KR100938499B1 - The watermark embedding and detecting method using b-spline curve modeling and mesh-spectral transform - Google Patents

Abstract

The present invention relates to a method for embedding and extracting watermarks using B-spline curve modeling and mesh-spectrum transformation, comprising: (a) generating a curve by B-spline modeling and extracting control points of the curve; (b) calculating n mesh-spectrum coefficients based on the control points of the curve extracted through step (a); (c) inserting a watermark by changing coefficient values with respect to the mesh-spectrum coefficients calculated in step (b); And (d) calculating a control point in which a watermark is inserted by inversely transforming the mesh-spectrum coefficients inserted through the process (c), and reconstructing a B-spline curve based on the control point. It includes.

B-spline curves, watermarks, and mesh-spectrum transformations

Description

WATERMARK EMBEDDING AND DETECTING METHOD USING B-SPLINE CURVE MODELING AND MESH-SPECTRAL TRANSFORM}

The present invention relates to a method of inserting and extracting watermark information on a line of an image. More specifically, the conventional watermarking technique uses a B-spline curve to convert the curves of the image, unlike the conventional watermarking technique. The digital watermarking technology for copyright protection of digital content is extracted by extracting the B-spline control point, converting it into a frequency domain, inserting a watermark, and extracting the inserted watermark using an original line image. will be.

As is well known, as industrial development is accelerated, specialized devices such as computers, high performance scanners and printers are being distributed at low prices, and various contents are easily illegally copied using such digital devices and software.

At present, digital watermarking technology has been researched and put into practice as one of contents security technologies for illegal copying of video contents. Digital watermarking technology is a technology that can prevent copyright infringement by inserting copyright information into the original content unknowingly. The image watermarking method studied so far has inserted a watermark by finely adjusting the pixel brightness value of the image.

However, navigation systems, GIS digital maps, and cartoons represent images with straight lines and curves, which is difficult to apply to existing watermarking techniques in a way that is different from conventional image representation. Therefore, there is a need for a suitable watermarking technology.

As a related technique, Solachidis and Pitas proposed an algorithm that embeds a watermark into the size component of a Fourier descriptor for a line image, which is robust to geometric attacks. In addition, Gou, Wu and Ohbuchi et al. Proposed watermarking technique using B-spline curve model and watermarking technique using mesh-spectrum coefficients, respectively. Hereinafter, the methods and problems used in these techniques will be described.

1. Watermarking Using Control Points of B-Spline Curves

Gou and Wu studied the extraction of control points for line images on a map using a B-spline curve model and the insertion of watermarks. Control points are points outside the curve that control the shape of the curve. In order to find the control point, the set P of curve points can be expressed as a B-spline model as follows.

Where c is the control point, n is the number of control points, and t = 0,1,2 ..., n. B is a weight function and k is the order of the weight function.

의 차이를 최소화하는 n개의 제어점을 구할 수 있다. 마지막으로 노이즈 형태의 워터마크를 생성하여 제어점 좌표에 스프레드 스펙트럼 형태로 삽입하고 상관 계수 기반 추출 방법을 통하여 삽입된 워터마크를 확인한다. Modeled with original curves P and B-splines using the above equation

We can get n control points to minimize the difference between. Finally, a watermark in the form of noise is generated and inserted into the spread spectrum in the control point coordinates, and the inserted watermark is confirmed through a correlation coefficient based extraction method.

1 is an exemplary view illustrating a watermarking technique using a control point of a conventional B-spline curve. As shown in (A), an original curve is shown and a control point extracted from the curve is indicated by a circle. It can be seen that the control point of the curve controls the shape of the curve.

If the coordinate value is changed by inserting a watermark at the control point, the shape of the curve changes as shown in (B). However, it is possible to create a smooth curve that is difficult to recognize the distortion of the curve.

That is, the present method has a problem in that a point similar to the original control point must be found during the extraction process because the control point representing one curve is not unique. Therefore, an excellent synchronization algorithm and an iterative calculation process are required. When the watermark is simply inserted into the coordinates of the control point, the watermark extraction rate is relatively low in the print-scan attack, and the degree of corruption of the reconstructed image is also large.

2. Watermarking with 2D Mesh-Spectrum Transformation

Vector maps are represented by vertices and lines. Ohbuchi et al. Proposed a method of embedding a watermark using a mesh generated as a vertex. First, the n × n Laplacian matrix is calculated from the mesh generated by Delaunay triangulation for the vertices as follows.

The eigen analysis of R generated in this way yields n eigen vectors and eigen values. Projecting n points to an eigen vector yields n mesh-spectrum coefficients. After the map is expressed in the frequency domain, a random watermark of {+1, -1} is inserted into the extracted coefficient in the form of spread spectrum, and the map is reconstructed through inverse transformation. In extracting the watermark, the correlation coefficient between the coefficient in the frequency domain and the watermark is calculated to confirm the inserted watermark.

Experimental analysis of the technique using this mesh-spectrum showed the robust performance against geometric attack. However, this technique is applied to polygonal shapes such as buildings, houses, and roads of a map, and is not a model applied to curves. It has a limitation that it is complicated to calculate a mesh with all vertices.

The present invention has been made in view of the above-described problems, and after extracting the B-spline control point after modeling the image line of the B-spline curve, and inserting the watermark to the intermediate frequency coefficient through the mesh-spectrum transformation Watermark insertion and extraction method using B-spline curve modeling and mesh-spectrum transformation, which can protect intellectual property by extracting the watermark inserted for content that is illegally copied and distributed. Its purpose is to provide it.

The present invention relates to a watermark embedding method using B-spline curve modeling and mesh-spectrum transformation, comprising: (a) generating a curve by B-spline modeling and extracting control points of the curve; (b) calculating n mesh-spectrum coefficients based on the control points of the curve extracted through step (a); (c) inserting a watermark by changing coefficient values with respect to the mesh-spectrum coefficients calculated in step (b); And (d) calculating a control point in which a watermark is inserted by inversely transforming the mesh-spectrum coefficients inserted through the process (c), and reconstructing a B-spline curve based on the control point. Characterized in that it comprises a.

Preferably, the step (a) comprises the steps of: (a-1) generating a curve by assigning an index to each point of the curve and estimating a starting point and an ending point; (a-2) sequentially obtaining the coordinate values of the generated curve, sampling and extracting sample points to generate a B-spline curve; And (a-3) defining a sample point and a control point of the B-spline curve and calculating the control point using a least square method; Characterized in that it comprises a.

Also preferably, the step (b) includes (b-1) calculating an n × n Laplacian matrix (R) from a mesh generated through Delaunay triangulation to convert the control points of the B-spline curve into the frequency domain. ; (b-2) obtaining n eigen vectors and eigen values based on the matrix R calculated in the step (b-1); And (b-3) projecting the n control points to an eigen vector to obtain n mesh-spectrum coefficients r _i (1 ≦ _i ≦ n) = (r _{s, i} , r _{t, i} ); Characterized in that it comprises a.

Also preferably, the step (c) may include: (c-1) generating a watermark b for the mesh-spectrum coefficients calculated through the step (b); And (c-2) inserting the watermark b into a spread spectrum coefficient r to generate a mesh-spectrum coefficient r '; Characterized in that it comprises a.

Also preferably, in the step (c-2), the watermark b is inserted into the mesh-spectrum coefficient of the intermediate frequency domain.

Also preferably, the step (d) may include: (d-1) calculating a control point from the mesh-spectrum coefficient into which the watermark is inserted; And (d-2) calculating a sample point based on the control point calculated through the step (d-1), and extracting a watermark-inserted B-spline curve by interpolating the sample point. Characterized in that it comprises a.

And preferably, before the step (a), a preprocessing step of binarizing and thinning the original line image; It characterized in that it further comprises.

On the other hand, the present invention relates to a watermark extraction method using B-spline curve modeling and mesh-spectrum transformation, (g) curve synchronization of the original curve and the target curve; And (h) extracting a watermark by extracting a mesh-spectrum coefficient of the target curve and calculating a difference value from the original mesh-spectrum coefficient. Characterized in that it comprises a.

Preferably, the step (g) comprises: (g-1) extracting bending points of the original curve and the target curve to obtain a set of bending points; (g-2) calculating a distortion transform of the target curve; And (g-3) synchronizing all points on the object curve with the original curve position by taking an inverse transform of the distortion transform on the object curve; Characterized in that it comprises a.

Also preferably, the (h) process may include: (h-1) extracting a control point based on sample points of the original curve from the target curve; (h-2) mesh-spectrum analysis of the control points extracted through the step (h-1) to extract mesh-spectrum coefficients; And (h-3) extracting a watermark by calculating a difference value from a mesh-spectrum coefficient of the original curve; Characterized in that it comprises a.

Also preferably, after the step (h-3), (h-4) checking the inserted watermark by calculating a correlation coefficient between the watermark extracted through the step (h-3) and the inserted watermark; It characterized in that it further comprises.

And preferably, before the step (g), (e) pretreatment to binarize and thin the target curve; And (f) B-spline modeling the target curve; It characterized in that it further comprises.

According to the present invention as described above, by extracting the watermark inserted into the content that is distributed by unauthorized copying, there is an effect that can protect the intellectual property by authenticating the copyright owner.

In addition, according to the present invention, as a line-based watermarking technique that is robust against not only geometric attacks, but also print-scan attacks, there is an effect that can improve the robustness and image quality than conventional techniques.

Specific features and advantages of the present invention will become more apparent from the following detailed description based on the accompanying drawings. In the meantime, when it is determined that the detailed description of the known functions and configurations related to the present invention may unnecessarily obscure the subject matter of the present invention, it should be noted that the detailed description is omitted.

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

A method for embedding and extracting watermarks using B-spline curve modeling and mesh-spectrum transformation according to the present invention will be described with reference to FIGS. 2 to 9.

2 is a flowchart illustrating a method for embedding and extracting watermarks using B-spline curve modeling and mesh-spectrum transformation according to the present invention. The process takes place.

First, the process of inserting a watermark on an image line is as follows.

[Pretreatment (S110)]

In the preprocessing process, the original line image was converted into a 1-pixel thick line for the B-spline curve modeling. For this, the process of binarization and thinning was performed. In most cases, print-scanned images are not binary, so each pixel brightness value is distributed between 0 and 255. The curve is binarized to a threshold of 200 to extract the purified curve. In order to extract the ideal curve from the converted binary image, the thinning technique is applied to represent the lines of various pixel widths as 1-pixel thick lines.

B-Spline Modeling (S120)

FIG. 3A is a detailed flowchart of the B-spline modeling process S120 according to the present invention, and as shown, each point of the curve to estimate the shape of the 1-pixel curve in the preprocessed image through the step S110. An index is given to each other, the start point and the end point are estimated, and priority is set in eight directions of one point to generate a curve following the curve (S121).

In this case, a single curve is generated by correcting a case where a curve break occurs due to the distortion or when a closed curve is formed in the curve. [Table 1] below shows the priority of direction in curve estimation.

Previous direction Direction Estimation Priority One 2 3 4 5 6 7 8 9 10 E-> W - - W SW S SE E NE N NW NE-> SW SW W SW S SE E NE N NW W N-> S - - S SE E NE N NW W SW NW-> SE SE S SE E NE N NW W SW S W-> E - - E NE N NW W SW W SE SW-> NE NE E NE N NW W SW W SE E S-> N - - N NW W SW S SE E NE SE-> NW NW N NW W SW S SE E NE N

Table 1 Definition of Curve Estimation Priorities

After obtaining the coordinate values of the curve sequentially and sampled uniformly to extract the sample point (S122), it is used to model the B-spline curve. The sampling interval used in this embodiment is 6, and the sample point may be a B-spline curve as shown in Equation 1 below.

[Equation 1]

In this case, c is a control point, n is the number of control points, t = 0, 1, 2 ..., n, the weight function B is defined as shown in the following [Equation 2].

[Formula 2]

Here, t _i is called knots and means a position where the B-spline functions are connected together, thereby controlling the B-spline function and the control point. k denotes the order of the weight function, and in the present invention, cubic B-spline function is used by applying the fourth order.

Using the above two equations, a B-spline curve most similar to the original curve is generated (S123). In order to extract the control point from the above equation, a sample point and a control point are defined as shown in Equation 3 below (S124).

[Equation 3]

Then, using the least-squares method, the exact control point C is obtained through Equation 4 below (S125).

[Equation 4]

[Mesh-Spectrum Transformation (S130)]

3B is a detailed flowchart of a mesh-spectrum transformation process S130 according to the present invention, and as shown, n is obtained from a mesh generated through Delaunay triangulation to convert control points of a B-spline curve into a frequency domain. The xn Laplacian matrix is calculated as shown in Equation 5 below (S131).

[Equation 5]

In this case, I is a unit matrix, H is a diagonal matrix whose diagonal component is 1 / d _i , d _i is the number of intersections at the i-th point, and A is an adjacent matrix.

If the i and j points are adjacent, the element corresponding to row i and column j of A has a value of 1. Otherwise, assign a value of zero. Based on the generated R, eigen analysis can obtain n eigen vectors and eigen values (S132).

Subsequently, n control points are projected on the eigen vector to obtain n mesh-spectrum coefficient vectors r _i (1 ≦ _i ≦ n) = (r _{s, i} , r _{t, i} ) (S133). In this case, the subscripts s and t of the coefficient refer to a Cartesian coordinate system symmetrical with the x and y axes. The mesh thus created is basically robust to geometric attacks.

[Insert Watermark (S140)]

FIG. 3C is a detailed flowchart of the watermark embedding process S140 according to the present invention. As shown in FIG. 3C, by changing the coefficient value in a spread spectral manner with respect to the mesh-spectrum coefficient calculated through the process of S130, FIG. Insert a watermark.

At this time, the watermark for insertion follows a normal distribution having an average of 0 and is generated as a random sequence of length L composed of 1 and -1. The generated watermark w is spread by chip rate cr as shown in Equation 6 below to generate a watermark b (S141).

[Equation 6]

At this time, the length of the generated watermark b is cr · L. The watermark b is inserted into the spread spectrum coefficient r with the intensity α as shown in Equation 7 below to generate a mesh-spectrum coefficient r 'into which the watermark is inserted (S142).

[Formula 7]

Sonlanki et al. According to the research, color noise and nonlinear effects caused by the print-scan process change the high-frequency coefficients of the image. Thus, the watermark embedding method that survives the print-scan process uses a watermark in the coefficients of the low and medium frequencies. To insert it. Since the change of the coefficient in the low frequency region causes more distortion in the original line, in the present invention, the watermark is inserted into the mesh-spectrum coefficient in the intermediate frequency region.

[Image Reconstruction (S150)]

FIG. 3D is a detailed flowchart of an image reconstruction process S150 according to the present invention. As shown in FIG. 3D, a watermark-embedded control point is extracted by inversely converting a mesh-spectrum coefficient having an embedded watermark. To this end, the control point C 'at the mesh-spectrum coefficient r' _i = (r ' _{s, i} , r' _{t, i} ) to which the watermark is inserted is calculated through Equation 8 below (S151).

Equation 8

는 계수

에 해당되는 정규 eigen 벡터이다.here,

Is a coefficient

The regular eigen vector for.

The control point obtained through Equation 8 is a control point of the B-spline model in which the watermark is inserted, and by using Equation 9 below, a sample point having the watermark inserted can be obtained and the sample point is interpolated. The curve in which the watermark is inserted can be extracted (S152).

[Equation 9]

Next, the process of extracting the watermark is as follows.

The process of extracting the inserted watermark is similar to the process of embedding the watermark. As described above, the curve is curved after performing the preprocessing and the B-spring modeling process based on the curve into which the watermark is inserted. Use the dot to synchronize with the original line image.

[Line Sync (S160)]

4A is a detailed flowchart of a line synchronization process S160 according to the present invention.

For the extraction of watermarks, the original line image and the target line image for extraction must be synchronized to correct not only the geometric attack but also the minute curve position change caused by the print-scan process. The synchronization point of the two curves was conceived in the research of Mokhtarian and Mackworth and utilized the curve's bending point. The curve's bending point is robust to the conversion process, which does not change the shape of the curve, so curve synchronization can be performed. The synchronization process is as follows.

라고 하면 다음의 [수식 10] 을 통하여 대상 곡선에 가해진 왜곡 변환을 계산할 수 있다(S162).First, the bending points of two curves are extracted to obtain a set of the most similar bending points (S161). The curve point of the original curve is called (x, y) and the curve point of the target curve

In this case, the distortion transformation applied to the target curve may be calculated through Equation 10 (S162).

Equation 10

More easily expressing Equation 10 above, we can obtain a single matrix A that represents distortion transformation as shown in Equation 11 below.

[Equation 11]

라고 하고, 원본 곡선의 모든 점 집합을 (X, Y) 라고 할 때, 아래의 [수식 12] 와 같이 왜곡변환의 역변환을 취하면 대상 곡선 상의 모든 점을 원본 곡선 위치와 동기화할 수 있다(S163). Therefore, we need to

When the set of all points of the original curve is (X, Y), if the inverse transformation of the distortion transformation is performed as shown in [Equation 12] below, all points on the target curve can be synchronized with the original curve position (S163). ).

Equation 12

FIG. 4B shows the synchronization result of the original curve A and the target curve B subjected to the print-scan process, that is, the extraction of the mutually symmetrical bending points of the original curve and the target curve. The distortion transformation was calculated by selecting the most suitable set of bending points, and the inverse transformation of the distortion was applied to the target curve to synchronize with the original. This process is repeatedly performed until the difference value is the smallest from the original image, and FIG. 4B is repeated three times for the curve.

Figure 4c shows the difference between the original curve and the curve after the print-scan process (C) and the difference between the curve synchronized with the original curve (D), (C) overlaps the original curve and the target curve, ) Overlaps the target curve synchronized with the original curve. This confirms that the synchronized curve is more similar to the original curve.

[Watermark Extraction (S170)]

4D is a detailed flowchart of a watermark extraction process S170 according to the present invention.

The algorithm proposed in the present invention is a non-blind watermarking method using an original curve to extract a watermark. That is, the process of extracting the inserted watermark is processed similarly to the watermark embedding.

를 추출한다(S171).First, as described above, after performing curve synchronization, the control point is selected by selecting the sample point most similar to the sample point of the original curve in the target curve.

It is extracted (S171).

를 추출하고(S172), 원본 곡선의 메시-스펙트럼 계수와의 차이를 다음의 [수식 13] 을 통해 계산함으로써, 워터마크를 추출한다(S173).Then, mesh-spectrum analysis is performed on the extracted control points.

Is extracted (S172), and the watermark is extracted by calculating the difference with the mesh-spectrum coefficient of the original curve through the following [Equation 13] (S173).

Equation 13

As described above, the correlation coefficient between the extracted watermark and the inserted watermark is calculated to confirm the inserted watermark (S174).

는 추출한 워터마크를 의미한다.

는 삽입된 워터마크의 평균이고,

는 추출한 워터마크의 평균이다.In the present invention, a Z-statistic suitable for the verification of a robust watermark is used as the watermark verification method. A correlation coefficient between the inserted watermark and the extracted watermark is obtained as shown in Equation 14 below. Where L is the length of the watermark, w is the inserted watermark,

Denotes the extracted watermark.

Is the average of the inserted watermark,

Is the average of the extracted watermarks.

[Equation 14]

From the above [Equation 14], the Z statistical value can be obtained as shown in the following [Equation 15].

Equation 15

Hereinafter, experiments and results using watermark insertion and extraction methods using B-spline curve modeling and mesh-spectrum transformation according to the present invention will be analyzed.

The experiment was conducted on a total of nine different line images as shown in [Table 2]. The watermark embedding intensity is 0.5 and a 1000-bit watermark is inserted into all curved images except Curve.tif. For the Curve.tif curve, a 560-bit watermark is inserted.

Line image Size in pixels resolution Sample points Control points Apple.tif 756 × 756 96 675 686 Mountain.tif 591 × 591 96 1118 1230 Car.tif 629 × 245 96 578 599 Map.tif 770 × 708 96 535 543 Leaf.tif 756 × 756 96 919 976 Curve.tif 634 × 471 96 273 280 Topo1.tif 398 × 251 96 1188 1190 Topo2.tif 459 × 527 96 2411 2418 Duck.tif 634 × 447 96 1429 1432

Table 2 Sample line images used in experiment

To analyze the performance of the proposed algorithm, we compare it with a line watermarking algorithm that inserts a random noise-type watermark into the coordinate values of the control points presented by Gou and Wu. For comparison, the distance difference between the two curves was calculated and the Hausdorff Distance (HD) representing the distortion of the curve and the watermark extraction rate were used as measures of performance.

와

가 각각

,

로 정의될 때, HD는 다음의 [수식 16] 과 같이 계산된다. 따라서, HD는 서로 다른 두 곡선 간의 차이를 min-max 메트릭을 이용하여 정량화하는 방법이다.

Wow

Each

,

When defined as, HD is calculated as in Equation 16 below. Therefore, HD is a method of quantifying the difference between two different curves using the min-max metric.

[Equation 16]

FIG. 5 illustrates an example of reconstructing a watermarked image by extracting a control point from a Car.tif line image, converting it into a frequency domain, and inserting a watermark to identify an original image and a watermarked image with a human eye. It's hard to do. However, you can find subtle differences between the original mesh and the watermarking mesh.

1) Experiment on geometric attack

The robustness against the geometric attack of the mesh-spectrum coefficients was analyzed. For the accuracy of the extraction results for geometric attacks, the lines are not stored as images but use digital data and no line synchronization algorithm is applied. When the line is saved as an image, the cutting process occurs and data loss occurs, so the influence of the geometric attack on itself cannot be confirmed.

6 is a watermark extraction result for the geometric attack when the watermark is inserted in the existing control point coordinates. Of the 1000 random watermarks, the 500th watermark was inserted. In the Gou and Wu algorithms, the watermarks were not extracted from both the rotational attack, the enlarged attack, and the mobile attack.

The results for the proposed algorithm are shown in FIG. In case of converting to the frequency domain, it is shown that it is robust against geometric attack without applying the line synchronization algorithm.

2) Experiment on Print-Scan

Robustness to print-scans is essential for watermarking offline content. In this experiment, we compared the robustness of Gou and Wu's watermarking algorithm with the proposed watermarking algorithm for print-scanning on nine different line images. The printer used is the HP Color Laserjet 4650 and the scanner is EPSON PERFECTION 2400 PHOTO. In the process of saving the watermarked line image as a file before performing the printer-scanning, line information corruption occurs due to cutting of coordinate values.

8 and 9 show rough results of the Car.tif file. The Z-statistics were compared after the watermarked line image was saved as a file and additionally after the print-scan process. As can be seen from the results, the algorithms of Gou and Wu are lower than those of the proposed algorithm. Moreover, after the print-scan process, statistics are drastically lowered, making watermark extraction difficult. However, the proposed algorithm shows good extraction even after print-scan.

The same experiments were performed identically for the other images, and the results are summarized in Table 3. Here, digitization refers to a process of converting a watermarked line image into a file, and summarizes the maximum Z extraction statistics and HD analysis results for the images after the digitization and print-scan process.

Line image Gou and Wu's algorithm Proposed algorithm Digitization Print-Scan HD Digitization Print-Scan HD Apple.tif 8.51 6.06 4.0 9.02 8.74 4.0 Mountain.tif 11.24 4.70 4.0 11.32 7.46 4.0 Car.tif 9.41 3.85 5.2 9.98 6.60 4.0 Map.tif 9.26 3.05 4.0 11.18 6.14 4.0 Leaf.tif 8.70 4.08 4.5 11.40 5.42 4.0 Curve.tif 3.32 3.25 4.0 6.48 3.39 4.0 Topo1.tif 9.94 5.94 4.5 13.45 8.81 6.0 Topo2.tif 12.21 3.09 6.0 18.99 3.17 6.0 Duck.tif 9.39 6.03 6.0 13.45 12.26 4.5

[Table 3] Watermark Z extraction statistics and HD comparison results

When saving the watermarked line image as a file, the proposed watermarking algorithm showed 28.4% better extraction rate than the existing Gou and Wu algorithms. Seems. HD, which exhibits distortion of curves, also shows a 4.0% improvement over the existing watermarking algorithm.

As described above and described with reference to a preferred embodiment for illustrating the technical idea of the present invention, the present invention is not limited to the configuration and operation as shown and described as described above, it is a deviation from the scope of the technical idea It will be understood by those skilled in the art that many modifications and variations can be made to the invention without departing from the scope of the invention. Accordingly, all such suitable changes and modifications and equivalents should be considered to be within the scope of the present invention.

1 is an exemplary view of a watermarking method using a control point of a conventional B-spline curve.

2 is an overall flowchart of a watermark embedding and extraction method using B-spline curve modeling and mesh-spectrum transformation according to the present invention;

3A is a detailed flowchart of a B-spline modeling process S120 according to the present invention.

3B is a detailed flowchart of a mesh-spectrum transformation process S130 according to the present invention.

3c is a detailed flowchart of a watermark embedding process S140 according to the present invention;

3D is a detailed flowchart of an image reconstruction process S150 according to the present invention.

4A is a detailed flowchart of a line synchronization process S160 according to the present invention.

Figure 4b is an exemplary view showing a synchronization result of the original curve (A) and the target curve (B) after the print-scan process in accordance with the present invention.

Figure 4c is an exemplary view showing the difference between the original curve and the curve (C) after the print-scan process and the difference (D) of the curve synchronized with the original curve in accordance with the present invention.

4d is a detailed flowchart of a watermark extraction process S170 according to the present invention;

5 is an exemplary view illustrating a process of reconstructing a watermarking image using mesh-spectrum coefficients of a control point according to the present invention.

6 is a graph showing the result of a geometric attack when a watermark is inserted into the control point coordinates according to the conventional watermark method of Gou and Wu.

7 is a graph showing the results for a geometric attack in accordance with the present invention.

Figure 8 is a graph showing the results for the print-scan according to the watermark method of the conventional Gou and Wu.

9 is a graph showing the results for a print-scan attack in accordance with the present invention.

Claims (12)

In the watermark embedding method using B-spline curve modeling and mesh-spectrum transformation, (a) generating a curve by B-spline modeling and extracting control points of the curve; (b) calculating n mesh-spectrum coefficients based on the control points of the curve extracted through step (a); (c) inserting a watermark by changing coefficient values with respect to the mesh-spectrum coefficients calculated in step (b); And (d) calculating a control point in which a watermark is inserted by inversely transforming the mesh-spectrum coefficients inserted through the process (c) and reconstructing a B-spline curve based on the control point; Watermark embedding method using B-spline curve modeling and mesh-spectrum transformation comprising a. The method of claim 1, The (a) process, (a-1) generating a curve by assigning an index to each point of the curve, estimating a start point and an end point, and prioritizing the points; (a-2) generating a B-spline curve by sequentially obtaining coordinate values of the generated curve, sampling and extracting sample points; And (a-3) defining a sample point and a control point of the B-spline curve and calculating a control point using a least square method; Watermark embedding method using B-spline curve modeling and mesh-spectrum transformation comprising a. The method of claim 1, Step (b) is, (b-1) calculating an n × n Laplacian matrix R from a mesh generated through Delaunay triangulation to convert the control points of the B-spline curve into the frequency domain; (b-2) obtaining n eigen vectors and eigen values based on the matrix R calculated in the step (b-1); And (b-3) projecting the n control points to an eigen vector to obtain n mesh-spectrum coefficients r _i (1 ≦ _i ≦ n) = (r _{s, i} , r _{t, i} ); Watermark embedding method using B-spline curve modeling and mesh-spectrum transformation comprising a. The method of claim 1, Step (c) is, (c-1) generating a watermark b for the mesh-spectrum coefficients calculated in step (b); And (c-2) inserting the watermark b into a spread spectrum coefficient r to generate a mesh-spectrum coefficient r '; Watermark embedding method using B-spline curve modeling and mesh-spectrum transformation comprising a. The method of claim 4, wherein In the step (c-2), the watermark b is inserted into the mesh-spectrum coefficients of the intermediate frequency domain, wherein the watermark embedding method using B-spline curve modeling and mesh-spectrum transformation. The method of claim 1, (D) process, (d-1) calculating a control point from the mesh-spectrum coefficient having the watermark embedded therein; And (d-2) calculating a sample point based on the control point calculated through step (d-1), and extracting a watermark-inserted B-spline curve by interpolating the sample point; Watermark embedding method using B-spline curve modeling and mesh-spectrum transformation comprising a. delete In the watermark extraction method using B-spline curve modeling and mesh-spectrum transformation, (g) curve synchronization of the source and target curves; And (h) extracting a watermark by extracting a mesh-spectrum coefficient of the target curve and calculating a difference value from the original mesh-spectrum coefficient; Watermark extraction method using the B-spline curve modeling and mesh-spectrum transformation comprising a. The method of claim 8, (G) process, (g-1) extracting bending points of the original curve and the target curve to obtain a set of bending points; (g-2) calculating a distortion transform of the target curve; And (g-3) synchronizing all points on the object curve with the original curve position by taking an inverse transform of the distortion transform on the object curve; Watermark extraction method using the B-spline curve modeling and mesh-spectrum transformation comprising a. The method of claim 8, (H) process, (h-1) extracting control points based on sample points of the original curve from the target curve; (h-2) mesh-spectrum analysis of the control points extracted through the step (h-1) to extract mesh-spectrum coefficients; And (h-3) extracting a watermark by calculating a difference value from a mesh-spectrum coefficient of the original curve; Watermark extraction method using the B-spline curve modeling and mesh-spectrum transformation comprising a. The method of claim 10, After the step (h-3), (h-4) checking the inserted watermark by calculating a correlation coefficient between the watermark extracted through the step (h-3) and the inserted watermark; Watermark extraction method using the B-spline curve modeling and mesh-spectrum transformation further comprises. delete

Images