KR20010016866A - Watermark inserting and dectacting method for protecting the copyright of digital image - Google Patents

Abstract

PURPOSE: A method for inserting and detecting a watermark for protecting CONSTITUTION: A watermark which is a pseudo-random number is generated. According to a watermarking region(space or frequency region), an original image or converted image is scanned to generate an one-dimensional signal. The signals are classified by magnitude to reduce distribution. In addition, in order to insert the watermarks into the classified signals, the watermarks are classified by order. Thereafter, direct current elements of the classified signals of each class are removed. The watermarked signals are obtained from the classified signals and watermarks. When the watermarking region is a frequency region, inverse scanning and transforming processes are performed. When the watermarking region is a space region, the inverse scanning process is performed. Thus, a watermarked image is formed. A test image is converted into the watermark-inserted region, and the converted signal is scanned to an one-dimensional signal. The test image is classified into classes by using the information from the original image. A detection response is obtained from the classified signals and watermarks. Then, existence or absence of the watermark is confirmed by comparing the detection response with a threshold value.

Description

Watermark insertion and detection method to protect copyright of digital image {WATERMARK INSERTING AND DECTACTING METHOD FOR PROTECTING THE COPYRIGHT OF DIGITAL IMAGE}

The present invention relates to a watermarking technique for confirming the copyright of a digital image.

In order to protect the digital video information, only the authorized user can use the video data by encrypting and transmitting the given video data using the existing public key encryption algorithm. However, in a network environment such as the Internet, providing information on the premise of a specific user is meaningless, and there is a problem in that the size of image data increases by encryption. Thus, a watermarking technique has been proposed as a method of efficiently protecting ownership and copyright of image data. Watermarking is a method that allows all users on the network to freely use the information, while allowing the information provider to claim the right to use the information later. Digital watermarking makes an arbitrary change directly on the image by using the visual characteristics of the image, and provides a basis for claiming the copyright or ownership of the image by detecting the watermark when the copyright is required.

Current research on watermarking is divided into two categories. There is a watermarking method in a spatial domain that directly changes pixel values of an image and a watermarking method in a frequency domain that changes frequency coefficient values of an image.

According to the conventional watermarking technology, the correlator illustrated in FIG. 1 obtains a correlation between a watermark and an image to be tested for the presence or absence of a watermark, and then detects the watermark by comparing it with a threshold value. . However, since the correlation between the watermark and the original image to be watermarked can be so large that it can not be ignored, there is a problem that a detection error occurs when the watermark is detected.

As described above, according to the conventional watermarking technology, a watermark detection error may occur because the correlation between the watermark and the original image to be watermarked can be a large value. Therefore, it is necessary to minimize the probability that the watermark exists but the detector does not detect the watermark (false negative probability) and the probability that the watermark does not exist but the detector exists (false positive probability). Accordingly, an object of the present invention is to minimize the watermark detection error by removing the average of the original image to be watermarked and minimizing the dispersion of the original image to be watermarked so as to minimize the above two probabilities.

1 is a block diagram of a watermark detection correlator to which the present invention is applied;

2 is a diagram of an embodiment of dividing a signal to be watermarked into classes in order to minimize dispersion of the signal to be watermarked according to the present invention;

3 is a watermark embedding diagram for improving watermark detection reliability according to the present invention.

4 is a watermark detection diagram for improving watermark detection reliability according to the present invention.

According to the watermarking technique of the present invention for achieving the above object, it removes the average of the original image to be watermarked, and inserts and detects the watermark to minimize the dispersion of the signal to be watermarked. According to stochastic process theory, when the threshold value is 0.5 from the analysis result of the correlation distribution between the watermark and the image to be tested for watermark and the watermark and the image to be watermarked, the watermark is A probability that exists but does not detect the watermark (false negative probability: P-) and the probability that there is no watermark but the detector is present (false positive probability: P +) is given as in Equation (1).

[Equation 1]

here And Are each of the signals to be watermarked. Expected value and variance, where E _w is the energy of the watermark w (i), i.e. Indicates. Therefore, to minimize the detection error from Eq. (1), E _w is as large as possible. This should be as small as possible. The present invention firstly, And secondly, Make the size smaller.

In the present invention Watermark signal to remove Let the DC component of be 0. That is, it becomes like Equation 2.

[Equation 2]

Where N is a signal unit for embedding and detecting a watermark. And the signal to watermark To Distribution of signal to watermark by dividing into several classes according to the size of It is possible to reduce the That is, for each i Is Things of similar size are classified to have the same class. This means that the variance of the signals belonging to that class is small. After this sorting operation, the DC component of the signal belonging to each class is removed by the first method (the method of setting the DC component of the watermark signal to 0). The total number of classes is M, the maximum and minimum values of x (i) Wow Let's say. At that time, the class classification process is performed by Equation 3 in a process very similar to the quantization process.

[Equation 3]

Here [·] means truncation of the floating point value to an integer value. Is the quantization step Δ = Can be considered to be quantized to an integer in the range [1, M]. One example where a signal is classified by the above process is shown in FIG. The set of signals belonging to class k is defined as in Equation (4).

[Equation 4]

Therefore, in the present invention, in order to minimize the detection error of the watermark signal, firstly, the average of the signals to be watermarked And secondly, the dispersion of the signal to be watermarked. Make the size smaller. Watermark embedding process for this step, generating a watermark; Converting (scanning) an original image (two-dimensional) or an image (two-dimensional) converted into a frequency domain into a one-dimensional signal according to an area where the watermark is inserted; Classifying into classes to minimize variance of the signal transformed in one dimension; Dividing the watermark signal into classes sequentially; Removing the direct current component of each class of watermark signal; Embedding a watermark signal; If the area where the watermark is inserted is a frequency domain, converting the signal into a signal of the spatial domain and performing an inverse scan process.

In addition, in the watermark detection method for protecting the copyright of the digital image,

Converting the test image into a watermark-embedded region (frequency domain or spatial domain) and scanning the converted signal into a one-dimensional signal; Classifying the test image into classes using information classified into classes from the original images to be watermarked; Obtaining a detection response from the classified signal and the watermark from which the DC component used in the watermark embedding process is removed; And determining the presence or absence of a watermark by comparing the detection response with a threshold value.

In addition, the present invention provides a watermark embedding and detection method for protecting copyright of a digital image.

Generating a watermark that is a pseudo-random number; Generating a one-dimensional signal by scanning an image converted into an original image or a frequency region according to a watermarked region (spatial region or frequency region); Classifying a watermarked signal into a class having a similar size to reduce dispersion of the watermarked signal; Classifying watermark signals in order to insert watermarks into the classified and watermarked signals; Removing direct current components of the watermark signals of each class; Obtaining a watermarked signal from the classified and watermarked signal and a watermark; If the watermarked region is a frequency domain, a reverse scan process and an inverse transform process are performed. If the watermarked region is a spatial domain, a watermark is inserted. Converting the watermark into a region (frequency region or spatial region) embedded therein, and scanning the converted signal into a one-dimensional signal; Classifying the test image into classes using information classified into classes from the original images to be watermarked; Obtaining a detection response from the classified signal and the watermark from which the DC component used in the watermarking insertion process is removed; And determining the presence or absence of a watermark by comparing the detection response with a threshold.

Hereinafter, the configuration and processing of the present invention will be described with reference to the accompanying drawings.

2 is an embodiment divided into classes by a signal to be watermarked in order to minimize dispersion of a signal to be watermarked according to the present invention.

3 is a mean square of a signal to insert a watermark according to the present invention; Remove and disperse Is a diagram showing the overall configuration of a watermark embedding method for reducing the size of a.

First, watermark signal of pseudo-random number of length N Is generated according to a given variance (301). The generation of this watermark signal is generated from the random number generator and the secret key. Watermark embedding includes a method of directly inserting into a spatial domain (an image representing brightness and color values) and a method of converting an image into a frequency domain and inserting the image into a frequency domain. The present invention can be applied to any case. A signal obtained by converting a two-dimensional signal into one-dimensional (scanned) image from the original image or the image converted into the frequency domain according to the region (spatial region or frequency region) where the watermark is inserted. (302). When the watermark is inserted in the frequency domain, the frequency transform may be a discrete cosine transform (DCT), a discrete Fourier transform (DFT), a discrete wavelet transform (DWT), or the like. Then, in order to minimize dispersion of the watermarked signal, Classify into the corresponding class. The class number M is determined by the user. Then the signals classified into classes (303). And a signal to be watermarked classified into the generated watermark w (i) as a class. Watermark signals classified by separating them in order to be inserted into the (304). Method of classifying a watermark signal is out allocate a watermark signal having a length of N in order as many as the number N _k of each class. At this time to be. In order to zero the expected value of the signal to be watermarked, The direct current component of, k = 1,2, ... M is removed (305). Then the DC component has been removed To Let's say Then, a watermark is inserted from the classified signal and the watermark signal from which the DC component is removed (306). The insertion method is made of class units, as shown in Equation (5).

[Equation 5]

here Is a coefficient for adjusting the energy of the watermark. The walkmarked signal y (j) is obtained from the sum of the watermarked signals y _k (i) in units of classes. In other words, Finally, if the region into which the watermark is inserted is the frequency domain, inverse scanning and inverse transformation are performed on y (j) to obtain a watermarked image J (i, j) in the spatial domain (307). ). In other words, a watermarked image is obtained. If the area in which the watermark is inserted is a spatial area, the frequency inverse transform is not necessary, and an image J (i, j) in which the watermark is inserted is obtained only through inverse scanning.

4 is a diagram showing a configuration of detecting an inserted watermark according to FIG.

According to the present invention, the process of detecting a watermark may include converting an image to be tested for watermark presence into a watermark insertion region and converting (scanning) a two-dimensional signal into a one-dimensional signal; Classifying a signal to be tested for the presence of a watermark into a class using the original image used to insert the watermark; Obtaining a detection response using the classified signal and the watermark signal from which the DC component is removed; And determining the presence or absence of a watermark from the detection response.

If the watermark embedding area is the frequency domain, the 2D signal is converted into the one-dimensional signal by converting the image to be tested to the frequency domain and scanning the converted image according to the frequency conversion method used. (401). If the watermark embedding area is a spatial area, only the test image is scanned as a one-dimensional signal without performing frequency conversion. Test signal scanned in one dimension by obtaining information on class classification according to equation (3) from original image Classify as a class (402). In other words From class classification information for To Classify as Test signals classified into classes And classified watermark signals Response detected using (403). Watermark signal classified here with DC component removed Should use the same one used for embedding the watermark. This can generate a worker mark signal if the same random number generator and the same secret key as used in the watermark insertion process are also included in the watermark detection process. Finally, the watermark is compared with a threshold (404). If the detection response z is larger than the threshold value, it is determined that there is a watermark, otherwise it is determined that there is no watermark in the test image.

The present invention made as described above has the effect of minimizing the influence of the correlation between the original image and the watermark on the performance of the watermark detector, thereby minimizing the watermark detection error during watermark detection, thereby improving the reliability of the detection.

Claims (7)

In the watermark embedding method for protecting the copyright of the digital image, (1) generating a watermark that is a pseudo-random number; (2) scanning the image converted into the original image or the frequency domain according to the watermarked region (spatial region or frequency region) to generate a one-dimensional signal; (3) classifying the watermarked signals into classes of similar magnitude to reduce dispersion of the watermarked signals; (4) classifying watermark signals in order to insert watermarks into the classified and watermarked signals; (5) removing direct current components of the watermark signals of each class; (6) obtaining a watermarked signal from the classified and watermarked signal and a watermark; (7) if the watermarked region is a frequency domain, undergoing an inverse scan process and an inverse transform process; and if the watermarked region is a spatial region, creating a watermarked image through an inverse scan process; Insert method. In the watermark detection method for protecting the copyright of the digital image, (1) converting the test image into a watermark-embedded region (frequency domain or spatial domain), and scanning the converted signal into a one-dimensional signal; (2) classifying the test image into classes using information classified into classes from the original images to be watermarked; (3) obtaining a detection response from the classified signal and the watermark from which the DC component used in the watermarking insertion process is removed; (4) determining the presence or absence of a watermark by comparing the detection response with a threshold. In the watermark embedding and detection method for protecting the copyright of the digital image, (1) generating a watermark that is a pseudo-random number; (2) scanning the image converted into the original image or the frequency domain according to the watermarked region (spatial region or frequency region) to generate a one-dimensional signal; (3) classifying the watermarked signals into classes of similar magnitude to reduce dispersion of the watermarked signals; (4) classifying watermark signals in order to insert watermarks into the classified and watermarked signals; (5) removing direct current components of the watermark signals of each class; (6) (obtaining a watermarked signal from the classified and watermarked signal and a watermark; (7) if the watermarked region is a frequency domain, undergoes an inverse scan process and an inverse transform process; and if the watermarked region is a spatial region, creating a watermarked image through an inverse scan process, and inserting a watermark, (8) converting the test image into a watermark-embedded region (frequency domain or spatial domain), and scanning the converted signal into a one-dimensional signal; (9) classifying the test image into classes using information classified into classes from the original images to be watermarked; (10) obtaining a detection response from the classified signal and the watermark from which the DC component used in the watermarking insertion process is removed; (11) A watermark embedding and detection method comprising the step of determining the presence or absence of a watermark by comparing the detection response with a threshold value. The method of claim 3, wherein The probability (P−) that the watermark exists but does not detect the watermark and the probability (P +) that the watermark does not exist but are determined to exist are represented by the following detection error equation, Watermark embedding and detection method, characterized in that. The method of claim 4, wherein Watermark energy in order to minimize the detection error from the detection error equation Is as large as possible, and the square of the expected value of the signal to be watermarked Distributed Sum of The method of embedding and detecting watermarks, characterized in that should be as small as possible. The method according to claim 4 or 5, Expected value of the signal to watermark And a direct current component of the watermark signal to zero to remove the watermark. The method according to claim 4 or 5 Dispersion of the signal to be watermarked And dividing the signal to be watermarked into a plurality of classes according to size to reduce the number of watermarks.