JP2006186509A - Watermark system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce a reading error in a watermark. <P>SOLUTION: An error map formation 14 embeds test information generated in random in embedding object information where additional information is embedded, forms a watermark image in an embedding image in which additional information is embedded, decodes data from the watermark image, and compares with the original test information, thereby detecting a bit position in which reading errors are generated. This inspection is repeated to the many test information, and asks for the position in which the reading errors statistically tend to generate. An encoder 12 avoids the bit position in which the errors tend to generate, rearranges the additional information, embeds the bit array obtained as this result in the embedding object information and forms the watermark image. A decoder 34 reads a bit string from the watermark image. In the bit string, the additional information is recovered by ignoring the bit of the position which errors tend to generate. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a technique for reducing errors in reading information from a digital watermark image in a digital watermark technique for embedding information in an image.

  Conventionally, as a technique for printing and recording digital data, a barcode and an evolved two-dimensional barcode are widely known and used. These codes clearly recognize the printed part where the data is recorded and do not look good. In some cases, it is possible to tamper with a portion where digital data is printed. For these technologies, digital watermark technology that adds a format that cannot be recognized by human eyes at first glance has been researched and developed. Many of these techniques have the feature that the target image in which the digital data is embedded can be made into an arbitrary photograph or picture according to the purpose and preference of the user. However, when embedding digital data, the embedding printing of the original target image and the digital data interferes with each other, such as the density and color of the image of the target image, and a sudden change in density, and an error occurs in the recording and reproduction process of the digital data. There is a case.

For example, the present applicant has proposed a digital data embedding method in Patent Document 1. In this method, different pattern images are assigned to the codes “0” and “1” of the digital data to be embedded, and the embedded images are arranged by arranging the pattern images according to the code sequence and the arrangement rule of the pattern images. The data is embedded by superimposing the embedded image on the target image data. Each pattern image is
(A) The result of adding the corresponding pixels of both pattern images is the same predetermined value (for example, “0”).
(B) The result of adding all the pixels in each pattern image is the same predetermined value (for example, “0”).
(C) Each pattern image has a discontinuous line (referred to as an edge) of pixel values passing through the center of the image and having different directions. The direction of the edge can be, for example, a direction along a vertical line and a horizontal line,
(D) The absolute value of the pixel value of each pattern image is the largest at the center and decreases as the distance from the center increases.
It has the characteristics.

  FIG. 10 shows an example of a pattern image having such features and having the edge directions horizontal and vertical. In FIG. 10, for example, the pattern (A) indicates the code “0”, and the pattern (B) indicates the code “1”.

  In this way, by expressing the code by a pattern having an area, the resistance to each image quality deterioration factor is acquired, while the density of the image is prevented from changing greatly after composition (depending on the characteristics of (a) and (b)). ), The detection of the pattern image is easy and the decoding process is simplified (according to the characteristics of (b) and (c)), and the occurrence of an edge between the patterns is prevented (according to the characteristics of (d)). Its presence becomes visually inconspicuous. This edge is used for estimation of a pattern period at the time of decoding.

  When decoding an image obtained by synthesizing this pattern image, a sum of pixel values after synthesis corresponding to each quadrant divided by edges in the pattern image (R1 to R4 for the first to fourth quadrants) is calculated. For example, when R1> R2 and R1> R4 and R3> R2 and R3> R4, it is determined that the code is “1”. That is, decoding can be performed by comparing the sum of pixel value groups for each quadrant.

  In the case of a data embedding method in which an edge is formed in a pattern image of each code as in Patent Document 1 and this edge is used for decoding, the digital data of the printed digital data is overlapped with a portion where the density change of the image to be embedded overlaps abruptly. Information has a property of easily causing a reading error (reproduction error). For example, when using a pattern image as shown in FIG. 10 in which the edge direction is horizontal and vertical, the pattern image has a feature along the diagonal direction. In some cases, the partially superimposed pattern image is difficult to decode.

  In this way, when digital data is embedded in a target image using digital watermark technology, the target image and digital data may interfere with each other, resulting in a reading error, and a device for recording and reproducing highly reliable digital data is realized. In order to do this, it is necessary to eliminate these reading errors.

JP 2004-140764 A

  The present invention provides a technique for reducing errors in reading embedded information in information embedding using a digital watermark.

  An electronic watermarking system according to the present invention is an electronic watermarking system including an information embedding device and an information reading device, and the information embedding device embeds additional information in an embedding target image and displays the electronic watermark image. Encoding means for generating; and error map generating means for detecting a portion that is likely to be erroneously recognized when extracting the data array from the digital watermark image in the data array of the additional information, and generating error map information indicating the part; The information reading side device obtains the additional information by processing the extracted data array based on the error map information, and a data extracting means for extracting the data array embedded in the digital watermark image Decoding means.

  The error map information is information indicating a portion that is easily misrecognized when the data sequence is extracted from the digital watermark image in the data sequence of the additional information. Are equivalent.

  In a preferred aspect, the error map generation means includes test data generation means for randomly generating a test data array, and encoding simulation for generating a test digital watermark image by embedding the test data array in the embedding target image. And a decoding simulation means for extracting a data array embedded in the test digital watermark image and comparing the data array with the test data array to obtain a portion in which erroneous recognition occurred during extraction. For each test data array generated at random, let the encode simulation means and the decode simulation means perform processing to obtain and record the portion where the misrecognition occurred, as a portion where the misrecognition occurred A portion where the number of detected times exceeds a predetermined number is erroneously recognized. It is extracted as the easy part.

  The best mode for carrying out the present invention (hereinafter referred to as “embodiment”) will be described below with reference to the drawings.

  As a digital watermarking method, there are a method of embedding data in an image space and a method of embedding data after converting an image into a frequency space.

  As shown in FIG. 1, the former is a multiplication or addition for each pixel or each block between the plane 1 of the embedding target image as the embedding destination of the additional information and the plane 2 of the additional image representing the additional information. By performing other operations, composite image data (image with embedded information) is obtained. The method shown in Patent Document 1 is an example of this. This method has a property that an error is likely to occur at a specific position of the additional image plane 2 depending on the property of the embedding target image plane 1. That is, there may be a place in the embedding target image that has a property of easily interfering with the additional image to be embedded, and information embedded in such a place is likely not to be read correctly.

  Similarly, in the latter method of superimposing the embedding target image plane and the additional image plane in the frequency space, there is also a property that an error is likely to occur in a specific bit position of the embedded information due to interference with the embedding target image. is there.

  In the present embodiment, in view of such properties, a position where a read error is likely to occur in advance (that is, a position where a bit value different from the embedded bit value is likely to be read) is detected, and error position information is generated. By using this on the encoder side and decoder side, decoding errors are reduced.

  FIG. 2 is a system configuration diagram of a digital watermark system according to such an idea. As shown in the figure, this system includes an information embedding side device 10, an error information management device 20, and an information reading side device 30.

  The information embedding side device 10 is a device that performs a process of embedding additional information in an embedding target image, and includes an encoder 12, an error map generation unit 14, and an image output unit 16.

  The encoder 12 is means for encoding additional information and embedding it in an embedding target image. As the encoder 12, various conventional digital watermarking encoders such as those shown in Patent Document 1 can be used. In one preferred embodiment, the encoder 12 uses the error map required by the error map generation unit 14 in addition to the data embedding process (referred to as basic process) performed by a conventionally known encoder. It is also possible to have an additional process of embedding additional information in a place where it is difficult to occur. Details of this addition processing will be described later. Note that it is not essential for the encoder 12 to use the error map, and an example of a processing method when the encoder 12 does not use the error map will be described later.

  The error map generation unit 14 is a means for detecting a place where an error in reading additional information embedded in the image is likely to occur in the embedding target image. In general, when embedding data in an image, each pixel of the embedding target image or each block formed by dividing the image into rectangular areas of a predetermined size is added in a predetermined order such as a raster scanning order. Embed each bit of information. Therefore, the error map generation unit 14 can also be regarded as means for detecting a bit position where a read error is likely to occur in the bit array of the additional information to be embedded. The error map generation unit 14 according to the present embodiment generates a random bit string as additional information for testing, and repeats a simulation of embedding it in a given embedding target image to read out a reading error in the embedding target image. A position where an error is likely to occur, that is, a bit position where a reading error is likely to occur in the additional information bit array is obtained, and a map (error map information) of the position where such an error is likely to occur is generated. In this way, by using a lot of random test data to obtain a statistically error-prone position, a number of different identification numbers and serial numbers, etc., such as tickets and certificates issued by government offices, can be obtained. It can cope well when embedding in the same pattern (embedding target image). The internal configuration and detailed processing contents of the error map generator 14 will be described later.

  The error information management apparatus 20 holds and manages error map information generated by the information embedding side apparatus 10 and provides it to the information reading side apparatus 30. The error information management device 20 can communicate with the information embedding side device 10 and the information reading side device 30 via, for example, the Internet, an intranet, or another data communication network.

  The error map generation unit 14 of the information embedding apparatus 10 sends the created error map information to the error information management apparatus 20 together with an embedding target image ID (identification information) that uniquely indicates the embedding target image that is the target of the error map. sign up. The error information management apparatus 20 associates the embedding target image ID with the error map information, and manages these sets by registering them in a table, for example. When the error management device 20 receives a request for error map information from the information reading side device 30, the error management device 20 obtains an embedding target image ID specified in the request and returns error map information corresponding to the embedding target image ID. The embedding target image ID may be assigned by the error map generation unit 14, or may be assigned by the error information management device 20 and notified to the error map generation unit 14.

  Further, the error map generation unit 14 notifies the encoder 12 of the embedding target image ID. The encoder 12 may embed the embedding target image ID in the embedding target image together with the additional information, or may incorporate it as a string of numbers or symbols visible to the embedding result image. In any case, the embedding target image ID is incorporated into the image in such a manner that the information reading side device 30 can read the embedding target image ID.

  The image output unit 16 outputs the information-embedded image 100 generated by the encoder 12 in this way. The image output format here may be printed on a medium such as paper or may be output as an electronic image file.

  In the information reading side device 30 that reads the information embedded image 100 generated in this way, the image input unit 32 reads the information embedded image 100. When the information-embedded image 100 is printed on paper, the image input unit 32 is a scanner that optically reads the image. When the information-embedded image 100 is input as a file, the image input unit 32 reads the image file. Open software.

  The decoder 34 receives the image data of the information embedded image 100 input from the image input unit 32, and decodes the additional information from the image data. The basic decoding process may be a conventionally known one such as that shown in Patent Document 1, but in this embodiment, in addition to this, the decoder 34 embeds an object to be embedded read from the information-embedded image 100. Additional processing such as passing the image ID to the error information management device 20 to obtain error map information corresponding to the image ID, and using the error map information, for example, ignoring the decoded bit value corresponding to a place where an error is likely to occur By doing this, reading errors (reproduction errors of additional information) are reduced. That is, the decoder 34 first reproduces a raw data bit string embedded therein from the information-embedded image 100 by a conventionally known basic decoding process. Then, the original additional information is restored by interpreting the raw data bit string according to the error map information by the additional processing. Details of the addition process will be described later.

  As described above, in this system, the information embedding side detects a portion that is likely to cause a data reproduction error in the embedding target image, and transmits the information to the information reading side for use. Is reduced.

  In the above example, the embedding target image ID for specifying the error map information necessary for decoding is used, but this is not essential. For example, if the relationship between an identification number or serial number displayed on a document such as a ticket and error map information for decoding the document is registered in the error information management device 20, the embedding target image ID is included in the image. Even if it is not incorporated, the information reading side device 30 notifies the error information management device 20 of the identification number of the document detected from the read image, thereby obtaining error map information necessary for decoding the additional information. I can do it.

  Next, details of the error map generator 14 will be described with reference to FIG. As shown in FIG. 3, a test information generator 12, an encoder 144, a decoder 146, and an error detector 148 are provided.

  The test information generator 142 generates a random bit string as test additional information. This bit string preferably includes 1 and 0 equally. As a specific means for generating a random bit string, a known method for generating a pseudo-random bit string such as an M series can be used. The generated bit string of the test additional information is supplied to the encoder 142 and the error detector 148.

  The encoder 144 embeds the additional test information generated by the test information generator 142 in the embedding target image. The processing performed by the encoder 142 is the same as the basic processing of the data embedding processing of the encoder 12 of the information embedding side device 10. Therefore, the encoder 144 may of course use the basic function of the encoder 12. The image in which the additional information is embedded by the encoder 12 is output in the form of printing or the like by the image output unit 16, but the test additional information output from the encoder 144 in the error map generation unit 14 is embedded. The image data is passed to the decoder 146 as data.

  The decoder 146 decodes the additional information from the received image data. The decoding process performed by the decoder 146 is the same as the basic decoding process of the decoder 34 of the information reading side device 30. By this decoding process, a bit string of the test additional information embedded in the image data is obtained. The obtained bit string is input to the error detector 148.

  The error detector 148 compares the bit string input from the decoder 146 (that is, data that can be read from the image) with the bit string of the test additional information supplied from the test information generator 142 in order from the first bit, Find and record erroneous bit positions.

  With the configuration as described above, a predetermined number of random bit strings are generated and embedded in an embedding target image, and bit positions having read errors are obtained and recorded. Based on this record, a bit position (or vice versa) where a reading error is likely to occur statistically is obtained, and an error map is created.

  In the simplest embodiment, the error detector 148 may obtain a bit position having a value different from that of the correct additional information for testing. However, as another embodiment, the polarity of the read error generated at the bit position ( (Directionality), that is, it is also preferable to determine and record whether 0 is mistaken as 1 or vice versa.

  Next, the processing operation of the error map generator 14 will be described with reference to FIG. In the following, for convenience of explanation, the size of additional information (bit string) embedded in one embedding target image is N (bits).

  In this process, first, various variables are initialized (S10). In this process, first, an error detection loop count T is set. In error map generation, a series of error detection operations may be performed at least once, but it is desirable to repeat the same operation several times in order to improve the accuracy of error information. The number of loops T is a numerical value for setting the number of repetitions, and can be set from a minimum of 1. However, it is desirable that the number of loops is determined by the number N of bits of additional information to be embedded, accuracy of error information to be obtained, processing time, etc. As a guideline, a numerical value equal to or close to the number of bits N of the additional information may be used. In this initialization operation, the value C of the loop counter indicating the number of error detection operations is initialized to 0, and all elements of the bit string error map E, which is an array for recording the bit position where the read error has occurred, are stored. Set to 0. This bit string error map E is an array for recording the number of occurrences of read errors in each bit of N-bit additional information, and has N elements from 0 to N-1.

  Next, the test information generator 142 generates a bit string Dt (the number of bits is N bits) of the additional test information (S12). The test information generator 142 is not regular, and it is desirable to generate a different bit string each time it operates. When the test information generator 142 generates an M-sequence pseudo-random number, if the sequence period is set to N-1, and the number of repetitions T = N-1, the pseudo-randomness always different in each loop count. Since the additional test information can be generated, it is desirable in terms of error detection accuracy, and further has an advantage that the generation mechanism is simple.

  Next, the encoder 144 embeds this test additional information in the given embedding target image to obtain an error detection electronic image G (S14). Here, since the encoder 144 performs encoding for the purpose of detecting an error in reading additional information due to interference from the embedding target image, an error occurs more than an encoder that actually embeds additional information for digital watermarking. It is desirable to use encoding parameters that are easy to do. As an example, as shown in FIG. 1, the embedding target image plane 1 and the additional information plane 2 are subjected to a certain kind of arithmetic processing, for example, the embedding processing from each pixel value P of the embedding target image plane 1 and the pixel value Q of the additional information plane 2. Consider a case where the pixel value R of the subsequent image is obtained as shown in Equation (1).

          R = P + a × Q (1)

  In this equation (1), a is a positive coefficient indicating the embedding strength. As this value increases, the image of the additional information is more conspicuous in the image after embedding processing than the embedding target image, that is, decoding is performed if the view is changed. It becomes easy. Therefore, if the value of the coefficient a is made smaller than the value used when the digital watermark is actually inserted, an error detection image G that is relatively likely to cause an error in later decoding processing can be obtained.

  Next, the error detection image G obtained electronically in this way is decoded (decoded) by the decoder 146 (S16). An N-bit bit string obtained by this decoding process is called a detection data bit string Dd.

  Next, the error detector 148 compares the bit string Dt of the test additional information with the detected data bit string Dd for each bit (S18). If a mismatch is detected at a bit position in the bit string by this comparison, 1 is added to the value E (i) of the bit position i (i is an integer from 0 to N−1) in the bit string error map E.

  In the embodiment for determining the polarity of the error, for the error in which Dt (i) = 0 and Dd (i) = 1, the value of E (i) is increased by 1, and conversely Dt (i) = For an error that is 1 and Dd (i) = 0, the value of E (i) may be reduced by 1. Since the reading error due to the interference with the embedding target image is caused by the characteristics of the image at the location in the embedding target image, a polarity error that reads 0 as 1 or a polarity that reads 1 as 0 Either one of the errors tends to prevail. Therefore, by increasing or decreasing the value of each element of the error map E in this way, the nature of the error at the bit position corresponding to that element can be shown. Of course, an error map may be used in which the number of error occurrences of each polarity is individually integrated and counted individually.

  In any case, the first purpose is that the error map indicates the position of an error that occurs on the bit string of the additional information.

Next, in S20, the value of each bit of the test additional information generated in S12 is inverted (that is, 1 is converted to 0, and 0 is converted to 1), and the bit string resulting from the inversion is embedded by the encoder 144 using the embedded image Dt ^*. The error detection electronic image G is obtained by embedding in FIG. Then, the electronic image G is decoded (S22), the decoding result is compared with the inverted bit string obtained in S20, and the bit position that is inconsistent as a result of the comparison is recorded in the bit string error map E (S24). The reason why the error is detected by inverting the bit string in this manner is that there are two types of polarities in the manner of error as described above, and error detection may not be possible unless the bit is inverted. However, it is not essential to perform error detection by inverting this bit string. If a simulation is performed by generating a sufficient number of additional test images sufficiently randomly, an equivalent result can be obtained without inversion. Therefore, this inversion process may be used when the error detection effect is improved by inversion of the bit string in principle with the accuracy of error detection.

  When the processing up to S24 is completed, the value of the loop counter C is then incremented by 1 (S26). If the loop counter C has not reached the value of the predetermined number of loops T set in S10 (S28). Returning to S12, new test additional information Dt is generated, and the above processing is repeated. When the value of the loop counter C reaches T, the series of processing operations ends.

  As a result of this processing, the bit string error map E indicates that an error occurs at each bit position of the additional information when various additional information is embedded in the given image to be embedded and decoded. Distribution is shown. Based on this bit string error map E, the error detection unit 14 determines a bit position in the additional information where a read error is likely to occur. For this determination, for example, a threshold H for the number of error occurrences is determined in advance, and a bit position indicating the number of error occurrences exceeding the threshold H among the elements of each bit position in the bit string error map E What is necessary is just to determine with a position. A set of bit positions that are likely to cause read errors as described above is error map information, which is provided to the error information management apparatus 20.

  Here, not only the bit position where an error is likely to occur but also the polarity of the error at the bit position is preferably included in the error map information. As the error map information in this case, for example, data having the following data structure can be used. In other words, this error map information is an array having elements as many as the number of bits of the additional information, and each element is composed of 2 bits, and the first bit of the 2 bits has an error in reading the bit position corresponding to that element. The subsequent bits indicate the polarity of read errors that are likely to occur.

  Note that the threshold value H may be determined by performing a simulation or the like as necessary in consideration of the nature of the digital watermark algorithm and the appearance frequency of errors in the range that can be taken by the element of E.

  In one preferred embodiment, this error map information is provided to the encoder 12 so that the additional information embedding process by the encoder 12 is highly resistant to read errors. That is, as an additional process to the basic process, for example, the encoder 12 performs a process of embedding after re-arranging a bit string of additional information to be embedded avoiding a bit position that is likely to cause a read error indicated in the error map information. . As a result, the additional information can be reproduced even if there is no data at the bit position where an error is likely to occur. Hereinafter, an example of such additional processing of the encoder 12 will be described.

  In order to explain this processing, as a specific example, the bit number N of the bit string to be embedded in the embedding target image is 64, the loop count T at the time of error detection is N−1, that is 63, and the test information generator 12 has a cycle of N−1. , And the bit string error map E obtained by the process of FIG. 4 is as shown in FIG.

  A number in each square in FIG. 5 indicates a value of each element of the bit string error map E. Here, in this example, a threshold value H that determines whether or not a bit position corresponding to each element of the bit string error map E is a read error occurrence point is assumed to be 10 in absolute value. That is, since | E (i) | ≦ H and H = 10, if the value of E (i) is within plus or minus 10, it is not regarded as a reading error. In the bit string error map E of FIG. 5, the cells determined to be prone to read errors by comparison with the threshold value H are hatched to be distinguished from others. From this, the bit position i = 12, 18, 23, 36, 49, 55 is determined as a place where an error is likely to occur, and information indicating these places is passed to the encoder 12 and the error information management apparatus 20 as error map information. It is.

  An example of encoding (additional information embedding) processing by the encoder 12 that has received this error map information will be described with reference to FIG. In this example, the size of additional information to be embedded in an embedding target image is 48 bits, and the additional information is represented as Y (1,0, ..., 47). In this example, the bit number N of the bit string that can be embedded is 64 bits. However, the size of the additional information that is actually embedded is set to be small in consideration of the case where there are 16 locations where errors occur. Since the bitmap E illustrated in FIG. 5 indicates that there are six places where errors are likely to occur, there is a margin of 10 bits in this example. The number of bits of the additional information to be embedded is set to a value smaller than the size N of the bit string of the information that can be embedded originally. Determine.

  In FIG. 6, a part of a bit string of 48-bit additional information Y (1,0,... 47) given as information to be embedded is shown in the upper stage, and error resistance is increased for the additional information Y in the lower stage. A corresponding portion of an embedded bit string X (0, 1,... 63) that is actually embedded in an embedding target image that can be added by processing is shown. As shown in this figure, Y (0) = X (0), Y (1) = X (1),... At bit positions other than those determined to be prone to errors. In addition, the value of the additional information Y is adopted as the value in the embedded bit string X as it is. Up to Y (11) can be assigned in this way. Since the next thirteenth bit position (that is, E (12)) is an error-prone bit position, the bit value Y (12) of the additional information is not assigned to X (12), and the first error that appears after that is not assigned. It is deferred and assigned to bit positions that are unlikely to occur (that is, bit positions that have not been determined to be “prone to error”). In this example, since 14th (i = 13) corresponds to it, the value of Y (12) is assigned to X (13). Thereafter, each bit of the additional information Y is shifted back one by one and set in the embedded bit string X until the bit position where an error is likely to occur next is reached. Since the bit position where an error is likely to occur is i = 18, Y (17) = X (18) is not assigned, and Y (17) is not the bit position where an error is likely to occur (the error is likely to occur). ) Assign to X (19).

  As described above, in the method illustrated in FIG. 6, each bit of the additional information Y is rearranged in order from the first bit in order, avoiding the bit position i that is determined to be prone to error. Is generated. In the example of the error map illustrated in FIG. 5, the last Y (47) of the bit string of the additional information is assigned to X (52) of the embedded bit string. The embedded bit strings X (53) to X (63) are unallocated, but are unnecessary for decoding, so dummy bit information is recorded.

  Also, bit position values X (12), X (18), X (23), X (36), and X (49) recorded as positions where errors are likely to occur in the error map information are decoded ( Since it is ignored at the time of (decoding), basically any value may be set. However, more preferably, each value is read according to the polarity (sign) of the map information of E (12), E (18), E (23), E (36), E (49), respectively. Bit information that is likely to appear as a result of an error is recorded. That is, X (12) = 0, X (18) = 1, X (23) = 0, X (36) = 1, X (49) = 0. This is for the following reason.

  That is, for example, in the information embedding in Patent Document 1, in a portion where a reading error in which 0 is mistakenly read as 1 is likely to occur, the embedding target image itself of the portion originally has an image feature close to the pattern image representing “1” of the additional information. Is likely to have. Therefore, if a pattern image indicating “0” is embedded in such a place, a pattern image having a characteristic opposite to that of the original embedding target image is superimposed, and the characteristics of the original embedding target image are weakened. It will be. On the other hand, the pattern image “1”, which is likely to occur as a reading error, has similar characteristics to the original image to be embedded, so that the image change due to embedding is small. As described above, it is preferable to assign bit information that is likely to be generated as a result of a reading error to a portion where an error is likely to occur, in that the change in the appearance of the image before and after embedding can be reduced.

  Next, another example of the encoding process using the error map information by the encoder 12 will be described with reference to FIG. Also in this figure, a part of the bit string of 48-bit additional information Y (1,0,... 47) to be embedded is added to the upper stage, and processing for increasing error resistance is added to the additional information Y in the lower stage. The corresponding part of the embedded bit string X (0, 1,... 63) is shown.

  In the method of FIG. 7, Y (0) = X (0), Y (1) = X (1), ..Y (47) = X (, regardless of whether or not the bit position is likely to cause a read error. 47) and the value of the additional information Y is assigned to the embedded bit string in order from the first bit. Then, for each bit after X (48), i = 12, 18,. . . The additional information Y (12) and Y (18) of each bit position is sequentially assigned. Here, for example, Y (18) is assigned to X (49), but since the bit position of i = 49 is a position where an error is likely to occur, Y (18) is assigned to the next X (50). . In this way, values are assigned in order, avoiding bit positions determined to be prone to errors.

  When the encoder 12 generates the embedded bit string X from the additional information Y by the method shown in FIG. 6 or FIG. 7, the embedded bit string X is embedded in the embedding target image. For example, when the method of Patent Document 1 is used as an embedding method, the additional information plane 2 is generated by replacing each bit value of the embedding bit string X with a corresponding pattern image and arranging them in order (see FIG. 1). . Then, the additional information plane 2 and the embedding target image plane 1 are combined to generate an electronic watermark image in which the additional information is embedded. This is merely an example, and the embedded bit string X may of course be embedded in the image by other digital watermarking methods.

  The example of the data embedding method using error map information has been described above. When the data embedding method shown in FIG. 6 is adopted, the decoder 34 of the information reading side device 30 acquires error map information from the error information management device 20 and reads a bit string read from the information embedded image 100 from the first bit. In sequence, the embedded additional information can be restored by interpreting the error map information while ignoring bits indicated as bit positions where errors are likely to occur.

  On the other hand, when the method of FIG. 7 is adopted, the information reading side device 30 adds the bit sequence read from the information embedded image 100 to the bit determined to be prone to error in the error map information in the bit sequence. The original additional information is restored by sequentially applying the value of each bit after the number of bits of information (except for the bit determined to be prone to error). In the case of the method of FIG. 7, it is necessary to know the number of bits of the additional information by some method. For this purpose, information on the number of bits of the additional information may be incorporated into, for example, a part of the error map information and transmitted to the information reading side device 30.

  As described above, if the decoding side shares the method of assigning the additional information bit string Y to the embedded bit string X at the time of encoding, the reverse process is easily performed, and an error is detected from the raw data bit string read from the image 100. It is possible to restore the additional information without.

  The above is an example when the error map information is applied to the data embedding process by the encoder 12. On the other hand, in the method of FIG. 8 described below, the encoder 12 does not use error map information. However, in this method, it is assumed that the additional information includes an error correction code (ECC). That is, when raw data to be embedded is given, the encoder 12 adds an error correction code to the data by a known method to form additional information.

  In this example, the error map information is the same as in the above example, but the number of bits of the additional information Y is 64 bits in consideration of the fact that an error correction code is included.

  In this method, the encoder 12 uses the additional information Y (i) incorporating the error correction code as an embedded bit string X (i) as it is. That is, the encoder 12 may be the same as the conventional one, and does not have to do anything special.

  The decoder 34 of the information reading side device 30 that has received the error adjusts the bit string read from the information-embedded image 100 using the error map information acquired from the error information management device 20, and then corrects the error. Apply error correction by code. In this embodiment, the error map information includes not only the bit position where an error is likely to occur but also information on the polarity of the error at the bit position.

  As this error map information, there are two data formats, that is, a bit indicating whether or not each bit position is a position where an error is likely to occur, and a bit indicating the polarity of an error likely to occur at that bit. When error map information including bits is used, it is as shown in FIG. This corresponds to the bit string error map of FIG.

  In this error map information, for a bit position that is likely to be erroneously set to 0 as 1, if the bit value of the bit position read from the image is 0, the bit value is 0 as it is, and if it is 1, a random bit generator Replaces the randomly generated value of 1 or 0 with the value of that bit. On the other hand, in the error map information, the bit position indicated as 1 is likely to be mistaken for 0 is set to 1 if the value of the bit read from the image is 1, and if it is 0, the random bit generator Replace with output value. In other words, in this method, when a bit value that is opposite to the value that tends to appear due to the polarity of the error is read, a value that hardly appears in the error appears in the first place. Trust. On the other hand, when a value that is likely to appear due to an error is read, the value is replaced with a random value because it cannot be trusted, and the error correction code is corrected by the error correction capability. By such processing, it is possible to reduce the rate at which errors occur at bit positions where errors are likely to occur, and to work efficiently in error correction. In addition, when a bit position where an error is likely to occur is known, error correction can be strengthened by using a technique called erasure error recovery, so that the possibility that correct data can be restored can be increased.

  In the method of FIG. 8, since the additional information Y includes an error correction code, the amount of information that can be expressed with the same number of bits is reduced. For example, the amount of information that can be expressed by 64-bit additional information Y is from several bits to 20 bits at most. Although there is such a disadvantage in terms of data efficiency, this method does not change the bit string itself of the additional information as shown in FIGS. 6 and 7, and thus is highly compatible with the conventional method. . That is, there is an advantage that even a conventional decoder that does not comply with the method of the present embodiment has a read error, but can be decoded properly.

  In the above example, the case of embedding additional information in the same embedding target image space as in Patent Document 1 is taken as an example, but the same may be applied to embedding in the frequency space. That is, in this case, the embedding target image is converted into frequency components, and bits of additional information are embedded in the values of the respective frequency components. Here, the error map generation unit 14 generates a digital watermark image by embedding a random bit string and inversely transforming the frequency component, and decodes this to obtain a bit position where there is a reading error. By repeating this, a bit position where a read error is likely to occur statistically is obtained, and by using this as error map information, the same processing as in the above example can be realized.

  In the above example, a map of bit positions where read errors are likely to occur is used as error map information. Conversely, a map of bit positions where read errors are unlikely to occur is created, and only such bit positions are created. Of course, additional information bits may be allocated.

  In the above example, the bit map generated at random is used as the test additional information in generating the error map information, but this is also an example. For example, if additional information that may be embedded in an image to be embedded is determined in advance and the number of such additional information is not so large, an error map is generated using the additional information itself as test data. Of course it is good.

  In the above example, the error map information is registered from the information embedding side device 10 to the error information management device 20 on the network, and the information reading side device 30 acquires and uses the error map information. It is only an example. Transmission of the error map information from the embedding side to the reading side may be performed by communication means such as e-mail, or may be performed by recording and sending it on a computer-readable portable recording medium. Further, the error map information itself may be embedded in the embedding target image, or may be embedded in the information embedded image 100 as visible information.

It is a figure which shows an example of the embedding method of the data to an image. 1 is a system configuration diagram of a digital watermark system according to the present invention. It is a functional block diagram which shows the internal structure of an error map production | generation part. It is a flowchart which shows the process sequence of an error map production | generation part. It is a figure which shows the specific example of the bit string error map produced | generated by the process of FIG. It is a figure for demonstrating an example of the additional information embedding process using error map information. It is a figure for demonstrating another example of the additional information embedding process using error map information. It is a figure for demonstrating another example of an additional information embedding process. It is a figure which shows an example of the data content of the error map information also including the information of the polarity of an error. It is a figure which shows the example of the pattern image showing each code | symbol.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 Information embedding side apparatus, 12 Encoder, 14 Error map production | generation part, 16 Image output part, 20 Error information management apparatus, 30 Information reading side apparatus, 32 Image input part, 34 Decoder

Claims (9)

An electronic watermarking system comprising an information embedding device and an information reading device,
The information embedding device is:
Encoding means for embedding additional information in an embedding target image to generate a digital watermark image;
In the data array of the additional information, an error map generating means for detecting a part that is likely to be erroneously recognized when extracting the data array from the digital watermark image, and generating error map information indicating the part;
With
The information reading side device is:
Data extraction means for extracting a data array embedded in the digital watermark image;
Decoding means for obtaining the additional information by processing the extracted data array based on the error map information;
Comprising
An electronic watermarking system characterized by that. The error map generating means includes
Test data generation means for randomly generating a test data array;
Encoding simulation means for embedding the test data array in the embedding target image to generate a test digital watermark image;
A decoding simulation means for extracting a data array embedded in the test digital watermark image and comparing the data array with the test data array to obtain a portion in which erroneous recognition occurred during extraction;
A part for which the misrecognition occurs, for each test data array generated at random, by causing the encode simulation unit and the decode simulation unit to perform processing to obtain and record the part where the misrecognition occurred 2. The digital watermark system according to claim 1, wherein a part in which the number of detected times exceeds a predetermined number is extracted as the part that is easily misrecognized.   The error map generation unit causes the encode simulation unit and the decode simulation unit to perform processing on the data array obtained by inverting the value of each bit of the test data array generated by the test data generation unit. 3. The digital watermark system according to claim 2, wherein a portion where erroneous recognition has occurred is obtained and recorded. The encoding unit of the information embedding side device generates an embedded data string by rearranging the data of the additional information based on the error map information, avoiding the erroneously recognized portion, and Embed in the target image,
The decoding means of the information reading side device performs the data arrangement changing process opposite to the encoding means on the data arrangement extracted from the digital watermark image with reference to the error map information. Seeking
The digital watermark system according to claim 1, wherein the digital watermark system is a digital watermark system. The error map generation means generates error map information including error polarity information indicating whether the misrecognized portion is easily misrecognized as 0 as 1 or vice versa, as the error map information.
The encoding means determines and sets a value that is easily recognized as a result of erroneous recognition, based on error polarity information corresponding to the part, in the erroneously recognized part of the embedded data string.
The digital watermark system according to claim 4. The encoding means embeds embedded information obtained by adding an error correction code to the additional information in the embedding target image,
The error map generation means generates error map information including error polarity information indicating whether the misrecognized portion is easily misrecognized as 0 as 1 or vice versa, as the error map information.
In the data array extracted from the digital watermark image, the decoding means determines the erroneously recognized portion from the error map information, and determines whether the value of the portion is a value that is easily recognized as a result of erroneous recognition. Judgment based on polarity information, if it is a value that is easily recognized as a result of misrecognition, replace that value with a randomly generated value, otherwise adopt the value without changing the value of that part The additional information is obtained by performing error correction processing on the reproduced embedded information that is reproduced,
The digital watermark system according to claim 1, wherein the digital watermark system is a digital watermark system. An encoding unit that embeds additional information in an embedding target image and generates a digital watermark image;
In the data array of the additional information, an error map generating means for detecting a portion that is easily misrecognized again by extracting the data array from the digital watermark image and generating error map information indicating the portion;
An information embedding device. The error map generating means includes
Test data generation means for randomly generating a test data array;
Encoding simulation means for embedding the test data array in the embedding target image to generate a test digital watermark image;
A decoding simulation means for extracting a data array embedded in the test digital watermark image and comparing the data array with the test data array to obtain a portion in which erroneous recognition occurred during extraction;
A part for which the misrecognition occurs, for each test data array generated at random, by causing the encode simulation unit and the decode simulation unit to perform processing to obtain and record the part where the misrecognition occurred 8. The information embedding side device according to claim 7, wherein a portion where the number of detected times exceeds a predetermined number is extracted as the portion that is likely to be erroneously recognized. Data extraction means for extracting a data array embedded in a given digital watermark image;
The error map information indicating a portion that is likely to be erroneously recognized when extracting the data sequence from the digital watermark image is acquired, and the data sequence extracted by the data extraction unit is processed based on the error map information to add the data Decoding means for obtaining information;
An information reading side device comprising:

Images