DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0072] The preferred embodiments of the present invention will now be described while referring to the accompanying drawings.

[0073] [First Embodiment]

[0074] (1. Electronic Watermark Embedment Apparatus)

[0075] An overview of an electronic watermark embedment apparatus according to this embodiment will now be given while referring to the drawings.

[0076] The electronic watermark embedment apparatus for this embodiment is shown in FIG. 1 . As is shown in FIG. 1 , the electronic watermark embedment apparatus comprises: a color component extraction unit 0101 ; a registration signal embedment unit 0102 ; a pattern arrangement determination unit 0110 ; an embedment position determination unit 0103 ; an additional information embedment unit 0104 ; a color component synthesis unit 0105 ; a JPEG compression encoding unit 0106 ; a memory 0107 ; a JPEG decompression decoding unit 0108 ; and a printer 0109 .

[0077] Image data I, multi-valued image data wherein predetermined multiple bits are allocated for each pixel, are input to the electronic watermark embedment apparatus.

[0078] In this embodiment, the apparatus can cope with the input image data I, regardless of whether the data are gray scale image data or color image data. For the gray scale image data, a pixel has only a single component, while for the color image data, one pixel has three components. In this embodiment, the three components are a red component (R), a green component (G) and a blue component (B), however, other color component combinations can also be employed.

[0079] When the image data I are input to the electronic watermark embedment apparatus, they are first transmitted to the color component extraction unit 0101 .

[0080] There, if the input image data I are color image data, the color component extraction unit 0101 separates the blue component from the color image data, and at the succeeding stage, outputs it to the registration signal embedment unit 0102 .

[0081] At the succeeding stage, the other color components are output to the color component synthesis unit 0105 . That is, only the color component in which the electronic watermark information is to be embedded is extracted and transmitted to the electronic watermarking system.

[0082] In this embodiment, the electronic watermark information is embedded in the blue component. This is because, of the three components, red, green and blue, a human's eyes exhibit the least sensitivity to the blue component. Thus, when electronic watermark information is embedded in the blue component, image deterioration due to the presence of the electronic watermark information is less noticeable to human beings than it would be were the electronic watermark information embedded in one of the other color components.

[0083] When the image data I are gray scale image data, the color component extraction unit 0101 temporarily converts the gray scale data to pseudo color image data.

[0084] In this embodiment, each pixel of pseudo color image data comprises three components (R, G and B), and the three image data components have equal values.

[0085] For this embodiment, the gray scale image data are first converted into pseudo color image data; then, the blue component (B) of the color image data is extracted and is output to the registration signal embedment unit 0102 .

[0086] Subsequently, at the succeeding stage, the other color components are output to the color component synthesis unit 0105 . In this manner, relative to the color image data, electronic watermark information is not embedded in all the color components, but only in the blue component.

[0087] It should be noted that hereafter an explanation will be given in which, to the extent possible, image data I that are color image data are not distinguished from image data I that are gray scale image data. That is, in the following explanation, the color image data are not distinguished from the pseudo color image data.

[0088] A description will now be given for the registration signal embedment unit 0102 , for which, as a pre-process for the extraction of electronic watermark information, a registration signal is used to perform a geometrical correction.

[0089] The registration signal embedment unit 0102 receives image data, from the color component extraction unit 0101 , for the blue component in which it embeds a registration signal, one which is difficult for a human's eyes to discern, using an electronic watermarking technique. Details of this process, including the method used to embed the registration signal, will be described later.

[0090] When the registration signal embedment unit 0102 outputs the image data, it includes the embedded registration signal.

[0091] The pattern arrangement determination unit 0110 employs an image resolution, represented by the input image data and the output resolution of the printer 109 , to determine a pattern arrangement to be used for embedding the electronic watermark information (the additional information), so that the electronic watermark information (the additional information) can be extracted (detected), even after the printer 109 has printed the image data containing the embedded electronic watermark, and even after a density-type tone has been changed to an area-type tone. The method for determining the pattern arrangement will be described later.

[0092] For printing, resolution is defined as the number of pixels contained in one inch of an image (a bit-mapped image), and is used when a predetermined size is employed for the printing of an image. Therefore, when predetermined sizes are used to print specific images, the image that has the most pixels has the highest resolution. To indicate the resolution for an image, the pixel/inch is used as a representative unit.

[0093] The output resolution is used to designate the number of dots the printer 109 deposits in one inch on a print medium. The printer that prints the most dots in one inch is the printer that has the highest output resolution.

[0094] The pattern arrangement determination unit 0110 transmits a selected pattern arrangement with the input image data to the embedment position determination unit 0103 .

[0095] The embedment position determination unit 0103 determines the position at which the additional information Inf is to be embedded in the image data when the registration signal is embedded.

[0096] The embedment position determination unit 0103 outputs, to the additional information embedment unit 0104 , control data representing the position at which the additional information Inf is to be embedded in the image data, together with the input image data and the pattern arrangement.

[0097] In addition to the image data, the pattern arrangement and the control data, the additional information embedment unit 0104 receives the additional information Inf (multiple bits of information). The method by which the additional information Inf is embedded using the electronic watermark technique will also be described later.

[0098] The image data included in the additional information Inf that is to be embedded is output by the additional information embedment unit 0104 to the color component synthesis unit 0105 .

[0099] The color component synthesis unit 0105 synthesizes the blue component processed at the preceding stage (by the additional information embedment unit 0104 ) and the red and green components received directly from the color component extraction unit 0101 to obtain the normal color image data.

[0100] The color image data obtained by the color component synthesis unit 0105 is transmitted to the JPEG compression encoding unit 0106 , whereat the received color image data, which consists of red, blue and green components, is converted into color image data consisting of color difference components, and JPEG compression encoding is performed for the obtained color image data.

[0101] The JPEG compression data obtained by the JPEG compression encoding unit 0106 is temporarily stored in the memory 0107 , and at the succeeding stage, in accordance with the timing for the transmission to an external device or the printing timing, the JPEG compression data is read from the memory and is transmitted to the JPEG decompression decoding unit 0108 . Subsequently, the JPEG decompression decoding unit 0108 decompresses the JPEG compression data and outputs the resultant color image data.

[0102] Finally, the color image data wI is transmitted by the JPEG decompression decoding unit 0108 to the printer 109 , whereat the input color image data is converted to CMYK color components and a half tone process is performed for the CMYK color components, and the resultant data, as printed material pwI, is printed on a print medium, such as paper.

[0103] The possibility exists that the printed material pwI may be obtained by a user other than the user of the apparatus, and an attack, such as geometrical editing like rotation or copying by a copier, is added to the printed material pwI.

[0104] Assume that the printed material that may have been altered is pwI′. The printed material pwI′ is again digitized by a scanner 2001 in FIG. 2 that will be described later.

[0105] The general processing performed by the individual units described above will now be described while referring to the flow chart in FIG. 31 .

[0106] At step 3102 the image data I is input to the color component extraction unit 0101 . This process includes a procedure during which a photo and printed material are read by the scanner 2001 to generate image data. In addition, the blue component is separated from the image that has been read, and is used to input a registration signal at the succeeding stage.

[0107] A registration signal is generated at step 3103 , and is embedded at step 3104 . The registration signal embedment process at step 3104 corresponds to the internal processing performed by the registration signal embedment unit 0102 in FIG. 1 . A detailed explanation for it will be given later.

[0108] At step 3105 , a mask is made, and the made mask is input at step 3106 to define the relation between embedded bit information and the embedded position. At step 3107 , the pattern arrangement determined at step 3111 is input, and the mask is enlarged while also referring to the input pattern arrangement. A detailed explanation for it will be given later.

[0109] At step 3108 , the additional information Inf is embedded in the image data in which the registration signal was embedded at steps 3103 and 3104 . During the additional information embedment process, the additional information Inf is repetitively embedded by the macro block units. This process will be described in detail while referring to FIG. 10 , which will be described later. A macro block is a minimum embedment unit, and all the data for one complete set of additional information Inf is embedded in the image area that corresponds to a macro block.

[0110] At step 3109 , JPEG compression and encoding is performed for the image data in which the additional information Inf has been being embedded, and the resultant image data is stored in the memory 0107 . Further, JPEG decompression and decoding is performed for the image data, and the obtained image data is output as the printed material pwI to the printer 109 .

[0111] (2. Electronic Watermark Extraction Apparatus)

[0112] An overview of an electronic watermark extraction apparatus according to this embodiment will now be given.

[0113] FIG. 2 is a diagram showing the electronic watermark extraction apparatus according to this embodiment. As is shown in FIG. 2 , the electronic watermark extraction apparatus comprises: a scanner 0201 , a color component extraction unit 0202 , a registration unit 0203 and an additional information extraction unit 0204 .

[0114] First, the printed material pwI′ is placed on the document table of the electronic watermark extraction apparatus, and the scanner 0201 scans the printed material pwI′ to generate digital image data wI′. As is described above, the printed material pwI′ may differ from the printed material pwI in FIG. 1 .

[0115] The image data wI′ receives attacks that cause various geometrical distortions of the image data wI. An attack includes scaling, rotation, printing and scanning, and in this embodiment, at least one attack, accompanied by printing and scanning, is {mounted against the image data.

[0116] Therefore, although ideally the image data wI′ and wI will have the same contents, in some cases the contents of these two image data sets will differ greatly.

[0117] The color component extraction unit 0202 receives the image data wI′ and extracts the blue component, and transmits the image data for the blue component to the succeeding registration unit 0203 . Of the image data wI′, color components other than the blue component, i.e., the red and the green components, are not necessary and are abandoned at this time.

[0118] The registration unit 0203 receives the image data wI 1 ′ for the blue component, which was obtained by the color component extraction unit 0202 . The image data wI 1 ′ for the blue component is then employed to generate image data wI 2 ′, for which geometrical distortion has been corrected.

[0119] As is described above, while there is a possibility that the image data wI′ will have a scale different from that of the image data wI, the image data wI 2 ′ always has the same scale as the image data wI. The reason this is true, and the process employed for matching the scales of the image data wI 2 ′ and the image data wI will be described in detail later.

[0120] When the additional information extraction unit 0204 performs a predetermined process that corresponds to the embedment process performed by the additional information embedment unit 0103 , it extracts and outputs the additional information Inf embedded in the image data wI 2 ′.

[0121] The general processing performed by the above described units will now be explained while referring to the flow chart in FIG. 32 . First, at step 3202 the image data wI′ is input. The image data wI′ is obtained by the scanner 0201 when it scans the image data that is predicted to be the printed material pwI′.

[0122] First, only the blue component of the image data wI′ is extracted, for use at the next step. At step 3203 , the scale of the extracted blue component image data wI 1 ′ is corrected. The scaling process used here, an internal process of the registration unit 0203 in FIG. 2 , will be described in detail later.

[0123] At step 3211 , the scaling rate output at step 3203 is employed to determine what pattern arrangement was used to embed the additional information Inf.

[0124] At step 3204 , the offset of the image data wI 1 ′ in the blue component is corrected.

[0125] Then, at step 3206 , for the extraction process using the first pattern arrangement, and at step 3205 , for the extraction process using the second pattern arrangement, the embedded additional information set Inf is extracted from the image data wI 2 ′, for which the scale and the offset have been already corrected.

[0126] At statistic authorization step 3207 , calculations are performed to determine the reliability of the extracted additional information set Inf. If it is ascertained that the correct additional information sets Inf have not been extracted, program control returns to step 3202 , and the image wherein the additional information Inf is assumed to have been embedded is again input. When it is ascertained that the additional information sets Inf are sufficiently correct, at step 3208 the additional information Inf is extracted by performing a comparison process for the two information sets, and at step 3210 , information indicating the reliability of the data is displayed using a reliability index D that will be described later.

[0127] The above described pattern arrangement determination process, offset adjustment process, extraction process using the first pattern arrangement and extraction process using the second pattern arrangement, statistic authorization process, and comparison process are performed internally by the additional information extraction unit 0203 in FIG. 2 , and a detailed explanation for them will be given later.

[0128] (3. Detailed Explanation for the Individual Units)

[0129] The individual units will now be described in detail.

[0130] First, an explanation will be given for the registration process performed at step 3203 by the registration unit 0203 of the electronic watermark extraction system.

[0131] The registration process is a pre-process for the extraction of the electronic watermark, and is performed so that the additional information Inf can be extracted from the image data wI′ received by the electronic watermark extraction apparatus. Generally, the term “registration process” includes not only the scale adjustment process but also the position adjustment process. However, in this embodiment, since the position information embedded as a part of the additional information Inf is employed for the position adjustment process, this process, together with the additional information extraction unit 0204 , will be explained later.

[0132] An explanation will now be given for the change applied to the image data that is processed by the printing system, and for the registration process for the change that is performed by the printing system.

[0133] Here we will discuss that portion of this embodiment wherein the image data wI is printed by an ink-jet printer loaded with yellow (Y), magenta (M), cyan (C) and black (K) inks, and the printed material is scanned by the scanner 0201 .

[0134] At this time, when the output resolution of the printer 0109 differs from the input resolution received from the scanner 0201 , the scale of the original color image data wI differs from the scale of the image data wI′ obtained by scanning. Therefore, there is only a small possibility that the electronic watermark information will be correctly extracted from the obtained image data wI′. Thus, means for correcting the difference in the scales is required.

[0135] In this embodiment, since both the input resolution and the output resolution are already known, the ratio of the scales can be calculated from the ratio of the resolutions. When, for example, the output resolution is 600 dpi (dots per inch) and the input resolution is 300 dpi, the ratio of the scale of an image before printing to the scale of an image after scanning is doubled. Therefore, the scaling is performed for the image data wI′ at the obtained scale ratio using an appropriate scaling algorithm As a result, the same scale can be used to represent the image for the image data wI and the image for the image data wI′.

[0136] However, the output resolution and the input resolution are not already known in all cases. And when the two resolutions are not known, the above method can not be employed. In this case, in addition to means for correcting the difference between the scales, means is required for obtaining the scale ratio.

[0137] When the image that is the source of the image data wI processed by the printing system is scanned by the scanner 0201 , the obtained image is as shown in FIG. 3 . In FIG. 3, a complete image 0301 is represented by the image data wI′. The image 0301 consists of an original image 0302 , represented by the image data wI, and a white margin portion 0303 . This margin is rendered incorrect when a user employs a mouse to cut and paste the image. And the position adjustment process for the position shift caused by scanning is performed by the additional information extraction unit 0204 during the offset adjustment process.

[0138] It is assumed that the above described points will be present in the image data wI′ obtained through the printing system. For when the image data wI is processed by the printing system, these points must be resolved.

[0139] An explanation has been given for a case wherein the image data is obtained after the printing process, as a pre-process for the extraction of an electronic watermark, has been performed at least once. This state can occur in the manual editing process.

[0140] An explanation will now be given for the registration signal embedment unit 0102 and the registration unit 0203 , which are provided in order to resolve the problem that arises due to a difference in the scales, while assuming the ratio of the input and output resolutions is unknown.

[0141] (3-1. Registration Signal Embedment Process)

[0142] First, the registration signal embedment unit 0102 (step 3104 ) will be described in detail.

[0143] The registration signal embedment unit 0102 is positioned at the stage preceding the additional information embedment unit 0104 . The registration signal embedment unit 0102 embeds, in advance, in the original image data a registration signal that is referred to for the registration of the image data wI′ by the registration unit 0203 . The registration signal is embedded in the image data (in this embodiment, the blue component of the color image data) as electronic watermark information that it is almost impossible for a human's eyes to discern.

[0144] FIG. 4 is a diagram showing the internal arrangement of the registration signal embedment unit 0102 . In FIG. 4 , the registration signal embedment unit 0102 comprises a block division unit 0401 , a Fourier conversion unit 0402 , an addition unit 0403 , an inverse Fourier conversion unit 0404 , and a block synthesis unit 0405 . These individual units will now be described in detail.

[0145] The block division unit 0401 divides the input image data into multiple blocks that do not overlap each other. The block size is defined as a power of 2 in this embodiment. Actually, other sizes can be employed; however, when the block size is a power of 2, the Fourier conversion unit 0402 provided at the succeeding stage of the block division means 0401 can perform fast processing.

[0146] The blocks obtained by the block division unit 0401 are sorted into two sets, I 1 , which is transmitted to the succeeding Fourier conversion unit 0402 , and I 2 , which is transmitted to the succeeding block synthesis unit 0405 . In this embodiment, of those blocks that are obtained by the block division unit 0401 , the block in the image data I that is positioned nearest the center is selected as I 1 , and other blocks are selected as I 2 .

[0147] This can be implemented in this embodiment by using a single block, and as there are fewer blocks, the processing time can be reduced. However, the implementation of the present invention is not thus limited, and two or more blocks may be selected.

[0148] Further, information concerning block size, which is used when dividing the image data, and information concerning the selection of a block, in which the registration signal is embedded, must be used in common by the electronic watermark embedment apparatus and the electronic watermark extraction apparatus.

[0149] Part I 1 , of the image data obtained by the block division unit 0401 , is transmitted to the Fourier conversion unit 0402 .

[0150] The Fourier conversion unit 0402 performs a Fourier conversion for the input image data I 1 . The original data form for the input image data I 1 is called the spatial domain, while the data form provided by a Fourier conversion, which is performed for all the input blocks, is called the frequency domain. In this embodiment, since the size of the input block is a power of 2, a fast Fourier conversion is employed to increase the processing speed.

[0151] A fast Fourier conversion is a conversion algorithm that can be executed using (n/2) log 2(n) calculations, while a Fourier conversion requires n×n calculations, wherein n is a positive integer. A fast Fourier conversion and a Fourier conversion differ only in the speed employed to obtain calculation results, and the same results can be acquired using either conversion. Therefore, in the explanation for this embodiment, no distinction is made between a fast Fourier conversion and a Fourier conversion.

[0152] The image data in the frequency domain obtained by a Fourier conversion is presented by using the amplitude spectrum and the phase spectrum, of which only the amplitude spectrum is transmitted to the addition unit 0403 , while the phase spectrum is transmitted to the inverse Fourier conversion unit 0404 .

[0153] The addition unit 0403 will now be described. The addition unit 0403 receives a signal r, called a registration signal, separately from the amplitude spectrum. An example registration signal is an impulse signal, shown in FIG. 5 .

[0154] In FIG. 5 the amplitude spectrum is shown for the two-dimensional spatial frequency components obtained by a Fourier conversion. The center represents the low frequency component, and the surrounding areas, the high frequency component. An amplitude spectrum 0501 represents a signal component included in the original image component, and for signals corresponds to a natural image, such as a photograph, wherein many large signals are concentrated in the low band.

[0155] In the explanation for this embodiment, it is assumed that the process sequence is performed for a natural image. However, the present invention is not thereby limited, and a process sequence may also be performed in the same manner for a document image and for computer graphics (CG). It should be noted that the embodiment of the invention is especially effective for processing a natural image having comparatively many portions and an intermediate density.

[0156] In the example in FIG. 5 , impulse signals 0502 , 0503 , 0504 and 0505 are added to the horizontal and vertical Nyquist frequency components of signals in the frequency domain of the signal 0501 of the original natural image. As is shown in this example, it is preferable that the registration signal be an impulse signal, because the electronic watermark extraction apparatus that will be described later can easily extract only the registration signal.

[0157] In FIG. 5 , the impulse signals are added to the Nyquist frequency components of the input signal; however, the application of the present invention is not thereby limited. That is, all that is required is that the registration signal not be removed, even when the image in which the electronic watermark information is embedded is attacked. Thus, when the impulse signals are embedded in the high frequency components that are to be compressed, the registration signal may be removed by data compression or decompression.

[0158] When the impulse signals are embedded in the low frequency component, compared with when they are embedded in the high frequency components, the impulse signals tend to be perceived as noise because of the visual characteristics of human beings. Therefore, in this embodiment, the impulse signals are embedded in a frequency at an intermediate level, equal to or higher than a first frequency, at which visual identification is difficult for human beings, and equal to or lower than a second frequency, that is not easily removed by irreversible compression and decompression. The registration signal is added to the blocks (one block in this embodiment) transmitted to the addition unit 0403 .

[0159] The addition unit 0403 adds the registration signal to the amplitude spectrum of the image data in the frequency domain, and outputs the resultant signal to the inverse Fourier conversion unit 0404 .

[0160] The inverse Fourier conversion unit 0404 performs an inverse Fourier conversion, for all the input blocks, to obtain the image data for the input frequency domain. Since as for the above Fourier conversion unit 0402 the size of the input block is a power of 2, a fast Fourier conversion is employed to increase the processing speed. The signal in the frequency domain input to the inverse Fourier conversion unit 0404 is converted by an inverse Fourier conversion into a signal in the spatial domain, and the obtained signal is output.

[0161] The image data at the spatial domain output by the inverse Fourier conversion unit 0404 is transmitted to the block synthesis unit 0405 .

[0162] The block synthesis unit 0405 performs a process that is the inverse of the division performed by the block division unit 0401 . Thus, as a result of the process performed by the block synthesis unit 0405 , the image data (blue component) is re-constructed and output.

[0163] The registration signal embedment unit 0102 in FIG. 1 has been described in detail.

[0164] While referring to FIG. 4 , an explanation has been given for the method used to embed a registration signal in a Fourier conversion domain. But there is also a method that is used for embedding the registration signal in the spatial domain. This method will now be described while referring to FIG. 29 .

[0165] The circuit shown in FIG. 29 comprises a block division unit 2901 , an addition unit 2902 , a block synthesis unit 2903 and an inverse Fourier conversion unit 2904 .

[0166] The block division unit 2901 and the block synthesis unit 2903 respectively perform the same operations as those performed by the block division unit 0401 and the block synthesis unit 0405 in FIG. 4 . The image data input to the registration signal embedment unit 0102 is first transmitted to the block division unit 2901 , which divides the image data into multiple blocks that it transmits to the addition unit 2902 . The registration signal r, which here is the signal in the frequency domain, as described in FIG. 5 , is transmitted to the inverse Fourier conversion unit 2904 , whereat an inverse Fourier conversion is used to convert it into a signal r′. The addition unit 2902 receives the blocks from the block division unit 2901 and the signal r′ from the inverse Fourier conversion unit 2904 , and adds them together. The signal from the addition unit 2902 is then transmitted to the block synthesis unit 2903 , and the image data (blue component) is re-constructed and output.

[0167] In the spatial domain, the configuration in FIG. 29 performs the same process as is performed by the configuration in FIG. 4 . But since, unlike the configuration in FIG. 4 , this configuration does not require the Fourier conversion unit, the processing can be performed rapidly.

[0168] Further, in FIG. 29 , the signal r′ is an independent signal for the input image data I. Thus, the calculation of the signal r′, i.e., the operation of the inverse Fourier conversion unit 2904 , need not be performed each time the image data I is input, and the signal r′ can be generated in advance. In this case, the inverse Fourier conversion unit 2904 can also be removed from the configuration in FIG. 29 , and in addition, the registration signal can be embedded at high speed. The registration process referring to the registration signal will be described later.

[0169] <<Patchwork Method>>

[0170] In this embodiment, a principle called the patchwork method is employed in order to embed the additional information Inf. The principle of the patchwork method will be first described.

[0171] According to the patchwork method, the embedment of the additional information Inf is implemented by generating a statistic deviation for an image.

[0172] This will be described while referring to FIG. 30 . In FIG. 30 , pixel subsets 3001 and 3002 and the complete image 3003 are shown, and two subsets, A 3001 and B 3002 , are selected from the complete image 3003 .

[0173] So long as the two selected subsets do not overlap each other, the patchwork method can be used to embed the additional information Inf. It should be noted that the sizes of the two subsets and the selection method employed greatly affect the robustness of the additional information Inf embedded using the patchwork method, i.e., the resistance that ensures the additional information Inf will not be lost as a consequence of an attack mounted on the image data wI. This will be described later.

[0174] Assume the subsets A and B each consist of N elements represented by A={a1, a2, . . . , aN} and B={b1, b2, . . . , bN}. The elements ai and bi of the subsets A and B are pixel values, or sets of pixel values. In this embodiment, a subset corresponds to a part of the blue component in the color image data.

[0175] The index d is defined as follows.

d =1 /N Σ( ai−bi )

[0176] This represents the expectation value of a difference between the pixel values of the two sets.

[0177] When appropriate subsets A and B are selected for a general natural image, and the index d is defined, the characteristic d≅0 is obtained. Hereinafter, d is called a reliability distance.

[0178] As the operation for embedding the individual bits that consist of the additional information Inf,

a′i=ai+c

b′i=bi−c

[0179] is performed. This operation is performed to add a value c to all the elements of the subset A and to subtract a value c from all the elements of the subset B.

[0180] Then, as in the previous case, the subsets A and B are selected from an image in which the additional information Inf has been embedded, and the index d is calculated.

d =1 /N Σ( a′i−b′i )

=1/ n Σ{( ai+c )−( bi−c )}

=1 /n Σ( ai−bi )+2 c

=2 c

[0181] is established, where d is not 0.

[0182] That is, since the reliability distance d is calculated for a specific image that is provided, it can be ascertained that when d≅0, the additional information Inf has not been embedded, and that when the d is a value separated from 0 by a distance equal to or greater than a specific value, the additional information Inf has been embedded.

[0183] The basic principle of the patchwork method has been explained.

[0184] In this embodiment, information consisting of multiple bits is embedded by using the principle of the patchwork method. According to this method, the selection method employed for the subsets A and B is defined by using the pattern arrangement.

[0185] According to the above method, the embedment of the additional information Inf is carried out by the addition or subtraction of the elements of the pattern arrangement relative to a predetermined element of the original image.

[0186] An example simple pattern arrangement is shown in FIG. 9 . In the pattern arrangement in FIG. 9, a change in the pixel value for the original image is represented when 8×8 pixels are referred to in order to embed one bit. As is shown in FIG. 9 , the pattern arrangement is formed of pattern elements composed of positive values, pattern elements composed of negative values, and pattern elements of 0.

[0187] In the pattern in FIG. 9 , the positions designated by the pattern elements +c indicate those whereat the pixel values of corresponding positions are incremented by c, and correspond to the position of the subset A. The positions designated by the pattern elements −c indicate those whereat the pixel values of corresponding positions are decremented by c, and correspond to the position of the subset B. The positions designated by Os indicate those other than the subsets A and B.

[0188] In this embodiment, in order to avoid changing the overall density of an image, the number of positive pattern elements equals the number of negative pattern elements. That is, the sum of all the elements in one pattern element is 0. This condition is always required for the extraction of the additional information Inf, which will be described later.

[0189] The operation for embedding bit information sets that consist of the additional information Inf is performed by using the above pattern arrangement.

[0190] In this embodiment, the pattern in FIG. 9 is arranged multiple times in areas that differ from each other in the original image data, and the pixel values are incremented or decremented, so that information for multiple bits, i.e., the additional information Inf, can be embedded. In other words, the additional information Inf for multiple bits is embedded while assuming not only a combination of the subsets A and B, but also multiple combinations of subsets A′ and B′, subsets A″ and B″, . . . .

[0191] It should be noted in this embodiment that when the original image data is large the additional information Inf is repetitively embedded. This is because, since the patchwork method employs the statistic characteristic, an appropriate number of data is required in order for the statistic characteristic to appear.

[0192] In addition, in this embodiment, the relative positions whereat the pattern arrangement is used for the mutual bits are determined in advance, so that areas wherein the pixel values are changed using the pattern arrangement do not overlap each other when multiple bits are to be embedded. That is, an appropriate relationship is determined for the position in the pattern arrangement for embedding the first bit information of the additional information Inf, and for the position in the pattern arrangement for embedding the second bit information.

[0193] When, for example, the additional information includes 16 bits, the positional relationship for the first to 16th bits in the 8×8 pixel pattern is relatively provided, so that image deterioration is reduced in an area larger than the 32×32 pixel pattern.

[0194] Furthermore, when the image data is large, the additional information Inf (the information consisting of constituent bits) is repetitively embedded, as many times as possible. This is because each bit of the additional information Inf should be correctly extracted. Especially in this embodiment, since the statistical measurement is performed by using the repetitious embedment of the same additional information Inf, information embedment repetition is important.

[0195] The above described selection of the embedment position controls the embedment position determination unit 0103 in FIG. 1 .

[0196] The method for determining the subsets A and B will now be described.

[0197] (3-2. Pattern Arrangement Determination Unit)

[0198] In the patchwork method, the determination of the subsets A and B greatly affects the robustness of the additional information Inf and the quality of an image in which the additional information Inf has been embedded.

[0199] In this embodiment, in FIG. 1 , the additional information Inf is embedded, the image data wI, which has been processed using JPEG compression and decompression, is output by the printer 0109 and is read by the scanner 0201 in FIG. 2 , and the image data wI′ is obtained. Between the processes for generating the image data wI and wI′,various attacks, including printing and scanning, are mounted against the data.

[0200] Here we will discuss how the additional information Inf embedded using the patchwork method obtains the robustness to resist an attack associated with printing.

[0201] According to the patchwork method, the shape of the pattern arrangement and the values of elements are parameters for determining a tradeoff between the embedding robustness of the additional information Inf and the image quality of the image data wI. Therefore, whether the additional information Inf can be extracted from the image data, against which the above attack has been mounted, can be optimized by manipulating the parameters. Later, this will be described in a little more in detail.

[0202] In this embodiment, the basic positional relationship for the patchwork method between the element ai of the subset A and the element bi of the subset B is fixed by the example matrix shown in FIG. 9 .

[0203] It should be noted that the elements ai and bi are not limited to one pixel value, but may also be constituted by sets of multiple pixel values.

[0204] Multiple pattern arrangements are allocated in the image, so that they do not overlap, and the pixels allocated in the image are changed based on the values of the elements in the pattern arrangement.

[0205] Assuming that the subset of the pixels that are changed to the positive value (+c) of the pattern arrangement is defined as A, and that the subset of the pixels that are changed to the negative value (−c) of the pattern arrangement is defined as B, it can be understood that the principle of the patchwork method is employed.

[0206] In the following explanation, the pixels (corresponding to the positions of the elements ai of a subset) having the positive value (+c) of the pattern arrangement are called positive patches, and the pixels (corresponding to the positions of the elements bi of a subset) having the negative value (−c) of the pattern arrangement are called negative patches.

[0207] Hereinafter, a case is presented wherein the positive patches and the negative patches are employed without being distinguished from each other, and a patch in this case indicates either a positive or a negative patch.

[0208] When the size of each patch in the pattern arrangement in FIG. 9 is increased, the value of the reliability distance d according to the patchwork method is increased, as is the robustness of the additional information Inf, but the quality of the image in which the additional information Inf has been embedded is greatly deteriorated, when compared with the original image.

[0209] When the value of each pixel in the pattern arrangement is reduced, the robustness of the additional information Inf is also reduced, but the quality of the image in which the additional information Inf has been embedded is not much deteriorated, when compared with the original image.

[0210] As described above, the optimization of the size of the pattern arrangement in FIG. 9 and of the values of the elements (±c) of the patches in the pattern is very important for robustness and for the quality of the image data wI.

[0211] First, the size of a patch will be explained. When a patch is enlarged, the robustness of the additional information Inf embedded using the patchwork method is increased. And when a patch is made smaller, the robustness of the additional information Inf embedded using the patchwork method is reduced. This is because that the irreversible compression and the printing process provide low-pass filter effects for the overall processing. When a patch is large, a signal that is biased for the embedding of the additional information Inf is embedded as a low frequency signal. Whereas, when the patch is small, a signal that is biased for the embedding of the additional information Inf is embedded as a high frequency signal.

[0212] When the additional information Inf embedded as a high frequency signal is processed by the printing system, a low-pass filter process is performed for the additional information Inf, and the additional information Inf may be deleted. Whereas, even though the printing process is performed, the probability is high that when the additional information Inf is embedded as a low frequency signal, it can be maintained and can be extracted.

[0213] As a result, in order to increase the robustness of the additional information Inf, it is preferable that a large patch be used. However, an increase in the patch size is inversely equal to the addition of a low-frequency signal to the original image data, and this causes the image quality of the image data wI to deteriorate. This is because the visual characteristics of human beings includes the VTF characteristic 1301 shown in FIG. 13 . As is apparent from the VTF characteristic 1301 in FIG. 13, a human's eyes are comparatively sensitive to noise at a low frequency, and comparatively less sensitive to noise at a high frequency. Therefore, it is preferable that the patch size be optimized in order to determine the resistance of the additional information Inf embedded by the patchwork method and the image quality of the image data wI.

[0214] The element values (±c) of the patch will now be described.

[0215] The value (±c) of each element constituting the patch is called a “depth”. When the depth of the patch is increased, the robustness of the additional information Inf embedded using the patchwork method is increased. But when the depth of the patch is decreased, the robustness of the additional information Inf embedded using the patchwork method is reduced.

[0216] The depth of the patch is closely related to the reliability distance d used to extract the additional information Inf. The reliability distance d is a value used to extract the additional information Inf, a process which will be described later. Generally, when the depth of the patch is increased, the reliability distance d is increased and the additional information Inf can easily be extracted. But when the depth of the patch is reduced, the reliability distance d is reduced and the additional information Inf can not easily be extracted.

[0217] As a result, the depth of the patch is also an important parameter when determining the robustness of the additional information Inf and the image quality of an image in which the additional information Inf has been embedded, and is preferably optimized. When patches having an optimal size and an optimal depth are constantly employed, it is possible to embed additional information Inf that has a satisfactory robustness and that can resist an attack, such as irreversible compression or printing, and that causes little deterioration of the image quality.

[0218] Specific path depths and path sizes used for this embodiment will now be described.

[0219] In order to simplify the explanation, a simple printing system process, a gray level transformation using a halftone process, is employed as an example printing process.

[0220] As is described above, the halftone process is a modification method for representing tone. Before and after the halftone process, a human's eyes perceives the same tone. However, since the input unit, such as the scanner 0201 , does not have ambiguous perception like that possessed by human beings, it does not always “perceive” the same tone before and after the halftone process.

[0221] That is, the scanner itself can not determine whether the tone represented by the area-type tone reproduction includes the tone information represented by the original density-type tone reproduction. Thus, here we will discuss which halftone process should be performed, so that the tone represented by the density-type tone reproduction can be represented by the area-type tone reproduction.

[0222] First, the relationship between the density-type tone representation and the area-type tone representation provided by the halftone process will be explained.

[0223] FIG. 43 is a diagram showing a 4×4 dither matrix and an example tone relationship that can be represented by using the matrix. In FIG. 43 , the matrix indicates a tone represented by the area-type tone reproduction, and the tone represented by the matrix is indicated by the number at the bottom.

[0224] There are 16 pixels in the 4×4 matrix. By turning the 16 pixels on or off, 4×4+1=17 levels can be represented.

[0225] Generally, m n dots obtained using the halftone process will represent (m×n+1) levels.

[0226] This will be explained while referring to FIG. 44 . In FIG. 44 , assume that a pixel 4401 is represented by the density level provided by a dynamic range of 0 to 16, and has a value of 8. Four pixels having the same value are arranged vertically and horizontally to generate a 4×4 block 4402 . The halftone process is performed for this block 4402 using an arbitrary 4×4 dither matrix, and binary data 4403 is generated, which is transmitted and output by the printer 0109 . Thereafter, the binary data is again input by the scanner 0201 at the same resolution as the output resolution of the printer 0109 . At this time, assuming that the ratio of the output resolution of the printer 0109 to the input resolution of the scanner 0201 is 1:1, the pixel that is output by the printer 0109 and input by the scanner 0201 is equal to the binary data 4403 . The thus generated image data is binary data 4404 . Using an appropriate interpolation method, the binary data 4404 is scaled down in size to 1/(4×4), and multi-valued data 4405 are generated. This multi-valued data 4405 has a value of 8. When the resolution of the scanner 0201 is not high enough to determine that the binary data 4403 is the binary data 4404 , the binary data 4403 is optically converted into the multi-valued data 4405 .

[0227] While referring to FIG. 44 , the process has been explained in which tone information represented by the density-type tone reproduction is converted into area-type tone information, and when the obtained information is to be represented by the density-type tone reproduction, the tone information is correctly transferred. Generally, when a halftone process is performed using area-type tone reproduction, whereby one pixel is represented by m×n pixels, and when the interpolation process is performed to change the binary data for the m×n pixels into one pixel, the tone information is transferred.

[0228] In this embodiment, to provide for the information the robustness to repel attacks, including the printing and scanning, the size and depth of the patch used for embedding the additional information Inf are designed while taking into account the relationship between area-type tone representation and density-type tone representation. In this embodiment, an explanation will be given for a case wherein for images of various sizes the printer 0109 outputs images having a predetermined size.

[0229] In examples in FIGS. 45A and 45B , two images, 4501 and 4504 , having different resolutions are converted into images, 4503 and 4506 , that have the same size, and are output by the printer 0109 . A process sequence performed for a low-resolution image is shown in FIG. 45 A, and a process sequence performed for a high-resolution image is shown in FIG. 45B .

[0230] First, the enlargement process is performed for the images 4501 and 4504 , so that one pixel corresponds to one dot. An interpolation procedure, such as the nearest neighbor method, is employed as the enlargement method. The nearest neighbor method is a method by which the value of a pixel is copied to a neighboring pixel for enlargement (when an image has a very high resolution, reduction (a thinning process) may be performed). As a result, the image 4501 is enlarged and the image 4052 is obtained, and the image 4504 is enlarged and the image 4505 is obtained. Thereafter, the enlarged images 4502 and 4505 are represented by dots as printed materials (print image data) 4503 and 4506 .

[0231] While in this case the printer 0109 will perform a CMYK conversion process and color matching, in order to simplify the description, no explanation for these processes will be given.

[0232] As is apparent from FIGS. 45A and 45B , when the resolution of an image is low, one pixel can be represented by many dots, and when the resolution of an image is high, one pixel must be represented by a small number of dots.

[0233] An explanation will now be given for an example wherein the embedment of the patches affects the conversion from the density-type tone to the area-type tone. In order for the explanation to be easily understood, the affect image resolution has on the process is not taken into account.

[0234] In FIG. 46 , image areas (subset A) 4601 and 4605 are those that are to be operated on using a positive patch when the additional information Inf is to be embedded in a specific image, and are set to the state preceding the halftone process.

[0235] Image areas (subset B) 4603 and 4607 are those that are to be operated on using a negative patch when the additional information Inf is to be embedded in a specific image, and are set to the state preceding the halftone process.

[0236] The image areas 4601 and 4603 in FIG. 46 are those for which a patch is not used when the additional information Inf is embedded, and the image areas 4605 and 4607 are those for which a patch is used when the additional information Inf is embedded.

[0237] Further, at this time, immediately before the halftone process is initiated, in the image areas 4601 , 4603 , 4605 and 4607 one pixel corresponds to one dot.

[0238] Then, following the initiation of the halftone process, and after area-type tone reproduction has been performed, the images in the areas 4601 , 4603 , 4605 and 4607 are represented by dots in the areas 4602 , 4604 , 4606 and 4608 .

[0239] When the additional information Inf is not embedded, generally it is ascertained that there is almost no change in the difference in the number of ink dots in the area 4602 and the area 4604 . When an image is large and the average of the differences in the ink dots in the patches is calculated, the average is substantially 0.

[0240] When the additional information Inf is embedded, it is anticipated that a difference will appear between the number of ink dots in the area 4606 and the number of ink dots in the area 4608 .

[0241] And when area-type tone reproduction is employed to represent the additional information Inf, the change in the ink dots can be controlled by the design of the patches. It can then be ascertained that the patchwork method provides the robustness required to repel attacks, such as are represented by printing and scanning.

[0242] Further, it can be intuitively understood from FIG. 46 that the number of ink dots is increased both when the area size for embedding the patch is increased and when the depth of the patch is increased.

[0243] The relationship between the patch and the increase in the ink dots will now be described while referring to FIG. 47 . FIG. 47 is a diagram showing a change in the ink dots in accordance with the size and depth of the patch.

[0244] In FIG. 47 , the horizontal axis represents the coefficient value of a dither matrix used to perform a halftone process for a subset A or B for which one pixel has been enlarged to one dot. The vertical axis represents the frequency appearance of the coefficient value of the dither matrix. And in order that the explanation can be easily understood, the horizontal axis also represents the average of pixel values, obtained using the halftone process, in the subset A or B for which one pixel has bees enlarged to one dot.

[0245] As is shown in FIG. 47 , generally, when the coefficient value of the dither matrix corresponds to a large sub-set A or B, the coefficient value is substantially not biased, and the appearance frequency can be regarded as substantially equal.

[0246] Assuming that when the additional information Inf is embedded, an average of 4703 pixel values before the embedment is changed to an average of 4704 pixel values after the embedment, it is understood that, through the binarization of the dither matrix, the ink dots are increased by a number equivalent to the area of a shaded portion 4702 .

[0247] That is, it is apparent that the depth of the patch and the increase in the ink dot count are proportional.

[0248] When the patch size is increased, the appearance frequency of the coefficient value of the dither matrix is further increased. Thus, the area 4702 , the shaded portion, is increased along the axis of the appearance frequency, and it is understood that the increase in the size of the patch is also proportional to the increase in the ink dot count.

[0249] While taking the above characteristic into account, for the entire image, (1) the embedment depth is proportional to the number of dots on the printed material, and (2) the size of the path is proportional to the number of dots on the printed material.

[0250] That is, when Δβ denotes a difference in the dot count in areas throughout the image, which are changed by embedding the patches, where the positive patches are embedded, and in areas where the negative patches are embedded,

Δβ=2×α× PA×C+γ (Equation 47-1).

[0251] In this equation, α denotes a proportional coefficient, γ denotes a margin, C denotes an embedment depth, and PA denotes the area of a positive or negative patch that corresponds to the one-pixel and one-dot relationship for the entire image. The values of α, β and γ are defined through experiment.

[0252] The principle of (Equation 47-1) can be employed not only for the halftone process using the dither matrix, but also for the error diffusion method ((1) and (2), described above, are established).

[0253] In (Equation 47-1), no consideration is given to the resolution of an image, the output resolution of a printer, and the input resolution of a scanner. An explanation will now be given for a case wherein the resolution of the image, the output resolution of the printer 0109 and the input resolution of the scanner 0201 are changed.

[0254] In this embodiment, in order to hold as much information as possible, the input resolution for the scanner 0201 is fixed at 600 ppi (pixels per inch), a satisfactory resolution for a flat-bed scanner.

[0255] Now, the output resolution of the printer 0109 and the resolution of the image will be described.

[0256] As was previously described while referring to FIGS. 45A and 45B , to print an image, the number of dots representing the density-type tone for one pixel is determined based on the resolution of an image. An example will be explained below.

[0257] In FIGS. 45A and 45B , as an example, for image 4501 the dimensions are 1000×800 pixels.

[0258] Assume that the image 4501 is to be reproduced by a printer having an output resolution of 1200 dpi in the main scanning and sub-scanning directions and that it will use 5 inches to print the long side. An enlargement process is performed so that before the halftone process is initiated one pixel equals one dot, and 1200 dpi×5=6000 dots are output for the long side. In this manner, the image 4501 is enlarged to provide the image 4502 , which is 6000×4800 pixels. Then, when the tone of the image 4501 is reproduced from the image 4503 for which the halftone process has been performed, 6×6 dots are used to represent one pixel.

[0259] On the other hand, let us assume that for image 4504 the dimensions are 3000×2400 pixels.

[0260] When the image 4504 is also to be reproduced by a 1200 dpi printer that will use 5 inches to print the long side, similarly, the image 4504 is enlarged to provide image 4505 , which is 6000×4800 pixels and in which one pixel corresponds to one dot. Thereafter, the image 4506 is obtained using the halftone process, and 2×2 dots are used to represent one image 4504 pixel.

[0261] Since it is assumed that the density of an ink dot is determined, when 5×5 dots are used to represent one pixel, the dynamic range for the density-type tone representation using one pixel is large. And when 2×2 dots are used to represent one pixel, the dynamic range for the density-type tone representation using one pixel is small.

[0262] The additional information extraction process will now be described in detail. According to the patchwork method, in the detection process

(the sum of pixel values in the areas wherein the positive patches are embedded)−(the sum of pixel values in the areas wherein the negative patches are embedded)

[0263] is calculated for each pattern arrangement, and an average is obtained for each pattern arrangement for the entire image. This average is called a reliability distance d, and as the reliability distance d is large, the extraction of additional information is ensured.

[0264] FIGS. 48A and 48B are diagrams showing the difference between the positive and negative patch areas for each pattern arrangement. In FIG. 48A, a positive patch area 4801 and a negative patch area 4802 are shown, for which the image resolution is low, and in FIG. 48B, a positive patch area 4803 and a negative patch area 4804 are shown, for which the image resolution is high.

[0265] Since a predetermined density is provided for each ink dot, and since each pixel in each of the patch areas 4801 and 4802 consists of many ink dots, the dynamic range of the density tone is large, even when the reliability distance d has a large value. On the other hand, since each pixel in each of the patch areas 4803 and 4804 is represented by only a few ink dots, the dynamic range of the density tone is small when the value of the reliability distance d is large.

[0266] Generally, when only a small number of dots is used to form one pixel (the image resolution is high), the dynamic range of the gray level for one pixel is small, so that a large reliability distance d can not be obtained, and the additional information Inf can not be extracted.

[0267] Therefore, when the image resolution is high, either the size of the patch or the embedment depth (±c) must be increased.

[0268] Generally, since a position shift constitutes a large problem when the output resolution is high, it is preferable that the patch size be increased.

[0269] When the number of dots in a pattern arrangement required for the detection of additional information is defined as Δβp, based on (Equation 47-1), the relationship of the pixel count P of the positive or negative patch, the embedment depth C of the patch and the dot count m×n representing one pixel is represented by

Δβ p =2 ×α′×P ×( m×n )×( C/ 255)+γ′ (Equation 47-2).

[0270] In this equation, (m×n)×(C/255) means that even when the embedment depth C is changed by the maximum tone levels 255, at the maximum, the number of dots allocated for one pixel is increased only by m×n.

[0271] In the equation, α′ is a proportional coefficient, and γ′ is a margin.

[0272] The m×n count of the dots used to reproduce one pixel is calculated by using

[0273] m×n=(the output resolution of a printer in the main scanning direction/the image resolution)×(the output resolution of a printer in the sub-scanning direction/the image resolution). The value m×n increases as the image resolution is high.

[0274] Therefore, when Δβp, α′ and γr are calculated through experiment, the embedment depth for each pattern arrangement required for the detection of the additional information Inf, and the size (the pattern arrangement size) and the embedment depth of the patch can be obtained from the output resolution of the printer and the image resolution.

[0275] Through the above observations, a method for changing the embedment depth (±c) and the patch size in accordance with the image resolution will now be proposed.

[0276] The apparatus using this method will now be explained.

[0277] The output resolution of the printer 0109 and the image resolution are transmitted to the pattern arrangement determination unit 0110 in FIG. 1 , and an appropriate pattern arrangement, for the extraction of the additional information Inf, is output.

[0278] As an example, assume that the output resolution of the printer 0109 is 120 dpi and the long side of the image is to be printed in a space of about 6 inches. Further, assume that there are images having dimensions ranging from 300 ppi to 600 ppi (1800 to 3600 pixels for the long side).

[0279] The pattern arrangement used for embedding the additional information Inf is selected in accordance with the image resolution. When the image resolution is less than 500 ppi, the pattern arrangement 4901 in FIG. 49 is employed, and when the image resolution is equal to or higher than 500 ppi, the pattern arrangement 4903 in FIG. 49 is employed.

[0280] Presume that (Equation 47-2) is used to determine an appropriate embedment depth.

[0281] After the pattern arrangement determination unit 0110 determines an appropriate pattern arrangement in this manner, based on the size of the pattern arrangement received from the pattern arrangement determination unit 0110 , the embedment position determination unit 0103 determines the embedment position at the succeeding stage. Thereafter, at the pattern arrangement embedment position received from the embedment position determination unit 0103 , the additional information embedment unit 0104 embeds the additional information Inf in the image.

[0282] When the pattern arrangement used for the embedment is unknown, the additional information Inf can not be extracted. Therefore, the pattern arrangement determination unit 2001 , which will be described later, uses the scaling rate output by the registration unit 0203 to determine the image resolution.

[0283] When the output resolution of the printer 0109 is fixed, so long as the image resolution is obtained from the scaling rate, the pattern arrangement determination unit 2001 can determine the pattern arrangement that was used for embedment.

[0284] Therefore, even when, in accordance with the image resolution, the patch and the pattern arrangement are variable, the additional information Inf can be extracted by using the information obtained from the registration signal.

[0285] (3-3. Embedment Position Determination Process)

[0286] FIG. 11 is a diagram showing the internal arrangement of the embedment position determination unit 0103 .

[0287] In FIG. 11, a mask making unit 1101 prepares a mask for designating the embedment positions of individual bit information forming the additional information Inf. The mask is a matrix that includes position information for designating the relative arrangement method for the pattern arrangement (see FIG. 9 ) that corresponds to each bit.

[0288] An example mask 1701 is shown in FIG. 17 . The coefficient values are allocated in the mask, and in the mask, have and equal appearance frequency. Using this mask 1701 , the additional information Inf, consisting of a maximum of 16 bits, can be embedded.

[0289] Next, a mask referring unit 1102 reads the mask 1701 prepared by the mask making unit 1101 , and correlates the coefficient values in the mask 1701 with information associated with the relationship between the bit information and the position of the bits, and determines what arrangement method to employ for the pattern arrangement to be used when embedding the individual bit information.

[0290] Further, at the locations of the coefficient values in the mask 1701 , a mask pattern arrangement correspondence unit 1103 develops the elements (e.g., an 8×8 block) of each pattern arrangement that is received from the pattern arrangement determination unit 0110 at the preceding stage. That is, each coefficient value (one block) of the mask 1701 in FIG. 17 is multiplied by 8×8, as is shown in a block 1703 in FIG. 17 , so that it can be referred to as the embedment position for each pattern arrangement.

[0291] The additional information embedment unit 0104 refers to the embedment start coordinates 1702 in FIG. 17 when embedding the bit information.

[0292] In this embodiment, the mask making unit 1101 prepares the mask 1701 each time image data (the blue component) is received. Thus, when image data for a large size is received, the same additional information Inf is repetitively embedded multiple times.

[0293] According to the above described method, the structure (the arrangement of the coefficient values) of the mask serves as a key for the extraction from the image of the additional information Inf. That is, only the owner of a key can extract the information.

[0294] It should be noted that in addition to a case for preparing a mask in real time, this invention also includes a case wherein a mask is prepared in advance and is stored in the internal storage device of the mask making unit 1101 , so that it can be called up as needed. In this case, the operation can be quickly shifted to the process at the succeeding stage.

[0295] The individual processes performed by the embedment position determination unit 0103 will now be described in detail.

[0296] (3-3-1. Mask Making Unit)

[0297] First, the mask making unit 1101 will be described.

[0298] For the embedment of the additional information Inf using the patchwork method, when the information is added while extensive manipulation of the pixel value is being effected in order to increase the robustness (for example, when a large value c is set for the pattern arrangement), the determination of the image quality, comparatively, is not noticeable at the edge portions whereat the pixel value is drastically changed, while in the flat portion, whereat the pixel value change is less, the portion whereat the pixel value is manipulated is noticeable as noise.

[0299] FIG. 13 is a graph showing the spatial frequency characteristi