KR100500144B1 - Method of embeding and extracting digital watermark for authenticating digitial image - Google Patents

Abstract

A watermark having both characteristics of a fragile watermark and a robust watermark is inserted in the digital image. In order to have strong watermark characteristics such as transition and cropping of an image, a template that is a standard for extracting a watermark is generated and inserted into the image in the form of a tile. The image into which the template is inserted is divided into detection block units, and the image is divided into the feature value portion and the watermarking portion, and the difference between the block average values is obtained from the block of the feature value portion and inserted into the block of the watermarking portion as a watermark. Extracting the inserted watermark extracts a template from an image and selects a detection block based on the extracted watermark. The detection block is divided into the feature value portion and the watermarking portion, and the feature block extracted from the feature value portion and the feature value extracted from the watermarking portion are compared. If the difference is within the tolerance, the detection block is not forged. If it is outside the tolerance, it can be confirmed that it is forged.

Description

Method of embedding and extracting digital watermark for digital image authentication {Method of embeding and extracting digital watermark for authenticating digitial image}

The present invention relates to a method of inserting and extracting a digital watermark for authenticating whether an image is forged or forged. More specifically, when the digital watermark is inserted, a template that is a standard for extracting watermarks is periodically inserted to be strong in cropping or moving the image, and at the same time, in order to resist deformation such as malicious forgery and compression. The present invention relates to a method of inserting and extracting a digital watermark and to an application device using the same to extract a feature value of an image and to authenticate a forgery of the image when the information is inserted and extracted as watermark information.

Various types of technologies for digital watermarking have been proposed since the mid-90s to verify the forgery / modulation of digital images. Digital watermarks to be used in the digital watermarking technology are divided into strong watermarks and fragile watermarks according to the purpose of use.

Robust watermarks are mainly used to protect the copyrights of watermarked digital content (audio, video, images, etc.) as the watermark survives malicious attacks. On the other hand, the soft watermark is used to determine that the image has been forged if the watermark inserted is lost when any forgery is applied to the image to be protected.

There are two types of watermarking techniques, depending on the method of insertion. The first method is to insert a watermark in the frequency domain. Usually, a watermark is inserted by manipulating a DCT coefficient obtained after a discrete cosine transform (DCT). The feature of this method is that it requires a lot of computation and takes a long time because the data obtained in the time domain has to be converted into the frequency domain, but it has some characteristics of compression.

In the second method, data is inserted into a watermark in a spatial domain. The watermark is inserted by changing the pixel values of the spatial domain, and it is determined by measuring the degree of change of the pixel value whether the image is forged or not. The feature of this method is that it inserts a watermark directly in the space domain, so it has a small amount of computation and can be processed in a short time, but it is weak in compression.

Algorithm for image authentication by digital watermarking is mainly made of soft watermark. However, the method of inserting a watermark by changing the least significant bit (LSB) of a pixel value in a spatial domain in units of blocks has a disadvantage in that the minimum bit is easily converted by compression. It was not suitable to be used for image authentication.

In order to solve the above problems, a semi-fragile watermarking algorithm has been proposed. Eugene T. Lin et al (Dectection of image alterations using semi-fragile watermarks, SPIE Security and Watermarking of Multimedia Contents II, pp. 152 to 163, Jan. 2000). We proposed a method for authenticating an image using the correlation. In this method, after dividing the image into blocks of a certain size in the spatial domain, the watermark is inserted, and when the watermark is detected, the correlation between the watermark and the block to be examined is measured. It was judged that they did not, and vice versa. If the block size is small, the length of the sequence for correlation measurement is small, so it is difficult to obtain reliable results. Therefore, a given pixel obtained by obtaining the difference between the next pixel value in the row and column directions without using the pixel value itself is obtained. Correlation was obtained using a sequence twice the value. Therefore, using the pixel values still gives more accurate results than when measuring the correlation. However, there is a disadvantage in that the watermark is incorrectly detected depending on the characteristics of the image when the image has a large number of edges or a pattern.

In addition, U. S. Patent No. 6,246, 777 divides pixels into blocks in a spatial domain, classifies the blocks into microblocks and macroblocks consisting of the microblocks, and obtains them by using a block average value of the microblocks. The block average value of the macroblock is inserted as a watermark. In this case, the forgery of the input image in the watermark inserted state is obtained by checking the average value of the microblocks of the input image to obtain the inserted watermark information and comparing the average value obtained from the macroblock of the input image with a difference of more than a certain threshold. In this case, it is determined that the data is forged.

However, in this method, since a watermark is formed by using a microblock, an error occurs due to compression, and error correction coding is used to correct it. Therefore, in order to correct errors with a lot of computational time, data of two to three times the amount of information are inserted. Therefore, convexity occurs due to the insertion of a watermark in a flat image. In addition, when a watermark is inserted in CCTV cameras and camcorders, and the watermarked video signal is converted into an analog video signal such as NTSC / PAL and transmitted, inevitably, the translation and cropping of the image occurs. . In this case, the watermarking algorithm by the block operation cannot be used, and there is a disadvantage in that it is incapable of attack such as image translation or cropping.

Accordingly, an object of the present invention is to provide a method of inserting and extracting a digital watermark, which is robust to transition or cropping of an image, which is the biggest disadvantage in blocking by periodically inserting a template in order to solve the above problems.

Another object of the present invention is to insert and extract a digital watermark capable of discriminating forgery for an image by inserting a feature value of an image capable of distinguishing a transformation such as malicious forgery and compression for a digital image as a watermark. To provide.

In order to achieve the above-described purpose, the method of embedding a digital watermark for authentication of a digital image according to the present invention

Converting the input image into a predetermined format;

Inserting a template having a predetermined size formed by a pseudo noise sequence into the converted input image;

Dividing the input image into which the template is inserted in units of a detection block having a predetermined size, and dividing the detection block into a feature value block representing a characteristic of the input image and a witter marking block into which a watermark is to be inserted;

Extracting a feature value representing a feature of the input image from the detection blocks representing the feature value block; And

And inserting the feature value into the watermarking block as a watermark.

In addition, the digital watermark extraction method for authentication of the digital image according to the present invention

Converting the input image into a predetermined format;

Extracting a template from the converted input image;

Dividing the input image into detection blocks having a predetermined size based on the extracted template, and dividing the detection block into an embedded witter marking block and a feature value block representing a characteristic of the input image;

Extracting feature values representing the feature of the input image from the feature value block and the watermarking block; And

And determining whether the input image is forged or altered according to the magnitude of the difference between the feature value block and the feature value block in the watermarking block.

Hereinafter, a method of inserting and extracting a digital watermark for authenticating a digital image according to the present invention will be described in detail with reference to the accompanying drawings.

First, a process of inserting a digital watermark for authentication of a digital image according to the present invention and a detailed configuration thereof will be described with reference to FIGS. 1 to 6. FIG. 1 is a schematic diagram illustrating an entire process for insertion, which includes inserting a template that is a reference when extracting a watermark, extracting a feature value by dividing an image according to the formed template, and extracting the extracted feature value. This process consists of inserting the watermark information as the watermark information.

As shown in FIG. 1, when an image is first input, the format of the input image is converted into a format of a predetermined form (step S10). In general, the input image has an RGB format, and the input image is converted into the YCbCr format, which is a standard format used when performing compression such as JPEG. The conversion at this time is converted by the following equation (1).

In the YCbCr format obtained by Equation 1, only the Y component is extracted and used. After the conversion of the input image to be watermarked is completed, the insertion and extraction of the watermark are performed in units of blocks. For this purpose, it is assumed that a watermark signal is used as a template to determine a standard of a method of setting a block. The template serves as a reference for extracting the watermark later. The size of the template block is 80 * 80 pixels and is generated using a pseudo-random sequence and periodically inserted into the input image through the process of FIG. 2 (step S20).

FIG. 2 schematically illustrates a process of inserting a pattern generated using a secret key, and looks at the process in detail with reference to this.

As shown in FIG. 2, a secret key input by a user is first provided to form a template block (step S100). The provided secret key is input to the pattern generator 10 generating pseudo noise to generate a two-dimensional pattern matrix of MxM size (step S110). The generated pattern matrix is periodically inserted up, down, left, and right in the form of a tile with respect to the input image (S120).

The above process will be described in more detail with reference to FIG. 3, which is a block diagram showing a configuration for inserting a template into an input image in order to implement a pattern insertion process.

As shown in FIG. 3, the pattern generation unit 10 generates a two-dimensional pattern matrix w (k) having a two-dimensional size of MxM by inputting a secret key. In addition, the input image is input to the image characteristic extraction unit 20 to extract λ (x), which is a scaling variable representing a local characteristic (local activity) of the input image. The scaling parameter λ (x) has a small value in an image portion having a small activity, for example, a small value in a flat image and a large value in an image portion having a large amount of activity such as an outline or a pattern region.

The reason for using scaling variables is to take advantage of the property that the human visual system (HVS) is less sensitive to high frequency components than to low frequency components. That is, the watermark pattern is strongly inserted in the less sensitive high frequency component region and weakly inserted in the low frequency component region of the image. The scaling parameter in the present invention uses an absolute value of a value obtained by performing Laplacian high frequency filtering on an image as shown in Equation 2 below.

From here, Is a convolution, and L is a Laplacian filter given by equation (3).

As such, when a scaling parameter considering a random number pattern using a secret key and local activity of an input image is obtained, these values are input to the modulator 30 to obtain a template (w (x)) considering local activity.

In addition, the template (watermark) must satisfy the condition required by Equation 4 to be independent of the movement of the image.

Here k is a multiple of M and M is tile size and M = 80 in the present invention. That is, in order to detect the watermark, the watermark is processed and inserted into the image in units of blocks. Therefore, in order to extract the watermark, it is possible to accurately detect the block position at the time of insertion. Therefore, when a part of the image is cut out, it is not easy to detect the watermark. In order to solve this problem, the present invention is a reference when a watermark is extracted by periodically inserting a template.

Subsequently, the multiplier 40 multiplies the scaling variable s representing the characteristics of the entire image with the template. The scaling variable here is related to the overall strength of inserting the template. If the compression is increased after the watermark is inserted, the scaling parameter value should be increased. On the contrary, if the image quality is important, that is, if the compression is weak after the watermark is inserted, the scaling parameter value is reduced. This value can be determined experimentally. Scaling variables use values greater than zero. The final template thus obtained is added by the input image and the adder 50 to obtain an image Y of the equation (5) including the template.

The final image Y is obtained by adding the watermark pattern to the entire region of the image by repeating the matrix {wi} like a tile and adding the input image.

The image into which the template is obtained is inserted is divided into detection blocks. In the present invention, the size of the detection block is 40 * 40 pixels. The size of the detection block is determined in consideration of the size of the template. Since the position of the watermark detection block is found based on the template block, knowing the reference point of the template block makes it easy to know the position of the watermark detection block. Therefore, the size of the watermark detection block is set smaller than that of the template block.

Each detection block is divided into a signature portion (block) and a watermarking portion (block) by a random value generated by the secret key. At this time, the two blocks are set so as not to overlap each other to prevent the feature value from being changed by the watermark when the extracted feature values are inserted using the watermarking method. As shown in Figs. 4A and 4B, the detection block according to the random number (0 or 1) can be divided and set, and this division method can be variously determined.

In FIG. 4A and FIG. 4B, the size of the small watermarking block in the watermarking block is 10 * 10 pixels, and the small watermarking block formed of such small pixels is composed of eight small watermarking blocks. Also, the feature value portion is divided into two small feature value blocks, and the size of each small feature value block is 20 * 20 pixels. The above-mentioned division method is mentioned as an example, and the feature value block and the watermarking block may be divided into arbitrary forms. That is, each block may be in the form of a circle or a rectangle, and may be in any form as long as only a block average value mentioned later for the block can be obtained.

After the image is divided into the feature value block and the watermarking block, the feature value of the image is extracted. The feature value is the difference between the mean values of two small feature block. This process will be described in more detail with reference to FIG. 5.

First, a block average value is obtained for each of the two feature block (feature value 0, feature value 1) divided into two as shown in FIGS. 4A and 4B (steps S210 and S220). The average value of the pixels xi, i = 1, ..., n (where n is the number of pixels of the partitioned signature block) constituting the feature block is calculated using Equation 6.

At this time, in order to ensure the robustness of the feature value, the average value of each feature block is modulated to be the median value of the corresponding section (step S230). In other words, it modifies so that the value change is small for non- malicious attacks (lossy compression, format conversion), so that it is strong for non- malicious attacks and easily broken for malicious attacks. Amount of change to modulate to the median ( ) Is obtained using Equation 7 below. Modulation to a median for robustness may cause the feature value to change only if more than one-half of the quantization size changes, so that the same value is always maintained for any change below it. Therefore, it is less likely that the feature value is changed by non- malicious attacks.

Δs is the magnitude of the quantization step, and '' 'is an operator for extracting the minimum integer value.

Changing the average value of the feature block to an intermediate value by the equation (7) means changing the pixel value of the feature block by the equation (8).

Where x * represents the changed pixel value.

The difference between the average value of each feature block obtained above is obtained, and the absolute value is set as a feature value as in Equation 9 (step S240).

Here, Ms1 represents an average value of the feature value 0 block, and Ms2 represents an average value of the feature value 1 block.

As described above, when the feature value is determined, it is inserted into the image as watermark information. The insertion process will be described with reference to FIG. The feature value is converted into a binary sequence (step S310). After conversion, additional conversion is performed to the gray code having an 8-bit size (S320). This additional conversion process is to protect the watermark embedding information more confidentially. If the binary sequence is represented by bi, i = 1, .., N and the gray code is gi, i = 1, ..., N, the conversion is given by the following equation (10).

Where denotes an exclusive-OR.

In addition, the equation for converting the gray code into a binary sequence is given by Equation (11).

The gray code value g _i obtained by the equation (10) is quantized and inserted into the average value of the small watermarking block. In detail, the small watermarking block w _i corresponding to the bit b _i to be inserted is selected (step S330). Selection of the watermarking block selects a small watermarking block corresponding to each bit string since the watermark information to be inserted is 8 bits. By using quantization of the mean value, b _i is inserted as a watermark bit (step S340). The amount u to be changed in order to insert b _i as a watermark is obtained using equations (12) and (13).

Where r is quantization noise and Δw is the magnitude of the quantization interval.

From here Is a quantization function and is given by the following equation (14).

The amount to be changed ) Using equation (15).

The above process is repeated until all 8 bits are inserted (step S350). If all 8 bits are not inserted, the process proceeds to the next bit (step S360), and performs steps S330 and S340 previously performed. By this repetitive operation, the feature of the input image is used as the watermark information, and the insertion process is performed on the watermark block portion to continue to the last detection block. It is possible to determine forgery for the input image based on the feature value inserted in the watermark embedding process.

After the watermark of the input image is inserted, the image is converted into the RGB format through reverse conversion of the format performed in step S10, and then recorded on the recording medium (step S60). That is, since only the Y component is processed in the YCrCb format for the watermarking process, after the Y component and the CrCb component are synthesized, the inverse matrix of Equation 1 is obtained, and then the YCrCb matrix and the calculation are performed in the RGB format. An output image having a watermark embedded therein is obtained and recorded on a recording medium or the like.

Hereinafter, a process of extracting a watermark from the image to examine whether the watermark is inserted into the image or the image is forged is described in detail with reference to FIG. 7.

7 is a flowchart schematically illustrating a process of extracting a digital watermark for authentication of a digital image according to the present invention.

In the same manner as in the above-described process of embedding the digital watermark, the format of the input image is converted into a predetermined form in order to extract the watermark (S400). It converts the input image of RGB format into YCbCr format which is a standard format used when performing compression such as JPEG. The conversion at this time is converted by the above equation (1).

After the conversion, only the Y component is extracted from the YCbCr component, and the template is extracted based on the Y component (step S410). The extraction process of the template will be described with reference to FIG. 8. First, since the template is repeatedly inserted in the input image to a predetermined size, the template is set to the size of M * M pixels and the input image is divided into M * M pixels (step S500). Each of the divided input images is added, and the added image B having the size of the M * M pixel is stored in the buffer (step S510). The image stored in the buffer is given by the following equation (16).

Where j ₁ and j ₂ represent the index of the tile.

In addition, the watermark pattern w (k) is generated by the secret key together with the processing for the input image as described above (S520). The watermark pattern is also formed in a two-dimensional pattern of the same size as the template.

The correlation between the watermark pattern w (k) and the added image stored in the buffer is measured (step S530). After dividing the input image into the template size and adding all the divided blocks without using one divided image, the correlation between the watermark pattern and the watermark pattern is higher than that of one block. A much larger value can be obtained, which makes it easier to determine the correlation (T. Kalker, G. Depovere, J. Haitsman, and M. Maes, "A video watermarking system for broadcast monitoring", Proceedings of the SPIE Security and Watermarking of Multimedia Contents, vol. 3657, pp. 103-112, San Jose, January 25-27, 1999). The correlation is measured by the following equation (17).

From here Ik, which is the index of, is calculated as the modulo of M (means tiling an image to the size of M and adding it). Equation 17 represents a two-dimensional cyclic convolution. Is In spatial flipping Equation 17 can be expressed as Equation 18 below. Since we want to do it again, we can find the original desired correlation through convolution.

From here " "Represents the cyclic convolution. The cyclic convolution can be performed more efficiently by simply performing multiplication in the frequency domain using Fourier transform than in the spatial domain. do.

Here, IFFT represents an inverse fast Fourier transform and FFT represents a fast Fourier transform.

An example of the result calculated by the above Equation 19 is shown in FIG. As shown in FIG. 9, the starting point of the detection block can be found by the position of the peak. At this time, it is searched that the value of the peak has the maximum peak above a predetermined value (S540). As a result of the search, it can be seen that the peak is located at 53 rows and 15 columns. From this, it can be seen that the first detection block starts after 15 rows 53 columns from the upper left of the input image (1 row 1 column). Step S550).

When the template is extracted as above, the image is divided (step S420). As described above, the image segmentation is divided into the feature value part and the watermarking part according to the random number generated by the secret key. Since the size of the template is 80 * 80, the first four random numbers generated by the secret key are used repeatedly. If the size of the template is increased or the distance between the templates is larger, the number of detection blocks can be included, and thus the value of the repeated random number is increased to increase the security. 10 shows the relationship between the template and the detection block used in the experiment in the present invention. Four detection blocks are located in one template.

When the image is divided into the feature value block and the detection block as described above, the feature value I ₁ is extracted in the same manner as the previous method of inserting the watermark. In the watermarking portion, the feature value I ₂ watermarked by the watermark inserting unit is extracted. Since the value extracted from the watermarking part is a gray code, it is converted into a binary sequence using Equation 11 and converted into an integer value again.

Once the values of I ₁ and I ₂ are determined, it is determined whether the difference between these two values is within the tolerance range according to Equation 20.

As such, if the difference between the two values is within the tolerance range, it is determined that the corresponding detection block is not forged, and if it is larger than the tolerance, it is determined as forged.

As described above, in the method of inserting and extracting watermarks for digital image authentication according to the present invention, the influence of the compression of the actual image, the change of the image quality according to the insertion of the watermark, the robustness to compression and the actual application The actual test was performed to detect the forgery of the image and the result is as follows.

1) Error test by JPEG compression of block mean value

After quantizing and inserting the block mean value, JPEG compression is changed by changing the QF (Quality Factor) from 90 to 50 (the closer to 0, the more compressed, the closer to 100, the less compressed). Measured. This measurement result is shown in FIG. The horizontal axis represents the change value of the block average value of the watermarking block WB, and the vertical axis represents the number of blocks of which the watermark block has been changed. As shown in FIG. 11, even when the compression ratio increases from QF = 90 to QF = 50, it can be seen that the change in the block average value is small. This means that the block average difference value used as the feature value is also small, and thus, the block mean difference value is less affected by non- malicious attacks such as compression. As a result, the block average value is effective as a means of watermarking.

2) Evaluation of image quality

It is important for the user not to know the change of the image before and after the watermark insertion. As a relatively objective value for image quality evaluation, PSNR (Peak Signal to Noise Ratio) is used in the image. There may be many changes depending on the inserted image, and PSNR is expressed by Equation 21 below.

X and Y are the number of columns and rows of the input image. px, y represents the original image, p'x, y represents the pixel value of the watermarked image. 12A to 12C obtained in the experiment, FIG. 12A illustrates an image having a size of 460 * 400 pixels, FIG. 12B to 320 * 410 pixels, and FIG. 12C to 560 * 380 pixels. Table 1 shows the PSNR before and after watermarking for the images having the pixel size as described above.

Drawing 12A Figure 12b Figure 12c PSNR (dB) 40.96 41.74 41.38

As shown in the table, the PSNR values performed in each case of FIGS. 12A to 12C are all 40 dB or more, which shows little loss in the image. If it is 30dB or more, there is no discomfort).

3) toughness against compression

The robustness against compression was tested to determine the suitability of semi-fragile watermarking. For robustness test, FPE (False Positive Error) which is judged to be forgery when the watermark of the compressed image is detected as it is without any manipulation of the image under compression with each QF, and only JPEG Table 2 shows the results of measuring the number of errors determined as forgery when compression is performed for various compression ratios. The number of blocks showing errors among the total detection blocks is shown.

12A Figure 12b Figure 12c QF = 90 0/110 0/80 0/126 QF = 80 0/110 0/80 0/126 QF = 70 0/110 6/80 0/126 QF = 60 3/110 13/80 0/126 QF = 50 1/110 11/80 0/126

As shown in Table 2, it can be seen that there are some differences depending on the figure. No error was detected with compression above QF = 80. The insertion strength used in the experiment was measured by the quantization size. The signature part was Δs = 2, and the watermarking part was inserted with a larger Δw = 3 because the size of the average value was smaller than the signature part.

4) Forgery detection of video

Results verified after modulating a portion of the image are shown in FIGS. 13A-13C. FIG. 13A is an example of an image having a size of 320 * 240 pixels, FIG. 13B is a case in which the '03 .05 'portion of FIG. 13A is modulated to '05 .05', and FIG. 13C is a diagram illustrating the detection of the modulated portion in FIG. 13B. Forgery-modulated areas are shown in white in the detection block unit. In FIG. 13C, the image is larger than the actual modulated image. This is because the forgery discrimination is made by the detection block unit (40 * 40).

FIG. 13D illustrates a change in Block Average Difference (BAD), which is a signature value between the original image and the forged image. In the forged region (fifth detection block), about 3.8 changes were measured. This is considered to be forgery because it is larger than 1.01, which is the tolerance in the experiment.

Fig. 13E shows the value of | original image BAD-forgery image BAD |. Here you can see that the change is quite large in the fifth detection block. FIG. 13F illustrates a histogram of the change amount obtained in FIG. 13E, and it can be seen that there is one number of blocks changed at about 3.8.

5) Application example of digital watermark insertion / extraction method according to the present invention

(1) Application to digital camera

The experiment was performed by applying the digital watermark insertion / extraction method to the digital camera. As implemented in FIG. 14A, the digital camera 100 may be applied in two ways to insert a digital watermark according to the present invention. First, a method of embedding an algorithm for implementing a digital watermark embedding method according to the present invention into a chipset 120 of a digital camera and inserting and storing a watermark directly into an image captured by the CCD module 110. to be. The advantage of this method is that in a digital camera, the storage device 130 is removable so that it can be used using an embedded watermark captured later, even if it is detachable from the digital camera. Second, a method of inserting a watermark at a device driver stage in the process of uploading a captured image after photographing with a digital camera to the personal computer 140. Currently implemented digital cameras are implemented in this way.

(2) Applied to DVR system

The DVR system stores video from digital camcorders or CCTV cameras in storage media such as hard disks or DAT. 15A and 15B illustrate examples of the structure of a DVR system for preventing forgery and alteration to be implemented in the present invention.

Referring to the example illustrated in FIG. 15A, an image is input from the camera 150 to the DVR system 200. At this time, the video is available in NSTC / PAL mode. The input image is separated into the YUV format by the signal converter 210 and then compressed strongly to store a lot of long time images. A compression codec unit (codec: compression and decompression) 220 compresses the input image data using a compression format such as MPEG or MJPEG as a compression method, and decompresses the compressed and stored image data to output device (monitor, TV). Etc.) to make the image data format immediately visible. The compressed data is encrypted and stored before being stored in the data storage 230. At this time, the encryption is performed using unique information (for example, a unique ID or a unique ID among components constituting the system) that can be extracted from the DVR system 200 itself. The reason for encrypting and storing is to prevent loss of the data storage 230 or forgery of the stored image in advance.

In order to view the stored encrypted image, a process of extracting from the DVR system to the external device 250 is required. At this time, in the extracting step, the encrypted image stored in the data storage unit 230 is decoded and decompressed. The watermark is inserted using the watermarking processor 240 in which an algorithm for the method of inserting and extracting the digital watermark for forgery authentication proposed in the present invention is implemented. By doing so, it is possible to determine whether the forgery may occur while moving to the external device 250 to be extracted and finally used from the DVR system. Thus, forged images can be prevented from being adopted as false evidence.

FIG. 15B illustrates another application example in which the configuration of FIG. 15A is modified. The difference from the configuration shown in FIG. 15A in the example of FIG. 15B is that the watermarking unit 240 that performs the watermarking process is installed before the compression codec unit 220 to the signal converted by the signal conversion unit 210. The watermarking process is performed immediately. The process of storing the data in the data storage unit 230 or extracting the data to the external device 250 by encrypting and decrypting the watermarked signal is performed as described above. Thus, in this example, the method proposed in the present invention has a robustness to compression, so that the watermark is alive even after the watermark processing unit 240 compresses the input image after processing the input image. Therefore, when any forgery or forgery occurs after being stored, the forgery and forgery can be determined.

In addition, the compression codec unit 220 encrypts using unique information that can be extracted from the DVR system 200 itself. The reason for encrypting and storing is to prevent data loss of the data storage unit or forgery and forgery of stored images in advance. Here, the encryption and decryption process can be omitted. This is because the watermark is inserted before it is stored, so if the data is forged or forged, it is easy to know the forgery and forgery. In order to view the stored encrypted image, a process of extracting from the DVR system to an external device is required. At this time, in the extracting step, the first encrypted image can be decrypted and decompressed to view the image using an external device. Here, the watermarking algorithm may use not only the algorithm proposed in the present invention but also other watermarking algorithms having semi-fragile characteristics. To implement this watermarking function, it can be implemented with FPGA or ASIC chip and DSP chip can be used.

As described above, the present invention makes it possible to use a template at the time of inserting a watermark as a reference when extracting the watermark, and insert the template into a video in the form of a tile, and then reference the watermark information to the template. By inserting it, it can have a strong characteristic against malicious attack such as image transition and cropping.

In addition, the image is divided into detection blocks, the detection block is divided into a feature value portion and a watermarking portion, and an average value difference between blocks forming the feature value portion is inserted as watermark information. It compares the inserted watermarking values to determine whether the image is forged or not. This process is performed in the spatial domain, so that it can be processed at high speed, and the average value and difference between blocks are less affected by non- malicious attacks such as compression. There is an effective advantage as a means of marking.

While the invention has been particularly shown and described with reference to the above embodiments, it has been used for the purpose of illustration and those of ordinary skill in the art, having the spirit and scope of the invention as defined in the appended claims. Various modifications can be made without departing.

1 is a flowchart schematically illustrating a process of inserting a digital watermark for image authentication according to the present invention.

2 is a flowchart schematically illustrating an insertion process of a pattern generated using a secret key.

FIG. 3 is a block diagram illustrating a configuration for inserting a template pattern into an input image to implement the pattern insertion process of FIG. 2.

4A and 4B illustrate examples of dividing an image into which a template is inserted in units of detection blocks.

5 is a flowchart illustrating a process of calculating a feature value in a detection block.

6 is a flowchart illustrating a process of inserting the feature value obtained in FIG. 5 as watermark information.

7 is a flowchart schematically illustrating a process of extracting a digital watermark after authentication of a digital image according to the present invention.

8 is a flowchart illustrating a process of extracting a template from a digital image.

9 is a diagram illustrating an example of an extracted value of a template.

10 is a diagram showing the relationship between the template and the detection block used in the experiment in the present invention.

11 is a graph illustrating a change in the watermarking block average value according to compression.

12A to 12C illustrate an example of images in which PSNR tests have been performed before and after watermarking. FIG. 12A shows 460 * 400 pixels, FIG. 12B shows 320 * 410 pixels, and FIG. 12C shows 560 * 380 pixels. It shows an image having a size.

FIG. 13A is an example of an image (original image) having a size of 320 * 240 pixels, FIG. 13B is a case where a part of FIG. 13A is modulated (modulated image), and FIG. 13C detects a modulated portion in FIG. 13B. One image is shown.

FIG. 13D is a graph showing a change in the block average difference which is a characteristic value of the image in FIGS. 13A and 13B, FIG. 13E is a graph showing a difference in the block average value between the original image and the modulated image, and FIG. Figure shows the histogram of the amount of change obtained in 13e.

14 illustrates an example of a configuration in which a digital watermark insertion method for authenticating forgery for a digital image is applied to a digital camera according to the present invention.

15A and 15B illustrate examples of a DVR system to which a method of inserting and extracting a digital watermark for authentication of forgery for a digital image according to the present invention is applied.

※ Explanation of codes for main parts of drawing

10 ... pattern generator 20 ... image characteristic extractor

30 ... Modulator 40 ... Multiplier

50 ... adder 100 ... digital camera

110 ... CCD Module 120 ... Chipset

130 ... Storage 140 ... PC

150 ... Camera 200 ... DVR System

210 ... signal conversion section 220 ... compression codec section

230 ... data storage 240 ... water marking processing

250 ... external

Claims (27)

In the method of inserting a digital watermark for authentication of a digital image, Converting the input image into a predetermined format; Inserting a template having a predetermined size formed by a pseudo noise sequence into the converted input image; Dividing the input image into which the template is inserted in units of detection blocks having a predetermined size, and dividing the detection block into a feature value block representing characteristics of the input image and a watermarking block into which a watermark is to be inserted; Extracting a feature value representing a feature of the input image from the detection blocks representing the feature value block; And Embedding the feature value as a watermark in the watermarking block. The method of claim 1, The image conversion step includes converting an RGB input image into a YCbCr format, and using the following conversion equation. The method of claim 2, The template inserting step Generating a two-dimensional pattern by generating a pseudonoise sequence using a predetermined secret key; And And repeatedly inserting the generated pattern in the form of a tile with respect to the entire area of the input image. The method of claim 3, wherein And applying a local scaling factor according to the frequency characteristic of the input image to the generated two-dimensional pattern. The method of claim 4, wherein And applying a scaling factor representing a characteristic of the entire image to the two-dimensional pattern to which the local scaling factor is applied. The method according to claim 4 or 5, And the local scaling factor uses an absolute value of a value obtained by performing Laplacian high frequency filtering on the input image. The method of claim 3, wherein And the size of the detection block is set to have a size smaller than that of the template. The method of claim 3, wherein The feature value extraction step Calculating a respective block average value from the feature block; And And setting the difference between the block average values as a feature value. The method of claim 8, And after the calculation of the block average value, modulating the block average value to be an intermediate value of a corresponding section. The method according to claim 8 or 9, The feature value insertion step Converting the feature value into a binary sequence; And And inserting the converted feature values by quantizing an average value of the watermarking block. The method of claim 10, And converting the converted feature values into a gray code. In the method for extracting the digital watermark inserted by claim 1, 2, 3, 4, 5, 7, 8 or 9 for the authentication of forgery of the digital image, Converting the input image into a predetermined format; Extracting a template from the converted input image; Dividing the input image into detection blocks having a predetermined size based on the extracted template, and dividing the detection block into a feature value block representing a characteristic of the input image and a watermark into an inserted watermarking block; Extracting feature values representing the feature of the input image from the feature value block and the watermarking block; And And determining whether the input image is forged or altered according to the magnitude of the difference between the feature value block and the feature value in the watermarking block. The method of claim 12, The image conversion step includes converting an RGB input image into a YCbCr format, and using the following conversion equation. The method of claim 13, The template extraction method Dividing the input image into a predetermined size and adding the same; Forming a watermark pattern formed of a pseudo noise sequence; Calculating a correlation between the added input image and the watermark pattern; And And extracting a template representing a start point of the detection block from the calculated value having a peak of a predetermined size or more. The method of claim 14, The feature value extraction step The feature value extraction step Calculating a respective block average value from the feature block; And And setting the difference between the block mean values as a feature value. The method according to any one of claims 13 to 15, And only a Y component is used as the input image in the converted YCbCr format. A computer-readable recording medium having recorded thereon a program for performing an operation by the method of inserting a digital watermark according to claim 1, 2, 3, 4, 5, 7, 8 or 9 for authentication of forgery of a digital image. 18. The computer program product of claim 17, wherein the program is implemented in an FPGA, an ASIC chip or a DSP chip. A computer-readable recording medium having recorded thereon a program for performing an operation by a method of extracting a digital watermark according to claim 12 for authenticating forgery of a digital image. Signal conversion means for converting an image signal captured by the camera into a predetermined format; Storage means for storing the video signal; Compression and decompression means for compressing the converted video signal in the format and storing the decoded video signal in the storage means and decompressing the video signal extracted from the storage means to output the stored signal to the outside; Watermarking processing means for inserting and outputting the watermark by performing the method of embedding the digital watermark according to claim 1, 2, 3, 4, 5, 7, 8 or 9 for authentication of the forgery for the extended signal Digital video recording system comprising a. The method of claim 20, The image conversion step includes converting an RGB input image into a YCbCr format, and using the following conversion equation. The method of claim 21, And said compression and decompression means encrypts the compressed video signal. The method of claim 22, And said compression and decompression means decodes the signal extracted from said storage means. Signal conversion means for converting an image signal captured by the camera into a predetermined format; Watermarking processing means by inserting a watermark by performing the method of inserting a digital watermark according to claim 1, 2, 3, 4, 5, 7, 8 or 9 for authentication of the forgery for the converted signal. ; Storage means for storing the video signal; And And compression and decompression means for compressing the watermarked video signal, storing it in the storage means, and decompressing the video signal extracted from the storage means to output the stored signal to the outside. system. The method of claim 24, The image conversion step includes converting an RGB input image into a YCbCr format, and using the following conversion equation. The method of claim 25, And the compression and decompression means encrypts the compressed video signal and stores the compressed video signal in the storage means. The method of claim 26, And said compression and decompression means decodes the signal extracted from said storage means to output the stored image signal to the outside.