KR20060081180A - Method and apparatus for embedding and detecting watermarks in jpeg compressed data - Google Patents

Abstract

A method of embedding and extracting watermarks in a quantization domain into JPEG compressed data is provided. Inserting a watermark selects some quantization values from each block of the plurality of quantization value blocks, converts the selected quantization values into quantization values corresponding to a predetermined goodness level, and inserts a watermark for the converted quantization values. The determination is performed by performing error control on the transformed quantization value and inserting a watermark into the transformed quantization value. Extracting the watermark selects a partial quantization value from each block of the plurality of quantization value blocks, converts the selected quantization value into a quantization value corresponding to a predetermined goodness level, and determines whether or not to extract the watermark with respect to the converted quantization value. And extracting the watermark from the transformed quantized value.

Watermark, quantization table, quantization value, quantization domain, goodness

Description

Method and apparatus for inserting and extracting watermarks into compressed data {METHOD AND APPARATUS FOR EMBEDDING AND DETECTING WATERMARKS IN JPEG COMPRESSED DATA}

1 is a flowchart schematically illustrating a watermark embedding and extraction method according to the present invention.

2 is a flowchart illustrating a watermark embedding method according to the present invention.

3 is a flowchart illustrating a watermark extraction method according to the present invention.

4 is a block diagram showing a watermark embedding apparatus according to the present invention.

5 is a block diagram showing a watermark extraction apparatus according to the present invention.

6A is an image in which a watermark is inserted by the watermark embedding method according to the present invention.

6B is an image in which modulation is applied to the image of 6a.

FIG. 7A is a graph showing cross correlation of the image of FIG. 6A. FIG.

FIG. 7B is a graph showing the correlation of the image of FIG. 6B.

8 is a graph showing the robustness to JPEG compression of the watermark embedding and extraction method of the present invention.

9 is a flowchart illustrating a general JPEG compression method.

10 shows an ISO standard quantization table.

* Description of the symbols for the main parts of the drawings *

400, 500: controllers 402, 502: Huffman decoder

404, 504: Quantization Value Converter 406: Error Regulator

408: watermark inserter 410: Huffman decoder

506: watermark extractor

The present invention relates to a method of embedding a watermark in JPEG compressed data, and more particularly, to a method of embedding a watermark in the quantization domain of JPEG compressed data.

In recent years, with the rapid development of computers and the Internet, digital data such as images, audio, and video are widely used and are transmitted among a large number of users. However, these data have a problem that the data is large in size, occupies a large storage space, and takes long time to transmit and receive. Various compression methods have been proposed to solve this problem, and the JPEG compression method is widely used for image data.

9 is a flowchart illustrating a general JPEG compression method. The JPEG compression method is largely divided into a discrete cosine transform (DCT) step 902, a quantization step 904, and an entropy encoding step 906.

In the DCT step 902, a discrete cosine transform is used to convert digital image data in the spatial domain into DCT coefficients in the spatial frequency domain. In this step 902 multiple DCT coefficient blocks are obtained.

Next, in quantization step 904, the DCT coefficients are quantized. Quantization is performed by a quantization table. The optimal quantization table may vary depending on the target image, and may be arbitrarily selected by a user according to a specific application. The ISO standard defines a standard quantization table, which is shown in FIG. By quantization step 904, the DCT coefficient blocks are transformed into blocks of quantization values. A detailed process of the quantization step 904 will be described later.

Finally, in entropy encoding step 906, the quantized value block is Huffman encoded and converted into a JPEG compressed code.

Restoration of JPEG compressed data is performed through the reverse process of the JPEG compression process.

Hereinafter, the quantization step will be described in detail.

Each quantum value of the quantization table is converted into a step size by a quality factor, which is expressed by the following equation (1).

In the formula (1), QT _{u, v} is a quantum value of the quantization table u rows v columns, s _{u, v} is a step size corresponding to QT _{u, v} , and QF is good. As can be seen from Equation (1), as the goodness value increases, the step size becomes smaller. As described later, the smaller the step size, the smaller the compression ratio and the better the image quality.

Next, the DCT coefficients are quantized using the obtained step size. One step size table obtained from one quantization table is applied to one block of DCT coefficients 702 to obtain a quantized value of one block. At this time, one step size is applied to the DCT coefficient at the corresponding position in the DCT coefficient block, and this quantization is expressed by the following equation.

Where DCT _{u, v} is the DCT coefficient at the position (i.e., row u column) corresponding to the step size (s _{u, v} ) in the DCT coefficient block, and Q _{u, v} is the quantization value from DCT _{u, v} .

As in Equation (2), the smaller the step size, the more finely quantized the DCT coefficients, thereby improving the quality of the image. However, because the quantization is fine, the compression rate is reduced. Through such quantization, the DCT coefficient block is converted into a block of quantization values.

By the JPEG compression method as described above, the amount of data in the image is greatly reduced, and storage, transmission, and editing of digital images are very easy. However, the JPEG compression method does not provide any means for guaranteeing the integrity of the digital image, and has the disadvantage of being very vulnerable to forgery and alteration.

In the past, a method of embedding a fragile watermark in digital image data is known to solve this problem. The fragile watermark is damaged when the data is forged, so it can be determined whether the data is forged or not by judging the damage of the watermark. However, since the soft watermark is also damaged by the compression process, there is a problem that the watermark is damaged even without forgery when JPEG compression is used.

To solve this problem, a method of embedding a semi-fragile watermark into digital data is known. The semi-soft watermark is not damaged by compression but is damaged by forgery, so it can be used to determine whether it is forgery.

A method of inserting a semi-soft watermark in relation to a digital image compression method includes a method of inserting a watermark in the spatial domain before image compression, a method of inserting a watermark in the frequency domain during image compression, and compression after compression of the image. It is divided into a method of embedding a watermark in a domain.

The method of embedding a watermark in the spatial domain is described in GC Langelaar et al. "Copy Protection for Multimedia Data based on Labeling techiques", and watermark in the frequency domain. The method of inserting the method is disclosed in the paper "Secure spread spectrum watermarking for multimedia" by Cox IJ et al.

A method of inserting a watermark in relation to the JPEG compression method, and a method of inserting a watermark after compressing image data with high quality is described in a paper by CY Lin et al. Semi-Fragile Water Authentication for Authentication JPEG Visual Content "and Kueno Maeno et al." New Semi-Fragile Image Authentication Watermarking Techniques Using Random Bias and Non-Uniform Quantization. " While this method has robustness to JPEG compression, it is impossible to extract watermark by ISO standard quantization table because it uses user's arbitrary quantization table with high goodness, and the amount of calculation by DCT encoding and decoding There are many problems.

As another watermarking method related to JPEG compression method, a method of embedding watermark information in the low frequency band of the frequency domain is described in Eugene T. Lin et al., "Detection of image alterations using semi-soft watermark. using Semi-Fragile watermarking). This method has the advantage of reducing the amount of computation, but has a problem that the degradation of the image is large because the watermark information is inserted into the DCT coefficients.

As another watermark embedding method, a method of embedding a watermark only in arbitrary DCT coefficients using the linearity of DCT is described in the paper "Semi-fragile warming using the linearity of discrete cosine transform" by Byun Young-bae et al. This method reduces computation and reduces image degradation. However, this method has a problem that the detection error of the watermark is maintained at 0% only up to the goodness value 85, so that the authentication range is limited to 85.

Accordingly, the present invention provides a method for embedding a semi-soft watermark that is robust to JPEG compression.

It is also an object of the present invention to provide a watermark embedding method which can be easily extracted even by an ISO quantization table and has a small amount of calculation.

Still another object of the present invention is to provide a watermark embedding method having a wide range of authentication while reducing deterioration of an image.

In order to achieve the above object, the watermark embedding method of the present invention selects some quantization values from each of a plurality of quantization value blocks and converts the selected quantization values into quantization values corresponding to a predetermined goodness level, wherein the transformed quantization Determining whether to insert a watermark for each block based on values, adjusting an error with respect to the transformed quantization values of a block into which a watermark is to be inserted, and inserting a watermark into the error-adjusted quantization values It includes a step. This makes it possible to embed a watermark that is robust to JPEG compression but reduces the amount of calculation and suppresses deterioration of image quality.

Preferably, the transforming step selects quantization values corresponding to quantizers having the same value in the quantization table.

Also, preferably, the determining may determine whether to insert a watermark based on the magnitude relationship between the converted quantization values. According to the present invention, the original image is not needed in the process of extracting the watermark.

In order to achieve the above object, the watermark extraction method of the present invention selects some quantization values from each of a plurality of quantization value blocks and converts the selected quantization values into quantization values corresponding to a predetermined goodness, the transformed quantization Determining whether to extract a watermark for each block based on the values, and extracting a watermark from the transformed quantization values of the block from which the watermark is to be extracted.

Preferably, the transforming step selects quantization values corresponding to quantizers having the same value in the quantization table.

Also preferably, the determining may determine whether to insert a watermark based on the magnitude relationship between the converted quantization values.

In addition, in order to achieve the above object, the watermark embedding apparatus of the present invention performs Huffman decoding on a JPEG compressed image to obtain a plurality of quantized value blocks, wherein each of the plurality of quantized value blocks is included in the quantization table. Quantization value conversion means for selecting quantization values corresponding to quantizations having the same value and converting the selected quantization values into quantization values corresponding to a predetermined goodness degree, for each block based on the transformed quantization values Determining means for determining whether to insert a watermark, error adjusting means for adjusting an error with respect to the quantization values of a block into which a watermark is to be inserted, and watermark embedding means for inserting a watermark in the error-adjusted quantization values. do.

The watermark extraction apparatus of the present invention performs Huffman decoding on the JPEG compressed image to obtain a plurality of quantized value blocks, and corresponds to quantum having the same value among the quantization tables in each of the plurality of quantized value blocks. A quantization value converting means for selecting quantization values and converting the selected quantization values into quantization values corresponding to a predetermined goodness degree, determining whether to extract watermarks for each block based on the transformed quantization values Means for extracting a watermark from the transformed quantization values of the block from which the watermark is to be extracted.

1 is a flowchart schematically illustrating a watermark embedding and extraction method according to the present invention. Insertion and extraction of the watermark of the present invention is performed in the quantization domain. In the case of performing the watermark embedding method of the present invention during the JPEG compression process, the watermark is inserted into the quantization value block obtained through the quantization step 104 (step 114), and the watermarked data is inserted. Entropy encoding (step 106) is performed to obtain JPEG compressed data with the watermark embedded therein.

When performing the watermark embedding method of the present invention on JPEG compressed data, entropy decoding is performed on the JPEG compressed data (step 108) to obtain a quantized value block. The watermark is then inserted into the quantization value block (step 114) and entropy encoded again (step 106) to obtain JPEG compressed data with the watermark embedded.

In the case of performing the watermark extraction method of the present invention on JPEG-compressed data embedded with watermarks, entropy decoding (step 108) is performed to obtain a watermark-embedded quantized block of values, and to generate watermarks. Extract (step 114). Subsequently, dequantization (step 110) and Inverse Decrete Cosine Transform (step 112) are performed on the quantized value block to obtain reconstructed image data.

2, a specific embodiment of the watermark embedding method according to the present invention will be described.

The watermark embedding method of this embodiment begins at step 202 of Huffman decoding JEPG compressed image data. By Huffman decoding JPEG compressed image data, a plurality of quantization value tables are obtained.

Next, in the obtained quantization value table, some quantization values are selected and converted into quantization values corresponding to a predetermined goodness level (step 204). This step 204 is accomplished using the equations (1) and (2) above. In this case, the quantization value is selected based on the quantization table used in JPEG compression, and it is preferable that the quantization value corresponding to the quantization having the same value among the quantization tables is selected.

As shown in Equation (1), since the same step size is derived from both of the same values, if the goodness is changed to the same, each quantization value is changed by the same ratio by Equation (2). It can be seen. Therefore, by converting the quantization values corresponding to both having the same value, the quantization values can be changed at the same ratio.

In the present embodiment, JPEG compressed data compressed using the ISO standard quantization table will be described, but the present invention is not limited thereto. As shown in FIG. 10, in the ISO standard quantization table, the quantum of three rows and one column (QT _3,1 ), the quantum of two rows and three columns (QT _2,3 ) and the quantum of four rows and one column (QT _4,1 ) are the same. Has a value (14). Therefore, the quantization values Q ₃ , ₁ , Q ₂ , ₃ and Q ₄ , ₁ corresponding to both of these are converted into predetermined goodness values. QT _1,5 , QT _3,4 and QT _6,1 also have the same values as 24, but in order to prevent degradation of image quality and to maintain the robustness against compression, the above Q _3,1 , Q _2,3 , Q ₄ It is preferable to use _{, 1} . Preferably, the goodness value used to transform the quantization value is 70.

Hereinafter, the case where the transformed quantization value is Q _3,1 = 0, Q _4,1 = 0 and Q _2,3 = 1 will be described as an example. However, one of ordinary skill in the art will readily recognize that the present invention is not limited thereto.

Next, in step 206, it is determined whether to insert a watermark into these values. The determination of whether to embed the watermark in step 206 is preferably made based on the mutual magnitude relationship between the transformed quantization values. In other words, the watermark is inserted only when the difference between the quantization values is within a predetermined value, otherwise the watermark is not inserted. By doing so, the selected quantization values of the watermarked block are within a predetermined range. Therefore, when the watermark is extracted, it is possible to extract the watermark by determining whether to insert the watermark even if there is no original image. In this embodiment, the watermark is applied only when the converted quantization values Q _3,1 , Q _2,3 , Q _4,1 satisfy all of the following equations (3), (4), and (5): Insert it.

B is a predetermined constant in the above formulas.

If the equation is satisfied, the process proceeds to step 210, where the watermark embedding process continues. If none of the equations are satisfied, it is determined that no watermark is inserted into the block, and the process proceeds to step 208. It is determined whether the next quantization value block exists. If the next quantization value block exists, step 204 is again performed on the block, and if the current quantization value block is the last block, the watermark insertion is terminated.

,

및

이므로, 워터마크를 삽입하도록 결정하며, 단계 (210) 로 진행한다.On the other hand, by changing the value of b, the number of quantization value blocks that satisfy the equations (3), (4) and (5) is changed to adjust the watermark embedding amount. In other words, when the value of b is increased, many quantization block of blocks satisfying the equations (3), (4) and (5) are found, and the watermark is inserted in many blocks. The number of quantized value blocks satisfying (5) becomes small, so that the amount of watermark embedding is reduced. In the present embodiment, the case where the b value is 2 is described as an example, but the value of b can be selected according to the application. In the example above

,

And

Therefore, it is decided to insert a watermark, and the flow proceeds to step 210.

In step 210, error control is performed to prevent quantization error caused by the quantization step after insertion of the watermark. As shown in Equation (2), since a round operation is performed in the quantization process, the quantization value conversion step 204 in the watermark embedding process and the quantization value conversion step in the watermark extraction process (to be described later) A quantization error may occur due to rounding. If a quantization error occurs, there is an error in determining whether to insert a watermark when extracting the watermark, so that the watermark is extracted from the quantization block containing no watermark or the watermark from the quantization block containing the watermark. You may get an error that doesn't extract. Therefore, error adjustment is performed in step 210 to prevent such errors.

Error adjustment of this embodiment is performed by following Formula (6)-(9).

인 경우에는 다음 식 (6) 및 (7) 에 따라 Q_4,1 및 Q_2,3 값이 변화된다.first,

In the case of, Q _{4, 1} and Q _{2, 3} values change according to the following equations (6) and (7).

인 경우에는, 다음 식 (8) 및 (9) 에 따라 Q_4,1 및 Q_2,3 값이 변화된다.Meanwhile,

In the case of, Q _{4, 1} and Q _{2, 3} values change according to the following expressions (8) and (9).

이므로, 식 (8) 및 (9) 에 의하여,

,

이 된다.In the formulas (6) to (9), t is a predetermined constant, which can be selected by the user according to the application. In the present embodiment, the case where t = 1 is described as an example, but the present invention is not limited thereto. In the example above,

Therefore, by the formulas (8) and (9),

,

Becomes

Next, the flow advances to step 212 to insert a watermark. Preferably, one bit of the watermark is inserted in each quantization value block, and more preferably one bit of the watermark is inserted only in one of the quantization values converted in step 204 above. In this embodiment, one bit of the watermark is inserted into _Q3,1 .

If the bit of the watermark to be inserted is 1, the following equation

By the way, a watermark is inserted at Q _3,1 .

On the other hand, if the bit of the watermark is 0,

The watermark is inserted by.

In Equations (10) and (11), Q ' _3,1 is a quantization value Q _3,1 with a watermark embedded therein, and Q' _2,3 and Q ' _4,1 are error adjustments in step 210. Quantized values.

가 되어 워터마크 삽입이 수행되고, 워터마크 삽입 후의 양자화 값들은 Q_3,1 = 2, Q_4,1 = 1 그리고 Q_2,3 = 0 이 된다. 이와 같이, 하나의 양자화 값 블록에 워터마크의 한 비트가 삽입될 수 있다.In the above example, when the watermark value to be inserted is 1,

The watermark embedding is performed, and the quantization values after the watermark embedding are Q _3,1 = 2, Q _4,1 = 1 and Q _2,3 = 0. As such, one bit of the watermark may be inserted into one quantization value block.

Proceeding to next step 214, it is determined whether there is a next quantization value block. If there are more quantization value blocks, step 204 or less is performed again for the next block, and if there are no more quantization value blocks, the flow proceeds to step 216.

In step 216, Huffman encoding of the quantized value blocks with or without the watermark is again performed to generate JPEG compressed data with the watermark embedded therein, and the watermark embedding process is terminated. Therefore, the JPEG compressed data having the watermark inserted according to the method of the present invention may be restored by a general JPEG reconstruction process.

The watermark embedding method according to the present invention performs watermark embedding only on a part of quantization values in the quantization domain, so that it is robust to JPEG compression, can reduce computation amount, and minimize image degradation.

Next, an embodiment of a watermark extraction method according to the present invention will be described with reference to FIG. The watermark extraction method is generally the reverse process of the watermark embedding method, and the watermark extraction method according to the present invention can extract the watermark only with the image into which the watermark is inserted without the original image.

The watermark extraction method of this embodiment begins by Huffman decoding JPEG compressed data in step 302. By Huffman decoding, blocks of quantization values are obtained.

Next, in the obtained quantization value block, some quantization values are selected and converted into quantization values corresponding to a certain degree of goodness (step 304). This step is accomplished using the above equations (1) and (2). In this case, the selected quantization value is a quantization value that is the same as the quantization value selected in the performance of the watermark embedding method, and it is preferable that a quantization value corresponding to both having the same value in the quantization table is selected. In the present embodiment, a method of using the ISO standard quantization table will be described, but the present invention is not limited thereto. The quantization value to be selected is a quantization corresponding to three rows and one column's quantum (QT _3,1 ), two rows and three columns' quantum (QT _2,3 ) and four rows and one column's quantum (QT _4,1 ) in the ISO standard quantization table. The values (Q _3,1 , Q _2,3 , Q _4,1 ). In addition, the goodness used in the conversion of the quantization value is also preferably the same as the goodness used in the watermark embedding process, and is 70 in this embodiment.

If the quantization value selected from the watermark-embedded image data is Q _3,1 = 2, Q _4,1 = 1 and Q _2,3 = 0, using the example used for the watermark embedding method described above. This will be described as an example.

The process then proceeds to step 306 to determine whether to extract the watermark from the selected quantization value. This is determined using the same criteria as was used when embedding the watermark, which is also preferably based on the mutual magnitude relationship between the transformed quantization values. That is, when the difference between the converted quantization values is within a predetermined range, it is determined that the watermark is embedded in the quantization values, and watermark extraction is determined. Otherwise, the watermark is not embedded in the quantization values. It is judged that the watermark is not extracted.

In this embodiment, it is determined to extract the watermark when all of the above expressions (3) to (5) are satisfied, and not to extract the watermark if any of the expressions are not satisfied. However, in the watermark extraction, the value b should be the same as the value used at the time of embedding the watermark, which is 2 in this embodiment. In this way, by using the same criteria as the one used at the time of embedding the watermark, the watermark can be extracted only with the image into which the watermark is inserted without requiring the original image.

If watermark extraction is determined, the process proceeds to step 310 and the watermark extraction process continues. If it is determined that no watermark is extracted, the process proceeds to step 308 to determine whether there is a next quantization value block. If there is a next quantization value block, the flow returns to step 304 to repeat the watermark extraction process for that block, and to terminate the watermark extraction process if the current quantization value block is the last block.

,

및

이므로, 워터마크의 추출이 결정되고 단계 (310) 로 진행한다. In the above example,

,

And

, The extraction of the watermark is determined and proceeds to step 310.

As described above, the error in the quantization value conversion process is suppressed by the error adjusting step (step 210, FIG. 2) at the time of embedding the watermark, so that accurate watermark extraction can be determined.

와 Q_3,1 값의 관계를 판단하여 이루어지는데, 다음 식 Next, a watermark is extracted from the quantized value converted in step 310. Preferably, one bit of the watermark is extracted from each quantization value block, and the quantization value from which the watermark is to be extracted is determined corresponding to the watermark embedding method. In this embodiment, a watermark is extracted from Q _3,1 , Q _2,3 and Q _4,1 . In this embodiment, the extraction of the watermark is

And Q _3,1 to determine the relationship between

If this holds true, watermark 1 is extracted,

If this holds, the watermark 0 is extracted.

이고 Q_3,1 = 2 이므로,

가 성립하고 워터마크 비트 1 이 추출된다. 이는, 워터마크 삽입 과정에서 삽입된 워터마크 비트 1 과 동일한 것으로, 본 발명에 의한 워터마크 추출 방법에 의하여 삽입된 워터마크가 추출된 것을 확인할 수 있다.As a result, one watermark bit is extracted from each quantization block. In the above example,

And Q _3,1 = 2,

Is established and the watermark bit 1 is extracted. This is the same as the watermark bit 1 inserted in the watermark embedding process, and it can be confirmed that the inserted watermark is extracted by the watermark extraction method according to the present invention.

Next, the process proceeds to step 312 to determine whether there is a next quantization block, and if there is a next quantization block, repeats the steps of step 304 or less for that block, and if no more blocks exist, extracts watermarks. Quit.

The watermark extraction method of the present invention can extract the watermark from the image into which the watermark is inserted without the original image, and the authentication range of the watermark can be increased as described below.

The present invention also provides a watermark embedding apparatus and a watermark extracting apparatus for executing the above watermark embedding and extraction methods.

4 is a block diagram showing a watermark embedding apparatus according to the present invention. The watermark embedding apparatus of FIG. 4 is combined with the Huffman decoder 402, the quantization value converter 404, the error adjuster 406, the watermark inserter 408, the Huffman encoder 410 and these to control each operation. The controller 400.

The watermark embedding apparatus receives, as an input of the Huffman decoder 402, JPEG compressed data (JPEG _O ) having no watermark embedded therein. Huffman decoder 402 performs Huffman decoding on JPEG compressed data (JPEG _O ) to output a plurality of quantized value (Q) blocks.

The quantization value converter 404 selects quantization values (Q) corresponding to quantization values of some of the plurality of quantization value (Q) blocks, preferably quantization tables having the same value in the quantization table, and selects these values by a predetermined value. Convert to a quantization value Q 'corresponding to goodness of mind. In this embodiment, the quantization table uses an ISO standard quantization table, and selects a quantization value corresponding to a quantum having a quantum value of 14. Further, the selected quantization value is converted into a quantization value corresponding to goodness of 70. However, those skilled in the art will appreciate that the quantization table, quantum and goodness are not limited to these values.

The controller 400 determines whether to insert a watermark based on the converted quantization value. Such watermark embedding determination is preferably made based on the magnitudes of the converted quantization values. In the present embodiment, the watermark embedding is determined when all of the above equations (3) to (5) are satisfied. The controller 400 causes the error adjuster 406 to perform error adjustment if it determines to insert the watermark, and if the watermark insertion is not determined, the quantization value converter 404 converts the quantization value for the next block of quantization values. To do this. If there are no more quantization block, the process does not proceed.

The error adjuster 406 then performs error adjustment on the transformed quantization value Q 'to obtain an error adjusted quantization value Q ". Such error adjustment is preferably performed according to the above formulas (6) and (7) or (8) and (9).

The watermark inserter 408 inserts a watermark for the error-adjusted quantization value under the control of the controller 400. The watermark inserter 408 preferably inserts one bit of watermark into each quantization value block, and watermark embedding is performed by the above equation (10) or (11).

The controller 400 determines whether the next quantization value block exists after the watermark embedding process is finished for one quantization value block, and if the next quantization value block exists, the quantization value converter 404 for the next quantization value block. Perform quantization value conversion, and if there are no more blocks, Huffman encoder 410 performs Huffman encoding to obtain JPEG image data JPEG _W with watermark embedded therein.

Next, a watermark extractor according to the present invention will be described. 5 is a block diagram illustrating a watermark extractor according to the present invention. The watermark extractor of the present invention includes a Huffman decoder 502, a quantization value converter 504, a watermark extractor 506, and a controller 500 coupled thereto to control each operation.

The watermark extractor receives JPEG image data (JPEG _W ) embedded with a watermark from the Huffman decoder 502. Huffman decoder 502 performs Huffman decoding on JPEG image data (JPEG _W ) to obtain a plurality of quantization value (Q _W ) blocks. Next, the quantization value converter 504 selects the quantization values of some of the quantization value (Q _W ) blocks and converts them into quantization values (Q _W ') corresponding to a predetermined goodness level. The figure depends on the quantization value and the goodness level used in embedding the watermark.

The controller 500 determines whether to extract the watermark. The determination of whether or not to extract the watermark is preferably performed using the same criteria as the watermark embedding determination criterion. In the present embodiment, the watermark extraction is determined when all of the above expressions (3) to (5) are satisfied. If the watermark extraction is determined, the controller 500 causes the watermark extractor 506 to extract the watermark from the transformed quantization value Q _W ', otherwise the quantization value converter 504 causes the next quantization value block ( Q _W ) to perform quantization value conversion. If there are no more quantization value blocks, the controller 500 does not go further.

The watermark extractor 506 extracts the watermark W from the transformed quantization value Q _W ′ under the control of the controller 500. The watermark is extracted according to the watermark embedding method. In the present embodiment, the watermark is extracted according to the above formula (12) or (13).

The controller 500 causes the quantization value converter 504 to perform quantization value conversion on the quantization value block if there is a next quantization value block after the watermark extraction process ends for one quantization value block. If there are no more quantization block, then the controller 500 does not proceed any further watermark extraction process.

In order to confirm the performance of the watermark embedding / extraction method according to the present invention, a watermark formed by a random random number was inserted into the test image to obtain the image of FIG. 6A. The b and t values used here were 2 and 1, respectively. Further, the image of FIG. 6A with the watermark embedded therein is modulated to obtain the image of FIG. 6B. In the image of FIG. 6B, a portion where the image is modulated is indicated by a rectangle.

As described above, the watermark is extracted and inserted into the watermark-embedded image (FIG. 6A) and the modulated image (FIG. 6B) by the watermark embedding method of the present invention using the watermark extraction method of the present invention. The cross correlation of the extracted watermark and the extracted watermark was calculated by the following equation.

Here, R _xy (u, v) is a correlation, D (x, y) is an extracted watermark, and E (x, y) is an inserted watermark.

7A and 7B show cross-correlation for an unmodulated image and a modulated image, respectively.

The correlation between the watermark extracted in FIG. 7A and the inserted watermark is 1, which shows that the extraction of the watermark was performed correctly. In contrast, the cross correlation in FIG. 7B indicates that the inserted watermark and the extracted watermark are different as 0.8. Thus, when modulation is applied to the image, the embedded watermark was not extracted correctly.

In addition, while FIG. 7A shows that the peak position of cross-correlation is zero, the peak position in FIG. 7B is shifted to the left rather than zero. Thus, from this, it is possible to know where the modulation occurred in the image.

8 is a graph showing the robustness to JPEG compression of the watermark embedding and extraction method of the present invention.

Referring to the graph of FIG. 8, the watermark extraction rate was maintained at 100%, that is, the watermark detection error was 0% until the goodness level was 77. Therefore, the authentication range of the watermark embedding / extraction method according to the present invention is 99-77, which is larger than that of the prior art 99-85.

Finally, PSNR (Peak Signal-to-Noise Ratio) was calculated between the original image and the watermark-embedded image in order to examine the degree of deterioration of the image according to the watermark embedding method of the present invention. The results are shown in Table 1 below.

Goodness PSNR (dB) Lena Image Babara Image Baboon image Goldhill image 70 42.83 42.69 42.93 42.61 75 42.00 41.86 42.19 41.97 80 42.29 41.85 42.30 41.46 85 42.64 42.27 42.08 42.21 90 42.68 42.16 42.18 42.10 95 42.67 42.16 42.28 42.23 98 42.50 41.96 42.34 41.94

As shown in Table 1, PSNR values of 40 or more appeared for all images, and it can be seen that the deterioration of image quality was very small.

According to the present invention, it is possible to insert and extract a watermark that is robust to JPEG compression, which is not deformed by JPEG compression but deformed by forgery.

In addition, since the present invention inserts and extracts a watermark using an ISO standard quantization table, it can be used universally.

In addition, since the present invention inserts a watermark only on a part of quantization values in the quantization domain, it is possible to reduce the amount of computation and to reduce image degradation.

Further, according to the present invention, the watermark may be extracted from only the watermark embedded image without the need for the original image, including the step of determining whether to insert the watermark.

In addition, according to the present invention, the amount of embedding the watermark can be adjusted by adjusting the b value.

Further, according to the present invention, the authentication range is improved compared to the conventional watermark embedding / extraction method, so that the watermark can be accurately inserted and extracted even with low goodness.

Claims (23)

A method for embedding a watermark in a quantization domain for a JPEG compressed image, Selecting some quantization values in each of the plurality of quantization value blocks and converting the selected quantization values into quantization values corresponding to a predetermined goodness of fit; Determining whether to insert a watermark for each block based on the converted quantization values; Adjusting an error with respect to the transformed quantization values of a block into which a watermark is to be inserted; And And embedding a watermark in the error-adjusted quantization values. The method of claim 1, And the converting step selects quantization values corresponding to quantizers having the same value in the quantization table. The method of claim 1, The determining may include determining whether to embed a watermark based on the magnitude relationship between the converted quantization values. The method of claim 1, The embedding of the watermark comprises inserting one bit of a watermark into each block. The method of claim 1, The converting step includes performing Huffman decoding on the JPEG compressed image to obtain the plurality of quantized value blocks. The method of claim 1, Wherein the converting step selects Q ₁ , Q _2, and Q ₃ , which are quantization values corresponding to quantum having a value of 14, from an ISO standard quantization table. The method of claim 6, The determining may include: for b, wherein the selected quantization values are a predetermined constant,

,

And

A watermark embedding method that determines to insert a watermark when all are satisfied. The method of claim 6, Adjusting the error, In the case of Q _3> Q _2, Q _3, and subtracting the threshold value from the threshold value and adds the Q _2, And if Q ₂ > Q ₃ , adding the threshold to Q ₃ and subtracting the threshold at Q ₂ . The method of claim 6, Inserting the watermark, If the watermark to be inserted for the block to which the quantum values belong is 1,

Replaces Q ₁ , If the watermark to insert is 0,

A method for embedding watermarks, replacing Q ₁ . The method according to any one of claims 6 to 9, Q ₁ corresponds to quantum in three rows and one column of the ISO standard quantization table, Q ₂ corresponds to quantum in two rows and three columns of the ISO standard quantization table, Q ₃ corresponds to both 4 rows and 1 columns of the ISO standard quantization table. In the method for extracting a watermark in the quantization domain for a JPEG compressed image, Selecting some quantization values in each of the plurality of quantization value blocks and converting the selected quantization values into quantization values corresponding to a predetermined goodness of fit; Determining whether to extract a watermark for each block based on the converted quantization values; And Extracting a watermark from the transformed quantization values of a block to extract a watermark. The method of claim 11, Wherein the converting step selects quantization values corresponding to quantizers having the same value in the quantization table. The method of claim 11, The determining may include determining whether to extract a watermark based on the magnitude relationship between the converted quantization values. The method of claim 11, The extracting of the watermark comprises extracting one bit of the watermark from each block. The method of claim 11, The converting step includes performing Huffman decoding on the JPEG compressed image to obtain the plurality of quantized value blocks. The method of claim 11, Wherein the converting step selects Q ₁ , Q ₂ and Q ₃ , which are quantization values corresponding to quantum having a value of 14, from an ISO standard quantization table. The method of claim 16, The determining may include: for b, wherein the selected quantization values are a predetermined constant,

,

And

The method of extracting watermark, which determines to extract a watermark when all are satisfied. The method of claim 16, Extracting the watermark, For the quantum values,

<Q ₁ , extract 1 as a watermark,

> If Q _1, 0 is extracted as a watermark. The method according to any one of claims 16 to 18, Q ₁ corresponds to quantum in three rows and one column of the ISO standard quantization table, Q ₂ corresponds to quantum in two rows and three columns of the ISO standard quantization table, Q ₃ corresponds to both 4 rows and 1 columns of the ISO standard quantization table. A computer-readable recording medium having recorded thereon a program for performing a method of embedding a watermark in a quantization domain for a JPEG compressed image, wherein the computer, when the program is executed on a computer, Selecting some quantization values in each of the plurality of quantization value blocks and converting the selected quantization values into quantization values corresponding to a predetermined goodness of fit; Determining whether to insert a watermark for each block based on the converted quantization values; Adjusting an error with respect to the transformed quantization values of a block into which a watermark is to be inserted; And A computer-readable recording medium having recorded thereon a program for performing a step of inserting a watermark into the error-adjusted quantization values. A computer-readable recording medium having recorded thereon a program for performing a method of extracting a watermark from a quantization domain for a JPEG compressed image, wherein the computer, when the program is performed on a computer, Selecting some quantization values in each of the plurality of quantization value blocks and converting the selected quantization values into quantization values corresponding to a predetermined goodness of fit; Determining whether to extract a watermark for each block based on the converted quantization values; And A computer-readable recording medium having recorded thereon a program for performing a step of extracting a watermark from the transformed quantization values of a block from which a watermark is to be extracted. An apparatus for embedding a watermark in a quantization domain for a JPEG compressed image, Huffman decoding means for performing Huffman decoding on the JPEG compressed image to obtain a plurality of quantized value blocks; Quantization value conversion means for selecting quantization values corresponding to quantizations having the same value in a quantization table in each of the plurality of quantization value blocks and converting the selected quantization values into quantization values corresponding to a predetermined goodness degree; Determining means for determining whether to insert a watermark for each block based on the converted quantization values; Error adjusting means for adjusting an error with respect to said quantization values of a block into which a watermark is to be inserted; And And watermark embedding means for embedding a watermark in the error-adjusted quantization values. An apparatus for extracting a watermark in a quantization domain for a JPEG compressed image, Huffman decoding means for performing Huffman decoding on the JPEG compressed image to obtain a plurality of quantized value blocks; Quantization value conversion means for selecting quantization values corresponding to quantizations having the same value in a quantization table in each of the plurality of quantization value blocks and converting the selected quantization values into quantization values corresponding to a predetermined goodness degree; Determining means for determining whether to extract watermark for each block based on the converted quantization values; And And extracting a watermark from the transformed quantization values of a block to extract a watermark.

Images