KR101725095B1 - Digital Watermark Detecting Method for Image file With Unknown Rotation Angle - Google Patents

Abstract

The present invention relates to a method for inserting a watermark in the frequency domain or for normalizing an image file to detect an already inserted watermark, and a method for efficiently detecting a watermark in an image file subjected to a rotational attack in a state where a watermark is inserted According to the watermark detection method of the present invention, it is possible to insert a watermark which is robust against a rotation attack according to the image normalization method of the present invention. According to the watermark detection method of the present invention, The watermark can be efficiently detected even if the rotation angle is unknown.

Description

[0001] The present invention relates to a watermark detection method for an image file,

The present invention relates to a method for inserting a watermark in the frequency domain or for normalizing an image file to detect an already inserted watermark, and a method for effectively detecting a watermark from an image file subjected to a rotational attack in a state where the watermark is inserted And a watermark detection method.

Recently, as information and communication technologies are developed, digital images on the Internet are copied and transmitted unintentionally, thereby infringing copyrights frequently. In order to track or prevent such copyright infringement, a method of inserting a watermark into a digital image is widely used. However, even if a digital image is embedded with a watermark, it is difficult to extract a watermark when subjected to a geometric attack, and a watermark to be extracted includes many bit errors. Especially, when the digital image is rotated, it is difficult to extract the watermark because the angle of rotation can not be accurately known.

Although a technique for extracting a watermark has been introduced in the past when a digital image has been rotated, a simple level of extracting a watermark after rotating the digital image back in the reverse direction assuming that the angle of rotation of the digital image is known It was not at all suitable if the copyright holder did not know the angle of rotation applied to his digital image.

Registered Patent No. 0644276 entitled " Method of Extracting Watermark from Geometric Deformation Image ", 2006. 11. 10. Registered Patent No. 1363577 "Apparatus and method for detecting digital watermark detection against re-sampling attack ", 2014. 02. 17.

The present invention relates to a method for inserting a watermark in the frequency domain or for normalizing an image file to detect an already inserted watermark, and a method for effectively detecting a watermark from an image file subjected to a rotational attack in a state where the watermark is inserted And a method for detecting a watermark.

In order to solve the above-described problems, the present invention provides a method of extracting an edge from an image file, (b) extending the edge; (c) calculating an angle by normalizing the image file by radon conversion; And (d) rotating the image file by the calculated angle.

(S1) calculating a normalization angle? Of the image file according to the normalization method of the image file, and normalizing the image file while reducing the image file from? +? To? -Ta by an angle? Extracting a watermark from the file to generate an ensemble matrix (1 & tilde &0); (S2) acquiring a position of a column of 1 and a position of a column of 0 in an arbitrary row of the ensemble matrix, extracting values of the column in another row to generate a new matrix, and calculating the number of 1's and 0's in a newly generated matrix Obtaining new labels for corresponding values after rearranging in descending order; (S3) obtaining the number of each new label in each column of the new labeled matrix and extracting values from the column of the ensemble matrix corresponding to the corresponding column if the number of the first labels is not the largest; And (S4) generating a three-dimensional binary positional matrix corresponding to each value in the ensemble matrix, calculating a sum of values at a position corresponding to the value extracted from the ensemble matrix column of the three-dimensional binary number positional matrix, Extracting only values of a predetermined value or more from the calculated values and deriving a sum of binary position values corresponding thereto and correcting the value of the position of the error in the ensemble matrix based on the index of the largest value; And a watermark detection method including the watermark detection method.

According to the method of normalizing an image file according to the present invention, it is possible to insert a watermark which is robust against a rotation attack. According to the watermark detection method of the present invention, when an image file is rotated, it is possible to efficiently detect a watermark even if the rotation angle is unknown.

1 is a flowchart of a method of normalizing an image file according to the present invention.
Figure 2 shows the ramp edges, peaks and zero crossings of the image file.
3 is an embodiment of a method of inserting a watermark into an image file.
4 is a flowchart of a watermark extraction method according to the present invention.
5 is an embodiment of a method of extracting a watermark from an image file.

Hereinafter, a method of normalizing an image file and a method of extracting a watermark according to the present invention will be described in detail.

1. How to normalize an image file

The method of normalizing an image file according to the present invention includes: (a) extracting an edge from an image file; (b) extending the edge; (c) calculating an angle by normalizing the image file by radon conversion; And (d) rotating the image file by the calculated angle. Normalization of these image files is a prerequisite for inserting watermarks that are robust to rotation attacks into image files or extracting watermarks from rotated image files.

(1) Step (a)

This step extracts an edge from the image file. Specifically, the step (a-1) may include applying Gaussian smoothing filtering to the image file; And a step (a-2) of converting an image file to which Gaussian smoothing filtering is applied from a lamp edge to a peak of an image through a first differentiation and detecting a part where a brilliance difference occurs through a second differentiation. The concept of lamp edge, peaks, and zero crossing of an image file can be seen in [Figure 2].

(2) Step (b)

This step is a step of extending the edge. Specifically, in this step, among the pixels in the range of the gradient vertical vertical bidirectional direction of the edge, pixels not belonging to the edge are converted into pixels belonging to the edge.

(3) Step (c)

This step is a step of calculating an angle by normalizing the image file by radon conversion. The progress of this phase will be described in detail as follows.

First, the step (c-1) of accumulating the two-dimensional coordinates of the respective pixels of the image file represented by x and y in a radon space represented by (?,?) By line segmentation according to the following equation It proceeds.

(Equation 1)

(-∞ <x, y <∞, -∞ <ρ <∞, 0 ≤ θ <π)

Next, a straight line component of an edge of the image file is detected using a local maximum value of a portion exceeding a threshold value in the radon space, and a slope value in a direction perpendicular to the slope of the straight line component is detected The step (c-2) for accumulating the normalized angle? In accordance with the equation (2) proceeds.

(Equation 2)

(X, y) is a function based on the direction and size of the slope of the image file, isedge is a function for determining whether (x, y) is an edge, and mod Calculated from 0 to 180 °)

(4) Step (d)

In this step, the image file is rotated by the angle calculated in the step (c). By rotating the image file in this way, the normalization of the image file is completed.

The method of normalizing an image file according to the present invention is also a premise for inserting a watermark which is robust to a rotation attack. An embodiment of a method of inserting a watermark after normalization of an image file is shown in FIG.

In watermark embedding, when the original image is a color image, it switches to the NTSC color space to acquire luminance components and convert them to the frequency domain. The technique of converting an image into the frequency domain when inserting a watermark is 2-Dimensional Fast Fourier Transformation (2D FFT). In the embodiment of FIG. 3, an image is transformed from a spatial domain to a frequency domain using a 2D FFT technique to insert a watermark. We use a secret key to generate random number macrics of size n / 2 x n / 2 and extend this matrix to size n x n. Apply the inverse Steerable Pyramid Transform (SPT) transformation to the matrix to generate a transformed matrix (TM). If the watermark length is N, select an embedding area whose size is n * N ^ 0.5 x n * N ^ 0.5 and divide the area into blocks of size n x n. TM is added to a block whose watermark bit corresponds to 1, and TM is subtracted from a block whose watermark bit corresponds to zero. The watermark information and TM are combined and inserted into the frequency domain converted image, and then the inverse 2D FFT transform is performed to obtain the marked image.

2. How to extract the watermark of an image file whose rotation angle is unknown

As shown in FIG. 4, the watermark extracting method according to the present invention includes the steps of (S1) calculating a normalization angle? In an image file whose rotation angle is unknown according to the image file normalization method, Generating an ensemble matrix consisting of 1s and 0s by extracting watermarks from each normalized image file while normalizing the files while reducing the angles? From? +? To? -Τ; (S2) acquiring a position of a column of 1 and a position of a column of 0 in an arbitrary row of the ensemble matrix, extracting values of the column in another row to generate a new matrix, and calculating the number of 1's and 0's in a newly generated matrix Obtaining new labels for corresponding values after rearranging in descending order; (S3) obtaining the number of each new label in each column of the new labeled matrix and extracting values from the column of the ensemble matrix corresponding to the corresponding column if the number of the first labels is not the largest; And (S4) generating a three-dimensional binary positional matrix corresponding to each value in the ensemble matrix, calculating a sum of values at a position corresponding to the value extracted from the ensemble matrix column of the three-dimensional binary number positional matrix, Extracting only values of a predetermined value or more from the calculated values and deriving a sum of binary position values corresponding thereto and correcting the value of the position of the error in the ensemble matrix based on the index of the largest value; .

(1) Step (S1)

In this step, a normalization angle [theta] of an image file whose rotation angle is unknown is calculated according to the image file normalization method according to the present invention, and an image file which does not know the rotation angle is normalized by decreasing angle [ And extracting a watermark from each normalized image file to generate an ensemble matrix consisting of 1 and 0 (Ensemble Matrix). The ensemble matrix has a size of 2τ / α × N when the length of the watermark is N.

There are various methods of extracting the watermark from each normalized image file, one of which is shown in FIG. If the watermark embedded image is a color image, the NTSC color space is converted to obtain a luminance component and converted to a frequency domain. If the watermark is a gray image, the image is converted to a direct frequency domain. Next, the 2D FFT transformation is performed, and then the insertion region is selected to proceed with the SPT transformation. Subsequently, a watermark is extracted from the direction sub-band output from the directional filter among the sub-bands output from the SPT. Finally, if the absolute value of the maximum positive value of the coefficient is larger than the absolute value of the minimum negative value, the watermark bit is set to 1, and if the absolute value of the maximum positive value of the coefficient is set to 0, the watermark bit is set to 1 .

(2) Step (S2)

In this step, a position of a column of 1 and a position of a column of 0 are obtained in an arbitrary row of the ensemble matrix, a new matrix is generated by extracting values of the column in another row, and the number of 1's and 0's in a newly generated matrix , Rearranging in descending order, and assigning new labels to the corresponding values.

This step can be proceeded as follows. First, all the elements of the ensemble matrix are firstly labeled with c _i ∈ {+ 1, -1} (i = 1, 2) by defining the positions of the elements labeled with LP _i ^{m, i} extracts the labels in all the columns LP ^{m, i} of the matrix ensemble and generates a subset of PL ^{m, i} of the label. Next, we define NL ^{m, i} as the number of elements labeled as c _i in the m-th row, define the sum as NS ^m , sort each NL contained in NS in descending order to obtain RN (Equation 3 ), The process is completed by temporarily assigning a new label, nc, to generate an NPL defined as follows.

(Equation 3)

(3) Step (S3)

In this step, the number of each new label is obtained in each column of the matrix to which the new label is assigned, and when the number of the first labels is not the largest, the values are extracted from the column of the ensemble matrix corresponding to the column.

Specifically, when the number of elements corresponding to nc existing in the position k of the NPL is defined as NP, the NPs of nc existing at the position k are SN ^{m, i, k} . When the maximum value is extracted from SN, Index values must be equal to 1 because they are sorted in descending order. If it is not equal to 1, k is defined as ke, and labels of the elements existing at the position where the error occurs are extracted as shown in (Equation 4) of the ensemble matrix. The values extracted from (Equation 4) are defined as OC.

(Equation 4)

(4) Step (S4)

In this step, a three-dimensional binary position matrix corresponding to each value is generated in the ensemble matrix, a sum of values at a position corresponding to a value extracted from the column of the ensemble matrix of the three-dimensional binary number position matrix is calculated, Extracts only values equal to or larger than a predetermined size among the values of the ensemble matrix and then derives the sum of the corresponding binary position values and corrects the sum to the value of the position at which the error occurred in the ensemble matrix based on the index of the largest value.

Specifically, the three-dimensional binary position matrix can be defined as Equation (5) below. The sum of the values at the positions of the BPs corresponding to the OCs is referred to as a degree of similarity between EM and OC, and a high value And the PC is defined as the sum of the binary position values corresponding to the candidates, the index of the largest value in the PC is defined as ic, and c _ic is corrected to the value of the position where the error occurs in the ensemble matrix do.

(Equation 5)

The present invention has been described in detail above with reference to specific examples. However, the present invention is not limited to the above embodiments, and can be modified or modified without departing from the gist of the present invention. It is therefore intended that the appended claims cover such modifications and variations.

none

Claims (5)

(A) extracting an edge from an image file, (b) expanding the edge, (c) calculating a normalization angle by radon-transforming the image file, and rotating the image file by the calculated angle calculating a normalization angle? of an image file whose rotation angle is unknown according to a normalization method of an image file including step d), and normalizing the image file while reducing the angle of the image file from? +? to? Extracting a watermark from each normalized image file to generate an ensemble matrix (S1) consisting of 1 and 0;
Acquiring a position of a column of 1 and a position of a column of 0 in an arbitrary row of the ensemble matrix and extracting the values of the column in another row to generate a new matrix and obtaining the number of 1's and 0's in a newly generated matrix, (S2) a new label is assigned to the corresponding values after reordering;
Calculating a number of each new label in each column of the matrix to which the new label is assigned and extracting values from the column of the ensemble matrix corresponding to the column if the number of the first labels is not the largest; And
Generating a three-dimensional binary position matrix corresponding to each value in the ensemble matrix, calculating a sum of values at a position corresponding to a value extracted from the column of the ensemble matrix of the three-dimensional binary number position matrix, Extracting only values equal to or greater than a preset size, deriving a sum of binary position values corresponding thereto, and modifying the sum to a value of a position at which an error occurs in the ensemble matrix based on an index of the largest value (S4); Wherein the rotation angle of the image file is unknown.
The method of claim 1,
The step (a)
(a-1) applying Gaussian smoothing filtering to the image file; And
(a-2) converting an image file to which Gaussian smoothing filtering is applied from a lamp edge to a peak of an image through a first-order differential, and detecting a part where an azimuth difference occurs through a second-order differential A watermark detection method of an image file whose rotation angle is unknown.
The method of claim 1,
Wherein the step (b) comprises converting the pixels belonging to the edge to pixels belonging to the edge among the pixels in the vertical bidirectional fixed range of the edge of the edge. Way.
The method of claim 1,
The step (c)
(c-1) accumulating the two-dimensional coordinates of each pixel of the image file represented by x and y in a radon space represented by (?,?) by line segmentation according to the following equation (1); And
(c-2) detecting a straight line component of an edge of the image file using a local maximum value of a portion exceeding a threshold value in the radon space, and calculating a slope value in a direction perpendicular to the slope of the straight line component Calculating a normalized angle? By accumulating in accordance with the following equation (2); And detecting a rotation angle of the image file.
(Equation 1)

(-∞ <x, y <∞, -∞ <ρ <∞, 0 ≤ θ <π)
(Equation 2)

(X, y) is a function based on the direction and size of the slope of the image file, isedge is a function for determining whether (x, y) is an edge, and mod Calculated from 0 to 180 °) delete

Images