JP4804556B2 - Tamper detection digital watermark embedding method and apparatus, program, falsification detection method and apparatus using digital watermark, and program - Google Patents

Description

  The present invention relates to a digital image processing technique, and more particularly to a technique for embedding a digital watermark for detecting tampering and a technique for detecting tampering using the digital watermark.

  In recent years, with the widespread use of digital cameras and camera-equipped mobile phones, anyone can easily take digital photographs. The digital image holding method in a digital camera or a camera-equipped mobile phone has a high compression ratio and a good trade-off between image quality, so the JPEG encoding format has become the de facto standard. Some models can be stored in JPEG2000 encoding or uncompressed image format, but the most popular is the JPEG encoding method.

  And since digital cameras, especially camera-equipped mobile phones, can be taken easily, digital images taken using them can be used as evidence of personal property, as evidence of traffic accidents, or as process proofs at construction sites, There is a need to make a proof that it has been properly processed at the garbage disposal site. Moreover, although it is not a digital camera, there is a drive recorder that is similar, and the captured video (image sequence) is used as reference evidence of a traffic accident.

  However, because digital images are easy to edit, it is difficult to prove that they have not been tampered with, and the law remains reference evidence, which may not be definitive evidence. is there. Therefore, in order to enhance the evidence, Japanese Patent Application Laid-Open No. H10-228707 proposes a technique for adding a digital watermark to a digital image and improving the authenticity of the image depending on whether the digital watermark is not broken.

JP 10-308943 A

Masaaki Mitani “Industrial Mathematics for Redoing, Information Communication and Signal Analysis / Cryptography, Error Correcting Code, Integral Transformation” CQ Publishing Co., Ltd., ISBN 4-7898-3318-6 Kazutaro Aoki et al. "128-bit block cipher Camellia algorithm specification version 2.0" Nippon Telegraph and Telephone Corporation, Mitsubishi Electric Corporation, September 26, 2001 “Announcing the ADVACED ENCRYPTION STANDARD (AES)”, Federal Information Processing Standards Publication 197, November 26, 2001. “SECURE HASH STANDARD”, Federal Information Processing Standards Publication 180-2, August 1 2002 Hiroshi Yuki “Introduction to Cryptographic Technology” SoftBank Creative Corp., ISBN 4-7773-2297-7

However, in the method of Patent Document 1, since watermark data having a large value is embedded immediately before the zero-run portion of the quantization coefficient, image quality degradation due to watermark embedding may increase. Such a large deterioration in image quality is undesirable because it reduces the credibility of evidence photographs .
The present invention has been made to solve the above-described problems of the prior art, and an object of the present invention is to provide a technique for detecting tampering while suppressing image quality deterioration.

  The falsification detection technique according to the present invention uses, as a combination, a device that embeds a digital watermark for detection in an image and a device that detects falsification of an image using the digital watermark. Here, the principle of JPEG encoding / decoding is closely related.

In JPEG encoding, as shown in FIG. 1A, an input image (S01) is subjected to YCbCr conversion (S02), Cb (a signal indicating hue / saturation of a blue system), Cr (hue / saturation of a red system). after down sampling (S03) signal) representative of the degree, discrete cosine transform in block units of the sampling pixels (8 × 8 pixels) (DCT: performing discrete cosine transform) (S04).

  After the information is dropped by quantizing the converted data (S05), entropy coding using Huffman code is performed and compression is performed (S06), and the data is shaped into the JFIF format (S09).

  In the JPEG decoding, as shown in FIG. 1B, the input JPEG image (S11) is shaped from a JFIF format into a data string (S12), and dequantized after entropy decoding (S13) (S14). ). Thereafter, inverse DCT is performed (S15), and if necessary, color space conversion is performed (S16).

  In the present invention, when an input image is divided into blocks of 8 × 8 sampling pixels (referred to as DCT blocks), 1 bit is assigned to each block and a bit obtained by error detection coding and encryption of digital watermark data is embedded.

  At this time, the bit of the digital watermark data is not made to correspond directly to the DCT block, but is made to correspond after error detection coding is performed once to provide redundancy. By using an error detection code in this way, it is possible to prevent obtaining the same result as if it was accidentally embedded without embedding anything, and whether or not there was an error (falsification) without knowing the original digital watermark data at the time of decoding The effect of knowing is obtained.

Further, since the total number of DCT blocks included in an image is generally large (4800 or more luminance components of a VGA image), DCT blocks are divided into groups and error detection coding is performed independently in one group. As a result, if there is an error, it is possible to specify a group including the error, and the effect of knowing the location of the error (falsification) is obtained. Further, in the modulation method of the quantization coefficient in the DCT block, the absolute value of the quantization coefficient is obtained. A value B obtained through a predetermined calculation from the maximum value A, a random number obtained from the random number generation means, and a value obtained by hashing the quantization coefficient is used.

  Here, “−K”, “0”, and “K (K = any positive / odd number)” are added to A to adjust the even / odd difference of B. As a result, only a small variation is given to the coefficient A that originally has a large value, so that the change of the quantization coefficient is small, and the influence on the aesthetics after embedding is minimized. Although there are a plurality of methods for modulating the quantization coefficient in the DCT block, including those of Non-Patent Document 1, the change in the quantization coefficient due to the modulation is large, which is aesthetically disadvantageous.

  According to the present invention, it is possible to detect falsification by embedding a digital watermark while suppressing deterioration in image quality.

(A) is a chart diagram of normal JPEG encoding, and (b) is a chart diagram of normal JPEG decoding. 1 is a configuration diagram of a tamper detection digital watermark embedding apparatus according to an embodiment of the present invention. FIG. The chart of the whole process. The flowchart which shows the group division | segmentation of the DCT block. The flowchart which shows the process which embeds encryption encoding data in the DCT block. The flowchart which shows the selection process of the maximum value A of the same absolute value. The schematic which shows the calculation process of B value. The flowchart which shows adjustment of the maximum value A with the said embedding bit. The example figure of the same quantization coefficient. The same figure of an example of addition of the even-numbered row parity bit. The same figure of DCT block group division. FIG. 10 is a diagram illustrating an example of the reverse zigzag scan of FIG. 9. FIG. 10 is a diagram illustrating an example of a raster scan in FIG. 9. The block diagram of the tampering detection apparatus which concerns on embodiment of this invention. The flowchart which shows the process which detects the encryption code data from the DCT block. An example figure which shows the state by which the error was not detected. An example figure which shows the display state of the alteration location. The figure which shows the example of a format of embedded data. The block diagram of the other example of the embedding apparatus of the electronic watermark for a detection. The chart of the whole process.

  Hereinafter, a digital watermark embedding device and a tamper detection device using a digital watermark according to the present embodiment will be described with reference to the drawings. Here, digital video (Motion JPEG) is the processing target.

1. Digital Watermark Embedding Device FIG. 2 shows a tamper detection digital watermark embedding device according to an embodiment of the present invention. The embedding device 1 is composed of a computer incorporated in a digital video recording device such as a digital camera, a digital video camera, or a camera-equipped mobile phone.

  Here, the embedding device 1 JPEG-encodes a video image taken by the video acquisition means 2 of the video storage device, that is, a frame image sequence, and simultaneously embeds a digital watermark for tampering detection. This video acquisition means 2 is composed of an optical part (optical sensor) such as a CCD or CMOS.

Specifically, the embedding device 1 has a pre-processing unit 3, a digital watermark input unit 4, an electronic component as a result of cooperation between computer hardware resources (for example, an image processor) and software resources (OS and application). Watermark data error correction encoding means 5, encryption means 6, DCT block group dividing means 7, maximum value A selecting means 8, B value calculating means 9, random number generator 10, encrypted code data embedding means 11, final It functions as a DCT block operation value header storage unit 12, an entropy compression unit 13, and a JFIF format shaping unit 14.

  The pre-processing unit 3 is transmitted with the frame image sequence captured by the video acquisition unit 2. Each frame image transmitted here is digital data obtained by A / D converting analog data output from the optical unit. At this time, the preprocessing means 3 performs processing up to DCT coefficient quantization of normal JPEG encoding for each frame image.

  The digital watermark input means 4 corresponds to a button (key) for inputting digital watermark data (for example, calendar date) to be embedded in each frame image.

The digital watermark data error correction encoding means 5 gives the input digital watermark data redundancy to the data by a predetermined error detection encoding method to obtain error detection code data. The encryption unit 6 converts the error detection encoded data into code data that has been encrypted by a predetermined encryption method.

  The DCT block group dividing means 7 divides the DCT blocks of three types of color components Y, Cb, and Cr constituting each frame image into groups of the same number as the number of bits of encoded data. To do.

  The maximum value A selecting means 8 selects the A value having the maximum absolute value from the quantization coefficients of one DCT block of each group. The B value calculation means 9 calculates a B value based on a result calculated from a random value obtained from the random number generation means 10 and a hash value of a numerical value obtained by concatenating the quantization coefficient of each DCT block.

  The encrypted code data embedding unit 11 embeds encrypted code data that adds any of −K, 0, K (K is an arbitrary positive / odd number) to the A value. Here, when “bit 0” of the encoded code data is embedded in the DCT block, the difference between the A value and the B value of the DCT block is set to an even number. On the other hand, when “bit 1” of the encoded code data is embedded in the DCT block, the difference between the value A and the value B of the DCT block is set to an odd number.

  The final DCT block operation value header storage means 12 is a numerical hash value obtained by concatenating the A value changed coefficients of the final DCT block of the (final group number-1) group, and the DCT block immediately before the final DCT block. The exclusive OR of the coefficient value and the calculated value is calculated, and the result obtained by encrypting the result by a predetermined encryption method is stored in the comment field in the header of the JPEG format.

  The entropy compression means 13 and the JFIF format shaping means 14 shape the quantized coefficients after embedding so as to conform to the entropy compression / JPEG format in the same manner as in the normal JPEG encoding process, and perform JPEG encoding. Complete. Hereinafter, specific processing contents of the embedding apparatus 1 will be described with reference to the chart of FIG.

  S21 to S25: When an object is photographed by the video acquisition means 2 (S21), frame images of the photographed video are continuously transmitted to the preprocessing means 3. Here, the pre-processing means 3 advances the same process as the normal JPEG encoding process for each frame image. That is, each frame image of digital video is subjected to YCbCr conversion (S22), downsampling (S23), and DCT (S24) to DCT coefficient quantization (S25).

  After this quantization (S25), as shown in FIG. 9, “8 × 8 = 64” quantization coefficients exist in each DCT block. Here, the quantization coefficient exists for each color component of Y, Cb, and Cr.

  At the time of downsampling (S23), the Cb (a signal representing the hue / saturation of the blue system) and the Cr component (a signal representing the hue / saturation of the red system) than the Y component (luminance signal). Is generally sampled, the number of DCT blocks has the largest Y component. In any case of Y, Cb, and Cr, 8 × 8 quantization coefficients are included in the DCT block.

  S26: The digital watermark data is obtained by the digital watermark input means 4. The acquired digital watermark data is S27. Processing is performed in S28. Here, as an example, the digital watermark data is assumed to be 448 bits. This 448 bits can be arbitrarily determined, and can be arbitrarily selected under the condition that the number of bits resulting from redundancy by the error detection coding in S27 is the same as the number of DCT blocks in one group divided in S29. obtain.

  S27: The watermark data error detection encoding means 5 performs error detection encoding in order to provide redundancy to the digital watermark data. Here, a case where even-numbered row parity is selected as the error detection code will be described.

  That is, as shown in FIG. 10, 448 bits of digital watermark data are arranged in 7 columns and 64 rows, and one even parity bit is added to the end of each row. The even parity bit is a method of setting the value of the parity bit to 1 or 0 so that the number of 1s in 8 data bits of each row including the parity bit is an even number.

  As a result of adding even parity bits to each row, 64 parity bits are added to the data subjected to error detection coding, and the total becomes 448 + 64 = 512 bits. The error detection code is not limited to even-numbered row parity, and other detection codes (see Non-Patent Document 1) may be used.

  S28: The encryption means 6 encrypts the 512-bit data that has been subjected to error detection coding, and generates 512-bit encrypted code data. Here, as an example, the Camellia cipher (see Non-Patent Document 2) is used as the cipher.

  The Camellia cipher is a 128-bit block cipher, and converts a sequence of 128-bit data into another sequence of 128-bit data (encoded encoded data). When converting, an encryption key of 128/192/256 bits is required. In order to decrypt the encoded data and obtain the original data code, a camellia decryption process is performed using the same encryption key as that used for encryption. Since the Camellia cipher is a 128-bit block cipher, when a 512-bit error detection code is encrypted, 512/128 = 4 processes are required. The encryption method is not limited to camellia encryption, and other encryption methods may be used.

  S29: The DCT block group dividing means 7 performs group division of the DCT block in the image frame. The group division processing procedure will be described with reference to the flowchart of FIG.

  Here, first, the DCT blocks of the Y color component, the Cb color component, and the Cr color component are connected in this order and arranged in one row (S41). Next, DCT blocks are extracted for each bit number (512 in this case) of the encoded data counted from the youngest in this column, and set as the first group (S42 to S45).

  At this time, the group number is assigned to the group, the first 512 being the first group and the next 512 being the second group. If less than 512 DCT blocks remain at the end, they are set as the final group (S46). The state of grouping is shown in FIG.

  S30: The encoded code data is embedded in the DCT block included in each group. Here, embedding of the encoded code data is performed independently for each group. Specific processing contents will be described based on the flowchart of FIG. This process is performed until all DCT blocks are completed (S51).

(1) Maximum value A selection of absolute value (S52)
First, the maximum value selection means 8 searches for the maximum value A of the absolute value from the 64 quantization coefficients. Here, as shown in FIG. 9, the quantization coefficients are arranged in an 8 × 8 square, and the upper left is a DC component, and the higher the frequency is, the lower the right. If there are multiple absolute maximum values, they are determined as follows.

  That is, when there are a plurality of maximum values, the first value that appears in the reverse zigzag scan order (reverse order of normal JPEG zigzag scan) in FIG. Here, when there are equivalent ones, the reason why the one appearing earlier in the reverse zigzag scan is adopted is to bring the quantization coefficient used for watermark embedding to the high frequency component side as much as possible and make it less noticeable. Let A be the maximum value selected in this way.

  FIG. 6 shows a flowchart of this algorithm. Here, A = 0 is set as an initial value (S61), and the quantization coefficient is advanced by inverse zigzag scanning to acquire the coefficient α (S62).

  Next, it is confirmed whether or not | α |> A is satisfied (S63). If true, A = | α | is determined (S64), and it is confirmed whether or not the quantization coefficient has been fully scanned (S65). If all scans have been completed, the maximum value A = | α | The process ends.

  On the other hand, if the entire scan has not been performed, the process returns to S62 to resume the reverse zigzag scan. At this time, since the value of A is updated (S64), even if the same quantization coefficient α is read in the next and subsequent scans, | α |> A of S63 does not hold, and the confirmation of S65 is performed as it is. Proceed to processing. As a result, the quantization coefficient α first scanned in the reverse zigzag scan order is set to the maximum value A.

(2) Calculation of B value (S53)
The B value calculation method of the B value calculation means 9 is shown in FIG. One B value is calculated for each DCT block. The value of B takes 0 or 1. Here, in order to calculate the B value, the hash value of the random number obtained from the random number generator 10 is used (see Non-Patent Document 5, p. 162 to p. 175). Let H be the bit length of the hash value calculated by the hash function. H varies depending on the hash function to be used, but here, as an example, the SHA-1 hash function (see Non-Patent Document 4. Non-Patent Document 5, p. 176 to p. 184) is used, so that H = 160. A hash function other than SHA-1 may be used.

  Specifically, the value B of the first DCT block is the least significant bit of the hash value of the random number generated by the random number generator 10, as shown in FIG. The B value of the nth DCT block is the 160-bit hash value used to calculate the previous (n-1) B value, and the A value adjustment in S54 is added to the previous (n-1) block. The value obtained by calculating the exclusive OR (XOR) for each bit with the hash value of the numerical value obtained by arranging the DCT coefficients in the order of the raster scan (see FIG. 13) is counted from the least significant bit to the nth bit value. And

  At this time, when the number of DCT blocks (n) is larger than the hash value bit length H of the random number, a value obtained by subtracting H from n until “n ≦ H” is used. For example, when the number n of DCT blocks exceeds 160, when the B value is calculated up to the 160th bit, the next block and subsequent bits are returned to the least significant bit for calculation.

  The header storage means 12 is a numerical value in which the 160-bit hash value used for calculating the B value of the final DCT block of the (final-1) group and the DCT coefficient after the A value adjustment of the block are arranged in raster scan order. A value obtained by taking an exclusive OR (XOR) for each bit is stored in the comment field of the JPEG header after encryption.

  It is assumed that “0” is added (padded) to the right side of the least significant bit so as to be a multiple length of 128 bits. Further, it is assumed that the encryption uses the same 128-bit block encryption method and the same encryption key as the method of encrypting the error detection code.

  The random number generator 10 uses a method of obtaining the same random number sequence if the seeds (initial values) of the random numbers are the same, for example, a linear congruence method. However, the random number generation method is not limited to the linear congruential method, and other methods may be used in which the random number sequence to be output is determined deterministically by determining the seed of the random number.

(3) A value adjustment (S54)
The encrypted code data embedding unit 11 adjusts the A value by the embedding bit. In other words, if the DCT block embeds the bit “1” of the encrypted code data generated in S28, the adjustment is made so that “AB = odd number”, while the bit “0” of the encrypted code data is adjusted. If the DCT block embeds A, A is adjusted so that “AB = even number”.

  Specifically, the adjustment of the A value is selected from three methods: (b) subtracting arbitrary positive / odd numbers, (b) adding arbitrary positive / odd numbers, and (c) doing nothing. Arbitrary positive / odd numbers have a better aesthetic appearance after embedding if a smaller value is used. The specific processing content of the adjustment method will be described based on the flowchart of FIG.

  First, it is confirmed whether or not “AB = even number” is established (S81). For example, assume that data “1” is to be embedded when “A = 15, B = 0”. In this case, since “A−B = 15” is already an odd number, the process proceeds to S82. Here, “embedding bit = 0” is confirmed (S82). In this example, since “1” is to be embedded, “(c) do nothing” is selected.

  In the case of “A = 15, B = 1”, since “A−B = 14” is an even number, the process proceeds to S83. Also in S83, “embedding bit = 0” is confirmed. Since “1” is to be embedded here, the process proceeds to S84. In S84, it is confirmed whether or not “A ≧ 0” is satisfied, that is, whether A is positive or negative. In this case, since it is a positive value, “(b) Add arbitrary positive / odd” is selected, and an operation of “A = A + 1” is performed (S85). In this case, A = 15 + 1 = 16 is adjusted.

  Further, it is assumed that data “0” is to be embedded when “A = −18, B = 1”. In this case, since “A−B = −19” is an odd number, the process proceeds to S82. Since it is desired to embed “0” here, the process proceeds to S84 and the sign of A is confirmed. In this case, since A is a negative value, “(A) subtract arbitrary positive / odd number” is selected, and an operation of “A = A−1” is performed (S86). In this case, A = −18−1 = −19. If “A = −18, B = 0”, since “A−B = −18” is an even number, S81. According to S83, “(c) Do nothing” is selected.

  As described above, after the encrypted code data embedding unit 11 embeds the bit of the encrypted code data in each DCT block in the first group, the encryption is similarly performed from the second group to the (final-1) group. Embedded code data. Here, the same value as that of the first group is embedded in the final group, and the missing DCT block is ignored. The encrypted code data may be the same data in all groups or may be different in each group. Whether the encrypted data is the same or different in each group may be determined by the convenience of the application.

  S31. S32: After embedding of the encoded code data bits for all groups is completed in S30, data is shaped so as to conform to Empiro compression (S31) and JFIF format in the same manner as in the normal JPEG encoding process (S32). Is output as a JPEG sequence video. Assume that the output video is stored in a storage means such as a memory card.

  In embedding a digital watermark into a quantization coefficient, the output video is given a variable with the smallest positive / odd number (1.0.-1) as the maximum absolute quantization coefficient A as shown in S54. Therefore, the digital watermark is embedded without impairing the aesthetics, thereby minimizing image quality degradation and improving the credibility as evidence.

  The digital watermark data to be embedded at this time is not only encrypted in S28, but also the embedded algorithm (S53.S54) defines a numerical value to be embedded using a random number. If the initial value of the generator 10 is not known, it is difficult for a person who is familiar with the algorithm of the embedding device 1 to forge.

  Therefore, if the encryption key and the initial value of the random number are kept secret, it is possible to prevent an attack that makes a false image a genuine image. In this case, since the embedding numerical value determined by the random number also changes depending on the content of the image, an attacker who analyzes a plurality of embedded images and estimates the random number generation pattern of the embedding device 1 to try to forge a digital watermark In contrast, the effect of making it difficult to estimate the random number generation pattern occurs.

  In addition, since a plurality of digital watermarks can be embedded at the same image position by embedding a digital watermark in a plurality of color components, an attacker who attempts to forge the digital watermark of the embedding device 1 has a plurality of color components. Both digital watermarks must be counterfeited so that there is no contradiction, and this also increases the difficulty of counterfeiting.

2. Tamper Detection Device FIG. 14 shows a tamper detection device according to an embodiment of the present invention. The detection device 20 detects whether the JPEG image sequence embedded with the digital watermark by the embedding device 1 is falsified.

The detection device 20 may be built in the video recording device, or may be configured by a separate device such as a personal computer (PC). Specifically, the detection device 20 detects the JPEG as a result of cooperation between hardware resources of a computer (for example, an image processor of the video storage device and a CPU of a personal computer) and software resources (OS and applications). Encoded video data input means 21, preprocessing means 22, DCT block group dividing means 7 , maximum value A selecting means 8, B value calculating means 9, random number generating means 10, bit data obtaining means 23, encryption / decryption means 24, comparison means 25, error detection decoding means 26, inverse quantity quantization means 27, inverse DCT means 28, color space conversion means 29, and error occurrence location superimposing display means 30.

  Here, the DCT block group dividing means 7, random number generator 10, encryption / decryption means 24, and error detection / decryption means 26 of the detection device 20 are the DCT block group division shapes used when embedding the electronic watermark in the embedding device 1. The error detection encoding method, the encryption method, the encryption key, the random number generation method, and the initial value of the random number generation are known.

  The detection processing procedure of the detection device 20 will be described. First, the video data input means 21 reads and acquires the video of the JPEG sequence stored in the storage means such as the memory card. The acquired digital video data is subjected to a normal JPEG decoding process through the preprocessing means 22. That is, each JPEG image frame is shaped into a data string from the JFIF format, the data string is entropy-decoded and converted to a DCT block quantization coefficient.

  Next, DCT block grouping means 7 performs the same DCT block grouping (see FIG. 11) as that used for embedding. Bits (1 or 0) of encrypted code data are detected for the DCT blocks of each group.

  This detection process will be described based on the flowchart of FIG. This process is performed independently for each group. First, the selecting unit 8 for the maximum value A examines 64 quantization coefficient values for the first DCT block and determines the maximum value A of the absolute value (S152). The method for determining A is exactly the same as that for embedding (see FIG. 6). Next, the B value calculation means 9 determines the B value by the same method as that used for embedding (S153). The configuration and initial value of the random number generator 10 used at this time are the same as those at the time of embedding. Each of the means 7 to 10 executes the same processing as that of the embedding device 1, and therefore may be used as the embedding device 1 when the detection device 20 is built in the video recording device.

  The bit data acquisition means 23 is S152. A and B determined in S153 are examined (S154), and if A-B is an even number, "0" is acquired (S155), whereas if A-B is an odd number, "1" is acquired (S156). S152 to S156 are applied to all blocks, and the detection process is terminated (S151).

  Then, the encryption encoding unit 24 performs S155. In S156, after acquiring embedded bits (1 or 0) from all DCT blocks in one group (in this example, 512 bits in total), the acquired data is decrypted. At the time of decryption, the same encryption method and encryption key as at the time of embedding are used. The error detection decoding means 26 performs an error detection decoding process on the 512-bit decoded data code obtained as described above. As a result, the presence or absence of an error is detected, and if there is no error, 448-bit embedded watermark data is obtained, while if there is an error, information indicating a group including the error is obtained.

  After detection of falsification of the first group, falsification is detected in the same processing procedure from the second group to the (final-1) group. In the final group, it is confirmed whether or not the sequence of detected bit strings is the same as that in the first group, and if it is the same, it is determined that there is no error. Since the number of DCT blocks in the final group is less than 512, a shortage occurs when compared with the first group, but this is ignored.

  The comparison means 25 executes the following processing procedure. That is, the exclusive OR (XOR) for each bit of the 160-bit hash value used for calculating the B value of the final DCT block of the (final-1) group and the numerical value in which the DCT coefficients of the DCT block are arranged in raster scan order. Find the value taken.

  The same 128-bit block encryption method and the same method as the method of encoding the error detection code are added to the value obtained by adding (padding) “0” to the right side of the least significant bit so as to be a multiple of 128. Encrypt with encryption key. After this encryption, it is compared with the numerical value in the comment field of the JPEG header.

  Further, the numerical value in the comment field may be decrypted with the same 128-bit block encryption method and the same encryption key as the error detection code encoding method and compared with the exclusive OR value.

  If the comparison results are the same, there is no error, and if they are different, it is determined that there is an error in the (final-1) group. Although this process is redundant, the final DCT block of the (final-1) group is most vulnerable to forgery, and this process makes forgery resistance robust.

  After that, the inverse quantization means 27 and the inverse DCT means 28 inversely quantize the quantized coefficients of each DCT block into the DCT coefficients in the same manner as the normal JPEG decoding procedure, and perform inverse DCT on the quantized coefficients. Further, the color space conversion means 29 performs color space conversion to produce an image.

  Therefore, according to the processing procedure of the detection device 20, the presence / absence of falsification and the falsification location are determined using the presence / absence of error detected by the error detection / decoding means 26, error group, and embedded watermark data as follows. be able to. That is, as shown in FIG. 16, if no error is detected from all the groups, it is determined that the video has not been altered after watermark embedding.

  On the other hand, it is determined that the video in the group in which the error is detected has been tampered with. In this case, as shown in FIG. 17, there is a possibility of falsification in any of the areas occupied by the DCT blocks constituting the group.

  Here, when it is detected by the error detection code that the watermark is broken, it can be determined that the video group (= video range) including the broken watermark data is falsified. This provides a means of knowing which part of the video has been tampered with after watermark embedding.

  The location where the error is detected in this way, that is, the altered video group is output to the monitor or the like by the superimposed display means 30. Here, a mark indicating the position of the falsified group is superimposed on the JPEG-decoded video, and the falsified location can be easily grasped at this point.

3. Variation (variation)
The present invention is not limited to the above-described embodiment, and the number of bits of digital watermark data to be embedded, the type of error detection code, the type of encryption method, the type of random number generation method, and the like are as follows. Various variations can be obtained by changing. Note that the present invention can be applied not only to digital video but also to still images.

<Number of embedded watermark data bits>
(1) The number of bits of the digital watermark data in S26 can take an arbitrary value under the condition that the error detection code can be configured in terms of the error detection code. For example, if it is desired to minimize the data, the data is 1 bit and the parity is 1 bit. The maximum value is determined by the size of the image. For example, when a 640 × 480 pixel VGA image is used, the total number of DCTs including the Y component, the Cb component, and the Cr component is 7200. Therefore, the maximum number of data bits and the error detection code in this case are added to 7200. It becomes the following.
(2) From the aspect of the encryption method, the number of bits is restricted by how the encryption method is selected. As described above, the Camellia cipher is called a block cipher, and converts data having a 128-bit length into one block into 128-bit encrypted code data. In this case, the number of bits including the error detection code is preferably a multiple of the block length 128. The reason why this is not essential and desirable is that even if it is not a multiple of 128, it can be made a multiple of 128 by adding (padding) data of 0 or 1 to the extent that the number of bits is insufficient. However, it can be said that it is more advantageous in terms of application to add some meaningful data than to add meaningless data by padding.
(3) What data to embed depends on the application. For example, when using video as evidence, the camera-specific ID, the shooting time (obtained from the camera clock, if any), the shooting location (by GPS, etc.) May be embedded). Further, in order to specify the alteration location, a serial number from 0 or 1 may be assigned to each group in which the video is grouped and embedded. An example of the data format to be embedded is shown in FIG. The data and format to be embedded are not limited to this example, and can be arbitrarily designed according to application requirements.

<Types of error detection codes>
(1) Row Parity First, a supplementary description will be given of the row parity used for the error detection code in S27. This row parity is the simplest error detection code. As shown in FIG. 10, the original bit data is arranged in a matrix form, and 1 bit of parity bit is added to each row. If the parity is even, the parity bit is selected so that the number of 1s in each row including the parity bit is an even number. For odd parity, the parity bit is selected so that the number of 1s is odd. By adding the parity bit, the total number of bits increases. In the detailed description, since the original data 448 bits are arranged in 64 rows, the total bits increase by 64 bits to become 448 + 64 = 512 bits. Similar to row parity is column parity. The only difference is that the addition of parity is performed in units of columns, and the others are the same as row parity. In addition to column parity, the following can be used for error detection codes such as S27.

(2) Matrix parity Matrix parity is obtained by arranging original bit data in the form of a matrix and adding a parity bit to each row and each column. Also called vertical and horizontal parity. This matrix parity adds more parity bits than the row parity and column parity alone, but if this is adopted in the present invention, the accuracy of error detection becomes higher.

(3) Hamming code According to the Hamming code (Non-patent Document 1, p47 to p53), if the original data bits are k pieces and the check bits (redundant bits) are m pieces, the relationship is 2 ^m −m ≧ k + 1. For example, assuming that the total number of bits is about 500 bits, there are combinations of m = 9 and k = 502. That is, the original data is 502 bits, the inspection bit is 9 bits, and the total bit is 511 bits. Although the Hamming code can not only detect errors but also correct errors, the present invention does not use an error correction function.

(4) Cyclic code (CRC code)
The cyclic code (Non-Patent Document 1 p54 to p65) has a feature that it is easily implemented as hardware. Although this cyclic code can correct not only errors but also errors, the present invention does not use an error correction function.

(5) BCH code A BCH code (Non-patent Document 1, p66 to p75) is a kind of cyclic code and has a high ability to correct random errors. However, since the error correction function is not used in the present invention, the BCH code is used only when the BCH encoding module already exists and there is a cost advantage by diverting it.

(6) RS code (Reed-Solomon code)
The RS code (Non-Patent Document 1 p76 to p81) is a kind of cyclic code and has high burst error correction capability. However, since the error correction function is not used in the present invention, the RS code is used only when the RS encoding module already exists and there is a cost advantage by using it.

<Type of encryption method>
First, a supplementary description of the Camellia cipher used for encryption such as S28 will be given. Camellia cipher is a common key 128-bit block cipher, and a common key is used for encryption and decryption. Here, the 128-bit block cipher means that 128-bit original data can be encrypted into 128-bit encrypted data, and 128-bit encrypted data can be decrypted into 128-bit original data. A common key is required for encryption and decryption. The common key can be 128-bit length, 196-bit length, or 256-bit length. The longer the key length, the higher the cryptographic strength (hardness to break the encryption). The encryption such as S28 is not limited to camellia encryption, and for example, AES encryption can also be used.

  The AES cipher (see Non-Patent Document 3) is a common key 128-bit block cipher, and a common key is used for encryption and decryption. The 128-bit block cipher indicates that 128-bit original data can be encrypted into 128-bit encrypted data, and 128-bit encrypted data can be decrypted into 128-bit original data.

  A common key is required for encryption and decryption. The common key can be 128-bit length, 196-bit length, or 256-bit length. The longer the key length, the higher the cryptographic strength (hardness to break the encryption).

<Type of random number generation method>
The linear congruence method used for the random number generator 10 will be specifically described. The linear congruential method is an algorithm that generates a pseudo-random number sequence. The algorithm is given by the recurrence formula Xn + 1 = (A * Xn + B) mod M. A, B, and M are constants, and M> A, M> B, A> 0, and B> 0. X ₀ is a seed of random numbers, and when this is determined, X ₁ , X ₂ ,. . . Is obtained in a reproducible form. In the random number generator 10, appropriate values for M, A, and B are determined in advance, and the information is shared between the embedding side and the detection side. The random number seed X ₀ may be determined as an appropriate value and shared between the embedding side and the detection side, but the random number generator 10 can easily use the value used for the encryption key as it is. Here, in addition to the linear congruent method, for example, a mixed congruent method is used.

  The mixed congruential method is an algorithm for generating a pseudo-random number sequence. In this algorithm, integer initial value a and value b. c is determined and substituted into the formula ab + c = a ′. A random number is obtained by extracting the same number of digits as a from the center of the obtained a '. Thereafter, the operation of substituting the random number obtained in the above equation a again is repeated to obtain a pseudo-random number sequence. The initial value a is a seed for random numbers. An appropriate value for the seed of the random number may be determined and shared between the embedding side and the detecting side, but the random number generator 10 can easily use the value used for the encryption key as it is.

<Variation of equipment configuration>
A digital video storage device such as a digital camera includes a type manufactured using a camera module in which the image acquisition means 2 (an optical unit such as an optical sensor) and an electronic circuit are integrated. According to this video storage device, each frame image of the video received by the optical unit of the image acquisition means 2 is output from the camera module as JPEG format data. In this case, the DCT coefficients in the middle of JPEG encoding (S22 to S25) cannot be taken out and processed. Therefore, as shown in FIGS. 19 and 20, the JPEG format data of the preprocessing unit 40 (after the processing of S22 to S32) is transmitted to the preprocessing unit 41, and the data shaping / formatting from the JFIF format is performed through the preprocessing unit 41. Entropy decoding (S33.S34) is performed to obtain DCT coefficients. Thereafter, similarly to the flowchart of FIG. 3, the processing steps after S26 of FIG. 20 are executed. According to the configuration change of the embedding device 1, it is possible to change the DCT coefficient similarly to the device configuration of FIG.

The present invention can also be constructed as a falsification detecting digital watermark embedding program for causing a computer to function as the embedding device 1. This program may be made to function as all or part of the respective means 2 to 14 and 2.4 to 14.40.41 of the embedding apparatus 1. According to this program, S21 to S32. S41-S46. S51-S54. S61-S65. S81~S8 executing 6 all or part of various processes such as a computer.

  The present invention can also be constructed as a falsification detection program using a digital watermark for causing a computer to function as the detection device 20. This program may function as all of the means 21 to 30 of the detection apparatus 20, or may function as a part thereof. According to this program, the computer executes all or part of various processes such as S151 to S156.

  Each program can be provided through a network such as a website or e-mail. Each program is recorded on a recording medium such as a CD-ROM, DVD-ROM, CD-R, CD-RW, DVD-R, DVD-RW, MO, HDD, Blu-ray Disk (registered trademark). It can also be stored and distributed. This recording medium is read using a recording medium driving device (such as an optical drive), and the program code itself realizes the processing of the above-described embodiment, so that the recording medium also constitutes the present invention.

DESCRIPTION OF SYMBOLS 1 ... Digital watermark embedding apparatus for alteration detection 2 ... Video acquisition means (photographing means)
3.22.40.41 Pre-processing means 4 Digital watermark data input means 5 Error detection encoding means 6 Encryption means 7 DCT block group dividing means (group dividing means)
8: Maximum value A selection means (maximum value A selection means)
9: B value calculating means (value B calculating means)
10 ... Random number generator (random number generator)
11: Encoded code data embedding means (embedding means)
12 ... Final DCT block calculation value header storage means (header storage means)
13: Entropy compression means (compression means)
14 ... JFIF format shaping means (shaping means)
DESCRIPTION OF SYMBOLS 23 ... Bit data acquisition means 24 ... Encryption / decryption means 25 ... Comparison means 26 ... Error detection decoding means 27 ... Inverse quantization means 28 ... Inverse DCT means 29 ... Color space conversion means 30 ... Error occurrence location superimposition display means

Claims (8)

A method of embedding a digital watermark for tampering detection in the photographed image at the same time as JPEG encoding of the image photographed by the photographing means,
A first step in which preprocessing means performs YCbCr transformation, downsampling , discrete cosine transformation, and DCT coefficient quantization on the captured image, and performs processing so that a quantization coefficient exists in each DCT block;
A second step an error coding means for generating an error detection coding data assigned with redundancy to bit number of the electronic watermark data,
A third step in which an encryption means generates encrypted code data obtained by encrypting the error detection encoded data;
Group dividing means divides each DCT block into groups of the same number as the number of bits of the encrypted code data according to Y, Cb, and Cr color components, and assigns a serial number to each group . Steps,
A fifth step in which a maximum value A selection means selects a value A having the maximum absolute value among the quantized coefficients in each DCT block;
A sixth step in which the value B calculating means calculates a B value based on a result calculated from a random value obtained from the random number generating means and a hash value of a hash function of a numerical value obtained by concatenating the quantization coefficients of the respective DCT blocks; ,
The embedding means sets the value A to “K” in order to make the difference between the value A and the value B even or odd according to the bit value (0 or 1) of the encoded code data to be embedded in each DCT block. , 0, −K ”is added to adjust the value A ;
The compression / shaping means has an eighth step of entropy compression and JFIF format shaping of the quantization coefficient of each DCT block in which the bit value of the encoded code data is embedded;
In the sixth step, when calculating the B value of the nth DCT block, the hash value of the hash function used to calculate the B value of the (n−1) th DCT block, and the DCT block A value obtained by calculating an exclusive OR for each bit with a hash value of a numerical value obtained by concatenating the adjusted quantization coefficients of the value A, from the least significant bit, and setting the nth bit value as the B value; ,
(Final sequence number- 1) a hash value of the hash function used to calculate the B value of the final DCT block of the group, and a numerical value obtained by concatenating the adjusted quantization coefficient of the value A to the DCT block and storing the XOR took the value of each bit of the hash value in the comments in the header of the JPEG format after encryption,
A method of embedding a digital watermark for tampering detection, comprising: A method of embedding a digital watermark for falsification detection in a JPEG format image photographed by a photographing means,
A first step in which preprocessing means performs data shaping and Empiror decoding from the JFIF format on the JPEG format image, and processes each DCT block with a quantized coefficient;
A second step an error coding means for generating an error detection coding data assigned with redundancy to bit number of the electronic watermark data,
A third step in which an encryption means generates encrypted code data obtained by encrypting the error detection encoded data;
Group dividing means divides each DCT block into groups of the same number as the number of bits of the encrypted code data according to Y, Cb, and Cr color components, and assigns a serial number to each group . Steps,
A fifth step in which a maximum value A selection means selects a value A having the maximum absolute value among the quantized coefficients in each DCT block;
A sixth step in which the value B calculating means calculates a B value based on a result calculated from a random value obtained from the random number generating means and a hash value of a hash function of a numerical value obtained by concatenating the quantization coefficients of the respective DCT blocks; ,
The embedding means sets the value A to “K” in order to make the difference between the value A and the value B even or odd according to the bit value (0 or 1) of the encoded code data to be embedded in each DCT block. , 0, −K ”is added to adjust the value A ;
The compression / shaping means has an eighth step of entropy compression and JFIF format shaping of the quantization coefficient of each DCT block in which the bit value of the encoded code data is embedded;
In the sixth step, when calculating the B value of the nth DCT block, the hash value of the hash function used to calculate the B value of the (n−1) th DCT block, and the DCT block A value obtained by calculating an exclusive OR for each bit with a hash value of a numerical value obtained by concatenating the adjusted quantization coefficients of the value A, from the least significant bit, and setting the nth bit value as the B value; ,
(Final sequence number- 1) a hash value of the hash function used to calculate the B value of the final DCT block of the group, and a numerical value obtained by concatenating the adjusted quantization coefficient of the value A to the DCT block and storing the XOR took the value of each bit of the hash value in the comments in the header of the JPEG format after encryption,
A method of embedding a digital watermark for tampering detection, comprising: A method for detecting falsification of a JPEG image in which a digital watermark is embedded by the method according to claim 1,
A first step in which preprocessing means performs data shaping and Empiror decoding from the JFIF format on the input JPEG encoded image, and processes each DCT block to have a quantized coefficient;
Group dividing means, each of said DCT blocks Y, Cb, depending on Cr color component is divided into groups for each bit number and the same number of encrypted tally data used during embedding of the electronic watermark, to each group A second step of assigning a serial number similar to that at the time of embedding the digital watermark ;
A third step in which a maximum value A selection means selects a value A having the maximum absolute value among the quantized coefficients in each DCT block in the same manner as when the digital watermark is embedded ;
Based on the result calculated by the B value calculation means from the random value obtained from the random number generation means in the same way as when the digital watermark is embedded and the hash value of the hash function of the numerical value obtained by concatenating the quantization coefficient of each DCT block. A fourth step of calculating a B value;
Bit data acquisition means acquires a bit value (0 or 1) of the encrypted code data embedded in each DCT block according to whether the difference between the value A and the value B is even or odd. Steps ,
A sixth step in which the encryption / decryption means decrypts the cipher using the same encryption method and encryption key as those used when the encoded code data is embedded with respect to the obtained value (0 or 1) in the fifth step. When,
Error detection decoding means, wherein applying error decoding process on the decryption data in the sixth step, a seventh step of detecting a group containing the generated DCT blocks of the presence or absence and error of an error,
The header stored value comparison means uses the hash value of the hash function used to calculate the B value of the final DCT block of the ( final sequence number- 1) group, and the adjusted quantum of the value A in the DCT block. An eighth step of obtaining a value obtained by performing exclusive OR for each bit with the hash value of the numerical value obtained by concatenating the quantization coefficients;
The header stored value comparing means encrypts the value obtained in the eighth step by the same method as the stored value in the comment field in the header of the JPEG format, and compares the encrypted value with the stored value. Accordingly, a ninth step of detecting an error (final sequence number -1) final DCT block of the group,
The inverse quantization and inverse DCT-color space conversion unit, the following ninth step back to DCT coefficients by inverse quantizing the quantized coefficients of each DCT block, the 10 steps of JPEG decoding through an inverse DCT and color space conversion When,
An eleventh step in which the error occurrence location display means displays the presence / absence of the error detected in the seventh step and the ninth step and the position of the group including the DCT block in which the error has occurred;
A tamper detection method using a digital watermark characterized by comprising: An apparatus that embeds a digital watermark for falsification detection in the photographed image at the same time as JPEG encoding of the image photographed by the photographing means,
Preprocessing means for subjecting the captured image to YCbCr transformation, downsampling , discrete cosine transformation, and DCT coefficient quantization, and processing in a state where quantization coefficients exist in each DCT block;
An error coding means for generating error detection coding data assigned with redundancy to bit number of the electronic watermark data,
Encryption means for generating encrypted code data obtained by encrypting the error detection encoded data;
Group dividing means for dividing each DCT block into groups of the same number as the number of bits of the encoded code data according to Y, Cb, Cr color components, and assigning a serial number to each group ;
Maximum value A selection means for selecting a value A having the maximum absolute value among the quantization coefficients in each DCT block;
A value B calculating means for calculating a B value based on a result calculated from a random number value obtained from the random number generating means and a hash value of a numerical hash function obtained by concatenating the quantization coefficients of the respective DCT blocks ;
In order to make the difference between the value A and the value B even or odd according to the bit value (0 or 1) of the encoded code data embedded in each DCT block, the value A is set to “K, 0, − An embedding unit that adjusts the value A by adding any of K ” ;
Compression / shaping means for entropy compression and JFIF format shaping of the quantization coefficient of each DCT block in which the bit value of the encoded code data is embedded,
The value B calculating means calculates the B value of the n-th DCT block, the hash value of the hash function used to calculate the B value of the (n−1) -th DCT block, and the DCT block A value obtained by calculating an exclusive OR for each bit with a hash value obtained by concatenating the adjusted quantization coefficients of the value A is counted from the least significant bit, and the nth bit value is defined as the B value.
(Final sequence number- 1) a hash value of the hash function used to calculate the B value of the final DCT block of the group, and a numerical value obtained by concatenating the adjusted quantization coefficient of the value A to the DCT block A digital watermark embedding device for alteration detection, wherein a value obtained by exclusive ORing each bit with a hash value is encrypted and stored in a comment field in a header of a JPEG format. An apparatus for embedding a digital watermark for falsification detection in a JPEG format image photographed by photographing means,
A preprocessing unit for the JPEG format images subjected to the data shaping and Enpiro decoding from JFIF format to be processed in a state where there is quantized coefficients to each DCT block,
An error coding means for generating error detection coding data assigned with redundancy to bit number of the electronic watermark data,
Encryption means for generating encrypted code data obtained by encrypting the error detection encoded data;
Group dividing means for dividing each DCT block into groups of the same number as the number of bits of the encoded code data according to Y, Cb, Cr color components, and assigning a serial number to each group ;
Maximum value A selection means for selecting a value A having the maximum absolute value among the quantization coefficients in each DCT block;
A value B calculating means for calculating a B value based on a result calculated from a random number value obtained from the random number generating means and a hash value of a numerical hash function obtained by concatenating the quantization coefficients of the respective DCT blocks ;
In order to make the difference between the value A and the value B even or odd according to the bit value (0 or 1) of the encoded code data embedded in each DCT block, the value A is set to “K, 0, − An embedding unit that adjusts the value A by adding any of K ” ;
Compression / shaping means for entropy compression and JFIF format shaping of the quantization coefficient of each DCT block in which the bit value of the encoded code data is embedded,
The value B calculating means calculates the B value of the n-th DCT block, the hash value of the hash function used to calculate the B value of the (n−1) -th DCT block, and the DCT block A value obtained by calculating an exclusive OR for each bit with a hash value obtained by concatenating the adjusted quantization coefficients of the value A is counted from the least significant bit, and the nth bit value is defined as the B value.
(Final sequence number- 1) a hash value of the hash function used to calculate the B value of the final DCT block of the group, and a numerical value obtained by concatenating the adjusted quantization coefficient of the value A to the DCT block A digital watermark embedding device for alteration detection, wherein a value obtained by exclusive ORing each bit with a hash value is encrypted and stored in a comment field in a header of a JPEG format. An apparatus for detecting tampering of JPEG image watermarked with a device according to any one of claims 4 or 5,
Subjected to a data shaping and Enpiro decoding from JFIF format to the inputted JPEG encoded image, a pre-processing means for processing the conditions existing quantization coefficient to each DCT block,
Each DCT block is divided into groups of the same number as the number of bits of the encoded code data used when embedding the digital watermark according to the Y, Cb, and Cr color components, and the digital watermark is embedded in each group Group dividing means for assigning serial numbers similar to the time ,
Maximum value A selection means for selecting the value A having the maximum absolute value among the quantized coefficients in each DCT block as in the case of embedding the digital watermark ;
B value is calculated based on the result calculated from the random number value obtained from the random number generation means and the hash value of the hash function of the numerical value obtained by concatenating the quantization coefficient of each DCT block as in the case of embedding the digital watermark. A value calculating means;
Bit data acquisition means for acquiring the bit value (0 or 1) of the encrypted code data embedded in each DCT block according to whether the difference between the value A and the value B is even or odd ;
Encryption / decryption means for decrypting a cipher using the same encryption method and encryption key as those used when the encoded code data is embedded with respect to the acquired value (0 or 1);
Error detection decoding means for performing error decoding processing on the decrypted data and detecting a group including a DCT block in which an error exists and an error has occurred;
( Final sequence number- 1) a hash value of the hash function used to calculate the B value of the final DCT block of the group, and a numerical value obtained by concatenating the adjusted quantization coefficient of the value A to the DCT block a value taking the exclusive OR of each bit of the hash value, serial encrypted with the stored value and the same techniques comment field in JPEG format in the header, comparing the stored value and dark Goka value by a header stored value comparison means for detecting an error in the DCT block,
Inverse quantization / inverse DCT / color space conversion means for dequantizing the quantization coefficient of each DCT block to return it to DCT coefficients, and performing JPEG decoding via inverse DCT and color space conversion ;
Error occurrence location display means for displaying the presence / absence of the detected error and the position of the group including the DCT block in which the error has occurred ;
A tamper detection device using a digital watermark, comprising:   The program for functioning a computer as each means which comprises the electronic watermark embedding apparatus for falsification detection of any one of Claim 4 or 5.   The program for functioning a computer as each means which comprises the tampering detection apparatus using the electronic watermark of Claim 6.

Images