KR102054347B1 - Method and apparatus for generating steganography analysis model - Google Patents

Abstract

The method for generating a steganography analysis model according to an embodiment of the present invention includes generating a second 3D mesh model obtained by smoothing a first 3D mesh model, an edge normal of the first 3D mesh model, and an edge of the second 3D mesh model. The total curvature and the first feature vector extracted based on the difference of the normal, the mean curvature of the first 3D mesh model, and the total curvature and the first feature vector extracted based on the difference of the mean curvature of the second 3D mesh model. 2 information extracted in the 3D mesh model based on the difference in the total curvatures, and machine learning based on the information about the 1st, 2nd and 3rd feature vectors. Generating a model for determining if a word exists.

Description

METHOD AND APPARATUS FOR GENERATING STEGANOGRAPHY ANALYSIS MODEL}

The present invention relates to a method and apparatus for generating a steganography analysis model, and more particularly, to a method and apparatus for generating a steganography analysis model for determining whether there is hidden information in a 3D mesh model.

Steganography is an encryption technique that inserts and transmits information to various types of digital content such as images, video, and audio so that it is difficult for a third party to recognize that special information is included.

This steganography technique is actively used in areas requiring security technology, but it can be used for distributing malicious code or a means of transmitting a message between terrorists, so that steganography technique is used to analyze what content is hidden. This is required.

In particular, in the case of the 3D mesh model among the contents to which the steganography technique can be applied, components such as vertices, edges, and faces that make up the 3D mesh model are different for each 3D mesh model. As a result, it is difficult to specify universal judgment factors for analyzing the existence of hidden information. In addition, the factors extracted from the 3D mesh model are linear dependent, so that the accuracy of judgment is applied when applied to machine learning. There is a difficulty in raising.

Korean Laid-Open Patent Publication No. 10-2017-0095508: Method of transmitting encrypted information using an image

The problem to be solved in the embodiment of the present invention is to provide a technique for analyzing whether there is hidden information in the 3D mesh model.

However, the technical problem to be achieved by the embodiment of the present invention is not limited to the above-mentioned problem, various technical problems can be derived within the scope apparent to those skilled in the art from the following description.

The method for generating a steganography analysis model according to an embodiment of the present invention may include generating a second 3D mesh model obtained by smoothing a first 3D mesh model, an edge normal of the first 3D mesh model, and the first 3D mesh model. A first feature vector extracted based on a difference in edge normal of the 3D mesh model, a second feature vector extracted based on a difference in mean curvature of the first 3D mesh model, and a mean curvature of the second 3D mesh model, and the second feature vector 1) extracting a third feature vector extracted based on the difference between the total curvature of the 3D mesh model and the total curvature of the second 3D mesh model; and based on the information about the first, second, and third feature vectors. Performing training to generate a model that determines whether there is hidden information in the target 3D mesh model.

In this case, the first 3D mesh model may include a cover mesh corresponding to the original and a stego mesh into which information hidden in the original is inserted.

In addition, the extracting may include generating a mean vector, a variance vector, a skewness vector, and a kurtosis vector for each feature vector.

In addition, the mean vector, the variance vector, the skewness vector, and the kurtosis vector generated from the respective feature vectors are inputted into kernel space to modify the number of dimensions of the input vector so as to have a linearly independent relationship with each other. The method may further include generating each of the transformed linear independent vectors, and the information about the first, second and third feature vectors may include the linear independent vectors.

In addition, the kernel space may be generated based on a homogeneous kernel map algorithm.

In addition, the edge normal may be a sum of vectors derived by multiplying the area of each mesh piece by a normal vector at the center of gravity of each mesh piece for each mesh piece sharing a common edge in a 3D mesh model.

In addition, the first feature vector may include l2-norm value of the edge normal of the first 3D mesh model and the second 3D mesh of the inner product of the edge normal of the first 3D mesh model and the edge normal of the second 3D mesh model. The ratio of the product multiplied by the l2-norm value of the edge normal of the model may be a value obtained by taking an inverse cosine.

In addition, the mean curvature may be an average of values derived by applying a principal component analysis (PCA) technique to a vertex of the 3D mesh model.

In addition, the total curvature may be a total of values derived by applying a principal component analysis (PCA) technique to a vertex of a 3D mesh model.

The extracting may include a fourth feature vector extracted based on a difference between an x-axis value of the first 3D mesh model and an x-axis value of the second 3D mesh model in a rectangular coordinate system, and the fourth 3D mesh model of the first 3D mesh model in the rectangular coordinate system. a fifth feature vector extracted based on a difference between a y-axis value and a y-axis value of the second 3D mesh model, based on a difference between a z-axis value of the first 3D mesh model and a z-axis value of the second 3D mesh model in the rectangular coordinate system A sixth feature vector extracted with a seventh feature vector extracted based on a difference between an x-axis value of the first 3D mesh model and an x-axis value of the second 3D mesh model in a laplacian coordinate system, and in the Laplacian coordinate system An eighth feature vector extracted based on a difference between the y-axis value of the first 3D mesh model and the y-axis value of the second 3D mesh model, the y-axis value of the first 3D mesh model, and the second in the Laplacian coordinate system A ninth feature vector extracted based on a difference in y-axis values of a 3D mesh model, a tenth extracted based on a difference in coordinates of a vertex of the first 3D mesh model and a vertex of the second 3D mesh model in the rectangular coordinate system; A feature vector, an eleventh feature vector extracted based on a difference between the coordinates of the vertices of the first 3D mesh model and the coordinates of the vertices of the second 3D mesh model in the Laplacian coordinate system, and the corners are shared by the first 3D mesh model. A twelfth feature vector extracted based on a difference in backside angles of two faces and backside angles of two faces that share edges in the second 3D mesh model, and a face of the first 3D mesh model. A thirteenth feature vector extracted based on an angle difference of a plane of the second 3D mesh model, a fourteenth feature extracted based on a difference between a vertex normal of the first 3D mesh model and a vertex normal of the second 3D mesh model vector, A fifteenth feature vector extracted based on a difference between the gaussian curvature of the first 3D mesh model and the gaussian curvature of the second 3D mesh model, the curvature ratio of the first 3D mesh model, and the curvature ratio of the second 3D mesh model And extracting a sixteenth feature vector extracted based on the difference of, and generating the model may perform machine learning by adding information about the fourth to sixteenth feature vectors.

According to an embodiment of the present invention, by matching the number of dimensions of the feature vectors derived from the edge normal, mean curvature, and total curvature of the 3D mesh model, even if the components such as vertices, edges, faces, etc. are different for each 3D mesh model Determining elements that can be used universally can be created.

In addition, the homogeneous kernel map algorithm can improve the accuracy of the machine-learned model by transforming feature vectors extracted from the 3D mesh model to have a linear independent relationship.

1 is an exemplary view for explaining the structure of a 3D mesh model according to an embodiment of the present invention.
2 is a flowchart illustrating a process of a method for generating a steganographic analysis model according to an embodiment of the present invention.
3 is a graph for explaining an area under the curve (AUC) index for measuring the performance of a steganographic analysis model according to an embodiment of the present invention.
4 is a graph illustrating a detection error of a steganography analysis model according to an embodiment of the present invention.

Advantages and features of the present invention, and methods of achieving them will be apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but can be implemented in various forms. The present embodiments merely make the disclosure of the present invention complete, and have the ordinary knowledge in the art to which the present invention pertains. It is provided to fully inform the scope of the invention, and the scope of the invention is defined only by the claims.

In describing the embodiments of the present invention, detailed descriptions of well-known functions or configurations will be omitted unless they are actually necessary in describing the embodiments of the present invention. In addition, terms to be described below are terms defined in consideration of functions in the embodiments of the present invention, which may vary according to intentions or customs of users or operators. Therefore, the definition should be made based on the contents throughout the specification.

The functional blocks shown in the figures and described below are only examples of possible implementations. Other functional blocks may be used in other implementations without departing from the spirit and scope of the detailed description. Also, while one or more functional blocks of the present invention are represented by separate blocks, one or more of the functional blocks of the present invention may be a combination of various hardware and software configurations that perform the same function.

In addition, the expression "comprising" certain components merely refers to the presence of the components as an open expression, and should not be understood as excluding additional components.

Further, when a component is referred to as being connected or connected to another component, it should be understood that there may be a direct connection or connection to that other component, but there may be other components in between.

In addition, an expression such as 'first' and 'second' is used only for distinguishing a plurality of components, and does not limit the order or other features between the components.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

1 is an exemplary view for explaining the structure of a 3D mesh model according to an embodiment of the present invention.

As shown in FIG. 1, a 3D mesh model is data of a geometric model generated from a set of polygons composed of vertices, edges, and faces, and is a computer-aided design. ) And three-dimensional printing. At this time, the steganography technique (for example, Korean Patent Application No. 10-2017-0089752: 3D mesh model watermarking method and apparatus using segmentation technique) makes it difficult for a third party to identify that the 3D mesh model includes special information. Some information can be inserted.

This steganography technique is actively used in areas requiring security technology, but it can be used for distributing malicious code or a means of transmitting a message between terrorists, so that steganography technique is used to analyze what content is hidden. This is required.

Accordingly, an embodiment of the present invention is to provide a technique for analyzing whether there is hidden information in the 3D mesh model, through the embodiment of the present invention and the method of implementing the embodiment of the present invention in conjunction with FIGS. Describe the performance of the model you created.

2 is a flowchart illustrating a process of a method for generating a steganographic analysis model according to an embodiment of the present invention. The method for generating a steganography analysis model according to an embodiment of the present invention may be performed by one or more processors, and each step will be described below.

First, a second 3D mesh model is generated by applying a smoothing technique to an original first 3D mesh model to which a smoothing technique is not applied (S210).

When steganography is applied to a 3D mesh model to insert specific information, it is difficult to identify the human eye, but some parts of the 3D mesh model are changed in shape. In this case, when the cover mesh in which the hidden information is not inserted is the first 3D mesh model, the stego mesh in which the hidden information is inserted is compared with the difference between the first 3D mesh model and the second 3D mesh model. In the case of the first 3D mesh model, the difference between the first 3D mesh model and the second 3D mesh model may be relatively large. Based on this point, an embodiment of the present invention generates a second 3D mesh model from the first 3D mesh model and uses the feature vector derived based on the difference between the first 3D mesh model and the second 3D mesh model as training data. use.

In this case, in order to learn a model to distinguish whether there is hidden information in the target 3D mesh model, the first 3D mesh model may include a cover mesh without hidden information inserted and a stego mesh with hidden information inserted.

Next, a plurality of feature vectors are extracted based on the difference between the first 3D mesh model and the second 3D mesh model (S220).

At this time, the embodiment of the present invention may extract the first feature vector based on the difference between the edge normal of the first 3D mesh model and the edge normal of the second 3D mesh model.

)를 공유하는 모든 메쉬 조각(

)의 각 메쉬 조각(f)에 대하여 각 메쉬 조각의 면적(Area(f)) 및 각 메쉬 조각의 무게 중심에서의 법선 벡터(

)를 곱하여 도출된 벡터들의 합(

)을 의미한다. Here, the edge normal of the 3D mesh model is a feature vector that can be obtained through Equation 1 below.

) Share all mesh pieces (

For each mesh piece f of), the area (f) of each mesh piece and the normal vector at the center of gravity of each mesh piece (

Sum of vectors derived by multiplying

).

)는 아래 수학식 2와 같이, 제1 3D 메쉬 모델의 edge normal(

)과 제2 3D 메쉬 모델의 edge normal(

)의 내적(<

>)에 대한 제1 3D 메쉬 모델 edge normal의 l2-norm(

)과 제2 3D 메쉬 모델 edge normal l2-norm(

)을 곱한 값의 역코사인(arccos)을 통해 도출할 수 있다. Accordingly, the first feature vector (

) Is an edge normal (1) of the first 3D mesh model as shown in Equation 2 below.

) And the edge normal of the second 3D mesh model

) Dot product (<

L2-norm () of the first 3D mesh model edge normal for

) And the second 3D mesh model edge normal l2-norm (

We can derive from inverse cosine (arccos) multiplied by).

In addition, the embodiment of the present invention may extract the second feature vector extracted based on the difference between the mean curvature of the first 3D mesh model and the mean curvature of the second 3D mesh model.

)는 3D 메쉬 모델의 정점(

)에 PCA(principal component analysis) 기법을 적용하여 도출된 값(

)들의 평균을 의미한다. 다만, 간소화를 위해 아래 수학식 3과 같이 PCA 기법을 통해 구해진 값의 최대값(

)과 최소값(

)의 평균으로도 구할 수 있다. 여기서, PCA 기법은 공지된 기술로서 자세한 설명은 생략한다. Where mean curvature (

) Is the vertex (

Value obtained by applying the principal component analysis (PCA) method to

Mean). However, for the sake of simplicity, the maximum value of the value obtained through the PCA method as shown in Equation 3 below (

) And minimum value (

It can also be obtained as an average of. Here, the PCA technique is a known technique and detailed description thereof will be omitted.

)는 아래 수학식 4와 같이, 제1 3D 메쉬 모델의 mean curvature(

)과 제2 3D 메쉬 모델의 mean curvature(

)의 차를 통해 구할 수 있다. Accordingly, the second feature vector (

) Is the mean curvature (1) of the first 3D mesh model, as shown in Equation 4 below.

) And the mean curvature of the second 3D mesh model

You can get it through the car.

In addition, the embodiment of the present invention may extract the third feature vector extracted based on the difference between the total curvature of the first 3D mesh model and the total curvature of the second 3D mesh model.

)는 3D 메쉬 모델의 정점(

)에 PCA(principal component analysis) 기법을 적용하여 도출된 값(

)들의 합을 의미한다. 다만, 간소화를 위해 아래 수학식 5와 같이 PCA 기법을 통해 구해진 값의 최대값(

)과 최소값(

)의 합으로도 구할 수 있다. Where total curvature (

) Is the vertex (

Value obtained by applying the principal component analysis (PCA) method to

) Means the sum of However, for the sake of simplicity, the maximum value of the value obtained through the PCA method as shown in Equation 5 below (

) And minimum value (

It can also be obtained as the sum of).

)는 아래 수학식 6과 같이, 제1 3D 메쉬 모델의 total curvature(

)와 제2 3D 메쉬 모델의 total curvature(

)의 차를 통해 구할 수 있다.Accordingly, the third feature vector (

) Is the total curvature () of the first 3D mesh model, as shown in Equation 6 below.

) And the total curvature (of the second 3D mesh model

You can get it through the car.

In addition, the embodiment of the present invention is a fourth feature vector extracted based on the difference between the x-axis value of the first 3D mesh model and the x-axis value of the second 3D mesh model in the Cartesian coordinate system, the y-axis value of the first 3D mesh model in the Cartesian coordinate system And a fifth feature vector extracted based on the difference in the y-axis values of the second 3D mesh model, the sixth feature vector extracted based on the difference in the z-axis values of the first 3D mesh model and the z-axis value of the second 3D mesh model in the Cartesian coordinate system. , A seventh feature vector extracted based on a difference between the x-axis value of the first 3D mesh model and the x-axis value of the second 3D mesh model in the laplacian coordinate system, and the y-axis value and the first value of the first 3D mesh model in the Laplacian coordinate system. 2 The eighth feature vector extracted based on the difference in the y-axis values of the 3D mesh model, The ninth feature vector extracted based on the difference in the y-axis values of the first 3D mesh model and the y-axis value of the second 3D mesh model in the Laplacian coordinate system, Orthogonal First in coordinate system A tenth feature vector extracted based on the difference between the coordinates of the vertices of the 3D mesh model and the coordinates of the vertices of the second 3D mesh model, the coordinates of the vertices of the first 3D mesh model in the Laplacian coordinate system, and the coordinates of the vertices of the second 3D mesh model Based on the eleventh feature vector extracted based on the difference of, the dihedral angle of the two faces sharing the edges in the first 3D mesh model, and the difference in the backside angles of the two faces sharing the edges in the second 3D mesh model. The 13 th feature vector extracted based on the angular difference between the face of the first 3D mesh model and the face of the first 3D mesh model, the vertex normal and the second vertex normal of the first 3D mesh model The fifteenth feature vector extracted based on the difference in the vertex normal of the 3D mesh model, the fifteenth feature vector and the first 3D extracted based on the difference of the gaussian curvature of the first 3D mesh model and the gaussian curvature of the second 3D mesh model Curvature rati of mesh model A sixteenth feature vector extracted based on a difference in curvature ratios of o and the second 3D mesh model may be generated. Herein, a detailed process of extracting the fourth to sixteenth feature vectors may be described in "Li, Z., Bors, AG: 3D mesh steganalysis using local shape features.In: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pp. 2144-2148. IEEE (2016).

Meanwhile, in step S220, a mean vector, a variance vector, a skewness vector, and a kurtosis for each feature vector extracted based on the difference between the first 3D mesh model and the second 3D mesh model. ) Generating the vector.

Accordingly, the embodiment of the present invention can significantly increase the training data by deriving four times the training data using the average vector, the variance vector, the skewness vector, and the kurtosis vector, which are additionally extracted from each vector.

In addition, the feature vector obtained from the 3D mesh model itself is different from the 3D mesh model because the number and shape of the vertices, edges, and faces that make up the 3D mesh model are different for each 3D mesh model. Although it is difficult to use it as a general purpose element for determining whether there is different hidden information for each, an embodiment of the present invention extracts each 3D mesh model using the mean vector, the variance vector, the skewness vector, and the kurtosis vector as training data. Since the dimensions of the vector to be changed can be changed to be the same, it is possible to extract general judgment elements for various 3D mesh models. On the other hand, the process of obtaining the mean vector, the variance vector, the skewness vector, and the kurtosis vector from a set of vectors refers to statistics ("https://en.wikipedia.org/wiki/Moment_(mathematics)") as a known mathematical method. can do.

Next, a mean vector, a variance vector, a skewness vector, and a kurtosis vector generated from each feature vector are inputted into a kernel space that transforms the number of dimensions of the input vector to have a linearly independent relationship with each other. Each vector may include generating linear independent vectors transformed to have a linear independent relationship with each other (S230).

The machine learning algorithm is characterized in that the accuracy of the model is poor when learning with the training data in a linear dependent relationship. In this case, since the feature vectors extracted from the 3D mesh model are linearly dependent on each other, in step S230, the feature vectors extracted in the step S220 in the kernel space are modified so that they have a linearly independent relationship with each other. By inputting them, a linear independent vector obtained by converting each feature vector may be generated. The kernel space is homogeneous kernel map algorithm (Vedaldi, A., Zisserman, A .: Ecient additive kernels via explicit feature maps.IEEE Trans.Pattern Anal.Mach.Intel. (TPAMI) 34 (3). , 480-492 (2012)), and the kernel space may be modified to have a higher dimension than the number of dimensions of the input vector so that the output vectors may have a linear independent relationship with each other.

On the other hand, S230 is a step that can be selectively performed to further improve the effect of the present invention, the processor may perform only steps S210, S220 and S240 to be described later except step S230.

Subsequently, machine learning is performed based on the information about the feature vector extracted in step S220 or S230 to generate a model for determining whether there is hidden information in the target 3D mesh model (S240).

In this case, the information about the feature vector used as the training data to train the model may be, for example, an average vector extracted from each of the first to sixteenth feature vectors, or the first to sixteenth feature vectors, and the variance. Vector, skewness vector and kurtosis vector (hereinafter referred to as 'LFS64'), or linear independent vector (hereinafter referred to as 'LFS64 + HKM') modified by inputting LFS64 into kernel space.

In addition, when the steps S210 and S220 are performed based on the first 3D mesh model corresponding to the cover mesh, a label representing information indicating that hidden information has not been inserted is set as an output variable of the model, and the first corresponding to the stego mesh. When the steps S210 and S220 are performed based on the 3D mesh model, the model may be trained by setting a label representing information indicating that hidden information is inserted as an output variable of the model.

3 is a graph illustrating an AUC (area under the curve) index for measuring the performance of a steganographic analysis model according to an embodiment of the present invention, Figure 4 is a stegano according to an embodiment of the present invention It is a graph for explaining the detection error of a graph analysis model.

3 and 4 show the results of experiments that test the performance of models generated by applying various techniques, using 3D mesh model data stored in The Princeton Mesh Segmentation database as training data. In this case, a technique of inserting hidden information into a cover mesh stored in a database includes a first steganography technique (Chao, MW, Lin, CH, Yu, CW, Lee, TY: A high capacity 3D steganography algorithm.IEEE Trans) Vis.Comput.Graph. (TVCG) 15 (2), 274-284 (2009)) and second steganography techniques (Cho, JW, Prost, R., Jung, HY: An oblivious watermarking for 3-D polygonal meshes using distribution of vertex norms.IEEE Trans.Sig.Process. (TSP) 55 (1), 142-155 (2007)).

At this time, the LFS52 shown in FIGS. 3 and 4 is (Li, Z., Bors, AG: 3D mesh steganalysis using local shape features.In: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2144-2148 LFS64 is an average vector, variance vector, and skewness extracted from each of the first to sixteenth feature vectors of the present invention. The result of using a model generated based on the vector and the kurtosis vector, LFS64 + HKM is the average vector, variance vector, skewness vector and kurtosis vector extracted from each of the first to sixteenth feature vectors through kernel space. Means the result using the model generated based on the modified linear independent vectors.

In this case, referring to FIG. 3, both the first steganography technique (a) and the second steganography technique (b) show that the embodiments of the present invention have a much wider lower area of the curve than the conventional technique, and thus the accuracy is greatly improved. 4, it can be seen that the embodiments of the present invention reduce the detection error rate in both the first steganography technique (a) and the second steganography technique (b).

According to the embodiment described above, by matching the number of dimensions of the feature vectors derived from the edge normal, mean curvature, and total curvature of the 3D mesh model, even if the components such as vertices, edges, and faces are different for each 3D mesh model It can create judgment elements that can be used as a target.

In addition, the homogeneous kernel map algorithm can improve the accuracy of machine-learned models by transforming feature vectors extracted from 3D mesh models to have a linear independent relationship.

Embodiments of the present invention described above may be implemented through various means. For example, embodiments of the present invention may be implemented by hardware, firmware, software, or a combination thereof.

For implementation in hardware, a method according to embodiments of the present invention may include one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), and Programmable Logic Devices (PLDs). It may be implemented by field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, and the like.

In the case of an implementation by firmware or software, the method according to the embodiments of the present invention may be implemented in the form of a module, a procedure, or a function that performs the functions or operations described above. The computer program in which the software code or the like is recorded can be stored in a computer readable recording medium or a memory unit and driven by a processor. The memory unit may be located inside or outside the processor, and may exchange data with the processor by various known means.

In addition, the combination of each block of the block diagram and each step of the flowchart attached to the present invention may be performed by computer program instructions. These computer program instructions may be mounted on an encoding processor of a general purpose computer, special purpose computer, or other programmable data processing equipment such that the instructions executed by the encoding processor of the computer or other programmable data processing equipment are not included in each block or block diagram. It will create means for performing the functions described in each step of the flowchart. These computer program instructions may be stored in a computer usable or computer readable memory that can be directed to a computer or other programmable data processing equipment to implement functionality in a particular manner, and thus the computer usable or computer readable memory. It is also possible for the instructions stored therein to produce an article of manufacture containing instruction means for performing the functions described in each block or flow chart step of the block diagram. Computer program instructions may also be mounted on a computer or other programmable data processing equipment, such that a series of operating steps may be performed on the computer or other programmable data processing equipment to create a computer-implemented process to create a computer or other programmable data. Instructions that perform processing equipment may also provide steps for performing the functions described in each block of the block diagram and in each step of the flowchart.

In addition, each block or step may represent a portion of a module, segment, or code that includes one or more executable instructions for executing a specified logical function. It should also be noted that in some alternative embodiments the functions noted in the blocks or steps may occur out of order. For example, the two blocks or steps shown in succession may in fact be executed substantially concurrently, or the blocks or steps may sometimes be performed in the reverse order, depending on the functionality involved.

As such, those skilled in the art will appreciate that the present invention can be implemented in other specific forms without changing the technical spirit or essential features thereof. Therefore, the above-described embodiments are to be understood as illustrative in all respects and not as restrictive. The scope of the present invention is shown by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. .

Claims (13)

In the method for generating a steganography analysis model performed by one or more processors,
Generating a second 3D mesh model that smoothes the first 3D mesh model;
A first feature vector extracted based on a difference between an edge normal of the first 3D mesh model and an edge normal of the second 3D mesh model, a mean curvature of the first 3D mesh model, and a mean curvature of the second 3D mesh model Extracting a third feature vector extracted based on the difference between the second feature vector extracted based on the difference and the total curvature of the first 3D mesh model and the total curvature of the second 3D mesh model; And
And performing machine learning based on the information about the first, second and third feature vectors to generate a model for determining whether there is hidden information in a target 3D mesh model.
The edge normal is,
For each mesh piece that shares a common edge in a 3D mesh model, the sum of the vectors derived by multiplying the area of each mesh piece and the normal vector at the center of gravity of each mesh piece
How to create a steganographic analysis model.
The method of claim 1,
The first 3D mesh model,
A cover mesh corresponding to the original and a stego mesh into which the information hidden in the original is inserted;
How to create a steganographic analysis model.
The method of claim 1,
The extracting step,
Generating a mean vector, a variance vector, a skewness vector, and a kurtosis vector for each feature vector.
How to create a steganographic analysis model.
The method of claim 3,
The average vector, the variance vector, the skewness vector, and the kurtosis vector generated from the respective feature vectors are inputted into kernel space to modify the number of dimensions of the input vector so as to have a linearly independent relationship with each other. Generating a linear independent vector, each transformed,
The information about the first, second and third feature vectors includes the linear independent vector.
How to create a steganographic analysis model.
The method of claim 4, wherein
The kernel space is
generated based on the homogeneous kernel map algorithm
How to create a steganographic analysis model.
delete The method of claim 1,
The first feature vector is
L2-norm of the edge normal of the first 3D mesh model and l2- of the edge normal of the second 3D mesh model with respect to the inner product of the edge normal of the first 3D mesh model and the edge normal of the second 3D mesh model the ratio of the product of norm to the ratio of inverse cosine
How to create a steganographic analysis model.
The method of claim 1,
The mean curvature is,
The average of the values derived by applying principal component analysis (PCA) to the vertices of the 3D mesh model
How to create a steganographic analysis model.
The method of claim 1,
The total curvature is,
The sum of the values derived by applying principal component analysis (PCA) to the vertices of a 3D mesh model.
How to create a steganographic analysis model.
The method of claim 1,
The extracting step,
A fourth feature vector extracted based on a difference between an x-axis value of the first 3D mesh model and an x-axis value of the second 3D mesh model in a rectangular coordinate system,
A fifth feature vector extracted based on a difference between the y-axis value of the first 3D mesh model and the y-axis value of the second 3D mesh model in the rectangular coordinate system,
A sixth feature vector extracted based on a difference between a z-axis value of the first 3D mesh model and a z-axis value of the second 3D mesh model in the rectangular coordinate system;
A seventh feature vector extracted based on a difference between an x-axis value of the first 3D mesh model and an x-axis value of the second 3D mesh model in a laplacian coordinate system,
An eighth feature vector extracted based on a difference between a y-axis value of the first 3D mesh model and a y-axis value of the second 3D mesh model in the Laplacian coordinate system,
A ninth feature vector extracted based on a difference between a y-axis value of the first 3D mesh model and a y-axis value of the second 3D mesh model in the Laplacian coordinate system,
A tenth feature vector extracted based on a difference between coordinates of a vertex of the first 3D mesh model and coordinates of a vertex of the second 3D mesh model in the rectangular coordinate system;
An eleventh feature vector extracted based on a difference between coordinates of a vertex of the first 3D mesh model and coordinates of a vertex of the second 3D mesh model in the Laplacian coordinate system,
A twelfth feature vector extracted based on a difference between a dihedral angle of two faces sharing edges in the first 3D mesh model and a back angle of two faces sharing edges in the second 3D mesh model;
A thirteenth feature vector extracted based on an angle difference between a face of the first 3D mesh model and a face of the second 3D mesh model;
A fourteenth feature vector extracted based on a difference between a vertex normal of the first 3D mesh model and a vertex normal of the second 3D mesh model,
A fifteenth feature vector extracted based on a difference between a gaussian curvature of the first 3D mesh model and a gaussian curvature of the second 3D mesh model, and
Extracting a sixteenth feature vector extracted based on a difference between a curvature ratio of the first 3D mesh model and a curvature ratio of the second 3D mesh model,
Generating the model,
Machine learning is performed by adding information about the fourth to sixteenth feature vectors.
How to create a steganographic analysis model.
An apparatus comprising a processor for performing the method of any one of claims 1 to 5 and 7 to 10.
A computer readable recording medium having recorded thereon a computer program comprising instructions for causing a processor to perform the method of any one of claims 1 to 5 and 7 to 10.
A computer program stored on a computer readable recording medium for causing a processor to perform the method of any one of claims 1 to 5 and 7 to 10.

Images