KR20030054101A - Method for embedding watermark which is not affected by image compression and judging the alteration or forgery of the image using thereof - Google Patents

Abstract

PURPOSE: A method for inserting a watermark that is not affected by image compression and a method for judging forgery of an image using the watermark insertion method are provided to judge only whether or not there is deformation of an image without being affected by image compression. CONSTITUTION: An input image is split into the first blocks each of which has a predetermined pixel size(S220). Each of the first blocks is divided into two second blocks each of which has a predetermined pixel size, and each of the second blocks is discrete-cosine-transformed(S230). A watermark is inserted into the transformed second blocks(S250). The second blocks having the watermark are inverse-discrete-cosine- transformed(S260), to output an image into which the watermark has been inserted.

Description

How to insert a watermark that is not affected by image compression, and how to determine forgery of an image using the same

The present invention relates to a method of inserting a watermark that is not affected by image compression and a method of determining whether the image is forged or altered. Specifically, the image is easily broken by deformation of an image applied from the outside, but the effect of image compression The present invention relates to a method of inserting a soft watermark that does not receive a mark, and to a method of determining whether an image on which a watermark is inserted has been forged by a method of inserting a strong watermark in compression.

In the conventional watermark embedding method, in inserting a series of information signals into a digital image, the image into which the information signal is inserted cannot be visually distinguished from the first original image without the information signal. The information signal is inserted at a very low intensity, and is set to break any deformation applied to the image from the outside. Such a method is generally referred to as fragile watermarking.

Conventional soft watermarking methods are for the insertion and extraction of an information signal (hereinafter referred to as a watermark) for authenticating that the image into which the watermark signal is inserted is in a state of no modification. That is, after a series of watermarks are inserted into the image, the image is forged only if the image is not compressed, filtered, or subjected to image transformation such as a geometric blank. In other words, the existing fragile watermark is used to authenticate the image forgery. The fragile watermark is so weak that it is to be broken in any image deformation. Such a method is described in detail in patent application No. 2000-64767 filed by the applicant.

However, devices that are designed to capture digital images recently, such as digital cameras, typically compress and store digital images when most pictures are taken. In other words, the compression process is usually set as the default (fixed function) due to the efficiency of image processing, storage space, and transmission capacity. In other words, unlike other variations, compression (lossy compression (e.g. JPEG)) corrupts the original image, but in the case of devices that involve data storage (e.g. digital cameras store images). In order to take a picture, it compresses and saves the image.)

Therefore, in the case of using the conventional soft watermark technology, since a series of watermarks embedded in an image are broken even by simple compression, "compression" is recognized as a forgery. That is, when the inserted watermark is extracted from the compressed image, the watermark is broken, and thus it is impossible to extract the watermark. Therefore, it is determined that the image is forged. In this case, the conventional weak watermarking method has a disadvantage in that it is difficult to use the image forgery.

In this case, in the information signal insertion method for authenticating the image forgery, the forgery is detected by breaking the other image deformation as in the conventional method. There is a need for the mark to remain alive. In other words, image compression needs to not belong to any image transformation.

The present invention provides a method for inserting a watermark that is robust to compression but weak for other image modifications according to the necessity of a new weak watermarking method, and a method for determining whether a forgery of an input image is provided using the embedding method of the present invention. It aims to do it.

1 is a block diagram showing an example of the configuration of a system in which the watermark embedding method of the present invention is executed.

2 is a flowchart illustrating a watermark embedding process of the present invention.

3 is a block diagram illustrating an example of a configuration of a system in which a method for determining whether a video is forged or not is executed.

4 is a flowchart illustrating a process of determining whether an image is forged or not.

<Description of the symbols for the main parts of the drawings>

110,310: image analyzer 120: first color image processor

130: first black and white image processing unit 140: watermark insertion unit

150: second color image processing unit 160: second black and white image processing unit

170,370: watermarking rule generation unit 320: color image processing unit

330: monochrome image processing unit 340: forgery and alteration determination unit

In order to achieve the above object, a method of inserting a watermark sensitive only to forgery of an image without being affected by image compression of the present invention comprises: dividing an input image into first blocks having a predetermined pixel size; Dividing the first blocks into two second blocks having a predetermined pixel size, and discrete cosine transforming each of the second blocks; Inserting a watermark into the converted second blocks according to a watermarking rule; And outputting a watermark-embedded image by performing inverse discrete cosine conversion on the second blocks having the watermark embedded therein, wherein the first block is an 8 * 16 pixel sized block. A block is a block of 8 * 8 pixel size.

Meanwhile, when the input image is a color image, the method further includes converting the input image into an image model that can be easily compressed before the step of dividing the input image into first blocks, wherein the second block is inserted with a watermark. After the step of inverse discrete cosine transform, the watermark embedding method further comprises the step of inversely transforming the image represented by the easy-compression image model into the form of the input image.

A method for determining whether an input image is forged or modified using a watermark of the present invention comprises the steps of: dividing an input image into first blocks having a predetermined pixel size; Dividing the first blocks into two second blocks having a predetermined pixel size, and discrete cosine transforming each of the second blocks; And determining whether the image is forged or altered according to an input watermarking rule. Preferably, the first block is a block of 8 * 16 pixel size, and the second block is a block of 8 * 8 pixel size. to be.

Meanwhile, when the input image is a color image, the determination method of the present invention further includes converting the input image into an image model that can be easily compressed before dividing the input image into first blocks.

Hereinafter, with reference to the accompanying drawings will be described an embodiment of the present invention;

1 is a block diagram showing an example of the configuration of a system in which the watermark embedding method of the present invention is executed. The watermark embedding system of the present invention includes an image analyzer 110 for determining whether an input image is a color image or a black and white image, converting the input color image into a model that is easily compressed, and dividing the input image into blocks of a predetermined size. The first color image processing unit 120 to perform a discrete cosine transform after dividing the input black and white image into blocks having a predetermined size, and then performs a discrete cosine transform (DCT) on the first black and white image processing unit 130 and water using user information. Watermarking rule generation unit 170 for generating a marking rule, watermark embedding unit 140 for inserting a watermark by adding or subtracting a predetermined watermark intensity in accordance with the watermarking rule to the DCT coefficient of the input image, watermark insertion A second color image processing unit 150 for performing inverse discrete cosine conversion on the color image and converting the color image into an image form at the time of input and outputting the watermark embedded image, and the watermark embedded therein. After performing the image back to an inverse discrete cosine they transform and a second black-and-white image processing unit 160 outputs the watermark embedded image.

A process of embedding a watermark in an input image in the watermark embedding system of the present invention will be described with reference to FIG. 1 and FIG. 2, which is a flowchart illustrating a process of the watermark embedding method of the present invention.

When an image is input to the system to be executed in the watermark embedding method of the present invention (S200), the image analyzer 110 first checks whether the input image includes size information and color information of the input image. It is determined whether the image is a color image or a black and white image (S205).

First, a case in which the input image is a color image will be described.

When the input image is, for example, a color image represented by an RGB model, the input image is output to the first color image processing unit 120 through the image analyzer 110. The input image represented by the color model is converted into a Ycbcr model, which is a model commonly used for JPEG compression in the first color image processing unit 120 (S210). In the Ycbcr model, a human visual characteristic has a brightness signal rather than a color difference signal. By using the fact that the sensitivity is more sensitive, the image signal is divided into a brightness signal (Y) and a chrominance signal (cb, cr) and digitally expressed to facilitate compression.

If the input color image is in the RGB format, for example, the RGB image is converted into the Ycbcr model using the following equation.

Y = 0.29900R + 0.58700G + 0.11400B

Cb = -0.16874R-0.33126G + 0.50000B

Cr = 0.50000R-0.41869G-0.08131B

Such a conversion equation represents an example of converting an input image from an RGB model to a Ycbcr model. In addition, various conversion methods are known to those skilled in the art.

When the color image is converted into the Ycbcr model, the image is divided into blocks of 8 * 16 units for the Y component (S220). The image divided into 8 * 16 unit blocks is further divided into 8 * 8 block units and subjected to discrete cosine transform (DCT) (S230). The DCT coefficients of the blocks generated by DCT conversion are output to the watermark inserting unit 140, and the predetermined watermarks are output according to the watermarking rule output from the watermarking rule generating unit 170. The watermark is inserted by adding or subtracting the mark intensity to the DCT coefficient of each block (S250).

Here, the generation process of the watermarking rule will be described in detail.

In order to generate a watermark condition to be used in the present invention, the watermarking rule generator 170 uses user information. The user information used to generate the watermarking rule may include a social security number, a name, a PIN, a password, and a user's information. Personal information such as a logo. The watermarking rule generator 170 applies a predetermined algorithm to the input user information to generate a watermarking rule column corresponding to the number of block pairs so as to correspond to the above-described block pairs divided into 8 * 16 units. Output to the insertion unit 140, wherein the watermarking rule string is composed of 1 or 0.

The generated watermarking rule literally means a rule at the time of embedding the watermark. For example, if the value of the watermark is 1, the watermarking rule is included in an 8 * 16 block input from the first color image processing unit 120. It means that the DCT coefficient value at a predetermined position of the first block of size 8 * 8 is larger than the DCT coefficient value at a predetermined position of the second block of size 8 * 8, whereas, if the value of the watermark is 0, the water Conversely, when the value of the mark is 1, it means that the DCT coefficient value at the predetermined position of the second block is larger than the DCT coefficient value at the predetermined position of the first block. On the other hand, such a rule may be defined inversely to the above-described example (S240).

The watermark inserting unit 140 maps the watermarking rule string input from the watermarking rule generation unit 170 to each block of the input image divided by 8 * 16 units input from the first color image processing unit 120. A watermark is inserted according to a rule. Referring to the process, first, an 8 * 16 block input from the first color image processor 120 and a watermarking rule sequence input from the watermarking rule generator 170 are described. Are respectively matched, and the values of the DCT coefficients of predetermined positions of the first block and the second block of the two 8 * 8 blocks included in the 8 * 16 block are compared.

At this time, suppose that the value of the watermarking rule corresponding to these blocks is 1, and this means that the DCT coefficient value at a predetermined position of the first block is larger than the DCT coefficient value of the second block. If the DCT coefficient value of the position is larger than the DCT coefficient value of the second block, the next block is examined without inserting a separate watermark.

However, if the DCT coefficient value of the predetermined position of the first block is smaller than the DCT coefficient value of the second block, a constant value (hereinafter referred to as watermark strength) is added to the DCT coefficient value of the predetermined position of the first block, or The watermark is inserted by subtracting a constant watermark intensity to the DCT coefficient value of the predetermined position to satisfy the watermark rule (S250). At this time, the watermark intensity used at the time of embedding the watermark is appropriately selected in consideration of the general DCT coefficient values of the specific positions to be compared between the two blocks. That is, if the watermark is to be maintained even at a high compression rate for the image, a large value will be selected as the watermark intensity added to or subtracted from the DCT coefficient. Even if is selected, there is no problem in detecting the watermark.

If the above-described process is performed for all blocks of 8 * 16 units of the input image, the watermark intensity is added or subtracted so that DCT coefficient values of each block satisfying the watermarking rule are received from the watermark inserting unit 140. It is output to the color image processor 150. The second color image processing unit 150 performs an inverse DCT conversion on the input DCT coefficient values of each block in units of 8 * 8 blocks to form a watermark-embedded Ycbcr model, and then uses the Ycbcr model in the system of the present invention. A watermark-embedded image is output by converting the color format, for example, the RGB model, into an RGB model and outputting the same.

In the watermark embedding process, the selection of a predetermined position of the first block and the second block to which the DCT coefficients are compared is related to the compression ratio of the image. For example, the value of the quantization table used in the JPEG compression method is used in A small value, a high frequency region having a relatively small influence on the image, is composed of a large value, and a low frequency region having a small quantization coefficient in order to prevent the DCT coefficient value satisfying the watermarking rule from being greatly lost through quantization. Select the location of. Therefore, in the case of applications with high compression ratios, the DCT coefficients in the low frequency region are compared with the DCT coefficients in the relatively high frequency region in the case of low compression regions, and a predetermined position for selecting or decreasing the watermark intensity is selected. .

Thus, by inserting the watermark in consideration of the position and size of the watermark intensity according to the position of the quantization table, the watermark of the present invention is weak against external deformation (falsification), but is resistant to compression. That is, if the watermark intensity of a certain size is added to or subtracted from the DCT coefficient of the low frequency region where the quantization table value is relatively small to satisfy the predetermined watermarking rule, the DCT coefficient value of the predetermined position is changed when deformation is applied to the image from the outside. It is possible to detect the forgery of the image by not satisfying the predetermined watermarking rule anymore, and even though it is quantized in the compression process of the image, the value of the DCT coefficient at the predetermined position satisfies the watermarking rule, so it is robust to the compression of the image. Done.

Next, the case where the input image is a black and white image will be briefly described mainly on the color image and other parts.

When the input image is determined to be a black and white image by the image analyzer 110, the input image is output to the first black and white image processor 130. The first black and white image processing unit 130 divides the received input image into blocks of 8 * 16 units (S225), DCT-converts these blocks into units of 8 * 8, and outputs them to the watermark inserting unit 140 ( S235).

The watermark inserting unit 140 maps the watermarking rule string inputted through the same watermarking rule generation process as in the color image to each of 8 * 16 blocks, and in the same manner as in the color image, The watermark is inserted by adjusting or satisfying the watermarking rule by adding or subtracting the DCT coefficient value at a predetermined position of the first block or the second block included in the size block by a predetermined watermark intensity (S250). The watermark inserting unit 140 outputs the DCT coefficient of each block of the input image into which the watermark is inserted to the second black and white image processing unit 160 to perform inverse DCT conversion in units of 8 * 8 blocks, thereby inserting the watermark. The image is output (S265).

Hereinafter, with reference to FIG. 3 illustrating an example of a configuration of a system in which a method for determining whether a forgery of an input image of the present invention is executed and FIG. 4 which is a flowchart illustrating a process of determining whether or not forgery of an input image is performed, A system and method for determining whether an image is forged or altered using a watermark inserted by a watermark embedding process will be described.

The system of the present invention for determining the forgery of the watermark-embedded image is an image analyzer 310 for determining whether the input image is a color image or a black and white image, converts the input color image into a model that can be easily compressed, Color image processor 320 for dividing cosine transform after dividing into blocks of constant size, black and white image processor 330 for discrete cosine transforming (DCT) after dividing input black and white image into blocks of constant size, user information Watermarking rule generation unit 170 for generating a watermarking rule by using, and receives the DCT coefficient value of the input image from the color image processing unit or monochrome image processing unit, and receives the watermarking rule column from the watermarking rule generator And a forgery determination unit 340 for determining whether the input image is forged or not.

When an image is input to the system for determining forgery and alteration (S400) of the present invention, the image analyzer 310 determines whether the input image is a color image by checking whether the input image includes size information and color information of the image. It is determined whether the image is a black and white image (S405).

First, a description will be made of the forgery determination in the case where the input image is a color image.

When the input image is, for example, a color image represented by an RGB model, the input image is output to the color image processor 320 through the image analyzer 310 as in the watermark embedding process. The input image represented by the color model is converted into a Ycbcr model, which is a model commonly used for JPEG compression, through the same process as the watermark embedding process in the color image processing unit 320 (S410). After converting into a Ycbcr model, the converted image is divided into blocks of 8 * 16 size for the Y component (S420), and the image divided into 8 * 16 block units is further divided into 8 * 8 block units. DCT conversion (S430).

The DCT coefficients of each block generated by DCT conversion are output to the forgery determination unit 340, and the DCT coefficient values of the blocks satisfy the watermarking rule corresponding to each block output from the watermarking rule generator 370. The watermarking rule used for the forgery determination is generated through the same process as the generation of the watermarking rule when the watermark is inserted using the same user information as the watermark insertion. Same as the watermarking rule. That is, in the watermarking rule 1 generated by inputting the same user information, the DCT coefficient value at a predetermined position of the first block is larger than the DCT coefficient value at a predetermined position of the second block in the same manner when the watermark is inserted and when the forgery is determined. Or 0 indicates the opposite meaning (S440).

The forgery determination unit 340 corresponds to the 8 * 16 block input from the color image processor 320 and the watermarking rule input from the watermarking rule generator 370, and includes the 8 * 16 block in the 8 * 16 block. It is determined whether the watermarking rule is satisfied by comparing the DCT coefficient values of the first block and the predetermined position of the second block among the two 8 * 8 blocks (S450). That is, it is determined that the image of the 8 * 16 block that satisfies the rule does not have a forgery, and that the image of the block that does not satisfy the rule is forgery (S460).

For example, the watermarking rule value of the corresponding block is 1, and this watermarking rule value 1 means that the DCT coefficient value at a predetermined position of the first block is greater than the DCT coefficient value at the same position of the second block, and the first in the actual block. If the DCT coefficient value of the block is greater than the DCT coefficient value in the second block, the image of the 8 * 16 block including the first block and the second block is determined as not forged. However, if the DCT coefficient value of the second block is larger than the DCT coefficient value of the first block, the image of the 8 * 16 block including the first block and the second block is determined to be forged. Therefore, according to the present invention, not only the forgery of the entire image, but also the forged region of the entire image can be known.

On the other hand, when the input image is a black and white image, the determination of whether or not the image is forged, the input black and white image is output from the image analyzer 310 to the black and white image processing unit 330, the black and white image processing unit 330 The processing in the forgery determination unit 340 is different from that of the color image except that the block is divided into blocks of size 16 and DCT is converted into blocks of 8 * 8 blocks and output to the forgery determination unit 340. same.

In the fragile watermarking method of weakly inserting an information signal among a method of inserting a series of information signals into a digital image through the above-described configuration, it is strong to compress the image and to be broken to other image deformations. A watermarking method can be implemented. Using this method, there is an effect that the image compression is not influenced in authenticating whether the digital image is forged or not, and only the other image may be judged and authenticated.

The embodiments of the present invention described so far are illustrative, and all modifications and changes thereto are to be understood as belonging to the claims set out below.

Claims (14)

As a method of inserting a watermark that is not affected by video compression and sensitive only to forgery of the video, (a) dividing an input image into first blocks having a predetermined pixel size; (b) dividing the first blocks into two second blocks each having a predetermined pixel size, and discrete cosine transforming each of the second blocks; (c) inserting a watermark into the converted second blocks according to a watermarking rule; And and (d) inverse discrete cosine transforming the second blocks having the watermark embedded thereon to output an image having the watermark embedded therein. The method of claim 1, And the watermarking rule indicates a magnitude relationship between DCT coefficient values at predetermined same positions of the two second blocks included in the same first block. The method of claim 2, wherein step (c) Comparing the DCT coefficient values at predetermined positions of the second blocks, and if the watermark rule is not satisfied, a predetermined watermark intensity is added to or subtracted from the DCT coefficient values. The method of claim 3, wherein Wherein the first block is a block of 8 * 16 pixel size and the second block is a block of 8 * 8 pixel size. The method according to any one of claims 1 to 4, Before the step (a), further comprising the step of converting the input image into an image model that is easy to compress, After the step (d), further comprising inversely converting the image represented by the image model that is easily compressed into the form of the input image. The method of claim 5, And the input image is a color image in RGB format, and the easy-compression image model is a Ycbcr model. The method of claim 6, wherein step (a) And dividing the Y component of the Ycbcr model. A method for determining whether a forged image is forged using a watermark, (a) dividing an input image into first blocks having a predetermined pixel size; (b) dividing each of the first blocks into two second blocks having a predetermined pixel size, and performing discrete cosine transform (DCT) on each of the second blocks; And and (c) determining whether the image is forged or altered according to an input watermarking rule. The method of claim 8, And the watermarking rule indicates a magnitude relationship between DCT coefficient values at predetermined same positions of the two second blocks included in the same first block. The method of claim 9, wherein step (c) By comparing the DCT coefficient values of the predetermined positions of the second blocks, if the watermarking rule is not satisfied, it is determined that the image of the first block is forged. If the watermarking rule is satisfied, the image of the first block is Determining that it is not forged. The method of claim 10, Wherein the first block is a block of 8 * 16 pixel size and the second block is a block of 8 * 8 pixel size. The method according to any one of claims 8 to 11, Before the step (a), further comprising the step of converting the input image into an easy to compress image model. The method of claim 12, The input image is a color image in an RGB format, and the easy-compression image model is a Ycbcr model. The method of claim 13, wherein step (a) Dividing the Y component of the Ycbcr model.