KR102532250B1 - Image correcting method correct image by using deep learning algorithm and image correcting system performing the same - Google Patents

Abstract

An image correction method for correcting an image using a deep learning algorithm and an image correction system for performing the same are disclosed. An image correction method according to one aspect of the technical idea of the present disclosure includes receiving an image including an object, extracting at least one reference point corresponding to the object by using a deep learning algorithm, the at least one reference point The method may include calculating an angle of a virtual line connecting the first reference point and the second reference point, and rotating the image using the calculated angle.

Description

An image correction method for correcting an image using a deep learning algorithm and an image correction system performing the same

The technical idea of the present disclosure relates to an image correction method and an image correction system for performing the same, and more particularly, to an image correction method for performing image correction using a deep learning algorithm and an image correction system for performing the same. .

A camera can be used to capture a single image or a series of images to be used as frames of a video signal. People use cameras to create images with people as subjects, and use the created images for various purposes, such as for storing memories or for public purposes such as attaching resumes and passports. However, these images may be captured so that the subject does not come to a desired location depending on the location of the camera, the skill of the photographer, etc., and problems such as the subject tilting or not capturing an image of a desired size may occur.

To solve this problem, image correction technology has appeared. This image correction technology is used to adjust the size, angle, position, etc. of the image, and to correct the subject to the desired face. Until now, image correction was possible only with professional software tools from experts, and ordinary people can do simple corrections without experts. A difficult situation has arisen.

An object to be solved by the technical idea of the present disclosure is to provide an image correction method capable of automatically correcting an image by correcting an image using a deep learning algorithm and an image correction system performing the same.

An object to be solved by the technical idea of the present disclosure is to provide an image correction method capable of automatically rotating an image according to an object by rotating an image using a deep learning algorithm and an image correction system that performs the same.

In order to achieve the above object, an image correction method according to one aspect of the technical idea of the present disclosure includes receiving an image including an object, using a deep learning algorithm to determine at least one reference point corresponding to the object The method may include extracting, calculating an angle of a virtual line connecting a first reference point and a second reference point among the at least one reference point, and rotating the image using the calculated angle.

According to an embodiment, the first reference point and the second reference point may correspond to positions symmetrical to each other with respect to the central axis of the object.

According to an embodiment, the object is a human face, the at least one reference point corresponds to lines constituting the human face, the first reference point is included in the left eye of the object, and the second reference point is the It may be included in the right eye of the object.

According to one embodiment, the first reference point may be located at the leftmost among points included in the left eye of the object, and the second reference point may be located at the rightmost among points included in the right eye of the object. .

According to one embodiment, the first reference point is located at the leftmost of points corresponding to the left eyebrow of the object, and the second reference point is located at the rightmost among points corresponding to the right eyebrow of the object. can

According to an embodiment, the first reference point may be located at the leftmost side among points corresponding to the lips of the object, and the second reference point may be located at the rightmost side among points corresponding to the lips of the object. .

According to an embodiment, the image correction method includes determining first average coordinates of at least some of the points included in the left eye of the object as the first reference point and at least some of the points included in the right eye of the object. The method may further include determining a second average coordinate for the second reference point.

According to an embodiment, the first reference point may be positioned at the top among points corresponding to the nose of the object, and the second reference point may be positioned at the bottom among points corresponding to the nose of the object.

According to an embodiment, the angle of the virtual line is an angle between a horizontal line and the virtual line, and rotating the image may include rotating the image so that the virtual line matches the horizontal line.

According to an embodiment, the image correction method may include receiving information about a bounding box, and adjusting a size of the image to correspond to the bounding box using a third reference point and a fourth reference point among the at least one reference point. and changing the position of the image so as to correspond to the bounding box using a fifth reference point among the at least one reference point, wherein the adjusting the size of the image comprises: Calculating a coordinate distance of a fourth reference point, obtaining a horizontal length of the bounding box, calculating a ratio between the coordinate distance and the horizontal length, and adjusting the size of the image based on the calculated ratio The step of changing the location of the image includes: calculating an X coordinate of a midpoint of the bounding box; calculating midpoints of the third reference point and the fourth reference point; Moving along the X-axis of the image so that the midpoint of the fourth reference point and the X-coordinate of the midpoint of the bounding box coincide, acquiring the Y-coordinate of the lower side of the bounding box, and the fifth reference point and the lower edge of the bounding box. and moving the Y-axis of the image so that the Y-coordinates of the sides coincide.

According to an image correction method and an image correction system performing the same according to an exemplary embodiment of the present disclosure, a user can automatically rotate an image using a reference point extracted using a deep learning algorithm, and accordingly, the user can Rotated images can be obtained according to the object without any special editing work.

1 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure.
Fig. 2 is a block diagram illustrating an image correction system according to an exemplary embodiment of the present disclosure.
3 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure.
Fig. 4 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure.
5 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure.
6 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure.
Fig. 7 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure.
8 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure.
Fig. 9 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure.
Fig. 10 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure.
Fig. 11 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure.
12 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure.

Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. The embodiments of the present invention are provided to more completely explain the present invention to those skilled in the art. Since the present invention can have various changes and various forms, specific embodiments are illustrated in the drawings and described in detail. However, it should be understood that this is not intended to limit the present invention to the specific disclosed form, and includes all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing each figure, like reference numbers are used for like elements. In the accompanying drawings, the dimensions of the structures are shown enlarged or reduced than actual for clarity of the present invention.

Terms used in this disclosure are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In the present disclosure, terms such as "comprise" or "having" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that it does not preclude the possibility of the presence or addition of numbers, steps, operations, components, parts, or combinations thereof. It is understood that ordinal expressions such as "first ~" and "second ~" used in the present disclosure do not limit that "first ~" precedes "second ~", and are intended to distinguish and express similar configurations differently. It should be. In addition, it should be understood that even though it is expressed as "first ~" in the claims, it is not construed as limited to "first ~" in the specification, and is expressed to refer to any one of "first ~" to "nth ~". do.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in this application, it should not be interpreted in an ideal or excessively formal meaning. don't

1 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure.

Referring to FIG. 1, the image correction method (S1) includes an image reception step (S10), a reference point extraction step (S20), an image rotation step (S30), an image size adjustment step (S40), and an image movement step (S50). can do.

In the image receiving step ( S10 ), the image correction system may receive an image including an object from a user. In one embodiment, the image correction system may operate a website or application capable of uploading images, and a user may transmit an image to the image correction system by uploading an image to a website or application operated by the image correction system. .

In the reference point extraction step ( S20 ), the image correction system may extract at least one reference point corresponding to an object included in the image using a deep learning algorithm. At least one reference point may be located along the outline or shape of an object included in the image, and in an embodiment in which the object is a human face, at least one reference point may be located on the eyes, eyebrows, nose, lips, and chin included in the face. It can be located along the outline.

In one embodiment, the image correction system may use an image analysis deep learning algorithm such as a convolutional neural network (CNN) or high resolution networks (HR Nets) as a deep learning algorithm, but the technical spirit of the present disclosure is not limited thereto.

In the image rotation step ( S30 ), the image correction system may rotate the image using at least one extracted reference point. In one embodiment, the image correction system may calculate a rotation angle of the image using the reference point and rotate the image according to the calculated angle.

In the image size adjustment step (S40), the image correction system may adjust the size of the image using at least one extracted reference point. In one embodiment, the image calibration system may utilize the specified bounding box size and reference point to enlarge or reduce the size of the image so that the image matches the size of the bounding box.

In the image movement step (S50), the image correction system may move the image using at least one extracted reference point. In one embodiment, the image correction system may utilize the location of the designated box and the reference point to move the image so that it matches the location of the bounding box.

According to the image correction method according to the technical idea of the present disclosure, by automatically rotating, resizing, or moving an image based on a reference point extracted using a deep learning algorithm, the user is distorted, the object is too large, or the object is too large. Even if an image that is small or the object is out of the designated location is uploaded, an image of a desired format can be automatically obtained, and accordingly, the image can be corrected efficiently without much effort.

1 shows an embodiment in which the image rotation step (S30), the image size adjustment step (S40), and the image movement step (S50) are performed in order, but this is an example, and in another embodiment, the image rotation step ( S30), image size adjustment step (S40), and image movement step (S50) may also proceed in a different order from the illustrated example.

Although not shown in FIG. 1 , the image after all steps may be cut to a predetermined size.

Fig. 2 is a block diagram illustrating an image correction system according to an exemplary embodiment of the present disclosure.

Referring to FIG. 2 , the image correction system 10 may include various software and hardware utilized to receive an image IMG and output an automatically corrected image IMG_C using artificial intelligence, In one embodiment, the image correction system 10 may include a management server that manages the image correction system 10 and an operating computer that operates the server. The management server may operate the image correction system using an application program such as a website or an application. An operation performed by each component of the image calibration system in the present specification may actually be an operation performed by a processor of the management server using an application program stored in a storage device of the management server or at least one command.

The image correction system 10 may include an artificial intelligence module 110 , a rotation module 120 , a scaling module 130 and a movement module 140 . Each operation of the artificial intelligence module 110, the rotation module 120, the size adjustment module 130, and the movement module 140 may mean an operation of a processor driven by configuration-specific software or at least one command.

The artificial intelligence module 110 may receive the image IMG and extract at least one reference point CP from the image IMG. As described above in FIG. 1, the artificial intelligence module 110 uses image analysis deep learning algorithms such as CNN (Convolutional Neural Network) and HR Nets (High Resolution Networks) to follow the outline or shape of an object from an image (IMG). It is possible to extract the reference point (CP) located. The artificial intelligence module 110 may output the extracted reference point CP to the rotation module 120, the size adjustment module 130, and the movement module 140.

The rotation module 120 may receive the image IMG and calculate a rotation angle for the image IMG based on at least one reference point CP. In one embodiment, the rotation module 120 determines a first reference point and a second reference point among at least one reference point CP, and a virtual line and a reference line (for example, a horizontal line, The rotation angle can be calculated by calculating the angle formed by the vertical line). In one embodiment, the first reference point and the second reference point may be points that exist at symmetric positions based on the center line of an object (eg, a human face) included in the image IMG. The rotation module 120 may generate the rotated image IMG_R by rotating the image IMG using the calculated rotation angle. The rotation module 120 may output the rotated image IMG_R to the scaling module 130 .

The size adjustment module 130 receives the rotated image IMG_R, and matches the size of the bounding box BB using the separately received size information of the bounding box BB and at least one reference point CP. The size of the rotated image (IMG_R) can be changed to In the present specification, the bounding box (BB) is a rectangular virtual box whose size and position are designated. In one embodiment, the bounding box (BB) can be designated according to the type of photo desired by the user, so as to fit the designated bounding box. Images can be corrected.

The size adjustment module 130 may adjust the size of the rotated image IMG_R using the third and fourth reference points among the at least one reference point CP. In one embodiment, the scaling module 130 calculates a coordinate distance between the third reference point and the fourth reference point, and calculates the coordinate distance and the length (eg, horizontal length or vertical length) of the bounding box BB. The size of the rotated image (IMG_R) can be adjusted so that it matches. The size adjustment module 130 may generate a size-changed image IMG_S by adjusting the size of the rotated image IMG_R and output the size-changed image IMG_S to the movement module 140 .

The movement module 140 receives the size-changed image IMG_S, and uses the location information of the bounding box BB and at least one reference point CP to change the size to match the position of the bounding box BB. The position of the image (IMG_S) can be changed. In one embodiment, the movement module 140 transfers the image IMG_S whose size has been changed so that midpoints of the fifth and sixth reference points of the at least one reference point CP coincide with the midpoint of the bounding box BB in the X-axis direction. , and the size-changed image IMG_S may be moved in the Y-axis direction so that the seventh reference point among the at least one reference point CP coincides with the lower side of the bounding box BB. The movement module 140 may generate a corrected image IMG_C as a result of moving the size-changed image IMG_S, and output the corrected image IMG_C to an external device (eg, a terminal operated by a user). can do.

The image correction system 10 according to the technical idea of the present disclosure extracts a reference point from an image, and rotates, scales, and moves using the extracted reference point, thereby efficiently correcting an image in a direction desired by a user.

3 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 3 shows an operation of extracting at least one reference point using a deep learning algorithm according to the reference point extraction step (S20) of FIG. 1 .

Referring to FIGS. 2 and 3 , the artificial intelligence module 110 receives an image IMG including a human face and extracts at least one reference point CP from the image IMG using a deep learning algorithm. can

3, reference points 1 to 17 corresponding to the jaw line, reference points 18 to 22 corresponding to the left eyebrow, reference points 23 to 27 corresponding to the right eyebrow, reference points 28 to 31 corresponding to the nose, nose Reference points 32 to 36 corresponding to the hole, reference points 37 to 42 corresponding to the left eye (left eye), reference points 43 to 48 corresponding to the right eye (right eye), and reference points 49 to 68 corresponding to the lips A total of 68 reference points of reference points can be extracted. The number and location of the reference points according to the technical idea of the present disclosure are not limited to the example of FIG. 3 and can be variously adjusted.

Fig. 4 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 4 shows the moving step (S30) of FIG.

Referring to FIGS. 2 and 4 , the image correction system 10 may calculate an angle of a virtual line connecting a first reference point and a second reference point among at least one extracted reference point ( S110 ). In one embodiment, the first reference point and the second reference point may be determined in advance by the image correction system 10 and may be determined to be symmetrical to each other with respect to the object.

In one example, the first reference point may be a leftmost reference point among points included in the left eye of the object, and the second reference point may be a rightmost reference point among points included in the object's right eye. In one example, the first reference point may be the leftmost reference point among points corresponding to the object's left eyebrow, and the second reference point may be the rightmost reference point among points corresponding to the object's right eyebrow.

In one example, the first reference point may be a leftmost reference point among points corresponding to the lips of the object, and the second reference point may be a rightmost reference point among points corresponding to the lips of the object. In one example, the first reference point may be the average coordinate of at least some of the points included in the left eye of the object, and the second reference point may be the average coordinate of at least some of the points included in the right eye of the object. In one example, the first reference point may be a reference point located at the top among points corresponding to the nose of the object, and the second reference point may be a reference point located at the bottom among points corresponding to the nose of the object.

According to an embodiment of the present disclosure, a reference point that can be a symmetrical point of a person can be determined, and the image can be efficiently and straightly rotated by rotating the image accordingly.

The image correction system 10 may generate the rotated image IMG_R by rotating the image IMG using the angle calculated using the reference point (S120).

The image correction system 10 according to an embodiment of the present disclosure can automatically rotate an object to look straight using the extracted reference point, and accordingly, the user can automatically and comfortably correct the rotation without any special correction technology. can be performed. In addition, according to an embodiment of the present disclosure, a reference point that can be a symmetrical point of a person can be determined, and the image can be efficiently and straightly rotated by rotating the image accordingly.

The example of determining the reference points listed in FIG. 4 is just one example, and the embodiment of determining at least two reference points to straightly rotate the image IMG and rotating the image IMG using them is the technical spirit of the present disclosure. Inclusion in should be understood.

5 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 5 shows an embodiment of rotating an image using both end points of an eye.

Referring to FIGS. 2 and 5(a) , the image correction system 10 may determine the 37th reference point located at the leftmost position among the reference points included in the left eye of the object as the first reference point CP1, and the second reference point With (CP2), it is possible to determine the 46th reference point located at the rightmost among the reference points included in the right eye of the object. The image correction system 10 may calculate an angle Deg formed by a virtual line CPL connecting the first reference point CP1 and the second reference point CP2.

Referring to FIG. 5(b) , the image correction system 10 may generate a rotated image IMG_R by rotating the image IMG using the calculated angle Deg.

6 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 6 shows various embodiments of determining a reference point. Contents overlapping with those of FIG. 5 are omitted.

Referring to FIGS. 2 and 6(a) , the image correction system 10 may determine the 18th reference point located at the leftmost position among the reference points included in the left eyebrow of the object as the first reference point CP1_1. As the reference point CP2_1, the 27th reference point located at the rightmost among the reference points included in the right eyebrow of the object may be determined. The image correction system 10 may calculate an angle Deg1 formed by a virtual line CPL1 connecting the first reference point CP1_1 and the second reference point CP2_1. The image correction system 10 may generate the rotated image IMG_R by rotating the image IMG using the calculated angle Deg1.

Referring to FIGS. 2 and 6(b) , the image correction system 10 obtains a first average coordinate using at least some coordinates of reference points 37 to 42 included in the left eye of the object, and obtains a first average coordinate. Coordinates may be determined as the first reference point CP1_2. In addition, the image correction system 10 obtains the second average coordinates using at least some coordinates of reference points 43 to 48 included in the right eye of the object, and determines the second average coordinates as the second reference point CP2_2. can The image correction system 10 calculates the angle Deg2 formed by the imaginary line CPL2 connecting the first reference point CP1_2 and the second reference point CP2_2, and calculates the image IMG using the calculated angle Deg2. A rotated image (IMG_R) can be created by rotating .

Referring to FIGS. 2 and 6(c) , the image correction system 10 may determine the 28th reference point located at the top among the reference points included in the nose of the object as the first reference point CP1_3, and the second reference point With (CP2_3), the 31st reference point located at the bottom among the reference points included in the object's nose can be determined. The image correction system 10 may calculate an angle Deg3 formed by a virtual line CPL3 connecting the first reference point CP1_3 and the second reference point CP2_3 . The image correction system 10 may generate the rotated image IMG_R by rotating the image IMG using the calculated angle Deg3.

An example of determining a reference point according to embodiments of the present disclosure is not limited to the example shown in FIG. 6, and in one example, both end points of the lips (reference point 49 and reference point 55) are the first reference point and the second reference point. , and points included in the left eye (reference points 37 to 42) and points included in the right eye (reference points 43 to 48) are corresponding pairs, and can be determined as the first reference point and the second reference point, respectively.

Fig. 7 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 7 shows the size adjustment step (S40) of FIG.

Referring to FIGS. 2 and 7 , the image correction system 10 may receive size information of the bounding box BB (S210). In one embodiment, the image correction system 10 may receive horizontal and vertical size information of the bounding box BB, and in another embodiment, the image correction system 10 may include four vertices constituting the bounding box. Size information of the bounding box BB may be obtained by receiving coordinate information for BB and calculating a distance to the coordinates.

The image correction system 10 may calculate coordinate distances of the third reference point and the fourth reference point among at least one reference point (S220). In one embodiment, the third reference point and the third reference point may be determined in advance by the image correction system 10 and may be determined to be symmetrical to each other around the object. In one example, the third reference point is a reference point located N (where N is a natural number) th from the left among the reference points corresponding to the jaw line of the image IMG, and the fourth reference point is a reference point located N th from the right among the reference points corresponding to the jaw line of the image IMG. can be

The image calibration system 10 may calculate a ratio of the coordinate distance between the horizontal length of the bounding box BB and the third reference point and the fourth reference point (S230). The image correction system 10 may adjust the image size by enlarging or reducing the image size based on the calculated ratio (S240).

The image correction system 10 according to an embodiment of the present disclosure can automatically adjust the size of an image so that an object can fit the size of a bounding box (BB) by utilizing the extracted reference point, and accordingly, a user may not have a special expertise. Resizing correction can be performed automatically and conveniently without any special correction technology.

The example of determining the reference points listed in FIG. 7 is an example, and an embodiment of determining at least two reference points to adjust the size of the image IMG_R to the bounding box BB and adjusting the size of the image IMG_R using the determined reference points. It should be understood that is included in the technical spirit of the present disclosure.

8 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 8 shows an embodiment of adjusting the size of an image using reference points.

Referring to FIGS. 2 and 8 , the image correction system 10 may determine a third reference point located third from the left among reference points 1 to 17 extracted along the jaw line as the third reference point CP3, and As the reference point CP4, a reference point 15 located third from the right among reference points 1 to 17 extracted along the jaw line may be determined.

The image correction system 10 may calculate the coordinate lengths of the third reference point CP3 and the fourth reference point CP4. In one embodiment, the image correction system 10 may calculate the x-axis coordinate difference between the third reference point CP3 and the fourth reference point CP4 as the coordinate length.

Also, the image calibration system 10 may obtain the horizontal length of the bounding box BB based on the information of the bounding box BB. The image correction system 10 may calculate the scaling ratio by dividing the horizontal length of the bounding box by the calculated coordinate length, and the scaled image by enlarging or reducing the size of the image IMG_R based on the scaling ratio. (IMG_S) can be obtained. Accordingly, the size of the image IMG_S may be adjusted so that the bounding box BB is positioned at a desired part of the object.

According to an embodiment of the present disclosure, a reference point for a user to cut an image may be determined based on the bounding box BB, and accordingly, the size of the image may be adjusted so that the user may use the image later. .

Fig. 9 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 9 shows the movement step (S50) of FIG.

Referring to FIGS. 2 and 9 , the image correction system 10 may move the image along the X axis using a fifth and a sixth reference point among at least one reference point (S310). The image correction system 10 may move the image along the Y-axis using a seventh reference point among at least one reference point (S320).

The image correction system 10 according to an embodiment of the present disclosure may automatically move the position of an image so that the object fits the position of the bounding box BB by utilizing the extracted reference point, and accordingly, the user may Positioning correction can be performed automatically and comfortably without professional calibration skills.

9 shows an example of moving an image in the X axis using two reference points and moving an image in the Y axis using one reference point, but this is an example and using more or less than two reference points. An example of moving the image in the X axis and moving the image in the Y axis using more than one reference point may also be included in the technical spirit of the present disclosure. As an example, the image correction system 10 may move the image along the X and Y axes to match the location of the bounding box BB using one reference point.

Fig. 10 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 10 shows the X-axis movement step (S310) of FIG.

Referring to FIGS. 2 and 10 , the image correction system 10 may calculate the X coordinate of the midpoint of the bounding box (S311). In one example, the image correction system 10 may calculate the X coordinate of the midpoint of the bounding box by using the average of the X coordinates of two vertices having different X coordinates among the vertices of the bounding box.

The image correction system 10 may calculate midpoints of the fifth and sixth reference points among at least one reference point. In one embodiment, the fifth reference point and the sixth reference point may be determined to be symmetrical to each other, the fifth reference point may be the same as the third reference point used for resizing, and the sixth reference point may be the fourth reference point used for resizing. can be the same as In one example, the fifth reference point may be located at the Nth reference point from the left (N is a natural number) among the reference points corresponding to the jawline, and the sixth reference point may be located at the Nth reference point from the right among the reference points corresponding to the jawline.

The image correction system 10 may move the image along the X axis using the X coordinates of the midpoints of the fifth and sixth reference points and the X coordinates of the midpoint of the bounding box (S313). In one example, the image correction system 10 may move the image in the X axis by a difference between the X coordinate of the midpoints of the fifth and sixth reference points and the X coordinate of the midpoint of the bounding box.

Fig. 11 is a flowchart illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, FIG. 11 shows the Y-axis movement step (S320) of FIG.

Referring to FIGS. 2 and 11 , the image correction system 10 may acquire the Y-coordinate of the lower side of the bounding box (S321). In one example, the image correction system 10 may acquire the Y-coordinate of a lower vertex among the vertices of the bounding box as the lower Y-coordinate.

The image correction system 10 may move the image along the Y-axis based on the distance between the seventh reference point among the at least one reference point and the Y-coordinate of the lower side (S322). In an embodiment, the seventh reference point may be determined as a reference point positioned at the bottom among at least one reference point. In one embodiment, the seventh reference point may be determined as a central reference point among reference points formed along the jaw line.

12 is a diagram illustrating an image correction method according to an exemplary embodiment of the present disclosure. In detail, Fig. 12 shows an image movement method.

Referring to FIGS. 2 and 12 , the image correction system 10 determines the third reference point located third from the left among the reference points formed along the jaw line as the fifth reference point CP5, and determines the jaw line as the sixth reference point CP6. Among the reference points formed along the chin line, the 15th reference point located third from the right may be determined, and the 9th reference point located in the middle among the reference points formed along the jawline may be determined as the seventh reference point CP7. The ninth reference point may be located at the bottom of at least one reference point.

The image correction system 10 uses the X coordinate (x1) of the fifth reference point CP5 and the X coordinate (x2) of the sixth reference point CP6 to determine the X coordinate ((x1+x2)/ 2) can be obtained. In addition, the image correction system 10 determines the X coordinate ((x4+x5)/2) of the second midpoint CP_B using the coordinates of the vertices CP_B1, CP_B2, CP_B3, and CP_B4 of the bounding box BB. can be obtained

The image correction system 10 calculates the difference between the X coordinate ((x1+x2)/2) of the first midpoint CP_C from the X coordinate ((x4+x5)/2) of the second midpoint CP_B, The image can be moved in the X-axis direction using the calculated difference ((x4+x5-x1-x2)/2).

The image correction system 10 may acquire the Y coordinate y3 of the seventh reference point CP7 and the Y coordinate y5 of the lower side of the bounding box BB, and calculate a difference therebetween. The image correction system 10 may move the image in the Y-axis direction using the calculated difference (y5-y3).

According to an embodiment of the present disclosure, movement correction for an image may be automatically performed to match the bounding box BB designated by the user by automatically moving the image using coordinates of reference points. According to an embodiment of the present disclosure, a reference point for a user to cut an image may be determined based on the bounding box BB, and accordingly, the location of the image may be moved so that the user may use the image later. .

Exemplary embodiments have been disclosed in the drawings and specifications as described above. Embodiments have been described using specific terms in this specification, but they are only used for the purpose of explaining the technical spirit of the present disclosure, and are not used to limit the scope of the present disclosure described in the meaning or claims. Therefore, those of ordinary skill in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Therefore, the true technical scope of protection of the present disclosure should be determined by the technical spirit of the appended claims.

Claims (10)

Receiving an image including an object;
extracting a plurality of reference points corresponding to the object using a deep learning algorithm;
calculating an angle of a virtual line connecting a first reference point and a second reference point among the plurality of reference points;
rotating the image using the calculated angle;
Receiving information about a bounding box;
calculating coordinate distances of a third reference point and a fourth reference point among the plurality of reference points;
obtaining a horizontal length of the bounding box;
Calculating a ratio of the coordinate distance and the horizontal length; And
Adjusting the size of the image based on the calculated ratio; includes,
The first reference point is a first average coordinate of at least two points included in the left eye of the object among the plurality of reference points,
The second reference point is a second average coordinate of at least two points included in the right eye of the object among the plurality of reference points,
The third reference point is the Nth point from the left (N is a natural number of 3 or more) among the reference points corresponding to the jaw of the object,
The image correction method according to claim 1 , wherein the fourth reference point is an Nth point (N is a natural number equal to or greater than 3) from the right among reference points corresponding to the chin of the object. According to claim 1,
The first reference point and the second reference point correspond to positions symmetrical to each other with respect to a central axis of the object. According to claim 2,
The object is a human face, the plurality of reference points correspond to lines constituting the human face,
The image correction method of claim 1 , wherein the first reference point is included in the left eye of the object, and the second reference point is included in the right eye of the object. According to claim 3,
The first reference point is located at the leftmost among points included in the left eye of the object, and the second reference point is located at the most right among points included in the right eye of the object. According to claim 2,
Wherein the first reference point is located at the most left of points corresponding to the left eyebrow of the object, and the second reference point is located at the most right of points corresponding to the right eyebrow of the object. According to claim 2,
The first reference point is located at the leftmost point among points corresponding to the lips of the object, and the second reference point is located at the rightmost side among points corresponding to the lips of the object. delete According to claim 1,
The first reference point is located at the top of points corresponding to the nose of the object, and the second reference point is located at the bottom of points corresponding to the nose of the object. According to claim 1,
The angle of the virtual line is an angle between a horizontal line and the virtual line,
The rotating of the image comprises rotating the image such that the virtual line matches the horizontal line. According to claim 1,
Changing the location of the image to correspond to the bounding box using a fifth reference point among the plurality of reference points;
The step of changing the location of the image,
calculating the X coordinate of the midpoint of the bounding box;
calculating midpoints of the third reference point and the fourth reference point;
moving along the X axis of the image so that midpoints of the third and fourth reference points coincide with X coordinates of the middle point of the bounding box;
obtaining the Y-coordinate of the lower edge of the bounding box; and
and moving the fifth reference point along the Y-axis of the image so that the Y-coordinate of the lower side of the bounding box coincides with the fifth reference point.

Images