KR102268981B1 - Original image detection apparatus using feature values in 360 degree images and method thereof - Google Patents

Abstract

The present invention relates to an apparatus for detecting an original image using feature values in a 360-degree image and method thereof. The apparatus for detecting an original image according to the present invention comprises: an image input unit for receiving an image to be judged that the image is taken in a 360-degree image form; a frame extracting unit for extracting a plurality of frames corresponding to a time point through the comparison of RGB variation between frames and characteristic information in the received image to be judged; a projection type conversion unit for converting the extracted frame into an equi-rectangular projection (ERP) type; a feature point extraction unit for extracting a feature point from the converted frame; and an image judgment unit for comparing the feature points extracted from the frame of the original image with the feature points extracted from the frame of the image to be judged to calculate the degree of agreement, and determining whether the original image matches the image to be judged using the calculated degree of agreement. According to the present invention, the identification recognition rate may be improved by providing a feature extraction method robust against attacks in consideration of the projection method and the reproduction method, which are the characteristics of a 360-degree image work.

Description

Original image detection apparatus using feature values in 360 degree images and method thereof

The present invention relates to an apparatus and method for detecting an original image using feature values in a 360-degree image, and more particularly, to a 360-degree original image of the corresponding judgment target image by analyzing the feature values extracted from the judgment target image. The present invention relates to an apparatus and method for detecting an original image using feature values in an image.

Media 　Technology has been steadily evolving to provide realism 　 such as 3D, UHD, panoramic, hologram, 360 VR, etc.

Here, the 360° image is an image that is stitched as if taken with a single camera without overlapping or disconnecting after shooting all 360-degree directions with multiple cameras at the same time, and using a head mounted display (HMD). If you watch using it, you can get a sense of immersion similar to the real world.

On the other hand, as 5G services from mobile carriers are fully serviced, large-capacity 360-degree video works can be used at a high speed anytime, anywhere, so they are illegally distributed in web hard, torrent, etc. through the dismantling of DRM and various image modifications.

Accordingly, there is a demand for a filtering technology supporting robustness for copyright protection.

Illegal copyrighted video filtering technology is a technology that blocks illegal Internet transmission of copyrighted materials by determining whether videos, music, games, etc. distributed online are illegally reproduced videos without the permission of the copyright holder.

This illegal video filtering technology should have a function to identify content basically. That is, after extracting feature information that can be a unique characteristic from each original video content, it is compared with the feature information of an illegally distributed work to determine whether the content is the same. Here, what kind of characteristic information is used and how it plays an important role in determining the performance of filtering technology, that is, the blocking rate of illegally copied images.

In particular, a 360-degree image must be robust to size change, brightness change, resolution change, and rotation change, as well as be robust against projection transformation and viewpoint transformation, which are new transformations in 360-degree images. However, the conventional illegal copyrighted image filtering technology is mainly used to determine whether a 2D image is illegally copied, and has a problem in that it cannot be applied to a 3D image such as a 360-degree virtual reality (VR) image.

The technology that is the background of the present invention is disclosed in Republic of Korea Patent Publication No. 10-1439475 (2014.09.17. Announcement).

An object of the present invention is to provide an apparatus and method for detecting an original image using feature values in a 360-degree image for detecting the original image of the corresponding judgment target image by analyzing the feature values extracted from the judgment target image. .

In the original image detection apparatus using a feature value in a 360-degree image according to an embodiment of the present invention for achieving this technical problem, an image input unit for receiving a judgment target image photographed in a 360-degree image form, the input judgment target A frame extraction unit that extracts a plurality of frames corresponding to the point in time at which a scene is changed through comparison of inter-frame RGB variation and feature information in an image, a projection type conversion unit that converts the extracted frames into an ERP (Equi-Rectangular Projection) type; a feature point extracting unit for extracting feature points from the converted frame, and comparing the feature points extracted from the frame of the original image with the feature points extracted from the frame of the judgment target image to calculate a degree of agreement, and using the calculated degree of agreement to calculate the match and an image determination unit that determines whether the image matches the image to be determined.

A frame transforming unit that generates each transformed frame by applying a plurality of transforming attacks to the original frame extracted from the original image according to the filtering performance evaluation items, and a feature point extracted from the original frame using the Euclidean algorithm and a plurality of transforms Create an original dataset that matches each extracted feature point with each other to extract the same existing reference feature point, and creates an original dataset including information on the reference feature point and copyright information of the original image for each of a plurality of original frames It may include more wealth.

The plurality of transformation attacks include logo insertion, subtitle insertion, image compression ratio conversion, codec conversion, aspect ratio conversion, resolution conversion, frame rate reduction, rotation, inversion, black and white conversion, brightness conversion, contrast effect change, projection conversion, monoscopic It may include at least one of cropping, stereoscopic cropping, and basic view movement.

If the input image is the original image, the frame extractor sequentially extracts a plurality of original frames corresponding to the time point at which the scene is changed, and if the input image is the image to be determined, a query corresponding to the time point at which the scene is changed Frames can be extracted as many as a randomly set number.

The feature point extraction unit down-samples the original frame, the modified frame, and the query frame converted to the ERP type at the same rate of resolution, and selects a central area corresponding to 45 degrees up and down from the center of the original frame, the modified frame, and the query frame. A feature point extraction region is selected, and a feature point can be extracted from an object included in the selected feature point extraction region.

The image determination unit, using a Euclidean algorithm, the reference feature points of the stored plurality of original datasets and A degree of agreement may be calculated by comparing feature points of the query frame, and a weight may be assigned to a corresponding query frame according to the calculated degree of agreement.

The image determining unit may initialize by assigning a weight of 1 when the degree of matching is equal to or greater than a reference value, and assigning a weight of 0 when the degree of matching is less than the reference value.

The image determination unit may be configured to determine that, if the query frames weighted by 1 among the plurality of query frames are consecutive, the determination target image and the original image are identical, and a query weighted by 1 among the plurality of query frames If the frames are non-consecutive, it may be finally determined that the judgment target image is deformed.

In addition, in the method for detecting an original image from a 360-degree image using the original image detection apparatus according to an embodiment of the present invention, the original frame corresponding to the time point at which the scene changes from the original image captured in the form of a 360-degree image is continuously generated. Extracting a plurality of frames, applying a plurality of transformation attacks to each of the extracted original frames to generate each transformed frame, extracted from the feature points of the object extracted from each original frame and a plurality of transformed frames corresponding to each of the original frames. After extracting the reference feature points that appear in common from the feature points of the object, storing the original dataset including the reference feature points and the copyright information of the 360-degree image in a database for each of a plurality of original frames, receiving a judgment target image from the user , extracting a query frame corresponding to a time point when a scene is changed by comparing the RGB variation between frames and feature information in the judgment target image, converting the extracted query frame to an ERP type, from the converted query frame extracting the feature points, and calculating the degree of agreement by comparing the reference feature points included in the original dataset with the feature points extracted from the query frame, and determining whether the original image matches the judgment target image using the calculated coincidence degree including judging.

As described above, according to the present invention, it is possible to improve the identification recognition rate by providing a feature extraction method that is robust against attacks in consideration of the projection method and the reproduction method, which are characteristics of a 360-degree image work.

In addition, according to the present invention, it is possible to provide a feature extraction method for filtering an attack processing in the form of a 2D image and an attack processing in the form of a 360-degree image, and changing the viewpoint information that is the standard of the image in the 360-degree image It is also possible to provide a feature extraction method for filtering attacks.

In addition, according to the present invention, a total of 16 transformed image frames are generated according to the items that can evaluate the filtering performance based on media features, and the dimension of the feature points extracted from the transformed image frame is reduced by using the Euclidean algorithm. , it has the effect of reducing the time and complexity required for filtering.

1 is a block diagram illustrating an original image detection apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating a method of detecting an original image in a 360 degree image using the original image detecting apparatus according to an embodiment of the present invention.
FIG. 3 is a flowchart for explaining step S250 shown in FIG. 2 .
FIG. 4 is an exemplary diagram for explaining step s251 shown in FIG. 3 .
FIG. 5 is an exemplary diagram for explaining step S252 shown in FIG. 3 .
FIG. 6 is a view for explaining step S270 shown in FIG. 2 .
FIG. 7 is a flowchart for explaining step S300 shown in FIG. 2 .

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings. In this process, the thickness of the lines or the size of the components shown in the drawings may be exaggerated for clarity and convenience of explanation.

In addition, the terms described below are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of the user or operator. Therefore, definitions of these terms should be made based on the content throughout this specification.

Hereinafter, the original image detection apparatus 100 according to an embodiment of the present invention will be described in more detail with reference to FIG. 1 .

1 is a block diagram illustrating an original image detection apparatus according to an embodiment of the present invention.

As shown in FIG. 1 , the original image detection apparatus 100 according to an embodiment of the present invention includes an image input unit 110 , a frame extraction unit 120 , a projection type conversion unit 130 , and a frame transformation unit 140 . , a feature point extraction unit 150 , an original data set generation unit 160 , and an image determination unit 170 .

First, the image input unit 110 receives an image captured in the form of a 360-degree image. Here, the image may be an original image or a judgment target image.

Next, the frame extraction unit 120 extracts a frame from the original image or the judgment target image. In other words, the frame extraction unit 120 extracts a plurality of frames corresponding to the time point at which the scene changes from the original image or the judgment target image by comparing the RGB variation between frames and the characteristic information. Hereinafter, a frame extracted from the original image is referred to as an original frame, and a frame extracted from the judgment target image is referred to as a query frame.

The projection type conversion unit 130 determines the projection type of the 360-degree image input to the image input unit 110 , and converts it into an ERP (Equirectangular Projection) type image.

The frame transforming unit 140 applies a transform attack to the ERP type original frame by using the filtering performance evaluation item to generate each transformed frame. Here, multiple transformation attacks include logo insertion, subtitle insertion, image compression ratio conversion, codec conversion, aspect ratio conversion, resolution conversion, frame rate reduction, rotation, inversion, black and white conversion, brightness conversion, contrast effect change, projection conversion, monoscopic crop. It includes at least one of ping, stereoscopic cropping, and basic view movement.

The feature point extraction unit 150 extracts a feature point from an object included in the frame. In other words, the keypoint extraction unit 140 selects a keypoint extraction area from the original frame, the query frame, and the modified frame converted to the ERP type. Next, the keypoint extraction unit 140 extracts a keypoint from an object included in the selected keypoint extraction area.

The original dataset generator 160 extracts the same existing reference feature points by matching the feature points extracted from the original frame with the feature points extracted from a plurality of deformed frames corresponding to the original frame. In addition, the original dataset generating unit 160 generates an original dataset including information on the extracted reference feature points and copyright information of the original image. Next, the original dataset generating unit 160 stores the original dataset generated for each of the plurality of original frames in the database.

Finally, the image determination unit 170 determines whether the determination target image matches the original image. In other words, the image determination unit 170 compares the feature points extracted from the query frame of the image to be judged with the reference feature points included in the plurality of original datasets, respectively, and calculates the degree of agreement. Then, the image determination unit 160 determines whether the determination target image and the original image match using the calculated degree of matching.

Hereinafter, a method of detecting an original image in a 360-degree image using the original image detecting apparatus 100 according to an embodiment of the present invention will be described in more detail with reference to FIGS. 2 to 6 .

2 is a flowchart illustrating a method of detecting an original image in a 360 degree image using the original image detecting apparatus according to an embodiment of the present invention.

A method of detecting an original image from a 360-degree image according to an embodiment of the present invention includes extracting a reference feature point and detecting an original image corresponding to an image to be judged by using the reference feature point. As shown in FIG. 2 , the step of generating the original dataset includes steps S210 to S260, and the step of detecting the original image includes steps S270 to S300.

First, the image input unit 110 receives an original image captured in the form of a 360-degree image (S210).

Here, the 360-degree image represents an image in which all 360-degree directions are simultaneously captured by multiple cameras, and then the captured images are stitched together as if they were captured by a single camera without overlapping or interruption.

Accordingly, the image input unit 110 receives an original 360-degree edited image from the user.

Then, the frame extraction unit 120 extracts a plurality of original frames from the input original image (S220).

In other words, the higher the rate of change between frames, the higher the probability of corresponding to a different scene. Therefore, the frame extraction unit 120 extracts the original frame corresponding to the point at which the scene is changed by comparing the RGB change amount between frames and the characteristic information. do. In this case, the frame extraction unit 120 extracts each original frame whenever a scene is changed in order to accurately determine whether the image to be judged is illegally copied. And the extracted frames have continuity.

When the step S220 is completed, the projection type conversion unit 130 converts the plurality of extracted original frames into an ERP (Equirectangular Projection) type (S230).

In other words, the projection type conversion unit 130 determines the projection type of the 360-degree original image input from the image input unit 110 . As a result of the determination, when the input original image is not an ERP (Equirectangular Projection) type, the projection type conversion unit 130 converts the plurality of frames extracted by the frame extraction unit 120 into the ERP type, respectively.

In the embodiment of the present invention, since frames are extracted based on the ERP type image, if the input 360-degree image is a CMP (cube-map projection) type, the frame is extracted by converting it to the ERP type, and a fisheye lens is used. If it is a type, after stitching each image, it is preferable to convert the image using the spherical coordinate system and then convert it to the ERP type to extract the frame.

Then, the frame transforming unit 140 generates a total of 16 frames by applying 16 transforming attacks to each original frame converted to the ERP type (S240).

The frame transforming unit 140 performs a transforming attack according to a filtering performance evaluation item based on immersive media features.

As shown in Table 1, the transformation attack consists of 16 types.

Among the variant attack items, logo Insert, Caption Insert, Severe compression, Codec Change, Aspect ratio Change, Resolution Change, Frame Frame-rate reduction, Rotation, Flip, Color to monochrome conversion, Brightness Change, and Contrast Change provide filtering performance in conventional 2D images. It corresponds to the item for evaluation.

Therefore, in the present invention, in addition to the 12 filtering performance evaluation items, four filtering performance evaluation items applicable to a 360-degree image are additionally applied.

To explain this in more detail, first, the projection conversion is used to filter attacks that may occur during the interconversion process between the projection method isotropic, Equirectangular Projection, and Cube Map Projection. attack method.

And monoscopic cropping (Monoscopic Cropping) is an attack method for filtering against an attack of recording a part of a 360-degree image and processing it in a 2D form.

And stereoscopic cropping (Stereoscopic Cropping) is an attack method for filtering against an attack that records a part of a 360-degree image and processes it in the form of a 360-degree image.

Lastly, move the base point is an attack method for filtering attacks that change viewpoint information that is a reference point of an image in a 360-degree image.

That is, the frame transforming unit 140 generates 16 transformed frames by applying 16 transforming attacks to each original frame.

Next, the feature point extraction unit 150 extracts each feature point from the original frame and 16 modified frames (S250).

Hereinafter, step S250 will be described in more detail with reference to FIG. 3 .

FIG. 3 is a flowchart for explaining step S250 shown in FIG. 2 .

As shown in FIG. 3 , the feature extraction unit 150 according to an embodiment of the present invention down-samples an original frame and 16 modified frames ( S251 ).

FIG. 4 is an exemplary diagram for explaining step s251 shown in FIG. 3 .

As shown in FIG. 4 , the keypoint extraction unit 150 improves the recognition rate by down-sampling the original frame and 16 modified frames at the same rate of resolution in order to reduce the keypoint extraction speed and the number of characteristic information.

Next, the feature extraction unit 150 selects a feature extraction region from the down-sampled original frame and 16 modified frames (S252).

FIG. 5 is an exemplary diagram for explaining step S252 shown in FIG. 3 .

As shown in FIG. 5 , when a sphere-shaped frame is spread in a planar shape by applying the ERP type, large distortion occurs at the upper and lower portions of the frame. Accordingly, the feature extraction unit 150 selects the central region in which the distortion is smallest and the most information is concentrated as the feature extraction region. That is, the feature point extraction unit 150 selects an area corresponding to 45 degrees up and down from the center of the frame as the feature extraction area.

Next, the feature point extraction unit 150 extracts an object from the selected feature extraction area, and acquires a feature point from the extracted object (S253).

The feature point extraction unit 150 extracts an object included in the selected feature extraction area, and removes a background other than the extracted object. Next, the feature point extraction unit 150 extracts a feature point from the object.

When step S250 is completed, the original dataset generating unit 160 extracts the reference feature points by matching the feature points extracted from the object of the original frame with the feature points extracted from the objects of 16 modified frames, and uses the extracted reference feature points to create an original dataset (S260).

In other words, the original data set generator 160 selects only matching feature information by comparing the feature points extracted from the original frame with each feature point extracted from 16 modified frames.

Next, the original data set generator 160 applies the Euclidean algorithm to the selected feature information to extract feature points that are identically present in the original frame and the 16 modified frames as reference feature points.

In addition, the original dataset generating unit 160 generates an original dataset including the extracted reference feature points and copyright information of the original image. The original dataset is repeatedly generated as many as the number of original frames extracted from the original image. Then, the original data set generating unit 160 stores the generated original data set in the database.

When the creation of the original dataset is completed, the image input unit 110 receives a judgment target image from the user and extracts five query frames from the input judgment target image (S270).

FIG. 6 is a view for explaining step S270 shown in FIG. 2 .

First, the frame extractor 120 cuts the first half of the image to be judged, which is an intro part, for about 20 seconds.

Next, as shown in FIG. 6 , the frame extraction unit 120 extracts a frame corresponding to a time point at which a scene is changed as a query frame by comparing the RGB change amount and feature information. In addition, the frame extraction unit 120 randomly extracts query frames of “a”, “b”, “c”, “d”, and “e” from the total media length of the image to be judged.

In the present invention, a total of five query frames are extracted to reduce the error probability, but the present invention is not limited thereto, and more or less than five query frames may be extracted.

When step S270 is completed, the projection type conversion unit 130 converts the extracted query frame into an ERP type (S280).

In other words, the projection type conversion unit 130 determines the projection type of the determination target image input from the image input unit 110 . As a result of the determination, when the input determination target image is not an ERP (Equirectangular Projection) type, the projection type conversion unit 130 converts the plurality of frames extracted by the frame extraction unit 120 into the ERP type, respectively.

Next, the keypoint extraction unit 150 extracts keypoints from five query frames (S290).

The keypoint extraction unit 150 extracts keypoints from five query frames in the same way as in step S250 . In other words, the feature extraction unit 150 down-samples five query frames, and then selects a central region of the query frame as the feature extraction region. Then, the feature point extraction unit 150 extracts a feature point from an object included in the selected feature extraction area.

When the feature point extraction is completed, the image determination unit 170 receives the original dataset stored in the database. Then, the image determination unit 170 compares the reference feature points of the original dataset with the feature points of the five query frames to determine whether the judgment target image matches the original image (S300).

Hereinafter, the step S300 shown in FIG. 2 will be described in more detail with reference to FIG. 7 .

FIG. 7 is a flowchart for explaining step S300 shown in FIG. 2 .

As shown in FIG. 7 , the image determination unit 170 receives a plurality of original data sets ( S301 ).

The image determination unit 170 receives a plurality of original data sets pre-stored in the database. For example, assuming that 50 original frames are extracted from the original image, the image determining unit 170 receives 50 original datasets.

Next, the image determination unit 170 compares each reference feature point included in the 50 original datasets with the feature points extracted from five query frames to calculate a similarity (S302).

That is, the image determination unit 170 compares the reference feature points included in the first original dataset with the feature points of the “a” frame using the Euclidean algorithm to calculate the similarity, and to the first original dataset. A similarity is calculated by comparing the included reference feature points with the feature points of the “b” frame. In addition, the image determining unit 170 compares the reference feature points included in the first original dataset with the feature points of the “c” frame to calculate the similarity, and compares the reference feature points included in the first original dataset with the reference feature points included in the first original dataset and the “d” frame. The similarity is calculated by comparing the feature points, and the similarity is calculated by comparing the reference feature points included in the first original dataset with the feature points of the “e” frame.

Next, the image determining unit 170 calculates similarity by comparing the reference feature points included in the second original dataset with the feature points of the “a” frame, and compares the reference feature points included in the second original dataset with the reference feature points included in the second original dataset and the “b” frame. The similarity is calculated by comparing the feature points of . In addition, the image determining unit 170 calculates a similarity by comparing the reference feature points included in the second original dataset with the feature points of the “c” frame, and compares the reference feature points included in the second original dataset with the reference feature points included in the second original dataset and the “d” frame. The similarity is calculated by comparing the feature points, and the similarity is calculated by comparing the reference feature points included in the second original dataset with the feature points of the “e” frame.

The similarity is calculated by iteratively performing until the 50th original dataset in the same way as above.

When step S302 is completed, the image determination unit 170 assigns a weight to the calculated similarity (S303).

The image determining unit 170 assigns a weight of 1 if the calculated similarity is 80% or more, and gives a weight of 0 if the calculated similarity is less than 80%.

For example, assuming that 80% similarity is obtained by comparing the reference feature points included in the first original dataset with the feature points of the “a” frame, the image determining unit 170 assigns a weight of 1 to the “a” frame .

Next, the image determination unit 170 determines whether the judgment target image matches the original image by using the weights assigned to each of the five query frames corresponding to the original dataset (S304).

In other words, if three or more query frames given a weight of 1 among the plurality of query frames are consecutive, the image determination unit 170 determines that the judgment target image matches the original image.

On the other hand, if the query frames given the weight of 1 among the plurality of query frames are non-consecutive, the image determination unit 170 finally determines that the determination target image is deformed.

For example, corresponding to the first original dataset, query frame "a", query frame "c", query frame "d", and query frame "e" have a weight of 1, and query frame "b" has a weight of 0. Assume that Then, since the number of query frames weighted by 1 is three or more, and the query frames of “c”, “d”, and “e” are continuous, the image determination unit 170 determines that the judgment target image matches the original image. judge to do

Also, it is assumed that the weight of the query frame "a", the query frame "b", the query frame "d" and the query frame "e" is 1, and the weight of the query frame "c" is 0. Therefore, although the condition that the number of query frames weighted by 1 is 3 or more is satisfied, the weighted query frames are non-consecutive, so the image determination unit 170 finally determines that the determination target image is deformed.

As described above, the original image detection apparatus according to the present invention can improve the identification recognition rate by providing a feature extraction method that is robust against attacks in consideration of the projection method and the reproduction method, which are characteristics of a 360-degree image work.

In addition, the original image detection device according to the present invention can provide a feature extraction method for filtering an attack processing in the form of a 2D image and an attack processing in the form of a 360-degree image. It is also possible to provide a feature extraction method for filtering attacks that change viewpoint information.

In addition, the original image detection apparatus according to the present invention generates a total of 16 types of modified image frames according to the items that can evaluate the filtering performance based on media features, and uses the Euclidean algorithm to determine the feature points extracted from the modified image frames. By reducing the dimension, there is an effect of reducing the time and complexity required for filtering.

Although the present invention has been described with reference to the embodiment shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible therefrom. will be. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the following claims.

100: original image detection device
110: video input unit
120: frame extraction unit
130: projection type conversion unit
140: frame deformation part
150: feature point extraction unit
160: original data set generation unit
170: image judgment unit

Claims (16)

In the original image detection apparatus using a feature value in a 360 degree image,
An image input unit for receiving an image to be judged, which is taken in the form of a 360-degree image,
a frame extracting unit for extracting a plurality of frames corresponding to the time point at which a scene is changed through comparison of the inter-frame RGB change amount and feature information from the input judgment target image;
A projection type conversion unit that converts the extracted frame into an ERP (Equi-Rectangular Projection) type,
a feature point extraction unit for extracting feature points from the converted frame;
An image determination unit that compares the feature points extracted from the frame of the original image with the feature points extracted from the frame of the judgment target image to calculate a degree of agreement, and determines whether the original image matches the judgment target image using the calculated degree of agreement includes,
a frame transforming unit for generating each transformed frame by applying a plurality of transforming attacks to the original frame extracted from the original image according to the filtering performance evaluation item; and
By using the Euclidean algorithm, the feature points extracted from the original frame and the feature points extracted from a plurality of modified frames are matched with each other to extract the same reference feature points, and information about the reference feature points and copyright information of the original image are included. Includes an original dataset generator that generates the original dataset for each of a plurality of original frames,
The image determination unit,
An original that compares the reference feature points of the generated plurality of original datasets with the feature points of the query frame using the Euclidean algorithm to calculate the degree of agreement, and weights the corresponding query frame according to the calculated degree of agreement image detection device. delete According to claim 1,
The plurality of transformation attacks are
Logo Insertion, Subtitle Insertion, Video Compression Rate Conversion, Codec Conversion, Aspect Ratio Conversion, Resolution Conversion, Frame Rate Reduction, Rotation, Invert, B&W Conversion, Brightness Conversion, Contrast Effect Conversion, Projection Conversion, Monoscopic Cropping, Stereoscopic Crop An original image detection apparatus comprising at least one of a ping and a basic viewpoint movement. According to claim 1,
The frame extraction unit,
If the input image is an original image, sequentially extracting a plurality of original frames corresponding to the time when the scene is changed,
If the input image is the image to be determined, the original image detection apparatus for extracting a randomly set number of query frames corresponding to the time when the scene is changed. According to claim 1,
The feature point extraction unit,
The original frame, the modified frame, and the query frame converted to the ERP type are down-sampled at the same rate of resolution, and the central area corresponding to 45 degrees up and down from the center of the original frame, the modified frame and the query frame is selected as the feature point extraction area. and
An original image detection apparatus for extracting a feature point from an object included in the selected feature point extraction area. delete According to claim 1,
The image determination unit,
The original image detection apparatus initializes by assigning a weight of 1 when the degree of agreement is greater than or equal to a reference value, and assigning a weight of 0 when the degree of agreement is less than the reference value. 8. The method of claim 7,
The image determination unit,
If the query frames given a weight of 1 among the plurality of query frames are continuous, it is determined that the judgment target image and the original image match, and among the plurality of query frames, if the query frames given a weight of 1 are non-consecutive , an original image detection apparatus that finally determines that the determination target image is deformed. In the original image detection method in a 360 degree image using the original image detection device,
A plurality of original frames corresponding to a scene change point are successively extracted from an original image taken in the form of a 360-degree image, and a plurality of transformation attacks are applied to each of the extracted original frames to generate each transformed frame, After extracting a reference feature point that appears in common in the feature point of the object extracted from the original frame and the feature point of the object extracted from a plurality of modified frames corresponding to each of the original frame, the reference feature point and the copyright information of the 360-degree image are extracted Storing the dataset in a database for each of a plurality of original frames;
receiving a judgment target image from a user;
extracting a query frame corresponding to a time point at which a scene is changed by comparing the RGB variation between frames and feature information in the judgment target image;
converting the extracted query frame into an ERP type;
extracting feature points from the transformed query frame; and
Comparing the reference feature points included in the original dataset with the feature points extracted from the query frame to calculate a degree of agreement, and using the calculated degree of agreement to determine whether the original image matches the judgment target image,
The step of creating and saving the original dataset for each of a plurality of original frames is:
Applying a plurality of deformation attacks to the original frame extracted from the original image according to the filtering performance evaluation item to generate each modified frame,
By using the Euclidean algorithm, the feature points extracted from the original frame and the feature points extracted from a plurality of modified frames are matched with each other to determine the same existing feature points as the reference feature points,
The step of determining whether the original image and the judgment target image match,
The original image in which the degree of agreement is calculated by comparing the reference feature points of the plurality of stored original datasets with the feature points of the query frame using the Euclidean algorithm, and weight is given to the query frame according to the calculated degree of agreement detection method. delete 10. The method of claim 9,
The plurality of transformation attacks are
Logo Insertion, Subtitle Insertion, Video Compression Rate Conversion, Codec Conversion, Aspect Ratio Conversion, Resolution Conversion, Frame Rate Reduction, Rotation, Invert, B&W Conversion, Brightness Conversion, Contrast Effect Conversion, Projection Conversion, Monoscopic Cropping, Stereoscopic Crop An original image detection method comprising at least one of ping and basic viewpoint movement. 10. The method of claim 9,
Extracting the query frame comprises:
An original image detection method that extracts a randomly set number of query frames corresponding to a scene change point. 10. The method of claim 9,
The step of extracting the feature point,
Downsampling the query frame converted to the ERP type at the same rate of resolution, and selecting a central area corresponding to 45 degrees up and down from the center of the query frame as a feature point extraction area,
An original image detection method for extracting a feature point from an object included in the selected feature point extraction area. delete 10. The method of claim 9,
The step of determining whether the original image and the judgment target image match,
An original image detection method in which a weight of 1 is given if the degree of agreement is equal to or greater than a reference value, and a weight of 0 is assigned if the degree of agreement is less than a reference value to initialize the original image. 16. The method of claim 15,
The step of determining whether the original image and the judgment target image match,
If the query frames given a weight of 1 among the plurality of query frames are continuous, it is determined that the judgment target image and the original image match, and among the plurality of query frames, if the query frames given a weight of 1 are non-consecutive , an original image detection method for finally determining that the judgment target image is deformed.

Images