KR20240003078A - Apparatus for automatic creation of korean landscape painting and control method thereof - Google Patents

Abstract

The present invention relates to a device for automatically generating Korean landscape paintings and a control method thereof. A method of controlling a Korean landscape painting modeling device according to the present invention includes receiving Korean landscape painting image data; extracting ridge lines from each of the received Korean landscape painting image data; Constructing a primary artificial intelligence model by performing primary machine learning using the extracted ridge line and the Korean landscape painting image data; generating temporary output image data by inputting the extracted ridge line into the first artificial intelligence model; It is characterized in that it includes the step of constructing a secondary artificial intelligence model by performing secondary machine learning using the temporary output image data and the Korean landscape painting image data.

Description

Device for automatically generating Korean landscape paintings and its control method {APPARATUS FOR AUTOMATIC CREATION OF KOREAN LANDSCAPE PAINTING AND CONTROL METHOD THEREOF}

The present invention relates to a device for automatically generating Korean landscape paintings and a method of controlling the same. More specifically, the present invention relates to an apparatus for automatically generating Korean landscape paintings and a method for controlling the same. More specifically, after building an artificial intelligence model through machine learning, the artificial intelligence model is used to automatically generate Korean landscape paintings according to user input. It relates to a device that performs and a method of controlling it.

Recently, various technologies for automatically generating specific images through machine learning have been proposed.

For example, through image learning of specific animals, the animal's shape can be automatically generated, a human face that does not exist in the world can be generated, or an abstract image can be generated.

In order to create new images through machine learning, a proper artificial intelligence model must be built through machine learning.

For example, when performing machine learning through deep learning, after placing the input layer, hidden layer, and output layer, related learning images are input and learned while modifying the parameters included in the hidden layer to create a proper artificial intelligence model. In this case, learning data is very important.

In other words, if learning data is not appropriately selected or input, the constructed artificial intelligence model will not operate properly and the results will be unsatisfactory.

Meanwhile, Korean landscape paintings have a unique painting style that is different from the landscape paintings of other countries.

In particular, Korean landscape painting has a unique painting style that anyone with knowledge in the related field can easily distinguish it from Chinese landscape painting.

However, while a relatively large amount of Chinese-style landscape paintings remain, the remaining amount of Korean landscape paintings is small, and the amount of additional production is not large, so it is not desirable to perform machine learning on them in the conventional manner.

For example, 1,000 images of Korean landscape paintings are needed to improve the performance of machine learning, but if there are only 100 images of Korean landscape paintings that currently exist, a high-performance artificial intelligence model cannot be built even if machine learning is performed on these alone. will be.

Therefore, there is a request for a method to build an artificial intelligence model for generating Korean landscape paintings with excellent performance even with an insufficient number of Korean landscape paintings.

Public Patent No. 10-2021-0097314

The present invention was developed to meet the above-described conventional requests, and is a device for constructing an artificial intelligence model for generating Korean landscape paintings with excellent performance even from an insufficient number of Korean landscape painting images and using this to generate new Korean landscape painting images, and the same. It provides a control method.

In order to achieve the above object, a control method of a Korean landscape painting modeling device according to the present invention includes the steps of receiving Korean landscape painting image data; extracting ridge lines from each of the received Korean landscape painting image data; Constructing a primary artificial intelligence model by performing primary machine learning using the extracted ridge line and the Korean landscape painting image data; generating temporary output image data by inputting the extracted ridge line into the first artificial intelligence model; It may include constructing a secondary artificial intelligence model by performing secondary machine learning using the temporary output image data and the Korean landscape painting image data.

Here, the step of selecting a region of interest based on each coordinate on the extracted ridge line; extracting a crop image from the Korean landscape painting image data corresponding to the selected region of interest; It may include constructing the first artificial intelligence model by performing machine learning using the extracted cropped image.

Here, a region of interest can be selected that includes each coordinate on the extracted ridge line and ensures that the ratio of the area above the ridge line and the area below the ridge line falls within a preset range.

Here, the steps include selecting coordinate groups having preset equal intervals on the extracted ridge line; It may include the step of selecting a square region of interest of the minimum size using each coordinate in the selected coordinate group as the center point and ensuring that the ratio of the area above the ridge line to the area below the ridge line is 3:7.

Here, for the specific coordinates in the selected coordinate group, if it is not possible to select a square region of interest with the specific coordinates as the center point and the ratio of the area above the ridge line and the area below the ridge line to be 3:7, the specific coordinates A square region of interest of the minimum size that is moved 20% in the vertical direction from the center point can be selected.

Here, for the coordinates included in the mountain peak area among the coordinates in the selected coordinate group, a square region of interest with the minimum size in which the corresponding coordinates are shifted 20% in the vertical direction upward from the center point can be selected.

Here, the step includes converting each of the received Korean landscape painting image data into grayscale; performing Gaussian blur filter processing on the grayscale converted image data for blurring to reduce the noise ratio; Converting the Gaussian blur filter-processed image to black and white based on a preset reference setting value; The method may include extracting a ridge line by applying a preset edge detection algorithm to the generated black-and-white converted image.

In addition, in order to achieve the above object, a method of controlling an apparatus for automatically generating Korean landscape paintings according to the present invention includes the steps of receiving Korean landscape painting image data; extracting ridge lines from each of the received Korean landscape painting image data; Constructing a primary artificial intelligence model by performing primary machine learning using the extracted ridge line and the Korean landscape painting image data; generating temporary output image data by inputting the extracted ridge line into the first artificial intelligence model; Constructing a secondary artificial intelligence model by performing secondary machine learning using the temporary output image data and the Korean landscape painting image data; After the above, when line image data is received from the user, the received line image data is input into the first artificial intelligence model, and the resulting image data is then input again into the second artificial intelligence model to create a Korean landscape painting image. It may include an automatic generation step.

In addition, in order to achieve the above object, a Korean landscape painting modeling device according to the present invention includes a learning data receiver that receives Korean landscape painting image data; a ridge extraction unit that extracts ridge lines from each Korean landscape painting image data received by the learning data receiver; a first machine learning unit that constructs a first artificial intelligence model by performing first machine learning using the ridge line extracted from the ridge extraction unit and the Korean landscape painting image data; a temporary output unit that generates temporary output image data by inputting the ridge line extracted from the ridge extraction unit into the first artificial intelligence model; It may include a second machine learning unit that builds a secondary artificial intelligence model by performing secondary machine learning using the temporary output image data of the temporary output unit and the Korean landscape painting image data.

Here, the first machine learning unit includes a region of interest selection unit that selects a region of interest based on each coordinate on the ridge line extracted from the ridge extraction unit; a partial image extraction unit for extracting a cropped image of the Korean landscape painting image data corresponding to the area of interest selected by the area of interest selection unit; It may include a first artificial intelligence model building unit that builds a first artificial intelligence model by performing machine learning using the cropped image extracted from the partial image extraction unit.

Here, the region of interest selection unit may select a region of interest that includes each coordinate on the ridge line extracted by the ridge extraction unit and ensures that the ratio of the area above the ridge line and the area below the ridge line falls within a preset range.

Here, the region of interest selection unit selects a group of coordinates having the same preset interval on the ridge line extracted from the ridge extraction unit, sets each coordinate in the selected coordinate group as the center point, and sets the area above the ridge line and the area below the ridge line. A square region of interest can be selected with the minimum size such that the ratio is 3:7.

Here, the region of interest selection unit may select a square region of interest for a specific coordinate in the selected coordinate group, with the specific coordinate as the center point and having the ratio of the area above the ridge line and the area below the ridge line be 3:7. If there is none, a square region of interest with the minimum size where the specific coordinates are shifted 20% in the vertical direction from the center point can be selected.

Here, the region of interest selection unit may select a square region of interest with the minimum size in which the corresponding coordinates are shifted 20% in the vertical direction upward from the center point for coordinates included in the mountain peak area among the coordinates in the selected coordinate group.

Here, the ridge extractor converts each Korean landscape painting image data received from the learning data receiver into grayscale, and performs Gaussian blur filter processing on the grayscale converted image data for blur processing to reduce the noise ratio, , After converting the Gaussian blur filter-processed image to black and white based on a preset standard setting value, a ridge line can be extracted by applying a preset edge detection algorithm.

In addition, in order to achieve the above object, an apparatus for automatically generating Korean landscape paintings according to the present invention includes a learning data receiver that receives Korean landscape painting image data; a ridge extraction unit that extracts ridge lines from each Korean landscape painting image data received by the learning data receiver; a first machine learning unit that constructs a first artificial intelligence model by performing first machine learning using the ridge line extracted from the ridge extraction unit and the Korean landscape painting image data; a temporary output unit that generates temporary output image data by inputting the ridge line extracted from the ridge extraction unit into the first artificial intelligence model; a second machine learning unit that constructs a secondary artificial intelligence model by performing secondary machine learning using the temporary output image data of the temporary output unit and the Korean landscape painting image data; After the secondary artificial intelligence model is built by the second machine learning unit, when line image data is received from the user, the received line image data is input into the first artificial intelligence model and the resulting image data is It may include a Korean landscape painting image generator that automatically generates a Korean landscape painting image by inputting it back into the secondary artificial intelligence model.

As described above, according to the present invention, it is possible to secure a significant amount of Korean landscape painting images for machine learning with just a small amount of Korean landscape painting images, and using this, the corresponding Korean landscape painting image is automatically generated when the user draws a simple line. can be created.

1 is a schematic configuration diagram of the entire system including a device for automatically generating Korean landscape paintings according to an embodiment of the present invention;
Figure 2 is a functional block diagram of the automatic Korean landscape painting generation device of Figure 1;
Figures 3 to 11 are diagrams to explain the process of using or generating the Korean landscape painting automatic generation device of Figure 1 in the process of performing its function;
Figure 12 is an overall control flow diagram of an automatic Korean landscape painting generation device according to an embodiment of the present invention;
Figure 13 is a diagram for explaining the processing performed by the automatic Korean landscape painting generation device according to another embodiment of the present invention.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

Each embodiment according to the present invention below is only an example to aid understanding of the present invention, and the present invention is not limited to these embodiments. In particular, the present invention may be comprised of a combination of at least one of the individual components, individual functions, or individual steps included in each embodiment.

In particular, for convenience, alphabet letters such as '(a)' are included in some claims, but these alphabet letters do not specify the order of each step.

A schematic configuration of the entire system including the automatic Korean landscape generation device 100 according to an embodiment of the present invention is shown in FIG. 1.

In the drawing, the user terminal 200 inputs basic data for machine learning according to the user's operation or inputs line image data to automatically generate a Korean landscape painting image as described later, for example, provided in the home. It may be a personal computer, or furthermore, it may be a portable wireless communication terminal such as a smartphone.

The Korean landscape painting automatic generation device 100 builds an artificial intelligence model through machine learning at the request of the user terminal 200, and also has a function of automatically generating Korean landscape painting images at the request of the user terminal 200. Perform.

To this end, the Korean landscape painting automatic generation device 100 includes a Korean landscape painting modeling device that builds a model for Korean landscape painting by conducting learning using learning data, and a Korean landscape painting model created in this way to create a Korean landscape painting model on a line entered by the user. It can be functionally divided into a model application device that generates a corresponding Korean landscape painting image. Hereinafter, the integrated state will be explained as an example.

The specific functional blocks of the automatic Korean landscape painting generation device 100 according to an embodiment of the present invention are as shown in FIG. 2.

As shown in the figure, the Korean landscape painting automatic generation device 100 includes a receiving unit 110, a ridge extraction unit 120, a first machine learning unit 130, a temporary output unit 140, and a second machine learning unit ( 150), and may be configured to include a Korean landscape painting image generator 160.

Here, the receiving unit 110 performs a function of receiving various data received from the user terminal 200.

For example, the receiver 110 can receive image data of Korean landscape paintings needed for machine learning, and can also receive line data corresponding to lines arbitrarily drawn by the user. That is, the receiver 110 not only functions as a learning data receiver 110 that receives data necessary for machine learning, but also serves as a user input line data receiver 110 that receives a user's arbitrary line during application. It can be done.

The ridge extractor 120 performs a function of extracting ridge lines from each Korean landscape painting image data received by the learning data receiver 110.

In other words, since Korean landscape painting image data is data about a picture depicting a type of landscape painting, at least one mountain is naturally drawn in the picture, and the ridge extraction unit 120 performs a function of automatically extracting the ridge line of this mountain. .

In order to extract the ridge line in this way, the ridge extractor 120 may use a preset algorithm. For example, the ridge extractor 120 converts each Korean landscape painting image data received from the learning data receiver 110 into grayscale. Convert, perform Gaussian blur filter processing on the image data converted to grayscale for blurring to reduce the noise ratio, and convert the Gaussian blur filter-processed image data to black and white based on preset reference settings. Later, the ridge line can be extracted by applying a predetermined edge detection algorithm.

Figure 3 is a diagram illustrating this process.

For example, when a Korean landscape painting image as shown in FIG. 3(a) is input, the ridge extractor 120 first converts it to grayscale as shown in FIG. 3(b) and then processes it again with Gaussian blur. Convert as shown in Figure 3(c), and based on the standard setting value (for example, brightness standard 170), the darker part becomes completely black (brightness standard 0), and the lighter part becomes completely black. After converting white (brightness standard 255) into a black and white image as shown in Figure 3(d), the ridge line can be extracted using a predetermined edge detection algorithm (for example, Canny Edge Detection), and the ridge line is extracted. The current state is shown in Figure 3(e).

At this time, the ridge extraction unit 120 can only extract ridges that are close to the sky, and this state is shown in FIG. 3(f). In the following description of this embodiment, the ridge line extracted by the ridge extraction unit 120 is shown in FIG. An example is the ridge line at the highest position, as shown in 3(f).

The first machine learning unit 130 performs the function of constructing a primary artificial intelligence model by performing primary machine learning using the ridge line extracted from the ridge extraction unit and Korean landscape painting image data.

To this end, the first machine learning unit 130 may be configured to include a region of interest selection unit 131, a partial image extraction unit 132, and a primary artificial intelligence model construction unit 133.

Here, the region of interest selection unit 131 performs the function of selecting the region of interest based on each coordinate on the ridge line extracted by the ridge extraction unit 120.

That is, the region of interest selection unit 131 selects a predetermined region of interest based on the coordinates of each point forming the ridge line. At this time, the region of interest selection unit 131 extracts each point coordinate at regular intervals, and extracts the extracted A region of interest containing point coordinates can be selected.

In particular, the region of interest selection unit 131 includes each coordinate (i.e., point coordinates) on the ridge line extracted from the ridge extraction unit 120, and the ratio of the area above the ridge line and the area below the ridge line is within a preset range. Each area of interest can be selected.

For example, the region of interest selection unit 131 selects coordinate groups having the same preset interval on the ridge line extracted from the ridge extraction unit 120, sets each coordinate within the selected coordinate group as the center point, and sets the area above the ridge line. A square of the minimum size such that the ratio of the area below the ridge line is 3:7 can be selected as each area of interest.

That is, in this case, the location of the midpoint of the square region of interest corresponds to the point coordinates previously extracted on the ridge line.

An example of this is shown in Figure 4, where it can be seen that a square of the minimum size with the coordinates on the ridge line as the center point was selected.

However, if a region of interest cannot be created due to the shape of the ridge line, such that the point coordinates are located at the center and form a square shape, and the ratio of the area above the ridge line to the area below the ridge line within the square is 3:7, the region of interest line The government 131 may select a square of the smallest size whose point coordinates are moved by a preset distance in the vertical direction from the center point of the square as the region of interest.

That is, in this case, the reference point corresponding to the point coordinates on the ridge line is not located at the center point of the square, but the reference point is moved upward or downward from the center point of the square.

For example, the point coordinates on the ridge line can be located at a location corresponding to 80% of the vertical length of the square.

In particular, the region of interest selection unit 131 selects a square of interest of the minimum size for coordinates included in the mountain peak area among the coordinates in the selected coordinate group, such that the coordinates are moved by a preset distance from the center point of the square in the upper vertical direction. You can also select an area.

In other words, referring to FIG. 5, a square located at 80% of the area of interest window based on the mountain peak coordinates can be extracted as the area of interest. At this time, whether it is a mountain peak is determined by calculating the slope at each coordinate and changing the sign of the slope. This can also be done by checking the branch.

Meanwhile, the partial image extraction unit 132 performs a function of extracting a crop image from the Korean landscape painting image data corresponding to the region of interest selected by the region of interest selection unit 131.

An example in which a crop image from Korean landscape painting image data is extracted by the partial image extraction unit 132 is shown in FIG. 6 .

In Figure 6, a square with a solid red line corresponds to the area of interest, and a crop image of the Korean landscape painting image data belonging to this area of interest is also displayed in Figure 6.

Although FIG. 6 shows only the crop image extraction process corresponding to three point coordinates for convenience, in reality, it is preferable to extract the crop image based on hundreds to thousands of point coordinates.

Hundreds to thousands of Korean landscape painting images can be extracted from one Korean landscape painting image by this partial image extraction unit 132.

The first artificial intelligence model building unit 133 performs a function of building a first artificial intelligence model by performing machine learning using the cropped image extracted from the partial image extraction unit 132.

As mentioned earlier, hundreds to thousands of Korean landscape painting image data have been extracted from a single Korean landscape painting image data, so it is possible to secure sufficient data for machine learning from only the few existing Korean landscape painting image data. .

When using an artificial neural network such as deep learning, the first artificial intelligence model building unit 133 determines the parameter value of each layer, including the hidden layer, using machine learning, and the parameter value of each layer is determined in this way. This can be said to be building the first artificial intelligence model.

For example, the first artificial intelligence model building unit 133 outputs a virtual Korean landscape painting image using the ridge line within each region of interest in FIG. 5, and outputs a virtual Korean landscape painting image and Learning can be done by comparing cropped images on Korean landscape painting image data. For example, machine learning can be done in the method disclosed in the so-called Pix2Pix paper.

Since this machine learning process itself corresponds to known technology, a more detailed description is omitted.

The temporary output unit performs the function of generating temporary output image data by inputting the ridge line extracted from the ridge extraction unit 120 into the primary artificial intelligence model.

In the case of the first machine learning unit 130 described above, learning is performed based on a plurality of regions of interest created along the ridge line, while the temporary output unit 140 is based on the original Korean before being divided into a plurality of pieces. It performs the function of generating temporary output image data by inputting the entire ridge line of the landscape image data into the primary artificial intelligence model.

Meanwhile, the second machine learning unit 150 performs secondary machine learning using the temporary output image data of the temporary output unit 132 and the Korean landscape painting image data received by the receiver 110 to create a secondary artificial intelligence model. Performs the construction function.

That is, the second machine learning unit 150 receives temporary output image data temporarily generated based on ridge lines extracted from an entire Korean landscape painting and creates a virtual output image data as previously performed in the first machine learning unit 130. After generating an output image, it can be compared with the first received image of all Korean landscape paintings and then learning can be performed according to the difference. For example, machine learning can be performed using the algorithm disclosed in the so-called Pix2Pix paper.

The only difference is that the image data input from the second machine learning unit may not be image data in the form of lines drawn by a user, etc., but may be an image generated through a primary artificial intelligence model (even if it is an overall blurry image).

As described above, the Korean landscape image generation unit 160 receives arbitrary line image data drawn by the user in the reception unit 110 after the first artificial intelligence model and the second artificial intelligence model are built. After inputting the line image data into the primary artificial intelligence model, the resulting image data is input again into the secondary artificial intelligence model to automatically generate a Korean landscape painting image.

To be more specific, temporary output image data output along an arbitrary line when the first artificial intelligence model is built is shown in FIG. 7.

As shown in the figure, temporary output image data as shown in FIG. 7(b) can be output for the line data in FIG. 7(a). Looking at FIG. 7(b), the ridge line in FIG. 7(a) can be output. Therefore, although a certain image was output, it can be seen that the image becomes blurry as it goes below the ridge line.

Therefore, since this blurry image alone cannot be considered a satisfactory result, an additional learning process is carried out in the present invention.

The overall process carried out in the present invention is shown in Figures 8 and 9.

Figure 8 shows the process of constructing the previously described first artificial intelligence model through machine learning and re-inputting the overall ridge line to generate temporary output image data, and Figure 9 shows the process of constructing the second artificial intelligence model. .

In particular, Figures 8 and 9 both show examples of using the Pix2Pix architecture.

As shown in the figures, it is desirable to perform 2,000 epochs in the case of primary machine learning, and 20,000 epochs in the case of secondary machine learning. In other words, it is desirable to perform secondary machine learning at a level of 8 to 10 times that of the primary.

FIG. 10 shows a state in which the first artificial intelligence model learned in FIG. 8 and the second artificial intelligence model learned in FIG. 9 are arranged in series, and as described above, the Korean landscape image generator (160) ) is such that the first artificial intelligence model and the second artificial intelligence model are connected side by side, and when random line image data drawn by the user is received by the receiver 110, the received line image data is connected to the first artificial intelligence model. After inputting it into the model, the resulting image data can be input back into the secondary artificial intelligence model to automatically generate Korean landscape painting images.

Accordingly, even if the user draws only arbitrary lines, a proper Korean unique landscape painting image can be created based on this, and Figure 11 shows this.

For example, when an image (FIG. 11(a)) containing simple lines drawn by the user as shown in FIG. 11 is received, the Korean landscape painting image generator 160 responds to such lines using a pre-built artificial intelligence model. It is possible to output a Korean landscape painting image (Figure 11(b)).

Hereinafter, the overall control flow of the automatic Korean landscape painting generation device 100 according to an embodiment of the present invention will be described with reference to FIG. 12.

First, the Korean landscape painting automatic generating device 100 receives Korean landscape painting image data (step S1) and extracts ridge lines from the received Korean landscape painting image data (step S3).

To extract ridge lines, the Korean landscape painting automatic generation device 100 can use a combination of various methods such as grayscale conversion, Gaussian blur processing, black and white conversion, and edge detection.

Next, the Korean landscape painting automatic generating device 100 selects the area of interest based on the coordinates included in the ridge line (step S5).

At this time, the apparatus 100 for automatically generating Korean landscape paintings may select a region of interest with the minimum size of a square so that the ratio of the area above and below the ridge line within the region of interest is a preset ratio (for example, 3:7).

Next, the Korean landscape painting automatic generating device 100 extracts an image (i.e., crop image) of the Korean landscape painting automatic generating device 100 corresponding to the region of interest (step S7).

For example, if 1,000 crop images are extracted per Korean landscape painting image, a total of 10,000 crop images can be extracted for 10 Korean landscape painting images.

The Korean landscape painting automatic generation device 100 performs machine learning using the plurality of cropped images extracted in this way to build a first artificial intelligence model (step S9).

Since the process of building an artificial intelligence model itself corresponds to a known technology, a more detailed description is omitted.

After the first artificial intelligence model is constructed in this way, the Korean landscape painting automatic generation device 100 inputs the entire ridge line before the area of interest is selected into this first artificial intelligence model (step S121) to generate temporary output image data. (step S13).

Subsequently, the Korean landscape painting automatic generation device 100 performs machine learning using each of these temporary output image data to build a secondary artificial intelligence model (step S15).

Here, the form of the secondary artificial intelligence model itself may be the same as the primary artificial intelligence model described above, but the parameter values within each model may be completely different due to machine learning.

In other words, the Pix2Pix architecture can be used when building both the primary and secondary artificial intelligence models, but since the inputs for machine learning are different, the parameters of the final built model will naturally vary for each model. will be.

Additionally, if there are 10 images of Korean landscape paintings that were actually drawn in the past (each referred to as an 'original Korean landscape painting image'), for each of these 10 Korean landscape paintings, a plurality of regions of interest based on the entire ridge line are extracted and 1 After performing secondary machine learning (i.e., the entire ridge line is also split), temporary output image data is generated by inputting the entire ridge line extracted from each of the 10 Korean landscape painting images into the primary artificial intelligence model constructed in this way. , A secondary artificial intelligence model can be built through machine learning using the temporary output image data generated in this way as input and labeling the corresponding original Korean landscape painting image.

After the first artificial intelligence model and the second artificial intelligence model are built in this way, when line image data is received from the user (step S17), the Korean landscape painting automatic generation device 100 converts the received line image data into the first artificial intelligence model. and a second artificial intelligence model, an integrated artificial intelligence model connected in series (step S19), to automatically generate and output a new Korean landscape painting image (step S21).

Therefore, users can have a diverse and new Korean landscape painting image that maintains the Korean landscape painting style by using only images with random lines drawn.

Meanwhile, in the above-described embodiment, when performing the first machine learning, the ridge line is extracted only once to select the region of interest as an example, but the ridge line is extracted twice to select a new region of interest each and extract the crop image. and machine learning may also be performed.

For example, as mentioned above, when a Korean landscape painting image for machine learning is input, the Korean landscape painting automatic generating device 100 uses the above-mentioned process, for example (i.e., the Korean landscape painting automatic generating device converts it into grayscale). After conversion, Gaussian blur processing is performed to convert, and based on the standard setting value, the darker part is completely black (brightness standard 0), and the lighter part is completely white (brightness standard 0). After converting it into a black and white image using the standard 255), the ridge line can be extracted using a predetermined edge detection algorithm (for example, Canny Edge Detection). At this time, the Korean landscape painting automatic generation device 100 uses the sky and the edge. Only the closest ridge line (i.e., the ridge line at the highest position) (hereinafter referred to as the first ridge line) is extracted and a first-order machine learning process is performed using this, and then the difference between the first ridge line and the second ridge line is performed. After deleting the image so that the second ridge line appears as if it touches the sky, first and second machine learning processes can be performed using this second ridge line.

At this time, if the second ridge lines are not connected to each other by a single line, a pre-connection process may be further included.

This process is shown in Figure 13.

That is, FIG. 13(a) is a diagram showing a state in which the initially extracted ridge line is extracted, and FIG. 13(b) is a diagram showing a state in which only the first ridge line described above is left in the initially extracted ridge line, and FIG. 13 (c) is a diagram showing a state in which broken second ridge lines are connected to each other, and Figure 13(d) shows a state in which only the second ridge line is left by removing the area between the first ridge line and the second ridge line. It is a drawing.

That is, the Korean landscape painting automatic generation device 100 performs 1-1 machine learning using the first ridge line in FIG. 13(b), and then uses the second ridge line in FIG. 13(d) to perform 1-1 machine learning. Secondary machine learning is performed, where the process of performing machine learning is to select a region of interest based on each coordinate on the currently selected ridge line (i.e., the first ridge line or the second ridge line), and It may include the process of extracting a crop image from Korean landscape painting image data corresponding to the selected area of interest and then performing machine learning using the extracted crop image to build an artificial intelligence model.

The interconnection of the second ridge lines that are broken in FIG. 13(c) can be accomplished by performing processing that takes into account the curvature of a straight line or an existing ridge line.

Meanwhile, of course, the process of performing each of the above-described embodiments can be performed by a program or application stored in a predetermined recording medium (eg, computer-readable). Here, recording media include electronic recording media such as RAM (Random Access Memory), magnetic recording media such as hard disks, and optical recording media such as CDs (Compact Disk).

At this time, the program stored in the recording medium can be executed on hardware such as a computer or smartphone to perform each of the above-described embodiments. In particular, at least one of the functional blocks of the apparatus for automatically generating Korean landscape paintings according to the present invention described above may be implemented by such a program or application.

In addition, the present invention is not limited to the specific embodiments described above, but can be implemented with various changes and modifications without departing from the gist of the present invention. It will be apparent that such changes and modifications are included in the present invention if they fall within the scope of the appended claims.

100: Korean landscape painting automatic generation device 200: User terminal
110: Receiving unit 120: Ridge extraction unit
130: first machine learning unit 140: temporary output unit
150: Second machine learning unit 160: Korean landscape painting image generation unit
131: Region of interest selection unit 132: Partial image extraction unit
133: Artificial intelligence model construction department

Claims (18)

(a) receiving Korean landscape painting image data;
(b) extracting ridge lines from each Korean landscape image data received in step (a);
(c) performing primary machine learning using the extracted ridge line and the Korean landscape painting image data to build a primary artificial intelligence model;
(d) generating temporary output image data by inputting the ridge line extracted in step (b) into the first artificial intelligence model;
(e) A control method of a Korean landscape painting modeling device, comprising the step of performing secondary machine learning using the temporary output image data and the Korean landscape painting image data to build a secondary artificial intelligence model. According to paragraph 1,
In step (c),
(c1) selecting a region of interest based on each coordinate on the ridge line extracted in step (b);
(c2) extracting a crop image from the Korean landscape painting image data corresponding to the region of interest selected in step (c1);
(c3) A control method of a Korean landscape painting modeling device, comprising the step of constructing the first artificial intelligence model by performing machine learning using the cropped image extracted in step (c2). According to paragraph 2,
In step (c1), a region of interest is selected that includes each coordinate on the ridge line extracted in step (b), and the ratio of the area above the ridge line to the area below the ridge line is within a preset range. Control method of Korean landscape painting modeling device. According to paragraph 3,
In step (c1),
(c11) selecting coordinate groups having preset equal intervals on the ridge line extracted in step (b);
(c12) selecting a square region of interest of the minimum size such that each coordinate in the selected coordinate group is used as the center point and the ratio of the area above the ridge line to the area below the ridge line is 3:7. Control method of Korean landscape painting modeling device. According to paragraph 4,
In step (c12), for a specific coordinate in the selected coordinate group, a square region of interest cannot be selected such that the specific coordinate is the center point and the ratio of the area above the ridge line to the area below the ridge line is 3:7. In this case, a control method of a Korean landscape painting modeling device, characterized in that a square area of interest of the minimum size in which the specific coordinates are shifted 20% in the vertical direction from the center point is selected. According to paragraph 4,
In the step (c12), for the coordinates included in the mountain peak area among the coordinates in the selected coordinate group, a square region of interest of the minimum size is selected where the coordinates are shifted 20% in the vertical direction upward from the center point. Control method of landscape modeling device. According to paragraph 1,
In step (b),
(b1) converting each Korean landscape image data received in step (a) into grayscale;
(b2) performing Gaussian blur filter processing on the grayscale converted image data of step (b1) for blurring to reduce the noise ratio;
(b3) converting the Gaussian blur filter-processed image of step (b2) to black and white based on a preset reference setting value;
(b4) A control method for a Korean landscape painting modeling device, comprising the step of extracting ridge lines by applying a preset edge detection algorithm to the black-and-white conversion image generated in step (b3). (a) receiving Korean landscape painting image data;
(b) extracting ridge lines from each Korean landscape image data received in step (a);
(c) performing primary machine learning using the extracted ridge line and the Korean landscape painting image data to build a primary artificial intelligence model;
(d) generating temporary output image data by inputting the ridge line extracted in step (b) into the first artificial intelligence model;
(e) performing secondary machine learning using the temporary output image data and the Korean landscape painting image data to build a secondary artificial intelligence model;
(f) After step (e), when line image data is received from the user, the received line image data is input into the first artificial intelligence model and the resulting image data is again inputted into the second artificial intelligence model. A control method of a device for automatically generating Korean landscape paintings, comprising the step of automatically generating a Korean landscape painting image by inputting . A computer-readable recording medium recording a program for executing the method of any one of claims 1 to 8. An application program stored in a computer-readable recording medium in combination with hardware to execute the method of any one of claims 1 to 8. a learning data receiving unit that receives Korean landscape painting image data;
a ridge extraction unit that extracts ridge lines from each Korean landscape painting image data received by the learning data receiver;
a first machine learning unit that constructs a first artificial intelligence model by performing first machine learning using the ridge line extracted from the ridge extraction unit and the Korean landscape painting image data;
a temporary output unit that generates temporary output image data by inputting the ridge line extracted from the ridge extraction unit into the first artificial intelligence model;
A Korean landscape painting modeling device comprising a second machine learning unit that constructs a secondary artificial intelligence model by performing secondary machine learning using the temporary output image data of the temporary output unit and the Korean landscape painting image data. According to clause 11,
The first machine learning unit,
a region of interest selection unit that selects a region of interest based on each coordinate on the ridge line extracted from the ridge extraction unit;
a partial image extraction unit for extracting a cropped image of the Korean landscape painting image data corresponding to the area of interest selected by the area of interest selection unit;
A Korean landscape painting modeling device comprising a first artificial intelligence model building unit that constructs a first artificial intelligence model by performing machine learning using the cropped image extracted from the partial image extraction unit. According to clause 12,
The region of interest selection unit includes each coordinate on the ridge line extracted from the ridge extraction unit, and selects a region of interest such that the ratio of the area above the ridge line and the area below the ridge line falls within a preset range. Landscape painting modeling device. According to clause 13,
The region of interest selection unit selects coordinate groups having the same preset interval on the ridge line extracted from the ridge extraction unit, sets each coordinate in the selected coordinate group as the center point, and sets the ratio of the area above the ridge line to the area below the ridge line. A Korean landscape painting modeling device characterized by selecting a square area of interest with a minimum size of 3:7. According to clause 14,
When the region of interest selection unit cannot select a square region of interest for a specific coordinate in the selected coordinate group, with the specific coordinate as the center point and the ratio of the area above the ridge line to the area below the ridge line being 3:7. A Korean landscape painting modeling device, characterized in that the specific coordinates are selected as a square area of interest of the minimum size shifted by 20% in the vertical direction from the center point. According to clause 14,
The region of interest selection unit is a Korean landscape painting characterized in that, for coordinates included in the mountain peak area among the coordinates in the selected coordinate group, a square region of interest of the minimum size is selected whose coordinates are shifted 20% in the vertical direction upward from the center point. modeling device. According to clause 11,
The ridge extractor converts each Korean landscape painting image data received from the learning data receiver into grayscale, and performs Gaussian blur filter processing on the grayscale converted image data for blur processing to reduce the noise ratio, A Korean landscape painting modeling device that converts a Gaussian blur filter-processed image to black and white based on preset reference settings and then extracts ridge lines by applying a preset edge detection algorithm. a learning data receiving unit that receives Korean landscape painting image data;
a ridge extraction unit that extracts ridge lines from each Korean landscape painting image data received by the learning data receiver;
a first machine learning unit that constructs a first artificial intelligence model by performing first machine learning using the ridge line extracted from the ridge extraction unit and the Korean landscape painting image data;
a temporary output unit that generates temporary output image data by inputting the ridge line extracted from the ridge extraction unit into the first artificial intelligence model;
a second machine learning unit that constructs a secondary artificial intelligence model by performing secondary machine learning using the temporary output image data of the temporary output unit and the Korean landscape painting image data;
After the secondary artificial intelligence model is built by the second machine learning unit, when line image data is received from the user, the received line image data is input into the first artificial intelligence model and the resulting image data is An automatic Korean landscape painting generation device comprising a Korean landscape painting image generator that automatically generates a Korean landscape painting image by inputting it back into the secondary artificial intelligence model.