the analysis or processing of the image may include indexing or searching in the image, object detection or recognition in the image, optical character recognition (OCR), data compression, and the like.

In order to distinguish each class in an image, accurate detection of a region (or boundary (layout)) corresponding to each class is required, and it may also be required to accurately identify each class.

In the conventional case, a method of detecting a layout or identifying a class based on a rule (algorithm, etc.) defined for distinguishing text or picture has been mainstream, but most of the rules correspond to generalized rules and thus a new pattern. Detection and identification accuracy for the class of may be low, and it may not be easy to improve or change rules according to the complexity of the system.

In addition, since most of the conventional methods express the layout of each class in a limited shape such as a rectangle, detection of the layout such that all areas of the corresponding class are not included in the layout or areas of other classes are included in the layout. Accuracy can be difficult to improve.

One problem to be solved by the present invention is to provide a method for more accurately detecting the layout (boundary) and class of a predetermined area from images (document images, etc.) having various classes such as text, table, picture, and photograph. is to do

In order to achieve the above object, an image layout analysis system according to an aspect according to the technical idea of the present disclosure includes an image acquisition module for acquiring an image including at least one class, and analysis of the acquired images a mask data acquisition module that acquires mask data specifying an area to be performed; and a layout analysis module that predicts a class of an area designated by the mask data and a layout of the class, among the acquired images, wherein the layout analysis module includes: includes a layout analysis network trained to receive the image and the mask data and output a prediction result of predicting the class and the layout based on the input image and the mask data.

Depending on the embodiment, the layout of the class may correspond to a boundary of an area occupied by a class existing in an area designated by the mask data in the image.

According to an embodiment, the layout analysis network outputs a plurality of coordinates corresponding to a plurality of points representing the layout of the class, and the layout analysis module determines the coordinates of the class based on the outputted plurality of coordinates. A layout generator generating layout information may be further included.

According to an embodiment, the layout generator may generate the layout information expressed in the form of a polygon connecting the plurality of points, and at least some of the plurality of points may correspond to vertices of the polygon.

According to an embodiment, the layout analysis system further includes a network learning module that performs learning of the layout analysis network based on the prediction result, and the network learning module includes the layout information, the correct answer layout information, and the predicted class. The layout analysis network may be updated based on information and correct answer class information.

According to an embodiment, the network learning module determines the layout analysis network based on at least one of an area of a difference set or an area of an intersection of a layout area corresponding to the layout information and a correct answer layout area corresponding to the correct answer layout information. Weights between nodes of the included neural network may be updated.

According to an embodiment, the at least one class represents a form of information included in the image, and the mask data represents a dot, line, or region representing at least one point corresponding to the region to be analyzed in the image. mask image may be included.

A layout analysis method of an image according to an aspect of the present disclosure includes obtaining an image including at least one class; Obtaining mask data designating a region to be analyzed among the acquired images; inputting the image and mask data to a layout analysis network trained to predict a class of a region designated by the mask data and a layout of the class among the acquired images; and acquiring a prediction result of predicting a class of a region designated by the mask data and a layout of the class in the input image through the layout analysis network.

The layout analysis method according to an embodiment of the present disclosure performs class identification and layout detection for a predetermined area in an image using a deep learning-based layout analysis network, thereby greatly reducing domain knowledge dependence in system design and providing accurate analysis results. can be obtained.

In addition, according to an embodiment of the present disclosure, the detected layout can be expressed in various shapes of polygons instead of a limited shape such as a rectangle, so that the layout detection accuracy can be maximized.

Effects according to the technical idea of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned can be clearly understood by those skilled in the art from the description below. There will be.

A brief description of each figure is provided in order to more fully understand the figures cited in this disclosure.
1 is a schematic block diagram of a layout analysis system in which a method for analyzing a layout of an image according to an exemplary embodiment of the present disclosure is implemented.
FIG. 2 is a block diagram illustrating an embodiment of a configuration included in the layout analysis module shown in FIG. 1 .
3 is a flowchart illustrating a method for analyzing a layout of an image according to an exemplary embodiment of the present disclosure.
FIG. 4 shows a document image as an example of an image acquired in the image acquisition step of FIG. 3 .
FIG. 5 shows examples of mask data acquired in the mask data acquisition step of FIG. 3 .
FIG. 6 shows an operation of the layout analysis network shown in FIG. 2 outputting a plurality of points related to layout and class information from image and mask data.
FIG. 7 illustrates an operation in which the layout generator shown in FIG. 2 generates layout information based on a plurality of points output from a layout analysis network.
8 is an exemplary diagram for explaining a learning operation of a layout analysis network included in a layout analysis module.
Fig. 9 is a schematic block diagram of a device performing a method for analyzing a layout of an image according to an exemplary embodiment of the present disclosure.

Exemplary embodiments according to the technical spirit of the present disclosure are provided to more completely explain the technical spirit of the present disclosure to those skilled in the art, and the following embodiments are modified in various forms. It may be, and the scope of the technical spirit of the present disclosure is not limited to the following embodiments. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the spirit of the invention to those skilled in the art.

Although terms such as first and second are used in this disclosure to describe various members, regions, layers, regions and/or components, these members, parts, regions, layers, regions and/or components do not refer to these terms. It is self-evident that it should not be limited by These terms do not imply any particular order, top or bottom, or superiority or inferiority, and are used only to distinguish one member, region, region, or component from another member, region, region, or component. Accordingly, a first member, region, region, or component to be described in detail below may refer to a second member, region, region, or component without departing from the teachings of the technical concept of the present disclosure. For example, a first element may be termed a second element, and similarly, the second element may be termed a first element, without departing from the scope of the present disclosure.

Unless defined otherwise, all terms used herein, including technical terms and scientific terms, have the same meaning as commonly understood by those of ordinary skill in the art to which the concepts of the present disclosure belong. In addition, commonly used terms as defined in the dictionary should be interpreted as having a meaning consistent with what they mean in the context of the technology to which they relate, and in an overly formal sense unless explicitly defined herein. will not be interpreted.

When an embodiment is otherwise embodied, a specific process sequence may be performed differently from the described sequence. For example, two processes described in succession may be performed substantially simultaneously, or may be performed in an order reverse to the order described.

In the accompanying drawings, variations of the shapes shown may be expected, eg depending on manufacturing techniques and/or tolerances. Therefore, embodiments according to the technical idea of the present disclosure should not be construed as being limited to the specific shape of the region shown in the present disclosure, and should include, for example, changes in shape resulting from the manufacturing process. The same reference numerals are used for the same components in the drawings, and duplicate descriptions thereof are omitted.

The term 'and/or' as used herein includes each and every combination of one or more of the recited elements.

Hereinafter, embodiments according to the technical idea of the present disclosure will be described in detail with reference to the accompanying drawings.

1 is a schematic block diagram of a layout analysis system in which a method for analyzing a layout of an image according to an exemplary embodiment of the present disclosure is implemented. FIG. 2 is a block diagram illustrating an embodiment of a configuration included in the layout analysis module shown in FIG. 1 .

Referring to FIG. 1 , in a system 100 (hereinafter referred to as a 'layout analysis system') in which a method for analyzing a layout of an image according to an embodiment of the present disclosure is implemented, various classes such as text, table, picture, and photo are mixed. A layout of a class corresponding to a predetermined point may be detected from an image (eg, a document image) and an analysis operation for identifying the corresponding class may be performed. The class may mean a property or form of data (or information).

The layout analysis system 100 may include at least one computing device. For example, each of the at least one computing device corresponds to a hardware-based device including a processor, memory, communication interface, input unit, and/or output unit. In this case, components included in the layout analysis system 100 may be implemented as hardware, software, or a combination thereof, and may be implemented by being integrated or divided into the at least one computing device.

The layout analysis system 100 according to an embodiment of the present disclosure may include an image acquisition module 110, a mask data acquisition module 120, a layout analysis module 130, and a network learning module 140.

The image acquisition module 110 may acquire an analysis target image for performing layout detection and class identification, or may acquire a learning image for a learning operation of the layout analysis module 130 to be described later. The image to be analyzed or the image for learning may correspond to an image (eg, a document image) in which various classes such as text, table, picture, and photo are mixed.

The image acquisition module 110 may acquire the analysis target image or training image from a user or manager through an input means or interface included in the layout analysis system 100 . Depending on the embodiment, the image acquisition module 110 may acquire the analysis target image or training image from a device, server, database, etc. connected through a network.

The mask data acquisition module 120 may acquire mask data designating a region to be analyzed among the acquired images. The mask data may include at least one point included in the region to be analyzed. For example, as described later in FIG. 5 , the mask data may be provided in the form of a mask image in which a point, line, or region representing the at least one point is expressed, or may include coordinate values of the at least one point. . The mask data may be input from a user or the like through an input means included in the layout analysis system 100, but is not limited thereto.

The layout analysis module 130 analyzes the acquired image and mask data, and includes class information identifying a class of a region designated by the mask data among the images, and layout information about a region where the class exists. The prediction result (analysis result) can be output.

A layout analysis method according to an embodiment of the present disclosure may perform class prediction of a region designated by the mask data and/or layout prediction of the class using a layout analysis network learned based on deep learning. can In addition, according to the present layout analysis method, when the layout of the area where the predicted class exists is indicated, restrictions on the layout form can be minimized. Specifically, the layout analysis method of the present disclosure is implemented to represent a layout not only in a limited shape such as a rectangle but also in an atypical polygon (see the embodiments of FIGS. 6 to 8), thereby improving the detection accuracy of the layout. can be maximized.

Referring to FIG. 2 in this regard, the layout analysis module 130 may include a layout analysis network 132 .

When the image obtained by the image acquisition module 110 and the mask data acquired by the mask data acquisition module 120 are input, the layout analysis network 132 predicts the layout of the region designated by the mask data among the images. Results and class prediction results of the region may be output. The area designated by the mask data in the image may mean an area occupied by a class (text, table, picture, photo, etc.) located at at least one point (coordinate value, etc.) included in the mask data.

The layout analysis network 132 may be implemented through deep learning-based learning, and this layout analysis network 132 may include a neural network structure. For example, the layout analysis network 132 may include, but is not limited to, a convolutional neural network (CNN). For example, the CNN includes at least one convolutional layer and a pooling layer that extracts features from an input image and forms a feature map, and applies the extracted features to a neural network to select the regions in the image. It may include a plurality of points representing the layout of classes existing in a region designated by mask data, and a fully-connected layer and a softmax layer outputting class information of the designated region.

The plurality of points may be included in a plurality of coordinates indicating a layout of a class existing in an area designated by the mask data in the image. For example, at least some of the plurality of points may correspond to vertices of the polygonal layout. The layout may mean a boundary between a class existing in the designated area and another class adjacent to the class. In other words, the layout may mean a boundary of an area occupied by a class existing in the designated area in the image.

Depending on the embodiment, the layout analysis module 130 may further include a layout generator 134. The layout generator 134 may generate layout information from a plurality of points output from the layout analysis network 132 . For example, the layout generator 134 may generate layout information having a polygonal shape by connecting the plurality of points based on a technique such as orthographic projection. In this case, the polygon may be generated to include all regions where the class exists.

The layout information includes information on coordinate values included in a layout (polygon) among coordinate values (pixel coordinates, etc.) included in an image, or each coordinate value corresponds to the predicted layout (or internal area of the layout). It may include information on whether or not it is included.

The class information may indicate a class (eg, text, table, picture, photo, etc.) of a region corresponding to the mask data in the image.

The network learning module 140 performs the layout analysis network 132 on the basis of the layout information and class information predicted by the layout analysis module 130 and the previously provided ground truth layout information and correct answer class information. learning can be performed.

For example, the network learning module 140 may include an objective function for learning the layout analysis network 132 . The objective function may be implemented to perform learning of the layout analysis network 132 by reflecting the difference between the predicted layout area and the layout area according to the correct answer layout information and the difference between the predicted class information and the correct answer class information. can For example, the objective function is the area of the difference between the predicted layout area and the correct layout area, the square of the area of the difference, the area of the intersection and the area of the union (eg, 1-(area of the intersection)/(of the union) It can be implemented to perform learning based on area)), etc. In addition, the objective function may be implemented to learn the layout analysis network 132 in a direction in which cross-entropy loss based on the predicted class information and the correct answer class information is minimized. However, since the implementation method of the objective function is not limited to the above-described example, the objective function may be implemented in various known ways for learning the layout analysis network 132 .

On the other hand, 'learning' used in this specification may have the same meaning as learning and training (or training), and the meaning of performing learning means that the network performs learning or the network learns both can include

3 is a flowchart illustrating a method for analyzing a layout of an image according to an exemplary embodiment of the present disclosure. FIG. 4 shows a document image as an example of an image acquired in the image acquisition step of FIG. 3 . FIG. 5 shows examples of mask data acquired in the mask data acquisition step of FIG. 3 . FIG. 6 shows an operation of the layout analysis network shown in FIG. 2 outputting a plurality of points related to layout and class information from image and mask data. FIG. 7 illustrates an operation in which the layout generator shown in FIG. 2 generates layout information based on a plurality of points output from a layout analysis network. 8 is an exemplary diagram for explaining a learning operation of a layout analysis network included in a layout analysis module.

Referring to FIG. 3 , the layout analysis method may include obtaining an image (S300) and acquiring mask data corresponding to a region to be analyzed among the acquired images (S310).

As described above, the layout analysis system 100 may obtain an analysis target image or an image for learning to perform layout detection and class identification. The image may be acquired from a user or the like through various input means or interfaces, or may be acquired from a device, server, or the like connected through a network.

The image may correspond to an image in which various classes such as text, table, picture, and photo are mixed. Referring to the document image 400 of FIG. 4 as an example of the image, the document image 400 includes a first text area 401, a second text area 402, a third text area 403, and a fourth text area. It can be divided into an area 404 and a picture area 405 . Each of the regions 401, 402, 403, 404, and 405 may be distinguished from each other according to a predetermined boundary (layout). Depending on embodiments, some of the regions may overlap each other.

Meanwhile, the layout analysis system 100 may obtain mask data designating a region to be analyzed among the acquired images. As described above, a mask included in the mask data may include at least one point (coordinates, etc.) for designating the region to be analyzed, and the mask may be input by a user or the like.

Referring to FIG. 5 with respect to examples of the mask data, a user or the like may input a mask for designating a region to be analyzed using an input means (eg, a touch panel, a touch pen, a mouse, etc.). For example, a user inputs a dotted mask 502 as shown in (a) of FIG. 5, and the layout analysis system 100 generates mask data 500 including the input mask 502. can Depending on the embodiment, the layout analysis system 100 generates mask data 510 including a mask 512 input in the form of a line, as shown in (b) of FIG. 5, which is input in various forms. Mask data including a mask may be generated.

Figure 3 will be described again.

The layout analysis method may include predicting a layout and class of a region corresponding to the mask data (region designated by the mask data) among the acquired images (S320).

Referring to FIG. 6 together, the layout analysis system 100 may input an acquired image (eg, a document image 400) and mask data 500 to the layout analysis network 132. According to an embodiment, the layout analysis system 100 obtains information (coordinate values, etc.) on at least one point based on the mask included in the mask data 500, and transfers the obtained information to the layout analysis network 132. can also be entered as

The layout analysis network 132 may be trained to analyze the image 400 based on a region designated by the mask data 500 among the input images 400 and output prediction results for layouts and classes. As described above, the layout analysis network 132 may be implemented as a CNN or the like. In the present specification, it has been described that the layout analysis network 132 is composed of one network and outputs prediction results for layouts and classes, but according to an embodiment, the layout analysis network 132 outputs prediction results for layouts. It may be divided into a first network and a second network that outputs a prediction result for a class.

The layout analysis network 132 may analyze the image 400 based on a region designated by the mask data 500 in the image 400 . Based on the analysis result, the layout analysis network 132 may output a plurality of points 610 representing the layout of the class existing in the designated area and class information 620 as the prediction result. As described above, the plurality of points 610 may be included in a plurality of coordinates representing the layout of the class. At least some of the plurality of points 610 may correspond to vertices of the polygonal layout. Meanwhile, the layout analysis network 132 may output a preset number of points, and the number of points may be set to 5 or more. Accordingly, the layout analysis system 100 may represent the layout of the class as various polygons of atypical shapes, rather than limited shapes such as rectangles.

Referring to FIG. 7 , the layout analysis system 100 (for example, the layout generator 134) may generate layout information 710 based on a plurality of points 610 output from the layout analysis network 132. . As described above, the layout analysis system 100 may generate polygonal layout information 710 by connecting a plurality of points 610 based on a technique such as orthogonal projection. The layout information 710 may include information on coordinate values included in the detected layout of the image 400 or information on whether each coordinate value of the image 400 is included in the layout. .

According to an embodiment, the layout analysis method may include outputting layout information and class information based on prediction results (S330).

The layout analysis system 100 may output the generated layout information 710 and class information 620 through an output means (display, etc.). Depending on the embodiment, the layout analysis system 100 may transmit the generated layout information 710 and class information 620 to other connected devices through a communication interface.

According to an embodiment, the layout analysis method may include performing learning of a layout analysis network based on a prediction result (S340).

Referring to FIG. 8 together, the layout analysis system 100 may analyze the training image 800 and the mask data 510 to output layout and class prediction results. The layout analysis system 100 may perform learning of the layout analysis module 130 or the layout analysis network 132 based on the output prediction result, the correct answer layout, and the correct answer class.

The network learning module 140 performs the layout analysis network 132 based on the predicted layout information 810 or 820 and the predicted class information 825, the correct answer layout information 830 or 840, and the correct answer class information 850. of learning can be performed. The predicted layout information 810 or 820 may be information related to the layout (polygon area) generated by the layout generator 134, but depending on the embodiment, a plurality of points output from the layout analysis network 132 and You may also include relevant information.

As described above in FIG. 1 , the network learning module 140 may include an objective function for learning the layout analysis network 132 . The objective function may be implemented to perform learning of the layout analysis network 132 by reflecting the difference between the predicted layout area and the layout area according to the correct answer layout information and the difference between the predicted class information and the correct answer class information. can For example, the objective function is the area of the difference between the predicted layout area and the correct layout area, the square of the area of the difference, the area of the intersection and the area of the union (eg, 1-(area of the intersection)/(of the union) It can be implemented to perform learning based on area)), etc. In addition, the objective function may be implemented to learn the layout analysis network 132 in a direction in which cross-entropy loss based on the predicted class information and the correct answer class information is minimized.

Based on the objective function, the layout analysis system 100 may perform learning of the layout analysis network 132 by changing (updating) weights between nodes constituting the neural network of the layout analysis network 132 .

The layout analysis method according to an embodiment of the present disclosure performs class identification and layout detection for a predetermined area in an image using a deep learning-based layout analysis network, thereby greatly reducing domain knowledge dependence in system design and providing accurate analysis results. can be obtained.

In addition, according to an embodiment of the present disclosure, the detected layout can be expressed in various shapes of polygons instead of a limited shape such as a rectangle, so that the layout detection accuracy can be maximized.

Fig. 9 is a schematic block diagram of a device performing a method for analyzing a layout of an image according to an exemplary embodiment of the present disclosure.

Referring to FIG. 9 , a device 900 according to an embodiment of the present disclosure may correspond to any one of at least one computing device constituting the layout analysis system 100 described above in FIG. 1 . In this case, the device 900 may correspond to a device that performs a layout analysis operation and/or a learning operation of the image (document image, etc.) described above with reference to FIGS. 3 to 8 .

Such a device 900 may include a processor 910 and a memory 920 . However, the components of the device 900 are not limited to the above examples. For example, the device 900 may include more or fewer components than the aforementioned components. Also, there may be at least one processor 910 and at least one memory 920 . Also, two or more of the processor 910 and the memory 920 may be combined into a single chip.

According to an embodiment, the processor 910 corresponds to at least one of the above-described image acquisition module 110, mask data acquisition module 120, layout analysis module 130, and network learning module 140, or the At least one of the modules may be executed or controlled.

The processor 910 includes hardware such as a CPU, an application processor (AP), an integrated circuit, a microcomputer, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), and/or a neural processing unit (NPU). can do.

According to an embodiment of the present disclosure, the memory 920 may store programs and data necessary for the operation of the device 900 .

Also, the memory 920 may store at least one of data generated or obtained through the processor 910 . According to an embodiment, the memory 920 stores data, commands, algorithms, etc. related to the image acquisition module 110, the mask data acquisition module 120, the layout analysis module 130, and/or the network learning module 140. can be saved

The memory 920 may include a storage medium such as ROM, RAM, flash memory, SSD, HDD, or a combination of storage media.

The descriptions of the above embodiments are only examples with reference to the drawings for a more thorough understanding of the present disclosure, and should not be construed as limiting the technical spirit of the present disclosure.

In addition, it will be clear to those skilled in the art that various changes and modifications are possible within a range that does not deviate from the basic principles of the present disclosure.

Claims (14)

In the image layout analysis system,
an image acquisition module that acquires an image including at least one class;
a mask data acquisition module that obtains mask data designating a region to be analyzed among the acquired images; and
Among the obtained images, a layout analysis module predicts a class of a region designated by the mask data and a layout of the class;
The layout analysis module,
A layout analysis network trained to receive the image and the mask data and output a prediction result of predicting the class and the layout based on the input image and the mask data,
The layout analysis network,
outputs a plurality of coordinates corresponding to a plurality of points representing the layout of the class;
The layout analysis module,
Further comprising a layout generator for generating layout information of the class based on the outputted plurality of coordinates;
The layout generator,
Generating the layout information expressed in the form of a polygon connecting the plurality of points,
layout analysis system. According to claim 1,
The layout of the class is,
Of the image, corresponding to the boundary of the region occupied by the class existing in the region designated by the mask data,
layout analysis system. delete According to claim 1,
At least some of the plurality of points correspond to vertices of the polygon,
layout analysis system. According to claim 1,
Further comprising a network learning module for performing learning of the layout analysis network based on the prediction result;
The network learning module,
Updating the layout analysis network based on the layout information, the correct answer layout information, and the predicted class information and correct answer class information;
layout analysis system. According to claim 5,
The network learning module,
A weight between nodes of a neural network included in the layout analysis network is updated based on at least one of the area of the difference set or the area of intersection of the layout area corresponding to the layout information and the correct answer layout area corresponding to the correct answer layout information. doing,
layout analysis system. According to claim 1,
The at least one class represents a type of information included in the image,
The mask data,
Including a mask image in which a point, line, or region representing at least one point corresponding to the region to be analyzed is expressed in the image,
layout analysis system. obtaining an image including at least one class;
Obtaining mask data designating a region to be analyzed among the acquired images;
inputting the image and mask data to a layout analysis network trained to predict a class of a region designated by the mask data and a layout of the class among the acquired images; and
Obtaining, through the layout analysis network, a prediction result of predicting a class of a region designated by the mask data in the input image and a layout of the class;
Obtaining the prediction result,
obtaining, from the layout analysis network, a plurality of coordinates corresponding to a plurality of points representing the layout of the class; and
Acquiring a prediction result of a layout represented by a polygon connecting a plurality of obtained coordinates,
How to analyze the layout of an image. According to claim 8,
The layout of the class is,
Corresponding to the boundary of the region occupied by the class existing in the region designated by the mask data, among the at least one class included in the image,
How to analyze the layout of an image. delete According to claim 8,
At least some of the plurality of points correspond to vertices of the polygon,
How to analyze the layout of an image. According to claim 8,
Further comprising the step of performing learning of the layout analysis network based on the prediction result,
The step of performing the learning is,
Updating the layout analysis network based on predicted classes and predicted layouts, correct answer class information and correct answer layout information,
How to analyze the layout of an image. According to claim 12,
Updating the layout analysis network,
Updating a weight between nodes of a neural network included in the layout analysis network based on an area of a difference between a layout area corresponding to the predicted layout and a correct answer layout area corresponding to the correct answer layout information ,
How to analyze the layout of an image. According to claim 8,
The at least one class includes at least one of text, picture, photo, figure, and table,
The mask data,
Including a mask image in which a point, line, or region representing at least one point corresponding to the region to be analyzed is expressed,
How to analyze the layout of an image.

Images