JP2006253839A - Apparatus for detecting watermark information, apparatus for embedding watermark information, method of detecting watermark information and method of embedding watermark information - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an apparatus for detecting watermark information for detecting a rotation angle of an image by using only electronic watermark information. <P>SOLUTION: The apparatus 300 has functions of an input section 310, a storage section 320a, a filter processing section 320b, a signal pattern specifying section 320c and a rotation angle detecting section 320d. The storage section 320a previously stores the kind of a signal pattern embedded in an image input by the input section 310, the section 320b applies filter processing to the image by using a plurality of kinds of filters, and the section 320c specifies the kind of the signal pattern embedded in the image on the basis of a result of processing. The section 320d obtains a rotation angle θ of the image from the relationship of angel information each represented by the signal pattern of the kind stored in the section 320a and the signal pattern of the specified kind. In this way, the rotation angle of the image can be highly accurately obtained. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention provides a watermark information detection device that detects a shift (rotation angle) in the rotation direction of an image that occurs when an image in which electronic watermark information is embedded is input, and allows the watermark information detection device to detect the rotation angle of the image. The present invention relates to a watermark information embedding apparatus that embeds digital watermark information into an image.

  When capturing an image in which digital watermark information is embedded using a camera or scanner, the image is often rotated and captured in an unintended direction. For this reason, a digital watermark detection apparatus that corrects unintentional image rotation by detecting the inclination of characters, ruled lines, etc. on a photographed printed material when reading an image with an embedded digital watermark has been proposed. (For example, refer to Patent Document 1).

JP 2004-15396 A

  By the way, the conventional watermark information detecting apparatus can logically detect a rotation angle from 0 degree to 180 degrees. However, the processing load when detecting the rotation angle is very large. Considering this, the conventional watermark information detecting apparatus detects only the rotation angle in the range up to about ± 3 degrees in practice. For this reason, an image inclined more greatly than this range could not be detected. As a result, the user has to carefully place the paper in place so that the image is not tilted and scanned.

  In addition, as described above, the conventional watermark information detection apparatus corrects unintentional image rotation by detecting the inclination of characters, ruled lines, and the like on the photographed printed material. For this reason, for example, in the case of an image having many ruled lines with inclinations other than the horizontal and vertical directions, such as a CAD image, or an image having extremely few characters and ruled lines, the inclination of the image can be detected with high accuracy. could not. As a result, a large error may occur in the detected rotation angle.

  Accordingly, the present invention has been made in view of such problems, and an object of the present invention is to provide a watermark information detection apparatus and watermark information for detecting the rotation angle of an image using only information in the background pattern of the digital watermark. An object of the present invention is to provide a watermark information embedding device and a method for embedding digital watermark information in an image that can cause a detection device to detect the rotation angle of the image.

  In order to solve the above-described problem, according to an aspect of the present invention, watermark information is obtained from a watermarked document image in which a watermark image including at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded. A watermark information detection apparatus for detecting, wherein an input unit for inputting the watermarked document image, a storage unit for preliminarily storing a signal pattern type embedded in the watermarked document image, and the embedded type Filtering the inputted watermarked document image using a plurality of types of filters including a filter capable of specifying a signal pattern, wherein the number of filters is larger than the number of types of the embedded signal pattern. A filter processing unit, and a signal embedded in the watermarked document image based on the result of each filter processing The watermark is determined based on the relationship between the signal pattern specifying unit for specifying the type of turn, the information on the angle represented by the stored type of signal pattern, and the information on the angle represented by the specified type of signal pattern. There is provided a watermark information detection device including a rotation angle detection unit for obtaining a rotation angle of an incoming document image.

  According to this, the rotation angle of an image within a predetermined range (that is, the rotation angle of an input image that is not less than several degrees and not more than 90 degrees) for each certain angle unit specified by the number of filters used for the filter processing. ) Can be detected in all directions (360 degrees). Further, the rotation angle of several degrees to several tens of degrees of the image can be detected using only the ground pattern signal pattern without using characters and ruled lines on the image. The certain angle unit specified by the number of filters used for the filter processing is, for example, 18 degrees (= 180 degrees / 10) when the number of filters is 10.

  In order to solve the above problems, according to another aspect of the present invention, a watermark image including a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded. A watermark information detecting apparatus for detecting watermark information from a watermarked document image, wherein the input is performed using an input unit for inputting the watermarked document image and a plurality of types of filters corresponding to the plurality of types of signal patterns. A filter processing unit for filtering each of the watermarked document images, and calculating a correlation for the special series pattern based on the result of each filter processing for a predetermined range of the input watermarked document image. The correlation calculation processing unit for obtaining the absolute value of the correlation value as the correlation absolute value, respectively, A rotation angle detector for determining the rotation angle of the watermarked document image, watermark information detecting device comprising the provided under the the correlation magnitude.

  According to this, the rotation angle is detected using the absolute value of the correlation value with respect to the synchronization code based on the filter output value. Therefore, it is more resistant to noise (such as dirt or wrinkles on the input image) than when the rotation angle is detected using the filter output value as it is. As a result, the rotation angle of the input image can be detected with higher accuracy. Further, by using only the synchronization code without using characters or ruled lines on the image, it is possible to correct rotation of the image by several degrees to several tens of degrees.

  The plurality of types of signal patterns respectively indicate the predetermined different angles depending on the propagation direction of the dot wave formed by arranging the plurality of dots at a predetermined position and either the wavelength or frequency of the dot wave. You may do it.

  The special sequence pattern may be configured by a code obtained by extending a one-dimensional special sequence to two dimensions.

  The one-dimensional special sequence may be a one-dimensional PN code.

  The special sequence pattern may be composed of a code based on the longest code.

  The special sequence pattern is a pattern in which the absolute value of the autocorrelation value of the special sequence pattern is larger than the correlation absolute value of the special sequence pattern and the sequence pattern obtained by rearranging the special sequence pattern in reverse order. Can be used.

  The special sequence pattern may be a fixed length.

  In addition, the special sequence pattern has an absolute value of the autocorrelation value of the special sequence pattern that is greater than the correlation absolute value of the special sequence pattern and a sequence pattern obtained by shifting the special sequence by a predetermined number of bits. You may be comprised from the big pattern.

  In addition, the rotation angle detection unit can determine the rotation angle of the watermarked image based on the maximum value of the plurality of calculated correlation absolute values.

  The input unit inputs a watermarked document image including a plurality of the special sequence patterns, and the rotation angle detection unit is configured to input the watermark input based on the plurality of obtained correlation absolute values. The positions of two or more special series patterns included in the document image may be specified. Thereby, an image shift (rotation angle) of about several degrees to several tens of degrees of the watermarked document can be obtained from the positional relationship between the two or more specified special series patterns.

  Further, the correlation calculation processing unit calculates a correlation with the sequence pattern in which the special sequence pattern is arranged in the forward direction based on the result of the filter processing for the predetermined range of the input watermarked document image. While calculating, the correlation with respect to the sequence pattern in which the special sequence pattern is arranged in the reverse forward direction may be calculated.

  Thereby, the correlation absolute values are obtained for the sequence patterns in which the special sequence patterns are arranged in the forward and reverse directions. The rotation direction of the watermarked document image can be obtained by comparing the magnitudes of the plurality of correlation absolute values obtained in this way.

  Further, the correlation calculation processing unit scans the special sequence pattern in the horizontal direction based on the result of the filter processing for a predetermined range of the input watermarked document image, thereby the special sequence pattern. You may calculate the correlation with respect to. The correlation calculation processing unit may calculate a correlation with the special sequence pattern by scanning the special sequence pattern in the vertical direction.

  Thus, the special sequence pattern correlation absolute value is obtained by scanning the special sequence pattern in the horizontal direction and the vertical direction. The rotation direction of the watermarked document image can be obtained by comparing the magnitudes of the plurality of correlation absolute values obtained in this way.

  In addition, the correlation calculation processing unit includes a sequence pattern in which the special sequence pattern is arranged in the forward direction and the special pattern based on a result of the filter processing on a predetermined range of the input watermarked document image. The correlation for each of the above series patterns may be calculated by scanning each of the series patterns in which various series patterns are arranged in the reverse direction in the horizontal direction. Further, the correlation calculation processing unit may calculate a correlation for each of the series patterns by scanning each of the series patterns in the vertical direction.

  Thus, the correlation absolute values for the respective sequence patterns are obtained by scanning the special sequence patterns in the forward and reverse directions in the horizontal direction and the vertical direction, respectively. By comparing the magnitudes of the plurality of correlation absolute values thus obtained, the rotation direction of the watermarked document image (rotation angle in units of 90 degrees of the watermarked document image) can be determined with higher accuracy.

  The special series pattern may be a pattern composed of one or more one-dimensional bit patterns generated by a pseudo-random number generator. At this time, the correlation calculation processing unit performs a correlation calculation in the horizontal direction and the vertical direction on one or more one-dimensional bit patterns generated by the pseudo-random number generator, and reverses the one-dimensional bit pattern. For the pattern inverted in the direction, correlation calculation may be performed in each of the horizontal direction and the vertical direction, and the rotation angle of the image may be detected from the magnitude relationship of the absolute values of the correlation values obtained as a result.

  The special series pattern may be a plurality of two-dimensional patterns obtained by combining a single or a plurality of one-dimensional bit patterns generated by a pseudo-random number generator in the horizontal and vertical directions. At this time, the plurality of two-dimensional patterns include a two-dimensional pattern in which the order of one-dimensional bit patterns in both the horizontal direction and the vertical direction or one of them is reversed. Each of them has a rotation angle corresponding to the order in the horizontal direction and the vertical direction, and determines the two-dimensional pattern having the largest absolute value of the correlation value with the pattern (the pattern read from the paper) among the plurality of two-dimensional patterns. Thus, the rotation angle may be detected.

  That is, the special series pattern is generated by combining a bit pattern in which one or more one-dimensional bit patterns generated by a pseudo random number generator are arranged in either the horizontal direction or the vertical direction. The two-dimensional bit pattern in which the arrangement order of the one-dimensional bit patterns arranged in the horizontal direction is reversed with respect to any first two-dimensional bit pattern included in the plurality of two-dimensional bit patterns. A second two-dimensional bit pattern composed of a pattern, and a third two-dimensional bit pattern composed of a bit pattern in which the order of the one-dimensional bit pattern arranged in the horizontal direction and the one-dimensional bit pattern arranged in the vertical direction is reversed. , From a bit pattern in which the order of the one-dimensional bit pattern arranged in the vertical direction is reversed. A fourth two-dimensional bit pattern, may be included.

  At this time, the storage unit has a unique rotation corresponding to each of the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern. A corner is stored in association with each bit pattern, and the correlation calculation processing unit performs the first two-dimensional bit pattern based on the result of the filter processing on a predetermined range of the input watermarked document image. , The second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern are respectively scanned in the horizontal direction to calculate the correlation with each of the series patterns. At the same time, each series pattern is scanned in the vertical direction so as to correspond to each series pattern. Correlating may be calculated, respectively.

  Accordingly, the correlation absolute values for the respective series patterns are obtained by scanning the four patterns in the horizontal direction and the vertical direction, respectively. In this way, the magnitudes of a plurality of correlation absolute values obtained by scanning in the horizontal direction are compared. Further, the magnitudes of the plurality of correlation absolute values obtained by the vertical scanning are compared.

  Based on these comparison results, a two-dimensional bit pattern is selected from the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern. One is identified. As a result, a unique rotation angle stored in the storage unit in association with the specified two-dimensional bit pattern can be obtained as the rotation direction of the watermarked document image.

  Further, the correlation calculation processing unit generates a sequence pattern in which the special sequence pattern is arranged in the forward direction and a sequence pattern in which the special sequence pattern is arranged in the reverse forward direction in the horizontal or vertical direction of the watermarked document image. The correlation absolute value in the forward direction and the correlation absolute value in the reverse direction may be obtained by calculating the correlation with respect to the special series pattern by scanning in either direction.

  The correlation calculation processing unit compares the average value of the correlation absolute value in the forward direction and the average value of the correlation absolute value in the reverse direction thus obtained, and as a result of the comparison, the larger correlation absolute value is set to 1. Dimensional correlation absolute value. Furthermore, the correlation calculation processing unit scans the sequence pattern used for calculating the primary correlation absolute value in either the horizontal direction or the vertical direction of the watermarked document image, and thereby performs the process on the sequence pattern. The correlation absolute value may be obtained by calculating the correlation, and the two-dimensional correlation absolute value may be calculated from the obtained correlation absolute value and the one-dimensional correlation absolute value.

  By comparing the magnitude of the two-dimensional correlation absolute value calculated in this way with the average value of the one-dimensional correlation absolute value, the rotation direction (90 of the watermarked document image is reduced while reducing the processing load. Rotation angle in degrees).

  Further, the signal pattern specifying unit may specify the type of signal pattern embedded in the watermarked document image based on each filter output value output as a result of the filtering process. That is, when each filter output value takes a high value, it can be determined that it is strongly responding to any kind of signal pattern embedded in the watermarked document image. The rotation angle of the watermarked document image can be obtained from the information regarding the angle included in the signal pattern thus determined.

  The plurality of types of filters may be filters that maximize the filter output value when the watermarked document image has different eigenvalues. Thereby, the rotation angle of the watermarked document image can be specified from the filter that has output the maximum filter output value.

  Note that the plurality of types of filters may be Gabor filters.

  In addition, the correlation calculation processing unit weights the correlation calculation for the special sequence pattern with a weight that varies in accordance with each filter output value output as a result of the filter processing, and performs the calculation for the special sequence pattern. A correlation absolute value may be obtained. According to this, the correlation absolute value can be calculated more accurately by weighting the correlation calculation corresponding to each filter output value. As a result, the rotation angle of the watermarked document image can be accurately obtained based on the obtained correlation absolute value.

  The correlation calculation processing unit may calculate a correlation for the special sequence pattern using a fixed value determined based on the result of the filter processing instead of the weighting. According to this, the load of the correlation calculation process is reduced. As a result, the entire processing can be speeded up, so that the rotation angle of the watermarked document image can be obtained at a higher speed.

  The watermark information detection apparatus further weights a plurality of filter output values output as a result of the filter processing by weights corresponding to the rotation angle of the watermarked image detected by the rotation angle detection unit. You may provide the synthetic | combination part which synthesize | combines a filter output value.

  In this case, the synthesis unit may newly generate a filter output value for at least one of the plurality of types of filters used for the filter processing.

  The synthesizing unit may newly generate a filter output value by a filter corresponding to at least one of the plurality of types of signal patterns among the plurality of types of filters used for the filter processing.

  The synthesizing unit may generate filter output values that are approximately equal to the filter output value output as a result of the filter processing by the filter corresponding to the plurality of types of signal patterns when the watermarked document image is not rotated. It may be.

  The synthesizing unit synthesizes filter output values at a plurality of points located in an arbitrary range of the input watermarked document image among the plurality of filter output values, so that the position is within the arbitrary range. One point of filter output value may be generated.

  According to this, in the process of synthesizing the filter outputs, the final filter output is estimated using the filter outputs near the position of interest. As a result, it is possible to compensate for the filter output value of a portion that cannot be read with respect to pixels such as dirt (degradation of the image itself) or characters, etc. As a result, it is possible to reduce errors included in the filter output value caused by deterioration of the watermarked document image itself or deterioration by pixels such as characters. As a result, watermark information can be extracted with high accuracy.

  Further, according to this, since it is not necessary to correct the rotation of the image, it is possible to reduce the amount of filter processing at the time of detecting the rotation for the rotation correction. As a result, the overall processing speed can be increased.

  In order to solve the above problem, according to another aspect of the present invention, there is provided a watermark information embedding device that embeds a watermark image in a document image in order to cause the watermark information detection device to detect watermark information embedded in the image, A document image creation unit for creating a document image, a sequence pattern generation unit for generating a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on predetermined different angles, and the generated A watermark image including the generated special sequence pattern is generated so that the watermark information detection device can detect the rotation angle of the watermarked document image input to the watermark information detection device using a special sequence pattern. A watermark image creation unit for superimposing the created document image and the created watermark image onto the watermarked document image A watermarked document image synthesizing unit to be generated, the watermark information embedding apparatus comprising a provided.

  According to this, it is possible to generate a watermarked document image for allowing the watermark information detecting apparatus to detect the rotation angle of the input image with high accuracy without using characters or ruled lines on the image.

  At this time, the special sequence pattern may have a fixed length.

  In addition, the special sequence pattern has an absolute value of the autocorrelation value of the special sequence pattern based on the correlation absolute value of the special sequence pattern and a sequence pattern obtained by shifting the special sequence pattern by a predetermined number of bits. Larger patterns can be used.

  In addition, the above special sequence pattern has a correlation absolute value in which the absolute value of the autocorrelation value of the special sequence pattern indicates the correlation between the special sequence pattern and the sequence pattern obtained by rearranging the special sequence pattern in reverse order. Larger patterns can be used.

  The special sequence pattern may be configured by a code obtained by extending a one-dimensional special sequence to two dimensions.

  The watermark information embedding device may further include an output unit that prints the created watermarked document image. According to this, a watermarked document image can be printed on paper or the like and output.

  In order to solve the above-described problem, according to another aspect of the present invention, watermark information is obtained from a watermarked document image in which a watermark image including at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded. A watermark information detection method for detecting a watermarked document image, wherein the watermarked document image is input, the type of signal pattern embedded in the input watermarked document image is stored in a storage unit in advance, and the embedded type Each of the input watermarked document images is filtered using a plurality of types of filters including a filter capable of specifying a signal pattern, wherein the number of filters is larger than the number of types of the embedded signal pattern. The type of signal pattern embedded in the watermarked document image is specified based on the result of each filtering process. A watermark for determining the rotation angle of the watermarked document image from the relationship between the information related to the angle represented by the signal pattern of the type stored in the storage unit and the information related to the angle represented by the signal pattern of the specified type An information detection method is provided.

  According to this, the rotation angle of the input image of several degrees or more and 90 degrees or less can be detected in all directions for every certain angle unit specified by the number of filters used for the filter processing. . Further, the rotation angle can be detected using only the ground pattern signal pattern without using characters or ruled lines on the image.

  In order to solve the above problem, according to another aspect of the present invention, a watermark image including a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on predetermined different angles. A watermark information detecting method for detecting watermark information from a watermarked document image in which a watermark is embedded, wherein the watermarked document image is input and input using the plurality of types of filters corresponding to the plurality of types of signal patterns. Each of the watermarked document images is filtered, and a correlation value is calculated by calculating a correlation with respect to the special sequence pattern based on the result of each filtering process on a predetermined range of the input watermarked document image. Are obtained as correlation absolute values, respectively, and the watermarked sentence is based on the plurality of calculated correlation absolute values. Watermark information detecting method for determining the rotation angle of the image is provided.

  According to this, the rotation angle is detected using the correlation value with the synchronization code. As a result, the rotation angle of the input image can be detected with higher accuracy because it is more resistant to noise than when the rotation angle is detected using the filter output value.

  In the watermark information detection method, when the correlation is calculated, the special sequence pattern is arranged in the forward direction based on the result of each filtering process on a predetermined range of the input watermarked document image. Each of the sequence patterns is calculated by scanning in the horizontal direction each of the sequence patterns in which the sequence patterns and the special sequence patterns arranged in the reverse forward direction are scanned in the horizontal direction. Correlation with respect to each series pattern may be calculated by scanning.

  Thus, the rotation direction of the watermarked document image can be obtained by comparing the magnitudes of the plurality of obtained correlation absolute values.

  The special sequence pattern used in the watermark information detection method is generated by arranging one or more one-dimensional bit patterns generated by a pseudo random number generator in either the horizontal direction or the vertical direction. The order of the arrangement of the one-dimensional bit patterns arranged in the horizontal direction is reversed with respect to any first two-dimensional bit pattern included in the plurality of two-dimensional bit patterns. A second two-dimensional bit pattern consisting of the bit pattern, a first two-dimensional bit pattern arranged in the horizontal direction, and a third two-dimensional bit consisting of a bit pattern in which the order of the one-dimensional bit pattern arranged in the vertical direction is reversed. It consists of a pattern and a bit pattern in which the order of the one-dimensional bit pattern array arranged in the vertical direction is reversed. A two-dimensional bit pattern 4 may contain.

  A unique rotation angle corresponding to each of the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern is set in each bit pattern. It may be associated and stored in the storage unit.

  When the correlation is calculated using the four two-dimensional bit patterns, the watermark information detection method uses the filter processing based on the result of each filtering process on a predetermined range of the input watermarked document image. By scanning each series pattern of one two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern in the horizontal direction, each series The correlation absolute value may be obtained by calculating the correlation with each pattern and calculating the correlation with each series pattern by scanning each series pattern in the vertical direction.

  In the watermark information detecting method, the magnitudes of the plurality of correlation absolute values obtained by the horizontal scanning are compared in this way, and the magnitudes of the plurality of correlation absolute values obtained by the vertical scanning are compared. Are compared, and a two-dimensional bit is selected from the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern according to the comparison result. One pattern is specified. The rotation direction of the watermarked document image can be obtained from the specific rotation angle stored in the storage unit in association with the two-dimensional bit pattern specified in this way.

  In the watermark information detection method, when calculating the correlation, a sequence pattern in which the special sequence pattern is arranged in the forward direction and a sequence pattern in which the special sequence pattern is arranged in the reverse forward direction are used as the watermarked document image. The correlation absolute value in the forward direction and the correlation absolute value in the reverse forward direction were obtained by calculating the correlation with the special sequence pattern by scanning either in the horizontal direction or in the vertical direction. In order to calculate the above-mentioned primary correlation absolute value by setting the correlation absolute value which is larger than the average value of the correlation absolute value in the forward direction and the average value of the correlation absolute value in the reverse direction as a one-dimensional correlation absolute value. The sequence pattern is scanned by scanning the used sequence pattern in either the horizontal direction or the vertical direction of the watermarked document image. The absolute value correlation by calculating a correlation for, may calculate the two-dimensional correlation absolute value from the absolute value and the 1-dimensional correlation magnitude correlation obtained.

  The rotation direction of the watermarked document image can be obtained by comparing the two-dimensional correlation absolute value calculated in this way with the average value of the one-dimensional correlation absolute value.

  In the watermark information detecting method, a plurality of filter output values output as a result of the filtering process are weighted by a weight corresponding to the rotation angle of the watermarked image detected by the rotation angle detecting unit. You may make it synthesize | combine a filter output value.

  According to this, in the process of synthesizing the filter output, it is possible to compensate for the filter output value of the portion that cannot be read for pixels such as dirt and characters of the scanned image. As a result, it is possible to reduce errors included in the filter output value caused by deterioration of the watermarked document image itself or deterioration by pixels such as characters. As a result, watermark information can be extracted with high accuracy.

  In order to solve the above problems, according to another aspect of the present invention, there is provided a watermark information embedding method for embedding a watermark image in a document image in order to detect the watermark information embedded in the image. Create and generate a special sequence pattern formed by at least one of multiple types of signal patterns including information on predetermined different angles, and input to the watermark information detection device using the generated special sequence pattern In order to cause the watermark information detecting device to detect the rotation angle of the watermarked document image, a watermark image including the generated special sequence pattern is created, and the created document image and the created watermark are A watermark information embedding method for generating a watermarked document image by superimposing the image is provided.

  According to this, a watermarked document image can be detected so that the watermark information detection device can detect the rotation angle of the input image using only the information in the background pattern of the digital watermark without using characters or ruled lines on the image. Can be generated.

  As described above, according to the present invention, it is possible to detect a rotation angle of an image by using the watermark information detection device and the watermark information detection device that detect the rotation angle of the image using only the information in the background pattern of the digital watermark. It is possible to provide a watermark information embedding device for embedding digital watermark information in an image and a method thereof.

  Hereinafter, in a preferred embodiment of the watermark information processing apparatus of the present invention, (A) when detecting the rotation angle of an input image (watermarked document image) of several degrees to 90 degrees, (B) several degrees When detecting the rotation angle of the input image to the extent, (C) When detecting the rotation angle of the input image in 90 degree units (the rotation direction of the input image), (D) Each filter using the detected rotation angle When combining the output, it will be described in order. In the present specification and the drawings, components having substantially the same configuration are denoted by the same reference numerals, and redundant description is omitted.

<A. When detecting the rotation angle of an input image of several degrees to 90 degrees>
(First embodiment)
First, the watermark information processing apparatus according to the first embodiment of the present invention will be described. The watermark information processing apparatus according to the present embodiment detects the rotation angle of an input image that is not less than several degrees and not more than 90 degrees.

(Configuration and function of watermark information processing apparatus)
The configuration and function of the watermark information processing apparatus according to this embodiment will be described with reference to FIG. The watermark information processing apparatus 10 includes a watermark information embedding apparatus 100 that embeds watermark information in a document image and a watermark information detection apparatus 300 that detects watermark information from a print document 220 in which the watermark information is embedded.

  First, the configuration and function of the watermark information embedding device 100 will be described. The watermark information embedding device 100 is a device that creates a watermarked document image based on the document data 200 and the watermark information 210 and prints it on paper as a print document 220. The watermark information embedding device 100 has a functional block indicated by a document image creation unit 110, a watermark image creation unit 120, a watermarked document image synthesis unit 130, an output unit 140, and a sequence pattern generation unit 150.

  The document data 200 is data that can be visually recognized when printed on paper as a print document 220. The watermark information 210 is data (for example, character string, image data, audio data, etc.) that is embedded in the print document 220 in a format other than characters and cannot be visually recognized in a normal state.

  Based on the document data 200, the document image creation unit 110 creates a document image that can be visually recognized when printed. This document image is a black and white binary image. On the image, white pixels (pixels having a value of 1) are backgrounds, and black pixels (dots, pixels having a value of 0) are character regions (regions to which ink is applied). ).

  The watermark image creation unit 120 digitizes the watermark information 210 and converts it into a numerical value. The result is shown as a bit string of information at the top of FIG. Next, the watermark image creating unit 120 creates a watermark image by assigning a signal pattern composed of a predetermined dot pattern to the converted numerical value (each bit of the bit string). The plurality of types of signal patterns shown in the center of FIG. 2 include information on predetermined different angles.

  The created watermark image is in accordance with a rule having at least one of a plurality of types of signal patterns (signal patterns s1 to s4) as shown in the lower part of FIG. It is arranged.

  The watermarked document image composition unit 130 creates a watermarked document image by superimposing the created document image and the watermark image (by embedding the watermark image in the document image).

  The output unit 140 prints the watermarked document image on paper and outputs it as a printed document 220. The print document 220 is printed by embedding watermark information 210 in the document data 200, and is physically stored and managed.

  The sequence pattern generation unit 150 generates a special sequence pattern formed by at least one of the plurality of types of signal patterns. The sequence pattern generation unit 150 will be described in detail in a second embodiment to be described later.

  In this way, the watermark information embedding device 100 creates and outputs a watermarked document image. The output unit 140 may be an internal function of the watermark information embedding device 100 or may be configured by an external printing device such as a printer.

  Next, the configuration and function of the watermark information detection apparatus 300 will be described. The watermark information detection device 300 is a device that reads the print document 220 as an image and restores the watermark information 210 embedded in the read image. The watermark information detection apparatus 300 has functional blocks indicated by the input unit 310 and the watermark detection unit 320.

  The input unit 310 inputs the print document 220 as a multi-value gradation input image. The watermark detection unit 320 performs a filtering process on the input image to detect a signal pattern embedded in the print document 220, and uses the watermark information 210 embedded based on information included in the detected signal pattern. Restore.

  As shown in FIG. 3, the watermark detection unit 320 includes functional blocks indicated by the storage unit 320a, the filter processing unit 320b, the signal pattern specifying unit 320c, and the rotation angle detection unit 320d.

  The storage unit 320a stores in advance the types of signal patterns embedded in the watermarked document image. The filter processing unit 320b filters each of the watermarked document images using a plurality of types of filters. The signal pattern specifying unit 320c specifies the type of the signal pattern embedded in the watermarked document image based on the result of each filter process. The rotation angle detection unit 320d is based on the relationship between the information related to the angle represented by the type of signal pattern stored in the storage unit 320a and the information related to the angle represented by the type of signal pattern identified by the signal pattern identification unit 320c. The rotation angle of the watermarked document image is obtained.

  In this way, the watermark information detection apparatus 300 detects the rotation angle of a watermarked document image (input image). Note that the input unit 310 may be an internal function of the watermark information detection device 300, or may be configured by an imaging device such as a camera or an input device such as a scanner.

  The watermark information embedding device 100 and the watermark information detection device 300 include a computer mainly composed of a CPU, a ROM, a RAM, and an interface (not shown), and the ROM is used for the watermark information embedding device 100 and the watermark information detection device 300. A program for executing each function is stored. The RAM corresponds to a storage unit 320a that stores signal pattern types, and temporarily stores data such as filter output values. The CPU achieves each function executed by the above-described functional block by executing each program stored in the ROM using data stored in the RAM.

Next, operations of the watermark information embedding device 100 and the watermark information detection device 300 will be described.
(Operation of watermark information embedding device 100)
First, the operation of the watermark information embedding device 100 will be described with reference to FIG. 4 showing the processing flow of the watermark image creation unit 120.

  In step S101, the CPU (watermark image creation unit 120) of the watermark information embedding device 100 converts the watermark information 210 into an N-dimensional code. N is arbitrary, but in the present embodiment, N = 2 (two-dimensional code) is used for ease of explanation. An example of the bit string of the watermark information 210 converted into a two-dimensional code is shown in the upper part of FIG.

  Next, in step S102, the CPU assigns a watermark signal to each dot of the watermark information 210 that has been two-dimensionally encoded. Here, the watermark signal is a signal pattern (see FIG. 2), and represents a wave having an arbitrary wavelength (or frequency) and propagation direction by the arrangement of dots (black pixels).

  In the following, assuming that a rectangle having a width and height of Sw and Sh is one signal pattern, FIG. 5 shows that the distance between dots is dense in the direction of arctan (3) (arctan is the inverse function of tan) with respect to the horizontal axis. And the wave propagation direction is arctan (−1/3).

  FIG. 6 is a cross-sectional view of the change in the pixel value in FIG. 5 as viewed from the direction of arctan (−1/3). In FIG. 6, the portion where the dots are arranged is the antinode of the minimum value of the wave (the point where the amplitude is maximum), and the portion where the dots are not arranged is the antinode of the maximum value of the wave.

  In addition, since there are two areas in which dots are densely arranged in one signal pattern, the frequency per signal pattern is 2 in this example. Since the wave propagation direction is perpendicular to the direction in which the dots are densely arranged, the signal pattern wave in FIG. 5 is arctan (−1/3) with respect to the horizontal direction. Note that a × b = −1 when the direction of arctan (a) and the direction of actan (b) are perpendicular.

  In this way, the signal pattern shows a predetermined angle depending on the wave direction (wave propagation direction) and wavelength (or frequency component) of dots formed by arranging a plurality of dots. Accordingly, the watermark information detection apparatus 300 detects the rotation angle of the input image based on the predetermined angle indicated by the signal pattern included in the input image.

  As described above, as an example of the signal pattern, there are four types of signal patterns shown in the center frame of FIG. (A) shows a signal pattern s1 for recording information "11". The angle θ indicated by the signal pattern s1 is 71.6 degrees. Also, (b) shows a signal pattern s2 for recording information “10”. The angle θ indicated by the signal pattern s2 is 108.4 degrees. Further, (c) and (d) show a signal pattern s3 and a signal pattern s4 for recording information of “01” and “00”. Each angle θ indicated by the signal pattern s3 and the signal pattern s4 is 161.6 degrees and 18.4 degrees. Thus, the signal pattern is a dot pattern having a certain frequency component in the direction of angles such as about 72 degrees, about 108 degrees, about 162 degrees, and about 18 degrees. As for the above-mentioned angles, in the processing of the CPU, the angles of 72 degrees, 108 degrees, 162 degrees, and 18 degrees are used as the angle information in consideration of the processing burden on the CPU.

  Returning to FIG. 4 again, next, in step S103, the CPU arranges the signal pattern assigned in step S102 on the watermark image corresponding to the bit string of the assigned watermark information 210. The watermark information after the placement is shown in the lower part of FIG. As a result, a watermark image that is dot-patterned with the signal pattern is generated based on the watermark information 210.

  On the other hand, the CPU (document image creation unit 110) creates a document image based on the document data 200. Next, the CPU (watermarked document image composition unit 130) creates a watermarked document image by superimposing the watermark image created in this way and the document image. Thereafter, the CPU (output unit 140) prints the watermarked document image on paper and outputs it as a printed document 220.

(Operation of Watermark Information Detection Device 300)
Next, the operation of the watermark information detection apparatus 300 will be described with reference to FIG.

  First, the CPU (input unit 310) of the watermark information detection apparatus 300 inputs a print document to a memory such as a RAM in step S301. The input document image (hereinafter also referred to as a watermarked document image or an input image) is a multi-valued image, and will be described below as a gray image with 256 gradations. Further, the resolution of the input image (resolution when read by the input unit 310) may be different from the print document created by the watermark information embedding device 100, but here it is created by the watermark information embedding device 100. In the following description, it is assumed that the resolution is the same as that of the generated image.

  Next, in step S302, the CPU (watermark detection unit 320) calculates how many signal patterns are embedded from the size of the input image and the size of the signal pattern.

  Next, in step S303, the CPU (watermark detection unit 320) specifies the position (coordinates) of the signal pattern embedded in the input image based on the number of signal patterns calculated in step S302, and extracts the input image. Divide into multiple blocks. Next, in step S304, the CPU (watermark detection unit 320) detects a signal for each signal pattern position, and in step S305, restores the signal pattern matrix. Thereby, the watermark information embedded in the print document 220 is detected.

  Hereinafter, a Gabor filter, which is one of the two-dimensional wavelet filters that can simultaneously define the frequency and direction of the wave of the signal pattern and the influence range in order to detect the signal pattern in step S304, will be described. Note that the two-dimensional wavelet filter does not necessarily need to be a Gabor filter as long as it has the same properties as the Gabor filter.

  The following formula (1) shows the Gabor filter G (x, y), x = 0 to gw−1, and y = 0 to gh−1. gw and gh are filter sizes, which are the same size as the signal pattern embedded by the watermark information embedding apparatus 100 here.

  In the signal detection of the present embodiment, four types of signal patterns (signal patterns s1, s2, s3, s4) embedded in a watermark image and four Gabor filters each having the same frequency, wave propagation direction, and size are included. Gabor filters are used. Specifically, the Gabor filters used at the time of detection include ten types of filters 0 to 9 (see FIG. 8) including Gabor filters a1 to a4 (see FIG. 8) that react strongly to the signal patterns s1 to s4 used at the time of embedding. 9 (a)) is used. The angles respectively indicated by the ten types of filters 0 to 9 are shown below.

Filter 0: about 0 degrees (18 degrees x 0)
Filter 1: about 18 degrees (18 degrees x 1)
Filter 2: about 36 degrees (18 degrees x 2)
Filter 3: about 54 degrees (18 degrees x 3)
Filter 4: about 72 degrees (18 degrees x 4)
Filter 5: about 90 degrees (18 degrees x 5)
Filter 6: about 108 degrees (18 degrees x 6)
Filter 7: about 126 degrees (18 degrees x 7)
Filter 8: Approximately 144 degrees (18 degrees x 8)
Filter 9: about 162 degrees (18 degrees x 9)

  As described above, the rotation angle indicated by each filter increases in order by 18 degrees with the filter 0 as 0 degree. Note that the filters on the detection side corresponding to the filters a1 to a4 on the embedding side in FIG. 8 are the filter 4, the filter 6, the filter 9, and the filter 1, respectively. For example, when the image is not inclined, the signal pattern s1 shown in FIG. 2 reacts strongly to the filter 4, the signal pattern s2 to the filter 6, the signal pattern s3 to the filter 9, and the signal pattern s4 to the filter 1.

  The storage unit 320a stores signal patterns s1 to s4 used at the time of embedding. Therefore, when the filters corresponding to these signal pattern types react strongly (that is, as indicated by the thick arrows in FIG. 9B), the filter output values of the filter 4, filter 6, filter 9 and filter 1 are Since the signal pattern corresponding to the filter 4, filter 6, filter 9, and filter 1 matches the signal pattern stored in the storage unit 320a, the CPU determines the rotation angle of the watermarked document image. Calculate as “0 degree”.

  On the other hand, when the filter 3, filter 5, filter 8 and filter 0 react strongly, information on the angle represented by the type of signal pattern stored in the storage unit 320a and the specified type of signal pattern (filter 3, The CPU calculates the rotation angle of the watermarked document image as “18 degrees” from the relationship with the information of FIG. 9C regarding the angle represented by the filter 5, the signal pattern corresponding to the filter 8 and the filter 0. .

  The number of filters used at the time of detection may be any number as long as it is one or more than the number of filters used at the time of embedding. However, if the number of filters used at the time of detection increases, the detection accuracy improves, but the processing load increases. If the number of filters used at the time of detection decreases, the detection accuracy decreases, but the processing load also decreases. Therefore, in the present embodiment, the number of filters used at the time of detection is 2.5 times the number of filters used at the time of embedding as described above. This numerical value is preferable because all 360 degrees of rotation angles can be detected in 18 degrees without excessively increasing the processing load on the CPU at the time of detection.

  Next, details of the signal detection processing in step S304 in FIG. 7 will be described. As described above, as shown in FIG. 10, in step S304, the CPU performs processing in the order of filter calculation processing (step S304a), rotation angle detection processing (step S304b), and filter output rotation processing (step S304c). Execute.

  First, the details of the filter calculation process (step S304a) will be described. The input image is divided into a plurality of blocks in step S303 in FIG. Therefore, the CPU (filter processing unit 320b) shifts each filter A (A = 0, 1, 2,..., 9) in the horizontal direction and the vertical direction of the input image as shown in FIG. Each block convolution (convolution operation) is calculated. From this result, the CPU (filter processing unit 320b) obtains the maximum filter output value in each block.

  The filter output value in each block is a real filter and an imaginary filter in the case of a Gabor filter (an imaginary filter is a filter whose phase is shifted by a half wavelength from the real filter). And

For example, if the convolution between the real filter of the filter a and the image is Ra and the convolution of the imaginary filter is Ia, the output value F (a) is calculated by the following equation (2).
F (a) = (R _a ² + I _a ² ) ^1/2 (2)

  Based on the ten types of filter output values obtained in this way, the CPU (signal pattern specifying unit 320c) embeds it at a specific position of the input image U (x, y) in the rotation angle detection process of step S304b. Determine the type of the signal pattern. Next, the CPU (rotation angle detector 320d) obtains the rotation angle of the image based on the determination result. The rotation angle detection process in step S304b will be specifically described below.

If the average value of the filter output values corresponding to the filters 0 to 9 is F _ave (0 to 9), and the weight for estimating the rotation angle in units of 18 degrees is w _θ , the weight w is given by It is shown by Formula (3).

For example, the weight when there is no image shift (rotation angle = 0, i = 0) is w ₀ = F _ave (4) + F _ave (6) + F _ave (9) + F _ave (1). In this way, the rotation angle is determined from the value of the weight w _0. Using this rotation angle, the CPU eliminates the inclination of the image by correcting the filter output value in step S304c.

  The details of the signal detection process (step S304) have been described above. Returning to the flowchart of FIG. 7 again, step S305 will be described. In step S305, the CPU (watermark detection unit 320) restores the original information by reconstructing the data code by concatenating the bit information of the signal pattern. FIG. 12 is an explanatory diagram showing an example of information restoration. The steps of information restoration are as follows.

  (1) The CPU uses the bit information (“00”, “01”, “10”) (signal pattern) based on each filter that has the maximum output value in each block of the input image (FIG. 12A). Or, “11”) is extracted (FIG. 12B). The relationship between the filter detected for each block and the bit information extracted from the filter is shown below.

Filter for detecting information “11”: F ((4 + i) mod10, x, y)

Filter for detecting information “10”: F ((6 + i) mod10, x, y)

Filter for detecting information “01”: F ((9 + i) mod10, x, y)

Filter for detecting information “00”: F ((1 + i) mod10, x, y)

  Here, for example, i = 0 when there is no shift in the rotation direction of the image, and i = 1 when there is a shift of 18 degrees in the rotation direction of the image.

(2) Next, the CPU concatenates the bit information detected in (1) to restore the data code (FIG. 12 (c)).
(3) Further, the CPU decodes the data code and takes out the embedded information (FIG. 12 (d)).

  In the above, the watermark information processing apparatus 10 has been described that detects a shift (rotation angle) in the rotation direction of an input image by using one or more filters more than the embedding filter as a detection filter.

  According to this, for every certain angle unit specified by the number of filters on the detection side, the rotation angle of an image in a certain range (that is, the rotation angle of the input image of several degrees to 90 degrees) is all set. It can be detected across the direction (360 degrees). Further, the rotation angle can be detected using only the ground pattern signal pattern without using characters or ruled lines on the image.

However, if i = 0, i = 5, i = 10, and i = 15 are substituted into the equation (3), w ₀ = w ₉₀ = w ₁₈₀ = w ₂₇₀ , so that w _0- the values of the respective values and _w 270 to w ₀ of the values and _w 180 _{to w 270} of each value and _w 90 _{to w 180} of w ₉₀ have the same value. As a result, for example, when the rotation angle is 5 degrees, the rotation angle is 95 degrees, the rotation angle is 185 degrees, and the rotation angle is 275 degrees, it cannot be distinguished. As a result, it is impossible to detect a shift in the rotational direction that is greater than 90 degrees and less than 360 degrees.

  In this embodiment, since the setting unit of the filter is 18 degrees, a rotation angle smaller than 18 degrees cannot be detected. Therefore, in order to detect the rotation angle of the image with higher accuracy, (B) a method of detecting the rotation angle of the input image of about several degrees, and (C) the rotation angle of the input image in units of 90 degrees (90 degrees or more) A rotation angle) is required. Each method of (B) and (C) will be described in the third and subsequent embodiments.

(Second Embodiment)
Next, a second embodiment will be described. In the watermark information processing apparatus 10 according to the present embodiment, a special sequence (PN code, PN: Pseudo-ramdom Number) included in the information obtained from the filter output value is extracted and rotated from the correlation value of this special sequence. It differs from the watermark information processing apparatus 10 according to the first embodiment in that the angle is obtained from the value of the weight w ₀ (filter type). Therefore, the watermark information processing apparatus 10 according to the second embodiment will be described focusing on this difference.

(Configuration of watermark information embedding device 100)
First, the configuration of the watermark information embedding device 100 of this embodiment will be described. As described above, the watermark information embedding device 100 of this embodiment has a functional block in which the sequence pattern generation unit 150 shown in FIG. 1 is added to the function of the watermark information embedding device 100 of the first embodiment. . The sequence pattern generation unit 150 generates a two-dimensional PN code (hereinafter also referred to as a synchronization code). As shown in FIG. 14, the two-dimensional PN code is an example of a special sequence pattern, which is a two-dimensional extension of a one-dimensional PN code sequence.

(Configuration of Watermark Information Detection Device 300)
Next, the configuration of the watermark information detection apparatus 300 of this embodiment will be described. As illustrated in FIG. 13, the watermark information detection apparatus 300 according to the present embodiment includes a functional block in which a correlation calculation processing unit 320 e is added to the watermark information detection apparatus 300 according to the first embodiment. The correlation calculation processing unit 320e obtains a correlation absolute value of a special series included in the information obtained from the filter output value.

(Operation of watermark information embedding device 100)
First, the operation of the watermark information embedding device 100 of this embodiment will be described. The CPU (sequence pattern generation unit 150) generates a two-dimensional PN code that is an extension of the one-dimensional PN code sequence.

  As the one-dimensional PN code sequence, a code sequence having a sufficiently low correlation with a code in which codes are arranged in reverse order in the front-rear direction is used. For example, when the one-dimensional PN code is 15-bit information consisting of “111101011001000”, the one-dimensional PN code in which the one-dimensional PN code is arranged in reverse order is 15-bit information consisting of “000100110101111”.

  Thus, for example, with respect to the bit pattern “111101011001000”, the pattern “111101011001000” as it is is called a bit pattern in the forward direction. A pattern “000100110101111” obtained by reversing the order of the bit pattern “111101011001000” is referred to as a reverse forward bit pattern. That is, the reverse forward bit pattern is a pattern in which forward forward bit patterns are arranged from the reverse direction.

The following formula (4) is used to correlate each combination of forward order and forward order and forward order and reverse order.

  As a result, the correlation value R between the normal order and the normal order is “1”. The correlation value R between the normal order and the reverse order is “0.3333...”. In this way, the correlation absolute values in the same order have higher signs than the correlation absolute values in different orders.

  Next, the CPU (watermark image creation unit 120) uses a special code formed by at least one of a plurality of types of signal patterns including information on predetermined different angles in the synchronization code composed of the two-dimensionally expanded PN code. A watermark image (synchronized code dot pattern in FIG. 15) is created by applying a series pattern. As shown in FIG. 15, the synchronization code itself is composed of a set of signal patterns (dot patterns), and is converted into a signal pattern in the same manner as digital information normally embedded in an image, and one or more embedded.

  Next, the CPU (watermarked document image composition unit 130) creates a watermarked document image by embedding the generated watermark image in the document image.

  For example, when the CPU embeds a synchronization code in a part of an image, for example, when one signal pattern represents 1-bit information, a signal pattern indicating information “0” and a signal pattern indicating information “1” Both may be embedded in combination, or only either a signal pattern indicating information “0” or a signal pattern indicating information “1” may be embedded.

When one signal represents multiple bits (for example, when there are 2 ⁿ signal patterns, n signal information can be represented by one signal pattern), 1 bit among the multiple bits is assigned to the synchronization code, Other bits can be used for information embedding. For example, when n = 2, as shown in FIG. 2, 2-bit information can be expressed by one signal pattern. Therefore, the LSB (Least Significant Bit) side (lower side) bits of each signal pattern may be assigned to the synchronization code. Further, the MSB (Most Significant Bit) side (upper side) bit of each signal pattern may be assigned to the synchronization code. When the LSB side (lower order) bits of each signal pattern are assigned to the synchronization code, the signal pattern s1 and the signal pattern s3 indicate information “1”, and the signal pattern s2 and the signal pattern s4 indicate information “0”. Show.

  At this time, the synchronization code does not change the absolute value of the correlation even when all bits are inverted, and the information when all bits are not inverted is "0", and the information when all bits are inverted is "1". it can. Alternatively, information when all bits are not inverted can be set to “1”, and information when all bits are inverted can be set to “0”.

  Therefore, the synchronization code can be used for embedding 1-bit information. That is, in the case of this embodiment, this 1 bit is diffused and embedded in the entire two-dimensional PN code. However, when the synchronization code is used only for rotation detection and synchronization detection, which are the purposes of this embodiment. The synchronization code may be spread and embedded or may be embedded without spreading.

  In this way, all information including the synchronization code is drawn on the image using the corresponding signal pattern of FIG. As described above, these signal patterns have a frequency of about 72 degrees for the signal pattern s1, about 108 degrees for the signal pattern s2, about 162 degrees for the signal pattern s3, and about 18 degrees for the signal pattern s4. A pattern with components is used.

(Operation of Watermark Information Detection Device 300)
Next, the operation of the watermark information detection apparatus 300 of this embodiment will be described. The CPU executes the signal detection process (step S304) after executing the image input process (step S301) to the signal detection coordinate specifying process (step S303) in FIG. In this signal detection process, the CPU (filter processing unit 320b) first performs the filter calculation process in step S304a of FIG.

  The filtering process will be described with reference to FIG. 16 showing details of the filter output value F (M, x, y) at each point. The filter output value F (M, x, y) at each point is calculated as follows. First, the CPU (filter processing unit 320b) performs a convolution operation while scanning the Gabor filter B on the image R that is a part of the input image. The result is shown as a filter output value F (B, x, y).

  The filter output value F (B, x, y) becomes the maximum value (indicating that the filter output value is higher as it becomes black) at the center position within the dotted line of the image R shown in the upper part of FIG. This is because the Gabor filter B reacts strongly to the characteristic portion within the dotted line of the image R.

  Similarly, the CPU (filter processing unit 320b) performs a convolution operation while scanning the image R with the Gabor filter A. The result is shown as a filter output value F (A, x, y). The reason why the filter output value F (A, x, y) reaches the maximum value at the center position in the solid line of the image R shown in the upper part of FIG. This is because it reacts strongly.

  As a result, the filter that outputs the maximum value of each point is shown in the lower left of FIG. The portion that reacts strongly with the filter B is shown in white, and the portion that reacts strongly with the filter A is shown in black. Also, the filter output value at each point regardless of the filter type is shown in the lower right of FIG. In this way, as a result of performing filter processing by each filter, filter output values corresponding to each filter are obtained.

  Next, the CPU (correlation calculation processing unit 320e) performs a two-dimensional correlation calculation corresponding to the filter processing result (filter output value) in the rotation angle detection processing in step S304b. For example, after the filter result is rotated by an angle corresponding to each rotation angle, a two-dimensional correlation with the synchronization code may be calculated in the horizontal direction and / or the vertical direction.

However, at the time of correlation calculation, the filter indicating the information “0” and the information “1” is switched as follows according to each rotation angle.
Filter for detecting information “11”: F ((4 + i) mod10, x, y)
Filter for detecting information “10”: F ((6 + i) mod10, x, y)
Filter for detecting information “01”: F ((9 + i) mod10, x, y)
Filter for detecting information “00”: F ((1 + i) mod10, x, y)

For example, when the synchronization code is embedded on the LSB side, the synchronization code information “0” or the synchronization code information “1” is as follows.
Filter for detecting information “1”: F ((4 + i) mod10, x, y) or F ((9 + i) mod10, x, y)
Filter for detecting information “0”: F ((6 + i) mod10, x, y) or F ((1 + i) mod10, x, y)

  Next, a two-dimensional correlation calculation method for extracting a special sequence (synchronization code) based on the filter output value thus obtained and obtaining the correlation absolute value of the special sequence will be described in detail.

  The CPU (correlation calculation processing unit 320e) extracts the PN code used at the time of embedding from the filter output value (output value of expression (1)) in the two-dimensional correlation calculation process, and calculates the correlation absolute value of this PN code. To do.

  At this time, if the filter output value can be obtained with multiple bits per signal pattern (for example, if one signal pattern has 2-bit information as shown in FIG. 2), the bit plane used for embedding (LSB side bit or MSB side bit) is used for correlation calculation.

  For example, when two types of filters, filter A and filter B, are used, the input values used for the two-dimensional correlation calculation are as follows. For example, when the filter output value F (A, x, y) is greater than or equal to the filter output value F (B, x, y), the input value used for the two-dimensional correlation calculation is F obtained using the equation (1). (A, x, y). On the other hand, when the filter output value F (A, x, y) is smaller than the filter output value F (B, x, y), the input value used for the two-dimensional correlation calculation is −F (B, x, y). .

  When there are four types of filters, four types of digital values of “11”, “10”, “01”, and “00” can be expressed depending on the type of each filter. Among the digital values corresponding to the largest filter output value, When the bit plane used to use the synchronization code is “1”, it may be F (M, x, y), and when it is “0”, it may be −F (M, x, y). However, F (M, x, y) indicates the output value of the filter M having the largest output value.

  Specifically, for the signal pattern indicating 2-bit information, the information indicated by the signal pattern s1 corresponding to the filter A is “11”, the information indicated by the signal pattern s2 corresponding to the filter B is “10”, and the filter C is indicated. Information indicated by the corresponding signal pattern s3 is “01”, and information indicated by the signal pattern s4 corresponding to the filter D is “00”. At this time, when the synchronization code is embedded in the MSB side bits of each of the four signal patterns (when the bit plane used for embedding is the MSB side bit), the bit plane is “1” indicated by the filters A and B. In this case, the output value F (M, x, y) of the filter M having the largest output value is used for the two-dimensional correlation calculation. On the other hand, when the bit plane is “0” indicated by the filters C and D, −F (M, x, y) is obtained by adding a sign to the output value F (M, x, y) of the filter M having the largest output value. y) is used for the two-dimensional correlation calculation.

Here, as shown in FIG. 17, when the interval of signals existing in the filter result is p, the correlation calculation is performed in units of sets that are separated by p in the filter result. For example, when calculating the correlation with an N-bit synchronization code, the filter output values to be used are {F (M _{x, y} , x, y), F (M _{x + p, y} , x + p, y), F (M _{x + 2p, y} , x + 2p, y),..., F (M _{x + Np, y} , x + Np, y)}.

  For example, as shown in FIG. 18, when one signal is generated by 15 pixels × 15 pixels and printed and scanned at the same resolution, one signal becomes 15 pixels × 15 pixels on the input image. If the filter processing is performed while shifting this pixel by pixel in the horizontal direction without thinning, the signal interval p in the filter processing result is “15”. By performing a correlation calculation at this interval p, a horizontal correlation value (one-dimensional correlation value) is obtained. The horizontal correlation absolute value is obtained as the absolute value of the horizontal correlation value.

  Next, as shown in FIG. 19, when one signal generated by 15 pixels × 15 pixels is subjected to filter processing while shifting in the vertical direction one pixel at a time without any thinning, the signal interval in the filter processing result is obtained. p is “15”. A two-dimensional correlation value is obtained by performing a correlation calculation in the vertical direction on the correlation calculation result in the horizontal direction at this interval p. The two-dimensional correlation absolute value is obtained as the absolute value of the two-dimensional correlation value.

  In this way, correlation calculation processing can be performed without performing synchronization processing for searching for the peak of the filter output value, and synchronization code loss due to peak search synchronization errors can be prevented.

Summarizing the above two-dimensional correlation calculation into an equation, the correlation value R (x, y) at the position (x, y) is calculated by the following equation (5).

However,
p: Signal interval M _{x + pi, y + pj} in the filter result: filter F (M _{x + pi, y + pj} , x + pi, y + pj) having the highest filter output value at the position (x + pi, y + pj): most filter output at the position (x + pi, y + pj) Output value of filter with high value

  The correlation value R (x, y) is a two-dimensional correlation calculation that combines a horizontal correlation calculation and a vertical correlation calculation. For the horizontal correlation calculation and the vertical correlation calculation shown below, Can be divided.

That is, the horizontal correlation calculation is expressed by the following equation.

Also, the correlation calculation in the vertical direction is expressed by the following equation.

Alternatively, for simplification of processing, the following equation (6) can be used.

  The input value used for the two-dimensional correlation calculation in the case of simplification as shown in this equation (6) is, for example, the filter output value F (A, x, y) when two types of filters, filter A and filter B, are used. Is greater than or equal to the filter output value F (B, x, y), the value “1” (= sgn (1)) is taken, and the filter output value F (A, x, y) becomes the filter output value F ( If it is smaller than B, x, y), it takes a value of “−1” (= sgn (0)).

  The two-dimensional correlation calculation using the input value thus obtained is performed by shifting the filter output result by one to several dots. The CPU (correlation calculation processing unit 320e) may perform the above-described two-dimensional correlation calculation by sequentially calculating each one-dimensional correlation calculation using equation (5) in step S304b of FIG. However, the two-dimensional correlation calculation described above may be performed by calculating at once using Equation (6).

In the case of an N × N sized synchronization code obtained by extending the longest N-bit code in two dimensions, if the phases (that is, the rotation angles of the images) completely match, the correlation absolute value | R (x, y) | If a value of “1” is taken and neither the horizontal direction nor the vertical direction match at all, the correlation absolute value | R (x, y) | takes a value of 1 / N ² . In other words, of the filter output values obtained by the CPU (filter processing unit 320b), only the portion that actually matches the synchronization code has a high correlation absolute value, and the non-matching portion has a low correlation absolute value.

  For example, as described above, in FIG. 18 and FIG. 19, the PN code of the synchronization code is scanned in the horizontal direction and the vertical direction bit by bit with respect to the result of the filter processing, respectively. As a result of calculating the correlation value for, the correlation value is “1” at the position that matches the PN code of the synchronization code, and the correlation is the highest. The correlation value decreases as the number of mismatches between the synchronization code and the PN code increases. Using the relationship between the correlation absolute value and the rotation angle of the image, a specific rotation angle is obtained from the correlation absolute value.

  Therefore, first, the CPU (correlation calculation processing unit 320e) calculates the absolute value | R (x, y) | of the synchronization code corresponding to each rotation angle by the above-described two-dimensional correlation calculation. Thereafter, the CPU (rotation angle detection unit 320d) displays the unique rotation angle corresponding to the maximum value of the correlation absolute value | R (x, y) | calculated based on the filter processing result for the predetermined range of the input image. It is detected as the rotation angle.

  Here, when one of a plurality of synchronization codes is embedded in the image, the predetermined range of the input image may be a range of the entire input image. On the other hand, as shown in FIG. 20, when a plurality of synchronization codes are embedded in an image, the predetermined range of the input image is a predetermined range calculated from the distance between adjacent synchronization codes (for example, the synchronization code is A rectangular range in which one side in the horizontal direction is indicated by Bw and one side in the vertical direction is indicated by Bh as a center.

  Next, the CPU eliminates the inclination of the image by correcting the filter output value based on the detected rotation angle of the image in step S304c of FIG.

Finally, in the information extraction (restoration) stage (step S305 in FIG. 7), the CPU extracts information corresponding to the filter as follows.
Filter for detecting information “11”: F ((4 + i) mod10, x, y)

Filter for detecting information “10”: F ((6 + i) mod10, x, y)

Filter for detecting information “01”: F ((9 + i) mod10, x, y)

Filter for detecting information “00”: F ((1 + i) mod10, x, y)

  As described above, according to the present embodiment, the rotation angle of the image of several degrees or more and 90 or less is detected by using only the synchronization code without using characters or ruled lines on the image. Thereby, the rotation of the image can be corrected from the detected rotation angle. In particular, since the rotation angle is detected using the correlation value with the synchronization code, it is more resistant to noise than when the rotation angle is detected using the filter output value as in the first embodiment. As a result, the rotation angle can be detected with higher accuracy.

<B. When detecting the rotation angle of an input image of several degrees>
(Third embodiment)
Next, a third embodiment will be described. In the watermark information processing apparatus 10 according to the present embodiment, one or more synchronizations embedded in an image are detected in that a rotation angle of about several degrees of the input image is detected based on a plurality of synchronization codes embedded in the image. This is functionally different from the watermark information processing apparatus 10 according to the second embodiment in which a rotation angle of several degrees to 90 degrees is detected based on the code. In the following, this difference will be mainly described.

(Operation of watermark information embedding device 100)
In the present embodiment, the CPU (watermarked document image composition unit 130) of the watermark information embedding device 100 places a plurality of synchronization codes in an image (output image) as shown in FIGS. The positions where the synchronization codes are arranged may be equally spaced or non-equally spaced. However, the detection side (watermark information detection apparatus 300) needs to be able to grasp each position of a plurality of synchronization codes embedded in an input image. In FIGS. 20 and 21, the synchronization codes are arranged at equal intervals. Other information may be embedded at a position where the synchronization code is not embedded.

  As shown in the middle frame of FIG. 21, when one signal pattern represents 1-bit information, the CPU combines a signal pattern indicating information “0” and a signal pattern indicating information “1”. Thus, the synchronization code may be generated, and the generated synchronization code may be embedded in the image.

When a single signal pattern represents information of a plurality of bits, one bit among the plurality of bits is assigned to the synchronization code, and the other bits are used for information embedding. For example, when there are 2 ⁿ signal patterns, one signal pattern can represent n-bit information. As an example, in the case of the signal pattern shown in the frame of FIG. 2, 2-bit information can be expressed by four signal patterns.

(Operation of Watermark Information Detection Device 300)
As described above, the CPU (correlation calculation processing unit 320e) of the watermark information detection apparatus 300 uses the two-dimensional correlation calculation shown in Expression (5) or Expression (6) to calculate the correlation absolute value | R (x , Y) | The detection of the synchronization position by the correlation calculation of the synchronization code can be executed without being affected by the rotation of the image several degrees. The reason will be described below.

  When a tilted image is filtered, the filter processing result is tilted in the same manner as the image tilt. FIG. 22 shows an example of the result of filtering an image with the rotation angle θ tilted by about 3 degrees. Assuming that the filter output peak values for each signal pattern are evenly arranged as a result of the filter processing, the horizontal direction PN code of length N has a width pN on the filter result, as shown in a simplified manner in FIG. The rectangle with the height p is inclined.

  Even if the CPU (correlation calculation processing unit 320e) calculates the correlation in the horizontal direction as in the case where the image is not inclined with respect to the input image in such a state, the CPU (correlation calculation processing unit 320e) A slight inclination does not extract the wrong signal. However, since the CPU (correlation calculation processing unit 320e) does not know the inclination of the image, it may happen that the output value of a different signal is referred to the filter output result depending on the location.

For example, if the center of the center signal in FIG. 22 is completely coincident, the inclination of the image is θ, the length of the synchronization code is N (N signal), and the signal interval is p, the reading positions at both ends are The deviation d from the signal center can be expressed by the following equation (7).

For example, it is assumed that a sufficiently large range of the filter output value is a range of αp / 2 (within a circle in each signal) from the center of the filter. However, α> 0 and in many cases α ≦ 1.0. At this time,
In the range where the above condition is satisfied, the filter output is sufficiently strong and the information accuracy is high, so that the information of the synchronization code can be read. As a result, the CPU (correlation calculation processing unit 320e) can obtain the correlation code absolute value | R (x, y) | of the synchronization code without being affected by the deviation caused by the rotation.

As described above, the synchronization code is two-dimensional. For this reason, as shown in FIG. 23, the signal at the position farthest from the center of the synchronization code becomes the signal at the four corners, and is positioned slightly far from the signal at the center. In this case, the deviation between the signal reading positions at the four corners and the center of the actual signal can be expressed by the following equation (8).

As described above, if the range where the filter output value is sufficiently large is the range of αp / 2 from the center of the filter (within the circle of each signal),
In the range where the above condition is satisfied, the filter output is sufficiently strong and the information accuracy is high, so that the information of the synchronization code can be read.

  As a result, the CPU (correlation calculation processing unit 320e) can obtain the correlation code absolute value | R (x, y) | of the synchronization code without being affected by the deviation caused by the rotation.

When the above equation (8) is calculated backward, the following equation (9) is calculated. Therefore, in the range where the rotation angle θ of the image satisfies Expression (9), the rotation of the image hardly affects the information extraction of the synchronization code.

  Substituting “0.8” into α in the above equation (9) and “15” into N (indicating that the synchronization code is 15 pixels × 15 pixels (15 signals × 15 signals)), the maximum rotation angle θ The value is about 2.3 degrees. That is, it can be seen that an image inclination of about 2.3 degrees has little effect on the extraction of the 15 × 15 signal size synchronization code. Note that the value of α is a parameter that depends on the printing environment and the scanning environment, and should be a value that allows practically sufficient filter output values assuming the printing environment and the scanning environment.

  For example, in the case of the rotation angle θ = 3.2 degrees, the central 11 signals × 11 signals among the synchronization code 15 signals × 15 signals (15 pixels × 15 pixels) are detected with almost no error. When the rotation angle θ = 4 degrees, the center 9 signals × 9 signals among the synchronization code 15 signals × 15 signals are detected with almost no error. This can be said to be long enough to detect the rotation angle when the longest code of 15 bits is used as the synchronization code. For the above reason, in the following description, the CPU (correlation calculation processing unit 320e) performs two-dimensional correlation calculation without regard to image rotation.

  Here, as a synchronization code for the two-dimensional correlation calculation, a code (code string) having an autocorrelation absolute value with the synchronization code larger than a correlation absolute value between the synchronization code and other sequences is used. The autocorrelation absolute value is usually the maximum value. However, when the printed synchronization code is scanned with a slight rotation, a part of the scanned synchronization code (for example, the peripheral part) is different from the original synchronization code. Therefore, it is detected as a partially incorrect synchronization code (that is, the autocorrelation absolute value is detected as a value smaller than the maximum value).

  In this embodiment, even if the autocorrelation absolute value is obtained by correlation calculation with such partially incorrect synchronization code, the synchronization code and other sequences (codes that are completely different or codes that are out of phase) A sequence pattern that is larger than the correlation absolute value obtained by the correlation calculation is used. As a result, the synchronization code can be detected with higher accuracy.

  Therefore, in order to detect the synchronization code accurately even when the image is greatly rotated, a code in which the autocorrelation absolute value of the synchronization code is “sufficiently” larger than the correlation absolute value of the synchronization code and other sequences. Is preferably used.

  The autocorrelation absolute value of the sync code is the value obtained by autocorrelation of the sync code itself, the value obtained by correlating the code with the sync code inverted and the sync code itself, and the sync code inverted. The value obtained by autocorrelating the codes as they are is included.

  Using the synchronization code having such a property, the CPU executes the signal detection process of step S304 without being aware of the rotation of the image after the process of steps S301 to S303 in FIG. Specifically, the CPU (filter processing unit 320b) executes the above-described filter calculation process in step S304a of FIG. Next, the CPU (correlation calculation processing unit 320e) executes a two-dimensional correlation calculation process in step S304d.

  For example, when a plurality of synchronization codes are equally embedded in the image as shown in FIG. 20, the result of the filter calculation process is as shown on the left side of FIG. 25, and the result of the two-dimensional correlation calculation is the right side of FIG. become that way. Among the results of the two-dimensional correlation calculation, the black portion indicates a portion with a high correlation absolute value, and the gray portion indicates a low portion. From this, it can be seen that the synchronization code is embedded at the position where the correlation value is high. Using this, the CPU (rotation angle detector 320d) detects the position of the synchronization code from the result of the two-dimensional correlation calculation.

  As shown in FIG. 25, the positions of the detected plurality of synchronization codes are inclined by the same angle as the angle at which the image is inclined. Therefore, the CPU (rotation angle detection unit 320d) detects the positions of a plurality of synchronization codes from the result of the two-dimensional correlation calculation process in the rotation angle detection process in step S304b. Then, the CPU (rotation angle detection unit 320d) obtains the inclination angle (rotation angle θ) of the image from the positions of the plurality of synchronization codes.

  Here, a process for detecting the position of the synchronization code from the result of the two-dimensional correlation calculation process will be described in detail. The CPU (rotation angle detection unit 320d) uses the synchronization code position having the largest correlation absolute value from the two-dimensional correlation calculation results as a search initial value, and as shown in FIG. 26, a part separated by left and right Bw and upper and lower Bh. For each, the maximum value of the correlation absolute value within a predetermined search range is obtained and used as the position of the adjacent synchronization code.

  For example, as shown in FIG. 26, assuming that the position of the reference synchronization code is (x1, y1), when searching for a synchronization code adjacent in the right direction from that position, the CPU is separated by Bw in the right direction. A predetermined search range is searched centering on this part. At this time, the predetermined search range is w in the x direction and h in the y direction, as shown in the search window of FIG. That is, the search range in the x direction is x1 + Bw− (w / 2) <x <x1 + Bw + (w / 2). The search range in the y direction is y1− (h / 2) <y <y1 + (h / 2).

  Further, when searching for a synchronization code adjacent in the upward direction from the position of the reference synchronization code, the CPU searches for the predetermined search range centering on a portion separated by Bh in the upward direction. That is, the search range in the x direction is x1− (w / 2) <x <x1 + (w / 2). The search range in the y direction is y1 + Bh− (h / 2) <y <y1 + Bh + (h / 2).

  In this way, the synchronization code adjacent in the right direction and the synchronization code adjacent in the upward direction are obtained based on the position of the absolute correlation value that is maximum in the predetermined search range. The position of the entire synchronization code is obtained by repeating the above search.

After determining the positions of all the synchronization codes as described above, the CPU calculates the rotation direction and rotation angle of the image based on the positions of the synchronization codes. For example, when the positions of any two of the embedded synchronization codes adjacent to the left and right are detected as the coordinates (x1, y1) and coordinates (x2, y2) shown in FIG. The rotation angle θ of the image is obtained using the following equation (10).

  As a result, when the rotation angle θ of the image does not exceed the predetermined range, the CPU (rotation angle detection unit 320d) determines “Yes” in step S304e in FIG. 24 and ends the signal processing. . On the other hand, when the rotation angle θ of the image exceeds a predetermined range, the CPU (rotation angle detection unit 320d) determines “No” in step S304e, proceeds to step S304f, and determines the rotation obtained. By using the angle θ to rotate the image in the direction opposite to the direction in which the image is shifted, the shift in the rotation direction of the image is corrected.

  Thereafter, the CPU returns to step S304a again, and repeats the processing of step S304a, step S304d, step S304b, step S304e, and step S304f using the corrected image until it is determined as “Yes” in step S304e. .

  An example of the predetermined range used in step S304e is ± 0.1 degrees. That is, when the image rotation angle θ is within a range of ± 0.1 degrees, the CPU ends the process without performing the image rotation process, and the image rotation angle θ exceeds ± 0.1 degrees. Only the image rotation process may be performed.

  According to the present embodiment, it is possible to correct rotation of the image by several degrees using only the signal pattern of the watermark pattern of the digital watermark without using characters or ruled lines on the image. This makes it possible to reliably detect information of digital watermarks embedded in various types of electronic documents, such as images with a lot of CAD and diagonal lines, and images with extremely few characters and lines. Can do.

<C. When detecting the rotation direction of an input image of 90 degrees or more>
As described above, when (A) the rotation angle of the input image of several degrees to 90 degrees is detected, and (B) the rotation angle of the input image of several degrees is detected, a digital watermark is embedded. When the image is rotated to 90 degrees, 180 degrees, 270 degrees, etc. (that is, when the image is tilted more than 90 degrees), the rotation angle cannot be detected.

  Therefore, in the following embodiment, the input image is inclined more than 90 degrees by calculating the correlation with the four two-dimensional bit patterns for the watermark information including the synchronization code embedded in the image. The watermark information processing apparatus 10 capable of detecting the rotation direction of the image when the image is being read will be described.

(Fourth embodiment)
First, the fourth embodiment will be described. In the watermark information processing apparatus 10 according to the present embodiment, the rotation direction of the image in units of 90 degrees is detected from the magnitude relationship between the correlation absolute values by obtaining the correlation absolute values of the forward order and the reverse order for the PN code. . The detection method will be described below.

(Operation of watermark information embedding device 100)
In the present embodiment, as shown in FIG. 13, a two-dimensional PN code obtained by extending a one-dimensional PN code sequence to two dimensions is used as the synchronization code.

  The CPU (watermarked document image synthesizing unit 130) embeds a synchronization code composed of the two-dimensionally expanded PN code in the image. The synchronization code itself is composed of a set of signal patterns, and is converted into a signal pattern and embedded in the same manner as digital information normally embedded in an image.

  In the present embodiment, the number of embedded synchronization codes may be plural or singular as shown in FIG. When synchronization codes are used for signal position synchronization, the greater the number of synchronization codes, the easier it is to synchronize (the easier it is to correct the image) and the deformation of the image (for example, dirt or wrinkles on the paper on which the image is printed). Etc.).

When embedding, for example, when one signal pattern represents 1-bit information, as shown in FIG. 28, the signal pattern indicating “0” and the signal pattern indicating information 1 are embedded as they are. But you can. In addition, when one signal represents a plurality of bits (for example, when there are 2 ⁿ signal patterns, n-bit information can be represented by one signal pattern), as shown in FIG. One bit may be assigned to the synchronization code, and the other bits may be used for information embedding.

  As described above, a code sequence having a sufficiently low correlation with a code in which codes are arranged in reverse order is used for the one-dimensional PN code. That is, the bit pattern of the PN code is arranged in the forward direction and the bit pattern arranged in the forward direction and the PN code in which the bit pattern of the PN code is not known. The correlation is calculated for each bit pattern. As a result, the correlation absolute value in the forward direction and the correlation absolute value in the reverse direction are obtained. In the correlation calculation, as shown in FIG. 29, the correlation absolute value in the direction of high correlation increases. Therefore, a bit pattern arranged in a direction in which a larger correlation absolute value is obtained can be specified as a PN code by comparing the magnitude relationship of the correlation absolute value.

  Further, the synchronization code does not change the absolute value of the correlation even when all bits are inverted, and represents information of “0” or “1” when all bits are inverted or when all bits are not used.

(Operation of Watermark Information Detection Device 300)
The operation of the watermark information detection apparatus 300 according to this embodiment will be described. FIG. 30 is a flowchart showing a signal detection processing routine executed by the CPU (watermark detection unit 320) in the present embodiment.

  After the filter calculation process in step S304a, the CPU (correlation calculation processing unit 320e) performs PN codes in the normal order and reverse order of the PN codes used at the time of embedding with respect to the filter output value sequence in the two-dimensional correlation calculation process in step 304d. The correlation value with the code is calculated. When the filter output value is obtained with multiple bits per signal pattern, the bit plane used for embedding is used for correlation calculation.

  When two types of filters, filter A and filter B, are used, input values for correlation calculation processing are as follows. For example, when the filter output value F (A, x, y) is greater than or equal to the filter output value F (B, x, y), the input value of the correlation calculation process is F (A, x, y). When the filter output value F (A, x, y) is smaller than the filter output value F (B, x, y), the input value is −F (B, x, y).

When the interval of signals existing in the filter result is p (see FIG. 17), the correlation calculation is performed in units of sets that are separated by p in the filter result. For example, when a correlation operation with an N-bit synchronization code is performed, the filter output values to be used are {F (M _{x, y} , x, y), F (M _{x + p, y} , x + p, y), F (M _{x + 2p, y} , x + 2p, y),..., F (M _{x + Np, y} , x + Np, y)}. For example, when one signal is generated with 18 pixels and printed and scanned with the same resolution, one symbol becomes 18 pixels on the input image.

  If the filter processing is performed while shifting this pixel by pixel without any thinning, the signal interval p = 18 in the filter processing result is obtained. By performing the correlation calculation at the interval p, the correlation calculation process can be performed without performing the synchronization process for searching for the peak of the filter output value, and the synchronization code can be prevented from being lost due to the peak search synchronization error. .

Summarizing the above two-dimensional correlation calculation into an equation, the correlation value R (x, y) at the position (x, y) is as shown in equation (11).

However,
p: Signal interval M _{x + pi, y + pj} in the filter result: filter F (M _{x + pi, y + pj} , x + pi, y + pj) having the highest filter output value at the position (x + pi, y + pj): most filter output at the position (x + pi, y + pj) Output value of filter with high value

  As described above, the correlation value is calculated while shifting the filter output result by one to several dots (see FIGS. 18 and 19). In the two-dimensional correlation calculation method, each one-dimensional correlation calculation constituting this calculation may be sequentially calculated using Expression (4) or may be calculated at once using Expression (11).

  The two-dimensional correlation operation is a total of four patterns in the forward direction and the reverse direction in each of the horizontal direction and the vertical direction, that is, (horizontal direction, vertical direction) = (forward order, forward order), (forward order, reverse order), Correlation calculation is performed in four patterns (reverse order, normal order) and (reverse order, reverse order).

In the case of an N × N sized synchronization code obtained by extending the longest N-bit code in two dimensions, the correlation absolute value | R (x, y) | If they do not match, the correlation absolute value | R (x, y) | = 1 / N ² is taken. For this reason, the correlation absolute value becomes high only in the part of the filter output value obtained by the watermark detection unit 320 that actually matches the synchronization code. That is, as shown in FIG. 29, the same order (normal order and normal order, reverse order and reverse order) has a high correlation absolute value. Also, the result of inverting the sign is the same as the result of not inverting. That is, the correlation absolute value is high in the normal order and the reverse normal order, and the reverse order and the reverse reverse order, and the correlation absolute value is low in the reverse order and the reverse normal order and the reverse order reverse to the normal order.

  For example, when the filter output value sequence is directed in the reverse forward direction, the correlation absolute value is the highest when the reverse PN code matches the synchronization code. The actual correlation absolute value fluctuates due to noise, but the correlation absolute value is often higher in the portion that matches the synchronization code than in other portions.

  After the correlation absolute value is obtained by the two-dimensional correlation calculation process in this way, the CPU (rotation angle detection unit 320d) proceeds to step S304b in FIG. 30 and executes the rotation angle detection process. Specifically, the CPU determines the relationship between the direction of the sequence based on the rotation angle of the image and the magnitude of the correlation between the normal order and the reverse order, depending on the magnitude of the correlation between the forward order and the reverse order of the PN code used in the synchronization code. Based on this, the rotation angle is detected.

  FIG. 31 shows the relationship between the rotated synchronization code and the rotation direction. Assuming that the synchronization code rotated to “0 degrees” (that is, not rotated) is the first two-dimensional bit pattern, an array of one-dimensional bit patterns arranged in the horizontal direction with respect to the first two-dimensional bit pattern The second two-dimensional bit pattern composed of the bit patterns in which the order is reversed (that is, the reverse bit pattern) indicates the synchronization code rotated to “90 degrees”. A third two-dimensional bit pattern comprising a bit pattern in which the order of the one-dimensional bit pattern arranged in the horizontal direction and the one-dimensional bit pattern arranged in the vertical direction is reversed with respect to the first two-dimensional bit pattern. Indicates a synchronization code rotated to “180 degrees”. In addition, the fourth two-dimensional bit pattern composed of a bit pattern in which the arrangement order of the one-dimensional bit pattern arranged in the vertical direction is reversed with respect to the first two-dimensional bit pattern is rotated to “270 degrees”. Indicates the synchronization code.

  The storage unit 320a has unique rotation angles (0 degrees, 90 degrees) corresponding to the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern, respectively. Degrees, 180 degrees, and 270 degrees) are stored in association with each bit pattern.

  In the detection process of the rotation direction, if the one having the larger correlation absolute value is in the normal order in both the horizontal direction and the vertical direction, it is determined that the synchronization code is the first two-dimensional bit pattern. As a result, the rotation angle of the input image is detected to be 0 degrees from the inherent rotation angle stored in the storage unit 320a in association with the first two-dimensional bit pattern.

  The case where the horizontal direction and the vertical direction are normal is the normal direction with respect to the synchronization code rotated with the rotation of the image (in this case, since the rotation angle is 0 degrees, the synchronization code is not rotated). When the PN code A arrayed in the horizontal direction is scanned in the horizontal direction, the correlation absolute value (the maximum value) obtained when the PN code A arrayed in the reverse forward direction is scanned in the horizontal direction with respect to this synchronization code. The absolute value of the correlation (the maximum value) obtained when the PN code A that is larger than (the maximum value of) the correlation code and is arranged in the forward direction with respect to the synchronization code is scanned in the vertical direction. , This synchronization code is greater than the absolute value of correlation (the maximum value) obtained when the PN code A arranged in the reverse forward direction is scanned in the vertical direction.

  Similarly, when the horizontal direction is reverse order and the vertical direction is forward order (= reversal of forward order), it is determined that the synchronization code is the second two-dimensional bit pattern, and the rotation angle is 90 degrees (ie, image Is determined to be 90 degrees clockwise). Here, when the horizontal direction is reverse order and the vertical direction is normal order, the horizontal correlation absolute value between the rotated synchronization code and the reverse order of PN code A is the difference between the synchronization code and the normal order of PN code A. The correlation absolute value in the vertical direction between the synchronization code and the PN code A is larger than the horizontal correlation absolute value and the synchronization code and the PN code A in the normal order (= reversal of the PN code A in the normal order) Is greater than the absolute correlation value in the vertical direction.

  When the horizontal direction and the vertical direction are in reverse order (= reverse of reverse order), the synchronization code is determined to be the third two-dimensional bit pattern, and the rotation angle is 180 degrees (that is, the image rotation direction is clockwise). 180 degrees). Here, the case where both the horizontal direction and the vertical direction are in reverse order means that the correlation absolute value in the horizontal direction between the rotated synchronization code and the reverse order of PN code A (= reversal of the reverse order of PN code A) is the sync code and PN. The correlation absolute value in the vertical direction between the synchronization code and the reverse order of the PN code A (= reversal of the reverse order of the PN code A) is larger than the correlation absolute value in the horizontal direction with the forward order of the code A. This is the case where the correlation absolute value in the vertical direction with the positive order of the code A is larger.

  Further, when the horizontal direction is normal order (= reversal of normal order) and the vertical direction is reverse order, it is determined that the synchronization code is the fourth two-dimensional bit pattern, and the rotation angle is 270 degrees (that is, image rotation) It is determined that the direction is 270 degrees clockwise. Here, when the horizontal direction is normal order and the vertical direction is reverse order, the correlation absolute value in the horizontal direction between the rotated synchronization code and the normal order of PN code A (= reversal of the normal order of PN code A) is , The correlation absolute value in the vertical direction between the synchronization code and the reverse order of the PN code A is larger than the horizontal correlation absolute value and the reverse order of the synchronization code and the PN code A (= reverse of the reverse order of the PN code A). This is a case where the absolute value of the correlation between the synchronization code and the normal order of the PN code A is larger than the absolute value of the correlation.

  Since the location of the peak of the correlation absolute value corresponds to the position where the synchronization code is embedded, the peak position of the correlation absolute value may be used for information synchronization. For example, when the synchronization code is arranged as shown in FIG. 20, since the position of the synchronization code is known by the correlation absolute value, information about Bw × Bh around the synchronization code position where the correlation absolute value of the synchronization code peaks is the center. Can be obtained with high accuracy even if the images are shifted.

  In this manner, after the rotation direction (rotation angle in units of 90 degrees) is obtained by the rotation angle detection process in step S304b of FIG. 30, the CPU (rotation angle detection unit 320d) proceeds to step S304c and is obtained. By executing the filter output rotation process using the rotation angle obtained, images in the correct order (no deviation) are obtained.

  According to the present embodiment, by using a two-dimensional PN code obtained by extending the one-dimensional PN code sequence to two dimensions, the normal order or the reverse order is detected in each of the horizontal direction and the vertical direction. Thereby, the rotation direction of the image in units of 90 degrees can be detected by a combination of forward order or reverse order. As a result, it is possible to correct the rotation of the image in units of 90 degrees based on the watermark information. Further, by using the filter output value for the correlation calculation, it is possible to prevent erroneous detection of the synchronization code in a portion where the filter output value is low (no signal).

(Fifth embodiment)
Next, a fifth embodiment will be described. In the watermark detection unit 320 of the present embodiment, both horizontal and vertical calculations are executed in either the forward or reverse correlation calculation process, and either the forward or reverse forward direction is performed. In the correlation calculation processing, only horizontal or vertical calculation is performed on one side, and in the normal forward and reverse forward correlation calculation processing, both horizontal and vertical correlation calculations are executed. The operation is different from the watermark detection unit 320 of the fourth embodiment. In the following, this difference will be mainly described.

(Operation of Watermark Detection Unit 320)
An operation of the watermark detection unit 320 according to the present embodiment will be described. In this embodiment, the correlation value between the forward and reverse PN codes of the PN code used at the time of embedding is calculated for the filter output value sequence obtained by the filter processing.

  In the present embodiment, in the signal detection process (step S304 in FIG. 7) executed by the watermark detection unit 320, after the filter calculation process in step S304a in FIG. 30 is executed, the CPU (correlation calculation processing unit 320e) Correlation calculation processing in the forward direction is executed in both the horizontal direction and the vertical direction. However, the CPU (correlation calculation processing unit 320e) performs the correlation calculation process from the reverse forward direction only in the horizontal direction.

  The correlation value is calculated while shifting the filter output result by one dot. As shown in FIG. 18 and FIG. 19, in the two-dimensional correlation calculation method, each one-dimensional correlation calculation constituting this calculation may be sequentially calculated using Expression (4), or Expression (11) You may calculate at once using.

  In the present embodiment, the correlation calculation in the forward and backward directions is performed only in the horizontal direction. That is, in the correlation calculation, two patterns of two-dimensional correlation calculations are performed in the total of the normal order and the reverse order only in the horizontal direction. Specifically, two patterns of (normal order, normal order) and (reverse order, normal order) are used in the horizontal direction and the vertical direction.

  Thus, after the correlation absolute value is obtained by the two-dimensional correlation calculation process, the CPU (rotation angle detection unit 320d) executes the rotation angle detection process in step S304b of FIG. In the rotation angle detection process, as shown in FIG. 32, the magnitude of the correlation between the normal direction and the reverse order of the horizontal direction of the PN code used in the synchronization code, and the sign of the correlation value of the two-dimensional correlation calculation result, A 90-degree rotation angle of the image is detected.

  That is, in the rotation angle detection process, the CPU compares the magnitudes of the correlation absolute values in the forward and backward directions obtained by scanning in the horizontal direction, and compares the two-dimensional correlation absolute value and the correlation absolute value in the horizontal direction. Compare with the mean.

  Next, the CPU determines a two-dimensional bit pattern from the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern according to the comparison result. Specify one. Next, the CPU obtains the rotation angle of the input image from the unique rotation angle stored in the storage unit 320a in association with the specified two-dimensional bit pattern.

  For example, if the horizontal direction is normal, the CPU detects that the rotation angle is 0 degree or 270 degrees as shown in FIG. 32, and if it is reverse, it detects that it is 90 degrees or 180 degrees. . Also, when the average value of the correlation absolute value in the horizontal direction for the height of the synchronization code is close to the correlation absolute value of the two-dimensional correlation calculation, it is detected as 0 degree or 90 degrees, and the correlation absolute value of the two-dimensional correlation calculation is detected. When the value is much smaller than the average value of the correlation absolute value in the horizontal direction by the height of the synchronization code (less than half), it is detected as 180 degrees or 270 degrees. The CPU detects the rotation angle of the image in units of 90 degrees by using both of these conditions.

  The horizontal correlation absolute value (that is, the one-dimensional correlation absolute value: the larger correlation absolute value between the forward correlation absolute value and the reverse correlation absolute value is set as the one-dimensional correlation absolute value) is synchronized. Whether the value averaged for the height of the code is close to the absolute value of the correlation of the two-dimensional correlation operation is determined by calculating the average value of the horizontal correlation absolute value for the height of the synchronization code and the absolute value of the correlation of the two-dimensional correlation operation. If the difference from the values is within a few percent (for example, about 5%), these values may be determined as close values. In addition, if the difference between the correlation absolute value of the two-dimensional correlation calculation and the average value of the correlation absolute value in the horizontal direction is equal to or less than half, the correlation absolute value of the two-dimensional correlation calculation is the horizontal correlation. It may be determined that the absolute value is considerably smaller than the average value of the synchronization code height.

  According to the present embodiment, the average value does not depend on the direction of the sequence, and the average value is always output. The two-dimensional correlation absolute value outputs an average when the sequences match, but weakens the output when the sequences do not match (that is, the value decreases). By detecting the weakness of the output of the two-dimensional correlation absolute value in comparison with the average value, the degree of coincidence (direction coincidence) of the sequence pattern is determined.

  As a result, the correlation calculation with high processing cost can be reduced in one direction in the vertical direction. As a result, the load on the correlation calculation process can be further reduced. As a result, the rotation angle can be detected significantly faster than the conventional process of extracting watermark information for images of various angles. As a result, the efficiency of the entire processing can be improved. In addition, the calculation cost can be halved.

  The correlation calculation processing unit 320e according to the present embodiment obtains a horizontal correlation absolute value and a vertical correlation absolute value for the PN codes arranged in the forward direction, respectively, and is arranged in the reverse forward direction. For the PN code, only the correlation absolute value in the horizontal direction was obtained.

  However, this embodiment is not limited to this, and the correlation absolute value in both the horizontal direction and the vertical direction is obtained in either the forward direction or the reverse direction, and the horizontal value is obtained in either the forward direction or the reverse direction. It is only necessary to obtain the correlation absolute value on one side of either the direction or the vertical direction. That is, in the above correlation calculation, the horizontal and vertical directions are not limited to (forward order, forward order) and (reverse order, forward order) patterns, but use two patterns (forward order, reverse order) and (reverse order, reverse order). The two patterns (normal order, normal order) and (normal order, reverse order) may be used, or the two patterns (reverse order, normal order) and (reverse order, reverse order) may be used. Also good.

(Sixth embodiment)
Next, a sixth embodiment will be described. The watermark information processing apparatus 10 according to the present embodiment is different from the fifth embodiment in that the correlation calculation processing executed by the correlation calculation processing unit 320e is further simplified. In the following, this difference will be mainly described.

(Operation of Watermark Information Detection Device 300)
The operation of the watermark information detection apparatus 300 according to this embodiment will be described. In the present embodiment, the correlation value between the forward and reverse PN codes of the PN code used at the time of embedding is calculated for the filter output value sequence.

  In the present embodiment, the CPU (correlation calculation processing unit 320e) applies the filter output value sequence obtained by the filter calculation process (formula (2)) in the two-dimensional correlation calculation process in step S304d shown in FIG. Thus, the correlation value between the forward PN code and the reverse PN code used at the time of embedding is calculated. When the filter output value is obtained with multiple bits per signal pattern, the bit plane used for embedding is used for correlation calculation.

  In the present embodiment, the input to the correlation calculation process is as follows when two types of filters, filter A and filter B, are used. For example, when the filter output value F (A, x, y) is greater than or equal to F (B, x, y), “1” and F (A, x, y) are smaller than F (B, x, y). , “−1” can be input.

Summarizing the above two-dimensional correlation calculation into an equation, the correlation value R (x, y) at the position (x, y) is as shown in equation (12).

However,
p: signal interval M _{x + pi, y + pj} in the filter result: filter with the highest filter output value at the position (x + pi, y + pj)

  After the correlation absolute value is obtained by the two-dimensional correlation calculation process in this way, the CPU (rotation angle detection unit 320d) proceeds to step 304b in FIG. 30 and executes the rotation angle detection process as described above. In step 304c, the input image is rotated in the direction opposite to the rotation direction according to the obtained rotation angle θ to obtain an image in the correct order (no deviation).

  According to the present embodiment, the correlation is calculated using a fixed value (that is, “1” or “−1”) determined based on the result of the filter processing instead of the weighting (see Expression (12)). As a result, the correlation calculation is weighted by a weight that varies in accordance with each filter output value output as a result of the filter process (see Expression (11)), and the correlation calculation process is further simplified compared to the fifth embodiment. . As a result, the load on the entire process can be further reduced.

<D. Correct image using detected rotation angle>
(Seventh embodiment)
As described above, conventionally, watermark information has been extracted from an image by the following methods (1) to (7).
(1) The print document 220 in which the watermark information is embedded is scanned to obtain a scanned image (input image). However, this is not necessary when watermark information is detected from an already read image.
(2) A filtering process that reacts to each signal pattern of the watermark is performed on the entire scanned image. (Fig. 16)
(3) The peak value of the filter that reacted strongly at the interval of the watermark signal pattern is searched. The search is performed centering on a portion separated from the adjacent signal pattern position by the interval p of the signal pattern used at the time of embedding. This search is performed on the entire filter processing result.
(4) A signal pattern is specified by comparing the magnitudes of peak values output from the filters, and information associated with the signal pattern (for example, a bit value of “0” or “1”) is extracted.
(5) The effective range where the information is embedded is specified from the obtained information.
(6) The information is combined in the same format as the embedding side, and the entire embedding information is acquired.
(7) Correcting incorrect information by performing a majority operation from the entire embedded information, and acquiring correct information.

  In such a process, when the scan image has a tilt, the conventional apparatus detects the watermark information after correcting the tilt of the image using information such as a character string or a ruled line in the scan image.

  However, in each embodiment described above, the inclination of the image is detected only from the background pattern. Thus, by combining the embodiments, it is possible to detect a rotation angle in an arbitrary direction of 360 degrees from a watermark background pattern at an arbitrary position in a scanned image.

  However, in the detection of the rotation angle in each embodiment, the accuracy of the filter is determined by the direction corresponding to the filter used for detection (for example, when filtering is performed with 10 filters, the rotation angle every 18 degrees is If the angle of the signal pattern is in the middle of the angle to which the filter is applied, the rotation angle of the image could not be detected accurately. For example, in the first embodiment, since the rotation angle of the image is detected using a filter with an interval of 18 degrees, if the rotation angle of the image is 9 degrees, the characteristic portion of the signal pattern is not necessarily output as the peak value of the filter output value. As a result, the rotation angle of the image could not be detected accurately.

  Therefore, if the image is rotationally corrected based on this rotation angle, the detection accuracy is not good, so that the image remains inclined despite the correction. For this reason, after performing the processes (2) to (7) using the corrected image, the rotation angle of the image is detected again, and the rotation of the image is corrected again based on the detected rotation angle. There is a case.

  However, the filtering process has the largest processing amount among the information detection processes executed by the watermark information detection apparatus 300, and the processing load on the CPU increases. Therefore, it is necessary to limit the filtering process to as few times as possible.

  In addition, any of the above rotation angle correction techniques assumes that the number of pixels such as characters existing on the scanned image is 5% or less of the whole, and there are a large number of character pixels on the scanned image. If the signal pattern is lost due to overlapping with the character pixel, the watermark information cannot be extracted correctly.

  Therefore, in the present embodiment, the processing load of the CPU when detecting a rotation shift of the image is reduced, and a large number of pixels such as characters are present on the watermark information. A method for correctly extracting embedded information by correctly detecting a watermark signal pattern even in a situation where the pixel overlaps with a pixel and is lost will be described.

  In the present embodiment, only the operation of the watermark detection unit 320 of the watermark information detection apparatus 300 is different from the second embodiment. Therefore, only the operation of the watermark detection unit 320 will be described.

  As shown in FIG. 33, the watermark detection unit 320 has a functional block in which a synthesis unit 320f is added to the watermark detection unit 320 of the second embodiment. The synthesizing unit 320f weights the plurality of filter output values output as a result of the filter processing by weighting according to the rotation angle of the watermarked image detected by the rotation angle detecting unit 320d, thereby obtaining the plurality of filter output values. It comes to synthesize.

  The filters used by the watermark detection unit 320 in this embodiment are the same ten types of filters as in the second embodiment, as shown in FIG. 9A.

Assuming that the angle of the filter N is (18 × N) degrees and the filter output value of the coordinates (X, Y) in the scan image is F (N, X, Y), the filter output value F is given by the equation (2). ). However, in Expression (1), tan ⁻¹ (u / v) = Nπ / 10 (rad).

  As a result of filtering by the CPU (filter processing unit 320b) using the above equation (2), as shown in FIGS. 9B and 9C, the filter that reacts strongly (the portion indicated by the thick arrow) is Varies depending on the rotation angle of the image. That is, as described above, when the rotation angle of the scanned image is θ (rad), the portion that reacts strongly to the filter is also inclined by θ. FIG. 34 shows the result of filtering a scan image in which the embedded signal pattern is also tilted by the rotation angle θ with 10 types of filters.

  Next, a method of synthesizing the filter output values of these filters while removing the inclinations of the images included in the respective filter processing results based on the plurality of filter processing results thus obtained will be described with reference to FIG. Will be described with reference to FIG.

With respect to the upper left origin (0, 0) of the filter processing result (right side in FIG. 35) after combining by removing the inclination of the image, the origin of mapping of the actual filter processing result (left side in FIG. 35) is (ox, oy). When the rotation angle θ of the filter output is rotated in the reverse direction and the filter output value is corrected so that the rotation angle of the combined filter output becomes “0”, the coordinates (x, y) of the combined filter output are On the actual filter processing result, it is mapped to the coordinates (X, Y) using the following equation (13).

  Using the positional relationship between the pre-mapping and post-mapping, the CPU (combining unit 320f) uses the ten types of filter output values on the left side of FIG. 36 for four patterns (signal patterns s1 to s4) on the right side of FIG. Composite to filter output value. The synthesis method will be described below.

  When each filter is scanned on an image having no inclination, the filter output value becomes the highest when the position of the filter completely coincides with any one of the signal patterns s1 to s4 in FIG. Decreases as you slide from position.

  However, if the scan image is deteriorated or the signal pattern is lost due to overlapping of pixels such as characters, the filter output of the deteriorated portion is lowered, so that the filter output is noisy (dirt and wrinkles on the printed document 220). (The signal pattern is buried in noise).

  However, the filter output of the entire signal pattern does not decrease, and many of the filter outputs in the vicinity of the signal pattern show values higher than noise. For example, if the size of the signal pattern is Bw × Bh, when the filter is scanned over the image, the filter processing outputs peak values for each adjacent signal pattern for each horizontal interval Bw and vertical interval Bh. During this period, a value attenuated according to the distance from the position where the peak value is output is output. This filter output is almost equal to or less than noise near the middle between different signal patterns. In other words, when one of the signal patterns is sufficiently closer than the other signal pattern, the output value of the filter shows a value higher than that of the noise.

  Therefore, in the present embodiment, the filter output of the target portion is synthesized from a plurality of filter outputs near the target portion to compensate for signal pattern deterioration and loss. For example, when the combined filter output is Fe (M, x, y) (where M∈ {0, 1, 2, 3}), the reference is made to generate the filter output value Fe (x, y). FIG. 36 shows the filter output value range (X ± sx, Y ± sy) to be performed. The CPU (synthesizing unit 320f) synthesizes the filter output values according to the following expression (14) using the ten types of filter output values in this range (X ± sx, Y ± sy).

Here, for example, the weight Wr of each rotation angle and the weight Wd of the distance are defined by the following equations (15) and (16).

However,

  As a result, four filter output results are synthesized from the weight Wd corresponding to the distance and the weight Wr corresponding to the rotation angle from the ten types of filter output results in the range of ± sx and ± sy respectively in the vertical and horizontal directions. The CPU (synthesizing unit 320f) calculates the filter output values at all positions on the synthesis filter output by calculating this with respect to all positions (M, x, y) on the synthesis filter output. FIG. 37 shows an example of how to take the coordinate system when the filter output is synthesized using the entire filter output image.

  The CPU searches for the peak position for the combined four filter output values obtained as described above, and extracts the watermark information by the method described in the watermark information extraction process described above.

  According to the present embodiment, in the process of synthesizing the filter outputs, the final filter output is estimated using the filter outputs near the target position. The four filter output results obtained in this way correspond to the filter outputs that react best to the respective signal patterns s1, s2, s3, and s4.

  As a result, it is possible to compensate for the filter output value of a portion that cannot be read with respect to pixels such as dirt (degradation of the image itself) or characters, etc. As a result, it is possible to reduce the influence of the filter output deteriorated due to deterioration of the image itself or pixels such as characters on information extraction. As a result, watermark information can be extracted with high accuracy.

  Further, according to the present embodiment, since it is not necessary to rotationally correct the image, it is possible to reduce the amount of filter processing when detecting rotation for rotational correction. As a result, the overall processing speed can be increased.

  In this embodiment, four types of filter outputs are synthesized in this way. However, the number of filters after synthesis may be less than 4, more than 4, and any integer of 1 or more. The number of filters after synthesis may be determined according to the number of signal patterns to be detected. Further, the number of filter processes used for detecting the rotation angle of the image need not be ten, but may be larger than the number of filters after synthesis.

  Expression (14) is an example of applying a weight according to the rotation angle and distance, and any form may be used as long as a plurality of filter output values can be synthesized to obtain the filter output of the target portion. .

  As described above, according to the watermark information processing apparatus 10 of each embodiment, when a watermarked document is tilted, the tilt can be accurately detected. In addition, the amount of processing required for the detection can be reduced. Furthermore, even when a lot of pixels such as characters are present on the watermark and many of the watermark patterns are lost, the watermark can be detected correctly and the embedded information can be extracted correctly.

  In the above embodiment, the operations of the respective units are related to each other, and can be replaced as a series of operations in consideration of the relationship between each other. And it can be set as embodiment of method invention by replacing in this way.

  Further, by replacing the operation of each unit with the processing of each unit, the program can be implemented. Further, by storing the program in a computer-readable recording medium in which the program is recorded, an embodiment of a computer-readable recording medium recorded in the program can be obtained.

  Therefore, in the embodiment of the watermark information detection method, watermark information detection for detecting watermark information from a watermarked document image in which a watermark image including at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded. A process for inputting the watermarked document image, a process for preliminarily storing a type of signal pattern embedded in the input watermarked document image in the storage unit, and a signal pattern of the embedded type. A plurality of types of filters including a filter that can specify a plurality of types of filters, and filtering each of the input watermarked document images using a larger number of filters than the type of the embedded signal pattern; , Embedded in the watermarked document image based on the result of each filter processing From the relationship between the process of specifying the type of the signal pattern, the information about the angle represented by the signal pattern of the type stored in the storage unit, and the information about the angle represented by the signal pattern of the specified type An embodiment of a watermark information detection program for causing a computer to execute a process for obtaining the rotation angle of a watermarked document image and a computer-readable recording medium storing the watermark information detection program can be provided.

  Further, the embodiment of the watermark information detection method described above is based on a watermarked document image in which a watermark image including a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded. A watermark information detection program for detecting watermark information, wherein the input watermarked document image is processed using a process of inputting the watermarked document image and a plurality of types of filters corresponding to the plurality of types of signal patterns. The absolute value of the correlation value is calculated by calculating the correlation for the special series pattern based on the filtering process and the filtering results for the predetermined range of the input watermarked document image. Based on the processing to calculate each absolute value of correlation and the above calculated multiple absolute values of correlation Watermark information detection program executed a process of determining a rotation angle of the watermarked document image, to a computer, and can be the embodiment of a computer-readable recording medium that stores the watermark information detection program.

  Further, the watermark information detection program and the computer-readable recording medium storing the watermark information detection program are used as a result of each filtering process for a predetermined range of the input watermarked document image in the correlation calculation process. On the basis of the above, by scanning each of the series pattern in which the special series pattern is arranged in the forward direction and the series pattern in which the special series pattern is arranged in the reverse forward direction in the horizontal direction, the correlation with each of the series patterns is obtained. And calculating each correlation pattern by scanning each series pattern in the vertical direction and calculating each correlation absolute value, and calculating the magnitudes of the plurality of calculated correlation absolute values. By comparing the above watermarked A process of determining the rotational direction of the writing image, may be configured to include.

  The special sequence pattern used for the watermark information detection program and the computer-readable recording medium storing the watermark information detection program is one or more one-dimensional bit patterns generated by a pseudo-random number generator. A plurality of two-dimensional bit patterns generated by arranging them in either the horizontal direction or the vertical direction, and for any first two-dimensional bit pattern included in the plurality of two-dimensional bit patterns, A second two-dimensional bit pattern comprising a bit pattern in which the arrangement order of the one-dimensional bit pattern arranged in the horizontal direction is reversed, a one-dimensional bit pattern arranged in the horizontal direction, and a one-dimensional bit arranged in the vertical direction A third two-dimensional bit pattern consisting of a bit pattern in which the pattern order is reversed , A fourth two-dimensional bit pattern comprising a bit pattern in which the arrangement order of the one-dimensional bit patterns arranged in the vertical direction is reversed, and the first two-dimensional bit pattern and the second two-dimensional bit In the process of storing the unique rotation angle corresponding to each of the pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern in the storage unit in association with each bit pattern, Based on the result of each filtering process on the predetermined range of the input watermarked document image, the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and Each series pattern is scanned by scanning each series pattern of the fourth two-dimensional bit pattern in the horizontal direction. And calculating the correlation absolute value by calculating the correlation for each sequence pattern by scanning each sequence pattern in the vertical direction, respectively, and scanning in the horizontal direction. And comparing the magnitudes of the plurality of correlation absolute values obtained by the above-mentioned method, comparing the magnitudes of the plurality of correlation absolute values obtained by the vertical scanning, and comparing the first two-dimensional bit pattern, A process of specifying one two-dimensional bit pattern from among the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern, and associated with the specified two-dimensional bit pattern A process for obtaining the rotation direction of the watermarked document image from the unique rotation angle stored in the storage unit. It may be configured to include

  Further, the watermark information detection program and a computer-readable recording medium storing the watermark information detection program include a sequence pattern obtained by arranging the special sequence pattern in the forward direction and the special sequence in the correlation calculation process. Correlation in the forward direction by calculating the correlation with the special sequence pattern by scanning the sequence pattern in which the patterns are arranged in the reverse direction in either the horizontal direction or the vertical direction of the watermarked document image. The absolute value of the correlation and the absolute value of the reverse forward direction are obtained, and the larger absolute value of the correlation is obtained by comparing the average value of the absolute value of the positive forward direction and the average value of the absolute value of the reverse direction. As the correlation absolute value, the sequence pattern used to calculate the primary correlation absolute value is The correlation absolute value is obtained by calculating the correlation with the sequence pattern by scanning either the horizontal direction or the vertical direction of the watermarked document image, and the obtained correlation absolute value and the one-dimensional correlation absolute value A process for calculating a two-dimensional correlation absolute value from the two-dimensional correlation absolute value, and a process for obtaining the rotation direction of the watermarked document image by comparing the magnitude of the two-dimensional correlation absolute value and the average value of the one-dimensional correlation absolute value. It may be configured to include.

  The watermark information detection program and the computer-readable recording medium storing the watermark information detection program further perform the filtering process according to the weight corresponding to the rotation angle of the watermarked image detected by the rotation angle detection unit. You may comprise the process which synthesize | combines a some filter output value by weighting the some filter output value output as a result.

  The embodiment of the watermark information embedding method is a watermark information embedding program for embedding a watermark image in a document image in order to detect the watermark information embedded in the image, and is different from a process for creating a document image by a predetermined amount. Processing for generating a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on angles, and watermarking input to the watermark information detection device using the generated special sequence pattern In order to cause the watermark information detection apparatus to detect the rotation angle of the document image, a process of creating a watermark image including the generated special sequence pattern, the created document image, the created watermark image, Processing to generate a watermarked document image by superimposing images and embedding watermark information Program, and it may be the embodiment of a computer-readable recording medium that stores the watermark information embedding program.

  Each processing in the embodiment of the program and the embodiment of the computer-readable recording medium on which the program is recorded is achieved by the computer executing the program.

  As mentioned above, although preferred embodiment of this invention was described referring an accompanying drawing, it cannot be overemphasized that this invention is not limited to this example. It will be apparent to those skilled in the art that various changes and modifications can be made within the scope of the claims, and these are of course within the technical scope of the present invention. Understood.

  For example, a one-dimensional PN code or a two-dimensional PN code obtained by extending a one-dimensional PN code sequence to two dimensions is an example of a special sequence. Any code sequence having a sufficiently low correlation with a reverse code arranged in order (a code obtained by arranging codes in reverse order) may be used.

  For example, a special sequence pattern is a sequence in which the absolute value of the autocorrelation value of this special sequence pattern is larger than the correlation absolute value of this special sequence pattern and a sequence pattern obtained by rearranging this special sequence pattern in reverse order. It may be. The special sequence pattern is preferably a fixed length.

  Further, the absolute value of the autocorrelation value of the special sequence pattern may be a sequence larger than the correlation absolute value of the special sequence pattern and a sequence pattern obtained by shifting the special sequence by a predetermined number of bits. . Further, the special sequence pattern may be composed of one or two or more one-dimensional bit patterns generated by a pseudo random number generator.

  A special sequence pattern is generated by combining a bit pattern in which one or more one-dimensional bit patterns generated by a pseudo-random number generator are arranged in either the horizontal direction or the vertical direction. A two-dimensional bit pattern may be used.

  In this case, the order of the arrangement of the one-dimensional bit patterns arranged in the horizontal direction is reversed with respect to any first two-dimensional bit pattern included in the plurality of two-dimensional bit patterns. A second two-dimensional bit pattern composed of a bit pattern, and a third two-dimensional bit pattern composed of a bit pattern in which the order of the one-dimensional bit pattern arranged in the horizontal direction and the one-dimensional bit pattern arranged in the vertical direction is reversed. And a fourth two-dimensional bit pattern composed of bit patterns in which the order of the arrangement of the one-dimensional bit patterns arranged in the vertical direction is reversed.

  Also, different sequences may be used for the special sequence of the horizontal sequence and the vertical sequence. In addition, a special sequence of a horizontal sequence and a vertical sequence may have different lengths. Further, the special series may be one in which random random numbers are two-dimensionally arranged. Furthermore, the special sequence may be composed of either a code obtained by extending a one-dimensional special sequence to two dimensions or a random two-dimensional code.

In addition, an arbitrary code string such as a longest code (M sequence) or a Gold code can be used as the PN code. The longest code means a code sequence having the longest length among code sequences that can be generated by feeding back an exclusive OR of the outputs of the shift register. When an n-stage shift register is used, the (longest) code length of the m sequence is 2 ⁿ −1. The Gold code is an example of a composite code sequence (longest code). It can be created most easily among the synthesized code sequences, and is synthesized by taking exclusive OR of two types of synchronized M sequences (M sequences).

  Each embodiment has been described on the assumption that printing is performed with black ink (single color) on white paper. However, the present invention is not limited to this, and the present invention is not limited to this. Applicable.

  The present invention embeds a watermark information processing apparatus that detects a rotation angle with high accuracy using only information in a ground pattern of the digital watermark, and digital watermark information that allows the watermark information detection apparatus to detect the rotation angle of the image in the image. The present invention can be applied to a watermark information embedding device.

1 is an overall configuration diagram of a watermark information processing apparatus according to a first embodiment of the present invention. It is a figure explaining the method of bit-patterning information with a signal pattern in the embodiment. FIG. 3 is a functional block diagram illustrating functions of a watermark detection unit in the same embodiment. 6 is a flowchart showing a routine executed by a watermark image creating unit in the embodiment. It is the figure which showed a part of signal pattern in the same embodiment. It is the figure which showed the waveform of the wave on the basis of the propagation direction of the wave shown by the dot pattern of the signal pattern in the embodiment. It is the flowchart which showed the routine which the watermark information detection apparatus performs in the same execution form. It is the figure which showed an example of the angle of the signal pattern used at the time of watermark information embedding in the embodiment. It is the figure which showed an example of the angle of the signal pattern used at the time of watermark information detection in the embodiment. It is the figure which showed the detail of the signal detection process in the same embodiment. It is a figure explaining the calculation method of a filter output value in the embodiment. It is a figure explaining the method to decompress | restor and extract information in the same embodiment. It is a figure explaining 2D PN encoding in 2nd Embodiment of this invention. 3 is a functional block diagram illustrating functions of a watermark detection unit in the embodiment. FIG. It is the figure which showed the example of the image which embedded the synchronization code in the same embodiment. It is the figure which showed the detail of the filter process result of the image in the same embodiment. It is the figure which showed the filter processing result of the image inclined several degrees in the same embodiment. It is the figure which showed the correlation calculation of the horizontal direction among the two-dimensional correlation calculation methods in the embodiment. It is the figure which showed the correlation calculation of the perpendicular direction among the two-dimensional correlation calculation methods in the same embodiment. It is the figure which showed the example of arrangement | positioning of a synchronous code in the same embodiment. It is the figure which showed the example of the image which embedded the synchronization code in 3rd Embodiment of this invention. It is the figure which showed the result of having filtered the image inclined several degrees in the same embodiment. It is the figure which showed the shift | offset | difference which arises with the signal farthest from the center in the same embodiment. It is the figure which showed the detail of the signal detection process in the same embodiment. It is the figure which showed the result of the filter calculation process and the two-dimensional correlation calculation in the same embodiment. It is a figure explaining the identification method of the peak position of the correlation with a synchronous code in the embodiment. It is the figure which showed the positional relationship of the adjacent synchronous code | cord | chord in the same embodiment. It is the figure which showed the example of the image which embedded the synchronization code in 4th Embodiment of this invention. It is the figure which showed the relationship of the correlation value of the normal order and reverse order in the same embodiment. It is the figure which showed the detail of the signal detection process in the same embodiment. It is the figure which showed the magnitude | size of the correlation absolute value of the direction of a series in the rotation angle of an image, and a horizontal direction and a vertical direction in the same embodiment. It is the figure which showed the magnitude of the correlation absolute value of the direction of a series in the rotation angle of an image, and a horizontal direction or a vertical direction in 5th Embodiment of this invention. It is a functional block diagram which showed the function of the watermark detection part in 7th Embodiment of this invention. It is the figure which showed the rotation angle of the image and the result of the filter process in the same embodiment. It is the figure which showed the synthetic | combination result of the filter output value in the same embodiment. It is the figure which showed the range of the filter output value referred in order to produce | generate the filter output value of a certain range in the same embodiment. It is the figure which showed an example of how to take a coordinate system in the case of synthesize | combining filter output using the whole filter output image.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Watermark information processing apparatus 100 Watermark information embedding apparatus 110 Document image creation part 120 Watermark image creation part 130 Watermarked document image composition part 140 Output part 150 Sequence pattern generation part 300 Watermark information detection apparatus 310 Input part 320 Watermark detection part 320a Storage part 320b Filter processing unit 320c Signal pattern identification unit 320d Rotation angle detection unit 320e Correlation calculation processing unit 320f Synthesis unit

Claims (40)

A watermark information detecting apparatus for detecting watermark information from a watermarked document image in which a watermark image including at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded:
An input unit for inputting the watermarked document image;
A storage unit for previously storing the type of signal pattern embedded in the input watermarked document image;
The inputted watermarked document using a plurality of types of filters including a filter capable of specifying the embedded type signal pattern, wherein the number of filters is larger than the type of the embedded signal pattern. A filter processing unit for filtering each image;
A signal pattern specifying unit for specifying the type of signal pattern embedded in the watermarked document image based on the result of each filtering process;
A rotation angle detection unit that obtains the rotation angle of the watermarked document image from the relationship between the information about the angle represented by the stored type of signal pattern and the information about the angle represented by the specified type of signal pattern A watermark information detection apparatus comprising: A watermark information detection device for detecting watermark information from a watermarked document image in which a watermark image including a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded. :
An input unit for inputting the watermarked document image;
A filter processing unit for filtering each of the input watermarked document images using a plurality of types of filters corresponding to the plurality of types of signal patterns;
Correlation for obtaining an absolute value of a correlation value as a correlation absolute value by calculating a correlation with respect to the special sequence pattern based on a result of each filtering process on a predetermined range of the input watermarked document image An arithmetic processing unit;
A rotation angle detection unit that calculates a rotation angle of the watermarked document image based on the plurality of obtained correlation absolute values. The plurality of types of signal patterns are:
3. The predetermined different angle is indicated by the propagation direction of the dot wave formed by arranging the plurality of dots at a predetermined position and either the wavelength or frequency of the dot wave. The watermark information detection apparatus described in 1. The special sequence pattern is
4. The watermark information detecting apparatus according to claim 2, wherein the watermark information detecting apparatus is composed of a code obtained by extending a one-dimensional special sequence into two dimensions.   The watermark information detecting apparatus according to claim 4, wherein the one-dimensional special series is a one-dimensional PN code. The special sequence pattern is
The watermark information detection apparatus according to any one of claims 2 to 5, wherein the watermark information detection apparatus includes a code based on a longest code. The special sequence pattern is
The absolute value of the autocorrelation value of the special sequence pattern is larger than the correlation absolute value of the special sequence pattern and a sequence pattern obtained by rearranging the special sequence pattern in reverse order. The watermark information detection apparatus described in the item.   The watermark information detection apparatus according to claim 2, wherein the special sequence pattern has a fixed length. The special sequence pattern is
9. The absolute value of the autocorrelation value of the special sequence pattern is larger than the correlation absolute value of the special sequence pattern and a sequence pattern obtained by shifting the special sequence by a predetermined number of bits. The watermark information detection apparatus according to one item. The rotation angle detector
The watermark information detection apparatus according to any one of claims 2 to 9, wherein a rotation angle of the watermarked image is obtained based on the obtained maximum value of the plurality of correlation absolute values. The input unit is:
Input a watermarked document image including a plurality of the special sequence patterns;
The rotation angle detector
Based on the obtained plurality of correlation absolute values, the positions of two or more special sequence patterns included in the input watermarked document image are specified, and the positional relationship between the two or more specified special sequence patterns The watermark information detection apparatus according to claim 2, wherein a rotation angle of the watermarked document image is obtained from the information. The correlation calculation processing unit
Based on the result of each filtering process on a predetermined range of the input watermarked document image, the correlation with the sequence pattern in which the special sequence pattern is arranged in the forward direction is calculated and the special sequence pattern is calculated. By calculating the correlation with the sequence pattern arranged in the reverse forward direction, the correlation absolute value is obtained respectively.
The rotation angle detector
The watermark information detection apparatus according to claim 2, wherein a rotation angle of the watermarked document image is obtained by comparing magnitudes of the plurality of obtained correlation absolute values. The correlation calculation processing unit
Based on the result of each filtering process on a predetermined range of the input watermarked document image, the special sequence pattern is scanned in the horizontal direction to calculate the correlation with the special sequence pattern, and the special sequence pattern By calculating a correlation with the special sequence pattern by scanning a specific sequence pattern in the vertical direction, a correlation absolute value is obtained,
The rotation angle detector
The watermark information detection apparatus according to claim 2, wherein a rotation direction of the watermarked document image is obtained by comparing the magnitudes of the plurality of obtained correlation absolute values. The correlation calculation processing unit
Based on the result of each filtering process on a predetermined range of the input watermarked document image, a sequence pattern in which the special sequence pattern is arranged in the forward direction and the special sequence pattern are arranged in the reverse direction. By calculating the correlation for each series pattern by scanning each series pattern in the horizontal direction, respectively, and calculating the correlation for each series pattern by scanning each series pattern in the vertical direction, respectively. Find the absolute value of each correlation,
The rotation angle detector
The watermark information detection apparatus according to claim 2, wherein a rotation direction of the watermarked document image is obtained by comparing the magnitudes of the plurality of obtained correlation absolute values. The special sequence pattern is
15. The method according to any one of claims 2, 3, 6, 7, 8, 9, 10, 11, 12, 13 or 14, comprising one or more one-dimensional bit patterns generated by a pseudo random number generator. Watermark information detecting apparatus. The special sequence pattern is
A plurality of two-dimensional bit patterns generated by combining bit patterns in which one or more one-dimensional bit patterns generated by a pseudo-random number generator are arranged in either the horizontal direction or the vertical direction, A second two-dimensional bit comprising a bit pattern in which the arrangement order of the one-dimensional bit patterns arranged in the horizontal direction is reversed with respect to an arbitrary first two-dimensional bit pattern included in the plurality of two-dimensional bit patterns A third two-dimensional bit pattern composed of a pattern, a one-dimensional bit pattern arranged in the horizontal direction, and a bit pattern in which the order of the one-dimensional bit pattern arranged in the vertical direction is reversed; and a one-dimensional arrangement arranged in the vertical direction A fourth two-dimensional bit pattern comprising a bit pattern in which the order of bit pattern arrangement is reversed. ,
The storage unit
A unique rotation angle corresponding to each of the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern is associated with each bit pattern. Remember,
The correlation calculation processing unit
The first two-dimensional bit pattern, the second two-dimensional bit pattern, and the third two-dimensional bit pattern based on the result of each filtering process on a predetermined range of the input watermarked document image And calculating a correlation for each of the series patterns by scanning each series pattern of the fourth two-dimensional bit pattern in the horizontal direction, and scanning each series pattern in the vertical direction. By calculating the correlation for each series pattern, the correlation absolute value is obtained,
The rotation angle detector
The magnitudes of the plurality of correlation absolute values obtained by the horizontal scanning are compared, and the magnitudes of the plurality of correlation absolute values obtained by the vertical scanning are compared. One specified two-dimensional bit pattern is selected from among the two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern, and the specified two-dimensional bit pattern The watermark information detection apparatus according to claim 2, wherein a rotation direction of the watermarked document image is obtained from a unique rotation angle stored in the storage unit in association with the watermark. The correlation calculation processing unit
Scanning the sequence pattern in which the special sequence pattern is arranged in the forward direction and the sequence pattern in which the special sequence pattern is arranged in the reverse forward direction in either the horizontal direction or the vertical direction of the watermarked document image; To calculate the correlation absolute value in the forward direction and the correlation absolute value in the forward direction, respectively, by calculating the correlation with the special sequence pattern according to The correlation absolute value that is larger than the comparison is defined as a one-dimensional correlation absolute value, and the sequence pattern used to calculate the primary correlation absolute value is either horizontal or vertical in the watermarked document image. The correlation absolute value is obtained by calculating the correlation with the sequence pattern by scanning in the other direction. Calculating a two-dimensional correlation absolute value from said one-dimensional correlation absolute value and relative value,
The rotation angle detector
The watermark information according to any one of claims 2 to 11, wherein a rotation direction of the watermarked document image is obtained by comparing a magnitude of the two-dimensional correlation absolute value and an average value of the one-dimensional correlation absolute value. Detection device. The signal pattern specifying unit is
The watermark information detection apparatus according to claim 1, wherein the type of signal pattern embedded in the watermarked document image is specified based on each filter output value output as a result of the filtering process. The plurality of types of filters are:
19. The watermark information detection apparatus according to claim 1, wherein the watermark information detection apparatus is a filter having a maximum filter output value when the rotation angles of the watermarked document images are different from each other.   The watermark information detection apparatus according to claim 1, wherein the plurality of types of filters are Gabor filters. The correlation calculation processing unit
The watermark information detection according to any one of claims 2 to 20, wherein the correlation calculation for the special sequence pattern is weighted with a weight that varies in accordance with each filter output value output as a result of the filter processing. apparatus. The correlation calculation processing unit
21. The watermark information detection apparatus according to claim 2, wherein a correlation with respect to the special sequence pattern is calculated using a fixed value determined based on a result of the filtering process instead of the weighting. The watermark information detection device according to any one of claims 1 to 22, further comprising:
A synthesizing unit that synthesizes a plurality of filter output values by weighting the plurality of filter output values output as a result of the filter processing with a weight according to the rotation angle of the watermarked image detected by the rotation angle detecting unit; A watermark information detection apparatus provided. The synthesis unit is
The watermark information detection apparatus according to claim 23, wherein a filter output value for at least one of a plurality of types of filters used for the filter processing is newly generated. The synthesis unit is
25. The watermark information according to claim 23, wherein a filter output value is newly generated by a filter corresponding to at least one of the plurality of types of signal patterns among the plurality of types of filters used in the filtering process. Detection device. The synthesis unit is
26. A filter output value substantially equal to a filter output value output as a result of filter processing by a filter corresponding to the plurality of types of signal patterns when the watermarked document image is not rotated is generated. The watermark information detection apparatus according to one item. The synthesis unit is
By combining a plurality of filter output values located in an arbitrary range of the input watermarked document image among the plurality of filter output values, a single filter output value located in the arbitrary range is obtained. The watermark information detection apparatus according to any one of claims 23 to 26, which is generated. A watermark information embedding device that embeds a watermark image in a document image to cause the watermark information detection device to detect the watermark information embedded in the image:
A document image creation unit for creating a document image;
A sequence pattern generation unit that generates a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on predetermined different angles;
In order to make the watermark information detecting device detect the rotation angle of the watermarked document image input to the watermark information detecting device using the generated special sequence pattern, the generated special sequence pattern is included. A watermark image creation unit for creating a watermark image;
A watermark information embedding device comprising: a watermarked document image synthesis unit that generates a watermarked document image by superimposing the created document image and the created watermark image.   The watermark information embedding apparatus according to claim 28, wherein the special sequence pattern has a fixed length. The special sequence pattern is
30. The absolute value of the autocorrelation value of the special sequence pattern is larger than the correlation absolute value of the special sequence pattern and a sequence pattern obtained by shifting the special sequence pattern by a predetermined number of bits. The watermark information embedding device described in the above. The special sequence pattern is
The absolute value of the autocorrelation value of the special sequence pattern is larger than a correlation absolute value indicating a correlation between the special sequence pattern and a sequence pattern obtained by rearranging the special sequence pattern in reverse order. The watermark information embedding device according to any one of the preceding claims. The special sequence pattern is
32. The watermark information embedding device according to any one of claims 28 to 31, wherein the watermark information embedding device includes a code obtained by extending a one-dimensional special series into two dimensions. The watermark information embedding device according to any one of claims 28 to 32, further comprising:
A watermark information embedding device comprising an output unit for printing the created watermarked document image. A watermark information detection method for detecting watermark information from a watermarked document image in which a watermark image including at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded:
Enter the watermarked document image,
The type of signal pattern embedded in the input watermarked document image is stored in advance in the storage unit,
The inputted watermarked document using a plurality of types of filters including a filter capable of specifying the embedded type signal pattern, wherein the number of filters is larger than the type of the embedded signal pattern. Each image is filtered,
Identifying the type of signal pattern embedded in the watermarked document image based on the result of each filtering process;
A watermark for determining the rotation angle of the watermarked document image from the relationship between the information about the angle represented by the type of signal pattern stored in the storage unit and the information about the angle represented by the specified type of signal pattern Information detection method. A watermark information detection method for detecting watermark information from a watermarked document image in which a watermark image including a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on predetermined different angles is embedded. :
Enter the watermarked document image,
Filtering each of the input watermarked document images using a plurality of types of filters corresponding to the plurality of types of signal patterns;
The absolute value of the correlation value is obtained as the correlation absolute value by calculating the correlation with respect to the special sequence pattern based on the result of each filtering process on the predetermined range of the input watermarked document image,
A watermark information detection method for obtaining a rotation angle of the watermarked document image based on the plurality of obtained correlation absolute values. The watermark information detection method,
When calculating the correlation, a sequence pattern in which the special sequence pattern is arranged in the forward direction and the special sequence based on a result of each filtering process on a predetermined range of the input watermarked document image Correlation with each series pattern is calculated by scanning each series pattern in which the patterns are arranged in the reverse forward direction in the horizontal direction, and correlation with each series pattern by scanning each series pattern in the vertical direction. To calculate the correlation absolute value,
36. The watermark information detection method according to claim 35, wherein the rotation direction of the watermarked document image is obtained by comparing the magnitudes of the obtained plurality of correlation absolute values. The watermark information detection method,
The special sequence pattern is
A plurality of two-dimensional bit patterns generated by arranging one or more one-dimensional bit patterns generated by a pseudo-random number generator in either the horizontal direction or the vertical direction, A second two-dimensional bit pattern composed of a bit pattern in which the order of the arrangement of the one-dimensional bit pattern arranged in the horizontal direction is reversed with respect to an arbitrary first two-dimensional bit pattern included in the bit pattern; And a third two-dimensional bit pattern composed of a one-dimensional bit pattern arranged in the vertical direction and a bit pattern in which the order of the one-dimensional bit pattern arranged in the vertical direction is reversed, and an arrangement of the one-dimensional bit pattern arranged in the vertical direction. A fourth two-dimensional bit pattern comprising a bit pattern with the order reversed,
A unique rotation angle corresponding to each of the first two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern is associated with each bit pattern. Storing in the storage unit,
When calculating the correlation, the first two-dimensional bit pattern, the second two-dimensional bit pattern, the second two-dimensional bit pattern, and the second two-dimensional bit pattern based on a result of each filtering process on a predetermined range of the input watermarked document image Each of the sequence patterns of the third two-dimensional bit pattern and the fourth two-dimensional bit pattern is respectively scanned in the horizontal direction to calculate the correlation with each of the sequence patterns, and each of the sequence patterns in the vertical direction. By calculating the correlation for each series pattern by scanning each, the correlation absolute value is obtained,
The magnitudes of the plurality of correlation absolute values obtained by the horizontal scanning are compared, and the magnitudes of the plurality of correlation absolute values obtained by the vertical scanning are compared. One specified two-dimensional bit pattern is selected from among the two-dimensional bit pattern, the second two-dimensional bit pattern, the third two-dimensional bit pattern, and the fourth two-dimensional bit pattern, and the specified two-dimensional bit pattern 37. The watermark information detection method according to claim 35, wherein a rotation direction of the watermarked document image is obtained from a unique rotation angle stored in the storage unit in association with the watermark. The watermark information detection method,
When calculating the correlation, a sequence pattern in which the special sequence pattern is arranged in the forward direction and a sequence pattern in which the special sequence pattern is arranged in the reverse forward direction are set in the horizontal or vertical direction of the watermarked document image. The correlation absolute value in the forward direction and the correlation absolute value in the forward direction are obtained by calculating the correlation with the special sequence pattern by scanning in either direction, and the obtained correlation absolute value in the forward direction. The correlation absolute value that is larger than the correlation absolute value in the forward direction is set as the one-dimensional correlation absolute value, and the sequence pattern used to calculate the primary correlation absolute value is the horizontal of the watermarked document image. By calculating the correlation with the sequence pattern by scanning in either the direction or the vertical direction, the correlation absolute The calculated, to calculate a two-dimensional correlation absolute value from the correlation calculated absolute value to the one-dimensional correlation with the absolute value,
37. The watermark information detection method according to claim 35, wherein the rotation direction of the watermarked document image is obtained by comparing the two-dimensional correlation absolute value and the average value of the one-dimensional correlation absolute value. . The watermark information detection method further includes:
36. A plurality of filter output values are synthesized by weighting a plurality of filter output values output as a result of the filter processing with a weight corresponding to a rotation angle of a watermarked image detected by the rotation angle detection unit. -The watermark information detection method described in any one of -38. A watermark information embedding method for embedding a watermark image in a document image to detect the watermark information embedded in the image:
Create a document image,
Generating a special sequence pattern formed by at least one of a plurality of types of signal patterns including information on predetermined different angles;
In order to cause the watermark information detecting device to detect the rotation angle of the watermarked document image input to the watermark information detecting device using the generated special sequence pattern, the watermark including the generated special sequence pattern is used. Create an image,
A watermark information embedding method for generating a watermarked document image by superimposing the created document image and the created watermark image.