JP2012147180A - Device and method for identifying image position, and device and method for embedding alignment signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To achieve speeding-up of alignment associated with cut editing.SOLUTION: A parallel displacement amount candidate selection section 8 substitutes a peak coordinate and a frequency spectrum, which are associated with the following combination, into an evaluation expression, in which an evaluation value is changed depending on the amount of parallel displacement of an input image with respect to an original image, in sequence from the combination in which an order determined by a processing order determination section 7 is earlier, and determines whether or not the evaluation expression is established. Thus, the amount of parallel displacement is narrowed down from among all candidates of the amount of the parallel displacement of the input image.

Description

  In the present invention, an editing process (for example, enlargement / reduction (scaling), rotation, inversion, cropping) is performed on a digital image in which a digital watermark is embedded, and the digital watermark is detected. In a case where the position cannot be specified, an alignment signal pattern embedded in the digital image is detected, and an image position specifying device and method for specifying the position of the image, and the alignment signal pattern are embedded in the digital image. The present invention relates to an alignment signal embedding apparatus and method.

When an editing process (for example, enlargement / reduction (scaling), rotation, inversion, cropping) is performed on a digital image in which the digital watermark is embedded, the digital watermark embedded in the digital image It may not be possible to specify the position of.
In order to cope with such a case, a registration signal pattern (registration signal) may be embedded in a digital image as a digital watermark separately from the digital watermark data.
Patent Document 1 below discloses an image position specifying device that detects the position of an image in which digital watermark data is embedded by detecting an alignment signal pattern when detecting a digital watermark embedded in a digital image. , 2.

In the image position specifying devices disclosed in Patent Documents 1 and 2, the alignment required in the editing of the cut-out is performed by performing correlation calculation or two-dimensional discrete Fourier transform.
In addition, the alignment necessary for the enlargement / reduction, rotation, or inversion editing is performed by performing a two-dimensional discrete Fourier transform.

JP 11-355547 A JP-T-2002-504272

  Since the conventional image position specifying apparatus is configured as described above, when performing alignment according to clipping editing, correlation calculation or two-dimensional discrete Fourier transform is performed, but correlation calculation or two-dimensional discrete Fourier transform is performed. In the method to be implemented, since the calculation amount is enormous, there is a problem that it takes a long time to complete the alignment.

That is, in Patent Document 2, two-dimensional discrete Fourier transform is used for alignment associated with editing of enlargement / reduction, rotation, or inversion, and correlation calculation is used for alignment associated with editing of cutout. In this case, for example, when the repetition period of the alignment signal pattern is set to horizontal M pixels and vertical N pixels, the calculation amount of the two-dimensional discrete Fourier transform is O (MN · log (MN)). Since the amount of calculation for searching for the peak position of the correlation value is O (M ² N ² ), the alignment associated with the editing of the cut becomes a bottleneck in the amount of calculation. For this reason, it is desirable to use a two-dimensional discrete Fourier transform for the alignment associated with the editing of the cutout.
In Patent Document 1, two-dimensional discrete Fourier transform is used for alignment associated with cut editing, but only mathematical content (determining a position that minimizes Equation (32)) is disclosed. However, a method for performing a high-speed operation in handling a digital image by a digital computer is not disclosed. Therefore, it is considered that the same expression must be evaluated at all positions (MN positions), and the calculation amount is O (MN). That is, when the number of frequency spectra included in the alignment signal pattern is s, the calculation amount is O (MNs). Since the calculation amount when using the correlation calculation is O (M ² N ² ), the calculation amount is somewhat reduced, but the expression (32) disclosed in Patent Document 1 includes trigonometric function calculation. Complex operations are required.

The present invention has been made to solve the above-described problems, and an object of the present invention is to provide an image position specifying apparatus and an image position specifying method capable of speeding up the alignment accompanying editing of cuts.
Another object of the present invention is to provide an alignment signal embedding device and an alignment signal embedding method that can embed an alignment signal pattern used for alignment associated with editing of a cutout in a digital image.

  An image position specifying device according to the present invention is embedded in an original image by inputting an image that may be edited on the original image and cutting out a part of the image from the input image. A partial image generating means for generating a partial image having the same size as the alignment signal pattern; and a Fourier transform means for performing a Fourier transform on the partial image generated by the partial image generating means and outputting a Fourier coefficient as a result of the Fourier transformation. Among the Fourier coefficients output from the Fourier transform means, the peak coordinate specifying means for specifying the peak coordinates, which are the coordinates of the Fourier coefficients whose absolute values are prominent, and the frequency spectrum of the frequency spectrum in the alignment signal pattern on the frequency plane Spectral coordinates corresponding to the peak coordinates specified by the peak coordinate specifying means in the spectrum coordinates that are coordinates When executing the evaluation process for each combination of the corresponding relationship specifying means and the peak coordinates and the frequency spectrum for which the corresponding relation is specified by the corresponding relation specifying means, the order of the combinations for executing the evaluation processes is determined by the Fourier in the peak coordinates. An evaluation order determining means for determining based on the coefficient, and the parallel movement amount specifying means is configured to generate the peak coordinates and the frequency spectrum relating to the combination in the order determined by the evaluation order determining means in order from the previous combination. By substituting into an evaluation formula whose evaluation value changes depending on the amount of translation of the input image relative to the image, and determining whether the evaluation formula is satisfied, it is possible to select a parallel translation from all the candidates for the translation of the input image. The movement amount is narrowed down.

  According to the present invention, an alignment signal pattern embedded in an original image is input by inputting an image that may be edited with respect to the original image and cutting out a part of the image from the input image. A partial image generating means for generating a partial image of the same size as the above, a Fourier transform means for performing a Fourier transform on the partial image generated by the partial image generating means, and outputting a Fourier coefficient as a result of the Fourier transform, and a Fourier transform means Among the Fourier coefficients output from, the peak coordinate specifying means for specifying the peak coordinates that are the coordinates of the Fourier coefficients with the protruding absolute value, and the spectrum that is the coordinates on the frequency plane of the frequency spectrum in the alignment signal pattern Among the coordinates, the correspondence that specifies the spectral coordinates that have a corresponding relationship with the peak coordinates specified by the peak coordinate specifying means. When executing the evaluation process for each combination of the peak coordinates and the frequency spectrum for which the correspondence relation is specified by the specifying means and the corresponding relation specifying means, the order of the combination for executing the evaluation process is based on the Fourier coefficient in the peak coordinates. An evaluation order determining means for determining, and the parallel movement amount specifying means is configured so that the order determined by the evaluation order determining means is in order from the previous combination, and the peak coordinates and frequency spectrum related to the combination are input to the original image. By substituting into an evaluation formula that changes the evaluation value depending on the amount of parallel translation of the input image and determining whether or not the evaluation formula is satisfied, the translation amount can be narrowed down from all the candidates for the translation amount of the input image. Therefore, there is an effect that it is possible to speed up the alignment accompanying the editing of the cutout.

BRIEF DESCRIPTION OF THE DRAWINGS It is a block diagram which shows the image position specification apparatus by Embodiment 1 of this invention. It is a flowchart which shows the processing content (image position specification method) of the image position specification apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows the example of cutting out a partial image. It is explanatory drawing which shows the example of the absolute value of a Fourier coefficient, and a peak coordinate. It is explanatory drawing which shows an example of a frequency spectrum, a peak, and spectrum correspondence information. It is explanatory drawing which shows an example of the coordinate of a spectrum information, a peak coordinate, spectrum correspondence information, expansion / contraction information, rotation information, and inversion information. It is explanatory drawing which shows the outline | summary of a parallel movement amount etc. judgment processing by the parallel movement amount candidate selection part. It is explanatory drawing which shows an example of the repetition of a positioning signal pattern, and the repetition of electronic watermark data. It is explanatory drawing which shows the outline | summary of a parallel movement amount etc. judgment processing by the parallel movement amount candidate selection part. It is a block diagram which shows the alignment signal embedding apparatus by Embodiment 3 of this invention. It is a flowchart which shows the processing content (alignment signal embedding method) of the alignment signal embedding apparatus by Embodiment 3 of this invention. It is explanatory drawing which shows an example of the frequency spectrum of the alignment signal pattern embedded with the alignment signal embedding apparatus of FIG.

Embodiment 1 FIG.
1 is a block diagram showing an image position specifying apparatus according to Embodiment 1 of the present invention.
In FIG. 1, an image cutout unit 1 inputs an image that may have been edited with respect to an original image that is a digital image, and cuts out a part of the image from the input image. A process for generating a partial image having the same size as the embedded alignment signal pattern is performed. The image cutout unit 1 constitutes a partial image generation unit.

The Fourier transform unit 2 performs a process of Fourier transforming the partial image generated by the image cutout unit 1 and outputting a Fourier coefficient that is a result of the Fourier transform. The Fourier transform unit 2 constitutes Fourier transform means.
The peak detection unit 3 performs processing for specifying the peak coordinates, which are the coordinates on the frequency plane of the Fourier coefficients whose absolute values protrude from the Fourier coefficients output from the Fourier transform unit 2. The peak detector 3 constitutes a peak coordinate specifying unit.

The spectrum correspondence calculation unit 4 is a process for specifying a spectrum coordinate corresponding to the peak coordinate specified by the peak detection unit 3 among the spectrum coordinates which are coordinates on the frequency plane of the frequency spectrum in the alignment signal pattern. To implement. The spectrum correspondence calculating unit 4 constitutes a correspondence specifying means.
The enlargement determination unit 5 performs processing for specifying the enlargement / reduction, rotation, and inversion of the input image with respect to the original image from the combination of the peak coordinates and the frequency spectrum for which the correspondence relationship is specified by the spectrum correspondence calculation unit 4. The enlargement determination unit 5 constitutes an enlargement identification unit.

The parallel movement determination unit 6 performs a process of specifying the parallel movement amount of the input image with respect to the original image and the presence or absence of point symmetry movement.
When executing the evaluation process for each combination of the peak coordinate and the frequency spectrum for which the correspondence relationship is specified by the spectrum correspondence calculation unit 4, the processing order determination unit 7 determines the order of the combination for executing the evaluation process as the Fourier coefficient in the peak coordinate. The process determined based on is executed. Note that the processing order determination unit 7 constitutes an evaluation order determination unit.

The parallel movement amount candidate selection unit 8 determines the peak coordinate and the frequency spectrum related to the combination in the order determined by the processing order determination unit 7 in the order from the previous combination, and the evaluation value is based on the parallel movement amount of the input image with respect to the original image. By substituting into a changing evaluation formula and determining whether or not the evaluation formula is satisfied, the parallel movement amount is narrowed down from all the parallel movement amount candidates of the input image.
The parallel movement amount candidate selection unit 8 also substitutes the peak coordinate and the frequency spectrum related to the combination specified by the spectral correspondence calculation unit 4 and the point symmetry to the evaluation formula, and the evaluation formula is established. By determining whether or not, the translation amount is narrowed down from all the candidates for the translation amount of the input image.
The parallel movement amount candidate selection unit 8 performs a process of determining whether or not there is a point-symmetric movement from the combination of the peak coordinates and the frequency spectrum related to the finally narrowed parallel movement amount.
The translation amount candidate selection unit 8 constitutes a translation amount specifying means.

In the example of FIG. 1, the image segmentation unit 1, the Fourier transform unit 2, the peak detection unit 3, the spectrum correspondence calculation unit 4, the enlargement determination unit 5, and the parallel movement determination unit 6 that are components of the image position specifying device. It is assumed that each is configured with dedicated hardware (for example, a semiconductor integrated circuit on which a CPU is mounted, or a one-chip microcomputer), but the image position specifying device is configured with a computer or the like. In this case, all of the programs describing the processing contents of the image cutout unit 1, the Fourier transform unit 2, the peak detection unit 3, the spectrum correspondence calculation unit 4, the enlargement determination unit 5, and the parallel movement determination unit 6 or A part may be stored in the memory of a computer, and the CPU of the computer may execute a program stored in the memory.
FIG. 2 is a flowchart showing the processing contents (image position specifying method) of the image position specifying apparatus according to Embodiment 1 of the present invention.

Next, the operation will be described.
In the first embodiment, a position embedded as a digital watermark with respect to an original image (uncompressed luminance image) that is a digital image that may be edited such as enlargement / reduction, rotation, inversion, or cropping An image position specifying device that performs alignment by detecting an alignment signal pattern will be described.

First, the theory of alignment based on the two-dimensional discrete Fourier transform as a premise will be described.
Here, it is assumed that the alignment signal pattern is a square pattern having M pixels in the horizontal direction and M pixels in the vertical direction, and includes s frequency spectra Si (iε [1, s]).
The alignment signal pattern may generally be a rectangle, but in many implementations it is a square, so in the following description, the alignment signal pattern is assumed to be a square.
Further, it is assumed that the frequency spectrum Si (Si is a complex number) is at a position (Xi, Yi) (Xiε [1, M / 2-1]) on the frequency plane. M is the pattern size.
Since the alignment signal pattern added to the image is a real number, a frequency spectrum of a conjugate complex number of the frequency spectrum Si is necessarily arranged on the frequency plane (−Xi, −Yi).

Assuming that the alignment signal pattern is f (x, y), the Fourier transform F (X, Y) of the alignment signal pattern f (x, y) is as follows.
f (x, y) (x∈ [−M / 2, M / 2-1], y∈ [−M / 2, M / 2-1])
F (X, Y) (X∈ [−M / 2, M / 2-1], Y∈ [−M / 2, M / 2-1])
F (X, Y) = Si ((X, Y) ε {(Xi, Yi) | iε [1, s]})
Conjugated complex number of Si ((X, Y) ∈ {(− Xi, −Yi) | i∈ [1, s]})
0 (other (X, Y))

The Fourier transform and inverse Fourier transform are as follows.
F (X, Y)
= (1 / M) Σ _{(x, y)} f (x, y) exp (−j2πxX / M) exp (−j2πyY / M)
f (x, y)
= (1 / M) Σ _{(X, Y)} F (X, Y) exp (j2πxX / M) exp (j2πyY / M)
Here, j is an imaginary unit, π is a pi, and exp () is an exponential function with the number of Napiers at the bottom.

In the first embodiment, the alignment signal pattern f (x, y) of horizontal M pixels and vertical M pixels is cyclically added to the original image. Depending on the characteristics of the original image, addition may be performed while adjusting the weight for each pixel in order to improve detection accuracy and subjective image quality.
The alignment is performed on a square portion of an arbitrary horizontal M pixel and vertical M pixel of the input image (an image that may be edited on the original image).
When the square portion is Fourier-transformed and the absolute value (amplitude) of the frequency spectrum is examined, the position of the frequency spectrum for alignment can be detected. This detection is an operation of picking up the coordinates where the absolute value protrudes in the absolute value map of the frequency spectrum, and can be sufficiently performed by a known image processing technique. Therefore, the detailed description of the method is omitted.

If the number of detected frequency spectra is t, the frequency spectrum is Tk (k∈ [1, t]), and the original image is rotated by a certain angle, the position of the frequency spectrum Si on the frequency plane is centered on the origin. If the original image is rotated about the same symmetry line as the axis, the position of the frequency spectrum Si on the frequency plane is inverted about the symmetry line passing through the origin at the same angle.
Further, if the original image is magnified at a certain magnification, the position of the frequency spectrum Si on the frequency plane is magnified with the magnification of the reciprocal of the magnification rate centered on the origin (frequency plane of the frequency spectrum Si). The distance from the origin above is the inverse of the image magnification).
Using this property, it is possible to determine how the input image has been scaled, rotated, or inverted with respect to the original image. Details of these determination methods are disclosed in Patent Documents 1 and 2 and the like, and thus the description thereof is omitted.

By correcting the input image based on the magnification, rotation angle, and reversal axis angle determined from the position of the detected frequency spectrum Tk, it is possible to correct the state in which only the uncertainty of translation remains.
Hereinafter, an image that has been corrected to a state where only the uncertainty of translation remains is referred to as a “half-corrected image”.
The half-corrected image has m pixels (m∈ [−M / 2, M / 2-1]) in the x-axis direction and n pixels (n∈ [−M / 2, M / 2−2) in the y-axis direction from the original image. 1]), it is assumed that obtaining m and n corresponds to alignment with respect to the parallel movement.

In addition to the case where the enlargement / reduction, rotation, and reversal are performed, even when the enlargement / reduction, rotation, and reversal are not performed, this alignment is necessary when the trimming is performed.
Here, assuming that the frequency spectrum corresponding to the frequency spectrum Tk is Si (the correspondence between Tk and Si has been clarified in the alignment procedure for enlargement / reduction, rotation, and inversion), the phase of the frequency spectrum Si is Pi ( When Si = | Si | exp (jPi)) and the phase of the frequency spectrum Tk is Qk (Tk = | Tk | exp (jQk)),
E (m, n, k) = mXi / M + nYi / M + (Qk−Pi) / 2π≈integer (any combination of corresponding Si and Tk)
Select m and n for which

Or if it expresses like patent documents 1,
Σ _k {1-cos (2πmXi / M + 2πnYi / M + Qk−Pi)}
(Σ in the corresponding Si and Tk combination)
Select m and n to minimize.
Thereby, the amount of parallel movement becomes clear, and the alignment of the input image and the original image is completed.

Next, the processing contents of the image position specifying apparatus in FIG. 1 will be specifically described.
The image cutout unit 1 performs a process of cutting out a partial image having the same shape as the alignment signal pattern (for example, a square of horizontal M pixels and vertical M pixels) from the input image (step ST1 in FIG. 2).
The position to cut out the partial image may be selected at random, may be selected with reference to the upper left corner, and various implementations are conceivable.
The pattern size M may be an arbitrary value, but is set to a power of 2 in a normal implementation.
The reason is that by setting the power to a power of 2, it becomes possible to perform fast Fourier transform (FFT) in the Fourier transform unit 2 described later, and the processing speed can be increased.
In addition, when the vertical or horizontal length of the input image is less than the vertical or horizontal length of the partial image to be cut out, an image with a fixed pixel value (for example, a zero value pixel) on the right side and the lower side of the input image. ), And cut out in accordance with the upper left corner to form a partial image. FIG. 3 shows an example of cutting out a partial image.

When the image cutout unit 1 generates a partial image, the Fourier transform unit 2 performs a two-dimensional discrete Fourier transform on the pixel value of the partial image, and outputs a Fourier coefficient that is a result of the two-dimensional discrete Fourier transform (step ST2).
The Fourier coefficient is a complex number of M ² , and is used by being separated into its absolute value (amplitude) and phase by each unit such as the enlargement determination unit 5. Therefore, when the Fourier transform unit 2 outputs the Fourier coefficient, the Fourier coefficient may be separated into an absolute value and a phase, or may be output by each unit such as the enlargement determination unit 5 using the Fourier coefficient. The coefficient may be separated into an absolute value and a phase.

Peak detecting unit 3 receives the M ² pieces of the Fourier coefficients are arranged from the Fourier transform unit 2 on the frequency plane, with reference to the absolute value map M ² pieces of Fourier coefficients, the peak (M ² pieces of Among the Fourier coefficients, a Fourier coefficient whose absolute value is prominent (for example, a Fourier coefficient having the maximum absolute value) is detected, and the peak coordinates which are the coordinates on the frequency plane of the peak are calculated (step ST3).
Since the peak detection process can be sufficiently performed by a known image processing technique, detailed description thereof is omitted.
FIG. 4 is an explanatory diagram showing examples of absolute values and peak coordinates of Fourier coefficients.
When the absolute value of the Fourier coefficient protrudes across a plurality of Fourier coefficients, peak coordinates that do not necessarily become integer values may be calculated as shown in FIG.
There are various methods of calculating real-valued peak coordinates, such as a simple average of coordinates indicating a peak and a weighted average (centroid calculation) using the absolute value of a Fourier coefficient as a weight.

When the peak detection unit 3 calculates the peak coordinates, the spectrum correspondence relationship calculation unit 4 is a spectrum having a correspondence relationship with the peak coordinates among the spectrum coordinates that are coordinates on the frequency plane of the frequency spectrum in the alignment signal pattern. The coordinates are specified, and spectrum correspondence information indicating the correspondence is output (step ST4).
That is, the spectrum correspondence calculation unit 4 inputs spectrum information, specifies the correspondence between the spectrum embedded in the original image and the spectrum detected from the partial image, and outputs the spectrum correspondence information indicating the correspondence To do.
Here, the spectrum information means the coordinates (spectrum coordinates) on the frequency plane of the frequency spectrum of the alignment signal pattern embedded in the original image and the phase of the frequency spectrum.
However, the spectrum correspondence calculation unit 4 uses only the spectrum coordinates in the spectrum information.

The spectrum correspondence calculation unit 4 determines which spectrum coordinate the peak coordinate calculated by the peak detection unit 3 corresponds to clarify the correspondence.
FIG. 5 shows an example of frequency spectrum, peak and spectrum correspondence information.
The distribution of the frequency spectrum of the alignment signal pattern embedded in the original image is always point-symmetric. Therefore, the uncertainty of point symmetry (180 degree rotation) remains (for example, depending on the implementation, T1 ← → S1 may be associated).
This uncertainty is eliminated by a determination unit 6 such as a parallel movement described later.

The spectrum correspondence information may be information that can identify the combination of the frequency spectrum Tk and the frequency spectrum Si that can be matched.
In the example of FIG. 5, there are four sets of spectra that can be handled.
As a specific method for associating points, a method using logarithmic polar coordinate transformation is known.
That is, when (X, Y) is converted into polar coordinates (r, θ) and each point is plotted in logarithmic polar coordinates (logr, θ) with the r axis as a logarithmic scale, Enlargement / reduction at the (logr, θ) plane corresponds to translation in the r-axis direction, and rotation on the (X, Y) plane is translation in the θ-axis direction at the (logr, θ) plane. Inversion on the (X, Y) plane corresponds to inversion on an axis of symmetry parallel to the r axis in the (logr, θ) plane.
Using this property, matching can be performed by matching the frequency spectrum Tk and the frequency spectrum Si by a known technique.

At the peak detection stage, the embedded spectrum may not be detected (missed), or may be detected as a peak even though the spectrum is not embedded (false detection), so there are two or more matching candidates. there is a possibility.
In such a case, the final determination is made by selecting the one with the smaller positional error of the frequency spectrum Tk.
In addition to the method using positional errors, there is also a method using Fourier coefficients.
That is, among the peaks, the peak with the larger absolute value of the Fourier coefficient can be said to be more reliable. Therefore, by selecting the one that includes the peak with the larger absolute value among the candidates listed as candidates, the final one is selected. Make a decision. Although the determination can be made without using the Fourier coefficient as in the method using the positional error, the use of the Fourier coefficient is expected to increase the success rate of matching.

When receiving the spectrum correspondence information from the spectrum correspondence calculation unit 4, the enlargement determination unit 5 specifies enlargement / reduction, rotation, and inversion of the input image with respect to the original image from the combination of the peak coordinates and the frequency spectrum having the correspondence ( Step ST5).
That is, the enlargement determination unit 5 refers to the spectrum correspondence information, the peak coordinates, and the spectrum information, and the input image is an image that has been subjected to any enlargement / reduction, rotation, or inversion of the original image. (For example, the input image is enlarged, reduced, rotated, or inverted while changing the enlargement ratio, rotation angle, or inversion angle. By determining whether they match, enlargement / reduction information, rotation information, and inversion information are output.
FIG. 6 is an explanatory diagram showing an example of coordinates, peak coordinates, spectrum correspondence information, enlargement / reduction information, rotation information, and inversion information of spectrum information.
Here, the enlargement / reduction information is information composed of a combination of the presence / absence of enlargement / reduction and the enlargement ratio (the enlargement ratio of the image is the reciprocal of the enlargement ratio on the XY plane) when enlargement / reduction is present.
The rotation information is information including a combination of presence / absence of rotation and a rotation angle (assuming rotation from the positive X-axis direction to the positive Y-axis direction) when rotation is present.
The inversion information is information composed of a combination of the presence / absence of inversion and the elevation angle of a perpendicular line from the origin to the inversion axis when the inversion is present (the elevation angle facing the positive direction of the X axis is 0).

Here, an example is shown in which the enlargement / reduction determination unit 5 outputs enlargement / reduction information, rotation information, and inversion information. However, if the contents of enlargement / reduction, rotation, and inversion can be expressed, other information is output. It may be.
For example, the elevation angle of the perpendicular line from the origin to the reversal axis may be an angle, not an arc degree, and in the reversal information, the elevation angle of the perpendicular line from the origin to the reversal axis may be the elevation angle of the perpendicular line. An expression form is possible.
Further, since the enlargement ratio means that there is no enlargement / reduction at the same time, it is possible to adopt a configuration in which a parameter for presence / absence of enlargement / reduction is not provided.
For the same reason, a configuration in which the parameter of presence / absence of rotation is not provided may be employed.
In addition, a configuration in which a parameter such as an elevation angle of a perpendicular line from the origin to the reversal axis is not provided is also possible. This is because, for example, the order of editing contents given up to the input image is determined in advance such that the original image is first inverted if it is inverted, and then scaled and rotated. This is because the angle information of the reversal axis is absorbed by the rotation angle, and any enlargement / reduction, rotation, and reversal can be expressed without information on the angle of the reversal axis.

Here, as in the spectrum correspondence calculation unit 4, the result depends on from which combination of the frequency spectrum corresponding to the peak the magnification ratio and the rotation angle are calculated due to the positional error of the peak. It is common to be different.
At this time, the final enlargement ratio and rotation angle may be calculated by simple averaging, but Fourier coefficients can be used.
That is, the evaluation is performed with priority given to the combination of the frequency spectrum corresponding to the peak having a large absolute value of the Fourier coefficient, or by giving a weight. Although the determination can be made without using the Fourier coefficient as in the method using the simple average, the accuracy of the calculated enlargement ratio and rotation angle can be increased by using the Fourier coefficient.

The parallel movement determination unit 6 specifies the parallel movement amount of the input image with respect to the original image and the presence / absence of point symmetric movement, and outputs parallel movement information and point symmetric information.
The translation information represents how many pixels are translated in the x-axis positive direction (rightward) and how many pixels are translated in the y-axis positive direction (downward). However, other representation methods such as a translation amount and a movement direction (angle) may be used.
In order to eliminate the uncertainty of the point target remaining in the spectrum correspondence calculation unit 4, whether the point symmetry information may be the correspondence as the spectrum correspondence information or not (the spectrum correspondence is the true correspondence) It is information indicating whether or not it is point-symmetric.
Hereinafter, the processing content of the parallel movement etc. determination part 6 is demonstrated concretely.

The processing order determination unit 7 of the parallel movement determination unit 6 evaluates each combination of the peak coordinate and the frequency spectrum whose correspondence is specified by the spectrum correspondence calculation unit 4 (evaluation processing in the parallel movement amount candidate selection unit 8). Is executed, the order of combinations for executing the evaluation process is determined based on the Fourier coefficients in the peak coordinates (step ST6).
For example, the absolute value of the Fourier coefficient at the peak coordinate point is referred to, and the order in which the evaluation process is executed is determined in order from the combination having the largest absolute value.

The parallel movement amount candidate selection unit 8 of the parallel movement etc. determination unit 6 determines the peak coordinates and the frequency spectrum related to the combination in the order determined by the processing order determination unit 7 in the order from the previous combination. By substituting it into an evaluation expression (the following expression (1)) whose evaluation value changes depending on the amount of parallel movement, and determining whether or not the evaluation expression is satisfied, all the candidates for the parallel movement amount of the input image are determined. The translation amount is narrowed down from the inside (step ST7).
In other words, the parallel movement amount candidate selection unit 8 is a combination of all parallel movement amount candidates (the number of moving pixels in the x-axis direction (M number of pixels) and the number of moving pixels in the y-axis direction (M number of pixels). The process of selecting each candidate is performed by performing threshold evaluation in the order of the processing order determined by the processing order determination unit 7 from (all M ² types)).
Similarly, for the correspondence relationship indicated by the spectrum correspondence information and the correspondence relationship that is point-symmetric, processing for selecting each candidate is performed.

Evaluation formula: D (m, n, k) ≦ Th (k) (1)
However,
D (m, n, k) = | E (m, n, k) −EI (m, n, k) |
E (m, n, k) = mXi / M + nYi / M + (Qk−Pi) / 2π
(Spectrum corresponding to Tk is Si)
EI (m, n, k) = E (m, n, k) nearest integer Th (k) is a threshold value Pi used for evaluating the peak Tk. Pi is a frequency spectrum corresponding to the peak Tk to be evaluated. Si phase Qk is Tk phase

The parallel movement amount candidate selection unit 8 evaluates whether or not the above evaluation formula is satisfied for all combinations (m, n) of m and n in the first evaluation.
After the second time, only the remaining candidate (m, n) is performed. In this way, candidates (m, n) are narrowed down.
If the candidate is unique before evaluating all the peaks Tk, the process is terminated.
If a plurality of candidates are left after evaluating all the peaks Tk, all of the remaining (m, n) are held as possible parallel movement amounts.
Regarding the removal of point symmetry uncertainties, the translation correspondence candidates are narrowed down in two ways, in the case of spectrum correspondence information and in the case of point symmetry with the spectrum correspondence information, and the remaining one of the translation displacement candidates is selected. Do that.

FIG. 7 is an explanatory diagram showing an outline of the parallel movement amount determination processing by the parallel movement amount candidate selection unit 8.
Even if the threshold Th (k) is a fixed value that does not depend on k, the image position specifying apparatus of FIG. 1 operates, but a reliable peak Tk is evaluated because the absolute value of the Fourier coefficient is large. In this case, it is possible to increase the narrowing-down efficiency by threshold evaluation by performing control so as to give a small threshold Th (k).
If no translation amount candidate finally remains, the value of the threshold Th (k) is increased, and the translation amount candidate of the correct answer may be picked up by narrowing down the translation amount candidates again by threshold evaluation. is there. If no candidate still remains, it is treated as a failure.
In addition, when many candidates remain after all threshold evaluations are completed, the threshold Th (k) is reduced for those remaining candidates and narrowed down by threshold evaluation again. You can narrow down with. Also, if s is set sufficiently large (in the example of FIG. 7, s = 3 but s = 30, etc.), the probability of narrowing down to one candidate is further increased, and the reliability of position specification is increased. Rise.

In this way, the threshold evaluation is repeated and the parallel movement amount candidates are narrowed down, so that it is possible to determine the parallel movement amount with an average calculation amount of O (M ² ).
The basis for this will be briefly described below.
However, for simplicity of explanation, the threshold value Th (k) is set to a constant value Th regardless of k.
If there is no particularly biased characteristic in the image, the phase of the Fourier transform coefficient of the image is randomly distributed. Therefore, the remaining translation amount candidates are reduced to about 2 Th by narrowing down by one threshold evaluation using the threshold Th. Can be doubled. Therefore, it is possible to narrow down to (2Th) ^s times by narrowing down ^s times.
The number of threshold evaluation, but the first time is ^{twice M,} about the second time ^M 2 (2Th) times, the third time will be the ^M 2 ^(2Th) about ^two times, s round of ^M 2 ^{(2Th) s-1} About once.

That is, when peaks corresponding to all s spectra can be detected, the threshold evaluation times are all,
M ² + M ² (2Th) + M ² (2Th) ² +... + M ² (2Th) ^s−1
≦ M ² / (1-2Th) = O (M ² )
And is suppressed to O (M ² ) regardless of the value of s.
Therefore, specifying the parallel movement amount by this method results in an average calculation amount of O (M ² ). Therefore, it is possible to process at high speed as compared with the parallel movement amount determination utilizing correlation operation that requires the computation of O (M ^4).
Compared with the method of determining the parallel movement amount by the calculation amount of O (M ² ) as in the method described in Patent Document 1, the four arithmetic operations and the threshold evaluation are performed regardless of complicated operations such as trigonometric functions. Since the amount of translation can be determined, processing can be performed at a somewhat higher speed.

As described above, enlargement / reduction information, rotation information, inversion information, parallel movement information, and point symmetry information are gathered. Also, correct rotation information can be obtained from the rotation information and point symmetry information.
Therefore, the correct enlargement ratio (when scaling is present), the correct rotation angle (when rotation is present), the vertical elevation angle from the correct origin to the reversal axis (if reversal is present), and the correct amount of translation are obtained. .
Since the object of the image position specifying device in FIG. 1 is alignment for detecting digital watermark data, the image position specifying device in FIG. 1 is usually mounted as a part of the digital watermark data detecting device.
The digital watermark data is usually embedded with regular repetition (see FIG. 8) so that it can be detected even when only a part of the image remains.

FIG. 8 shows an example in which digital watermark data having the same size (256 pixels × 256 pixels) as the alignment signal pattern is embedded in accordance with the repetition of the alignment signal pattern.
At this time, the electronic watermark data may be superimposed on the alignment signal pattern and embedded in the luminance, or may be embedded in other components such as a color difference component. The digital watermark data embedding method and detection method are out of the scope of the present invention, but if the repetition period is adjusted, the registration with the information obtained by the image position specifying device of FIG. Detection is possible.

In the description so far, the parallel movement amount candidate selection unit 8 of the determination unit 6 for parallel movement and the like corresponds to the case where the parallel movement amount candidate is in accordance with the spectrum correspondence information selected by the threshold evaluation, and the case where the spectrum correspondence information is point-symmetric. The parallel movement amount candidates are compared, and the remaining one is selected, and the point-symmetric uncertainty is resolved depending on which one remains.
However, there are cases where the translation amount candidate remains in both cases according to the spectrum correspondence information and point symmetry with the spectrum correspondence information.
In such a case, there is a method in which the threshold Th (k) is varied to adjust the number of remaining candidates, and the method is repeated until a result in which only one of the candidates remains is obtained, but there is also a method using a score.

For example, each parallel movement amount candidate clears the threshold evaluation, and how much the evaluation value D (m, n, k) has a margin with respect to the threshold Th (k) (for example, Th (k) −D ( m, n, k)) is added to give a score.
If a score is given in this way, the higher the score among the finally remaining translation amount candidates, the more the threshold evaluation is cleared, and there is a possibility that it is the correct translation amount. It will be expensive.
Finally, there is a method of finally selecting the one with the highest score as the translation amount among the candidates remaining both in the case of the spectral correspondence information and in the case of point symmetry with the spectrum correspondence information. . This method has the advantage of speeding up the processing because there is no need to reset the threshold and perform threshold evaluation again even if a plurality of parallel movement amount candidates remain.

In the description so far, the input image is a digital uncompressed luminance image, but it may be a color image typified by a YCbCr format or an RGB format.
Even in such a case, for example, by selecting one component such as Y or R and assuming that the image data composed of the one component is an uncompressed luminance image, the image position specifying device of FIG. 1 can be applied. However, it is necessary to know in advance which image component the alignment signal pattern is embedded in.
For example, when the alignment signal pattern is embedded in the G component in the RGB format, when aligning the image in the YCbCr format, the image is first converted into the RGB format, and then the G component is converted. On the other hand, by applying the image position specifying apparatus of FIG. 1 (giving an image consisting only of the G component as an input image to the image position specifying apparatus of FIG. 1), it becomes possible to perform alignment.

  In addition, when the input image is encoded (compressed) by some encoding method (compression method) such as JPEG or GIF, after decoding (expanding), pay attention to the target uncompressed component, The image position specifying device in FIG. 1 may be applied. As described above, the image position specifying apparatus of FIG. 1 can be applied regardless of the presence / absence of a color, the color space format, and compression / non-compression of an image.

In the description so far, the pattern size (M) and the spectrum information (s, Si, Xi, Yi, i∈ [1, s]) are set as defaults, and the setting method thereof is not particularly mentioned. The input image is processed on the premise that the alignment signal pattern is embedded with the fixed parameters.
As another mounting method, for example, a key (data of several bits) is defined, and the operation is performed using parameters such as M, s, Si, Xi, and Yi that are uniquely determined from the key. One image position specifying device can also be configured.
In this way, it is possible to realize a method that enables alignment only when the key used for embedding the alignment signal pattern is known.

When a key different from the key for generating the alignment signal pattern embedded in the input image is used, alignment is attempted with an incorrect parameter, so that a correct result cannot be reached.
The mapping from the key to the parameter group (M, s, Si, Xi, Yi) is preferably performed through a hash function in order to increase the randomness, which makes it difficult to predict the parameter group. Therefore, it is possible to suppress the occurrence of a situation where the matching of the parameters is expected and improve safety. If the alignment cannot be performed, of course, the detection of the digital watermark data after the alignment also fails. Therefore, by using the key in this way, the effect of increasing the confidentiality of the digital watermark data can be obtained.

In the description so far, the one that performs alignment from one partial image cut out by the image cutout unit 1 has been described. However, the image cutout unit 1 cuts out a plurality of different partial images, and each of the plurality of partial images. Alignment is performed by the method described so far, and the final result is obtained by comprehensively judging the obtained result to obtain final enlargement / reduction information, rotation information, inversion information, parallel movement information, and point symmetry information. You can also
In this way, even if the alignment signal pattern is severely degraded in some partial images and accurate information cannot be obtained, the majority decision of enlargement / reduction, rotation / non-rotation, reversal / non-point symmetry information is made. The accuracy of various types of information can be improved by taking the average of the enlargement ratio, the rotation angle, and the elevation angle of the perpendicular line from the origin to the reversal axis.
The parallel movement amount varies greatly depending on the position of the partial image, but the accuracy can be improved by taking an average after correcting the position fluctuation of the partial image.

In the description so far, the configuration in which the image position specifying device in FIG. 1 is mounted with the enlargement determination unit 5 is shown, but a configuration in which the image position determination device is not mounted with the enlargement determination unit 5 is also possible. It is.
In this case, since enlargement / reduction information, rotation information, and inversion information cannot be obtained, the image position specifying device can handle only cropping and 180-degree rotation.

In the description so far, the parallel movement amount candidate selection unit 8 has shown that the parallel movement amount is determined with a granularity of one pixel (M granularity with respect to M pixels). It is also possible to determine the parallel movement amount with a granularity of increments (2M granularity with respect to M pixels), determine with a finer granularity, or conversely with a coarse granularity.
In particular, when determining with a granularity of less than one pixel, it is possible to determine with the same algorithm using the same formula as in the case of integer precision, which is an advantage of the image position specifying apparatus of FIG. This is because the method of determining the parallel movement amount by the correlation calculation cannot immediately perform the correlation calculation at the decimal pixel difference between the image and the alignment signal pattern, and the calculation for calculating the pixel value at the real coordinate position by some filter calculation or the like is unnecessary. It is necessary for this.

  As is apparent from the above, according to the first embodiment, the processing order determination unit 7 executes the evaluation process for each combination of the peak coordinate and the frequency spectrum whose correspondence is specified by the spectrum correspondence calculation unit 4. At that time, the order of the combinations for executing the evaluation process is determined based on the Fourier coefficient in the peak coordinates, and the parallel movement amount candidate selecting unit 8 determines that the order determined by the processing order determining unit 7 is the order from the previous combination. By substituting the peak coordinate and frequency spectrum related to the combination into an evaluation formula whose evaluation value changes depending on the amount of translation of the input image with respect to the original image, it is determined whether or not the evaluation formula is satisfied. Since the parallel movement amount is narrowed down from all the parallel movement amount candidates, the alignment speed associated with the editing of the cutout is increased. An effect that can be achieved.

Embodiment 2. FIG.
In the first embodiment, the parallel movement amount candidate selection unit 8 has shown that the parallel movement amounts m and n on the X axis and the Y axis are simultaneously narrowed down. However, the parallel movement amount candidate selection unit 8 is parallel on the X axis. By further reducing the movement amount m and the parallel movement amount n on the Y axis, the processing speed may be further increased.
The configuration diagram and flowchart of the image position specifying device are the same as those in the first embodiment.
FIG. 9 is an explanatory diagram showing an outline of the parallel movement amount determination processing by the parallel movement amount candidate selection unit 8.

Compared with the first embodiment, the image cutout unit 1, the Fourier transform unit 2, the peak detection unit 3, the spectrum correspondence calculation unit 4 and the enlargement determination unit 5 have the same processing contents, but the parallel movement determination The processing content of the part 6 is different.
That is, in the first embodiment, the frequency spectrum corresponding to the peak Tk is Si, the phase of Si is Pi (Si = | Si | exp (jPi)), and the phase of Tk is Qk (Tk = | Tk | exp ( jQk)),
E (m, n, k) = mXi / M + nYi / M + (Qk−Pi) / 2π≈integer (any combination of corresponding Si and Tk)
M and n satisfying the above are selected.

On the other hand, in the second embodiment, the frequency spectrum corresponding to the peak Tk is Si, the phase of Si is Pi (Si = | Si | exp (jPi)), and the phase of Tk is Qk (Tk = | Tk). | Exp (jQk)), the frequency spectrum corresponding to Tq is Sp, the phase of Sp is Pp (Sp = | Sp | exp (jPp)), and the phase of Tq is Qq (Tq = | Tq | exp (jQq)) , I, pεSetX (X) = {i | Xi = X, iε [1, s]}, i ≠ p,
Hy (n, k, q) = n (Yi−Yp) / M + (Qk−Pi−Qq + Pp) / 2π
≒ integer (any combination of corresponding Si and Tk, Sp and Tq)
N that satisfies is selected.
When i, pεSetY (Y) = {i | Yi = Y, iε [1, s]}, i ≠ p,
Hx (m, k, q) = m (Xi−Xp) / M + (Qk−Pi−Qq + Pp) / 2π
≒ integer (any combination of corresponding Si and Tk, Sp and Tq)
M that satisfies is selected.
Thereby, the amount of parallel movement becomes clear, and the alignment of the input image and the original image is completed.

Hereinafter, the processing contents of the processing order determination unit 7 and the parallel movement amount candidate selection unit 8 of the parallel movement determination unit 6 in Embodiment 2 will be described in detail.
Similar to the first embodiment, the processing order determination unit 7 evaluates each combination of the peak coordinates and the frequency spectrum for which the correspondence is specified by the spectrum correspondence calculation unit 4 (evaluation in the parallel movement amount candidate selection unit 8). When the process is executed, the order of combinations for executing the evaluation process is determined based on the Fourier coefficient at the peak coordinates.
However, unlike the first embodiment, the processing order determination unit 7 determines the order of two combinations because there are two combinations of peak coordinates and frequency spectra used for one evaluation process.
For example, for the combination of the maximum absolute value of the Fourier coefficient at the peak point and the combination with other peak points, the peak point with the absolute value that is not maximum is combined in order from the largest absolute value of the Fourier coefficient. To do.

For alignment in the x-axis direction, T1 (corresponding to S1), T2 (corresponding to S2), T3 (corresponding to the conjugate complex number of S2), T4 (corresponding to S3) or T8 (corresponding to the conjugate complex number of S3) is used. Can be used. Among these, T2 has the highest absolute value of the Fourier coefficient. Therefore, a combination of T2 and other peaks is used.
In T1 and T4, the absolute value of the Fourier coefficient is the same. Thus, any order is acceptable.
For example, the alignment in the x-axis direction may be performed in the order of (T2, T1) and (T2, T4). On the other hand, T1 (corresponding to S1), T6 (corresponding to S5), and T7 (corresponding to S4) can be used for alignment in the y-axis direction. Among these, T7 has the highest absolute value of the Fourier coefficient. Therefore, a combination of T7 and other peaks is used.
In T1 and T6, the absolute value of the Fourier coefficient is larger in T6. Therefore, alignment in the y-axis direction is performed in the order of (T7, T6) and (T7, T1).

The parallel movement amount candidate selection unit 8 includes all candidates for the parallel movement amounts in the x-axis direction and the y-axis direction, that is, the number of moving pixels in the x-axis direction (M number of pixels) and the number of moving pixels in the y-axis direction ( The threshold value is evaluated in the order of the processing order determined by the processing order determination unit 7 from among the M number of pixels), so that each candidate in the x-axis direction is selected and each candidate in the y-axis direction is selected. Sorting is done separately.
Similarly, with respect to the correspondence relationship indicated by the spectrum correspondence information and the point-symmetric correspondence relationship, selection of each candidate in the x-axis direction and selection of each candidate in the y-axis direction are performed separately.

  Specifically, the parallel movement amount candidate selection unit 8 determines whether or not the following expression (2) is satisfied, thereby selecting each candidate in the x-axis direction. By discriminating whether or not the evaluation formula (3) holds, each candidate in the y-axis direction is selected.

Evaluation formula: Dx (m, k, q) ≦ Thx (k, q) (2)
However,
Dx (m, k, q) = | Hx (m, k, q) −HxI (m, k, q) |
Hx (m, k, q) = E (m, n, k) −G (m, n, q)
= M (Xi−Xp) / M + (Qk−Pi−Qq + Pp) / 2π
(Yi = Yp, iεSetY (Yi), pεSetY (Yi))
E (m, n, k) = mXi / M + nYi / M + (Qk−Pi) / 2π
(Spectrum corresponding to Tk is Si)
G (m, n, q) = mXp / M + nYp / M + (Qq−Pp) / 2π
(Spectrum Sp corresponding to Tq)
HxI (m, k, q) = nearest integer of Hx (m, k, q)

Evaluation formula: Dy (n, k, q) ≦ Thy (k, q) (3)
However,
Dy (n, k, q) = | Hy (n, k, q) −HyI (n, k, q) |
Hy (n, k, q) = E (m, n, k) −G (m, n, q)
= N (Yi−Yp) / M + (Qk−Pi−Qq + Pp) / 2π
(Xi = Xp, iεSetX (Xi), pεSetX (Xi))
HyI (n, k, q) = nearest integer of Hy (n, k, q)

Here, Thx (k, q) and Thy (k, q) are threshold values used for evaluating the combination of Tk and Tq when narrowing the parallel movement amounts in the x-axis direction and the y-axis direction, respectively.
When evaluating whether or not the evaluation formula of Expression (2) is satisfied with respect to the parallel movement amount in the x-axis direction, the first evaluation is performed for all m, and the second and subsequent tests are performed only for the remaining m.
When evaluating whether or not the evaluation formula (3) is satisfied with respect to the amount of translation in the y-axis direction, the first evaluation is performed for all n, and the second and subsequent tests are performed only for the remaining n.
In this way, candidate m and n are narrowed down separately.
If the candidate is unique before evaluating all combinations of Tk and Tq, the process is terminated.
If a plurality of candidates are left after evaluating all Tk, all the remaining m and n are held as possible parallel movement amounts.
Regarding the removal of point symmetry uncertainties, the translation correspondence candidates are narrowed down in two ways, in the case of spectrum correspondence information and in the case of point symmetry with the spectrum correspondence information, and the remaining one of the translation displacement candidates is selected. Do that.

In this way, the threshold value evaluation is repeated and the parallel movement amount candidates are narrowed down, so that it is possible to determine the parallel movement amount with an average calculation amount of O (M).
The basis for this will be briefly described below.
However, for simplicity of explanation, the threshold value Thx (k, q) and the threshold value Thy (k, q) are set to a constant value Th regardless of k.
If there is no particularly biased characteristic in the image, the phase of the Fourier transform coefficient of the image is randomly distributed. Therefore, the remaining translation amount candidates are reduced to about 2 Th by narrowing down by one threshold evaluation using the threshold Th. Can be doubled. Therefore, since the amount of parallel movement to be examined is M in both the x-axis direction and the y-axis direction, when this narrowing is repeated, the number of threshold evaluations is
M + M (2Th) + M (2Th) ² +... ≦ M / (1-2Th) = O (M)
Thus, O (M) is suppressed regardless of the value of s.
In the second embodiment, the x-axis direction and the y-axis direction are separately performed. However, the number of evaluations is merely doubled, and the calculation amount is still O (M). For this reason, it is possible to determine the amount of parallel movement faster than in the first embodiment.

In the description so far, the peak detection unit 3 operates in the same manner as in the first embodiment. However, as shown in FIG. 12B, a plurality of frequency spectrum coordinates exist on the same straight line. It is assumed that an input image in which an alignment signal pattern is set is handled.
Therefore, instead of using the peak detected by the peak detector 3 of the first embodiment as it is, a straight line that reduces the distance from the detected plurality of peaks (for example, a regression line obtained from the plurality of peaks). There is also a method of generating a peak coordinate having a peak at an intersection (perpendicular point) between the perpendicular line and the straight line of the peak drawn toward (). By doing in this way, the effect which correct | amends the peak detected by having shifted | deviated from the straight line on a straight line is anticipated, and the precision of a peak coordinate can be improved.

Embodiment 3 FIG.
FIG. 10 is a block diagram showing an alignment signal embedding device according to Embodiment 3 of the present invention.
In FIG. 10, an inverse Fourier transform unit 11 performs a process of performing inverse Fourier transform on spectrum information to generate an alignment signal pattern having an input pattern size.
In the s frequency spectra Si (i∈ [1, s]) indicated by the spectrum information subjected to inverse Fourier transform by the inverse Fourier transform unit 11, as shown in FIG. May be included (for example, the frequency spectra S1, S2, and S3 have the same X coordinate, and the frequency spectra S1, S4, and S5 have the same Y coordinate. ).
The inverse Fourier transform unit 11 constitutes an alignment signal pattern generation unit.
The intensity adjusting unit 12 performs a process of adjusting the intensity of the alignment signal pattern generated by the inverse Fourier transform unit 11 in accordance with the image characteristics of the original image. The strength adjusting unit 12 constitutes a strength adjusting means.
The embedding unit 13 performs a process of embedding the alignment signal pattern whose intensity is adjusted by the intensity adjusting unit 12 in the original image. The embedding unit 13 constitutes an alignment signal pattern embedding unit.

In the example of FIG. 10, each of the inverse Fourier transform unit 11, the intensity adjustment unit 12, and the embedding unit 13, which are components of the alignment signal embedding device, has dedicated hardware (for example, a semiconductor integrated circuit on which a CPU is mounted). In the case where the alignment signal embedding device is constituted by a computer or the like, an inverse Fourier transform unit 11, an intensity adjustment unit 12, and All or part of the program describing the processing content of the embedding unit 13 may be stored in the memory of a computer, and the CPU of the computer may execute the program stored in the memory.
FIG. 11 is a flowchart showing the processing contents (alignment signal embedding method) of the alignment signal embedding apparatus according to Embodiment 1 of the present invention.

Next, the operation will be described.
In the third embodiment, an alignment signal embedding device that generates a digital image in which an alignment signal pattern has been embedded, which can be aligned by the image position specifying device of FIG. 1, will be described.

When the inverse Fourier transform unit 11 receives the spectrum information and the pattern size, the inverse Fourier transform is performed on the spectrum information to generate an alignment signal pattern having the pattern size (step ST11 in FIG. 11).
Here, the spectrum information is the coordinates and phase on the frequency plane of the frequency spectrum in the alignment signal pattern.
In order for the inverse Fourier transform unit 11 to perform the inverse Fourier transform, a spectrum absolute value (amplitude) is also necessary, but since a fixed absolute value may be used for all spectra, it is not necessary to specify it.

The spectrum information is s frequency spectra Si (i∈ [1, s]), and the position of the frequency spectrum Si on the frequency plane is (Xi, Yi) (Xi∈ [1, M / 2-1]). And the phase of the frequency spectrum Si is Pi (Si = | Si | exp (jPi)), and when the pattern size is M, the Fourier transform F (X, Y) of the alignment signal pattern (X∈ [−M / 2, M / 2-1], Y∈ [−M / 2, M / 2-1]) is as follows.
F (X, Y) = Si = exp (jPi)
((X, Y) ε {(Xi, Yi) | iε [1, s]})
Conjugated complex number of Si ((X, Y) ∈ {(− Xi, −Yi) | i∈ [1, s]})
0 (other (X, Y))

Here, the absolute value of the frequency spectrum Si is unified with 1, but other values may be used. Since an alignment signal pattern is obtained by performing inverse Fourier transform on this, the alignment signal pattern f (x, y) (x∈ [−M / 2, M / 2-1], y∈ [−M / 2, M / 2-1]) is as follows.
f (x, y)
= (1 / M) Σ _{(X, Y)} F (X, Y) exp (j2πxX / M) exp (j2πyY / M)

When the inverse Fourier transform unit 11 generates the alignment signal pattern, the intensity adjustment unit 12 adjusts the intensity of the alignment signal pattern according to the image characteristics of the original image, and the intensity adjustment unit 12 outputs the alignment signal pattern after the intensity adjustment. Output as an embedding pattern (step ST12).
The original image is Image (x, y) (x∈ [0, W−1], y∈ [0, H−1]), and the embedding strength for the pixel (x, y) is intensity (x, y). Then, the embedding pattern emb (x, y) (x∈ [0, W−1], y∈ [0, H−1]) is as follows.
emb (x, y)
= Truncate {intensity (x, y)
Xf (((x + M / 2) mod M) -M / 2, ((y + M / 2) mod M) -M / 2)
/ Max _{(x ′, y ′)} | f (x ′, y ′) |}
However, truncate (z) is a function that returns an integer closest to z by rounding off, for example.

In the above equation, the embedding pattern f (x, y) is normalized by multiplying it by a real number so that the maximum absolute value of the embedding pattern becomes 1, and then the intensity is multiplied.
Thereby, the embedding pattern emb (x, y) adjusted to an appropriate intensity intensity (x, y) can be generated for the pixel (x, y).
The alignment signal pattern is a square having an M pixel length. By applying this to the original image cyclically, the embedding pattern can be applied to the original image image (x, y) of any size. emb (x, y) can be generated.
Intensity (x, y) is usually a function of x, y and image, a large value when judging that the image features of the original image image have a high tolerance for image quality degradation, and a small value otherwise. Value or 0. By adjusting the strength in this way, it is possible to achieve both the effect of suppressing the deterioration of the subjective image quality and the effect of improving the accuracy at the time of alignment by increasing the embedding amount.

When receiving the embedding pattern from the intensity adjustment unit 12, the embedding unit 13 performs a process of embedding the embedded pattern in the original image (step ST13).
Embedding the embedding pattern is usually a simple addition. However, when the result of the addition exceeds the range of the pixel value, a process of truncating within the range is added.
The truncation process may be performed after the addition, or before the addition, the truncation process may be performed by embedding all pixel values of the original image with the intensity of max _{(x, y)} intensity (x, y). It may be done in advance so that it is not necessary.

In the description so far, no particular restriction is imposed on the coordinates (Xi, Yi) of the frequency spectrum Si, but the coordinates are set so that a specific relationship is established between the coordinates (Xi, Yi) of the frequency spectrum Si. By adjusting (Xi, Yi), it is possible to increase the processing speed during alignment.
Here, the relationship between the coordinates (Xi, Yi) enabling high-speed processing is as follows. When the sets SetX and SetY are defined as follows, max _X | SetX (X) | and max _Y | SetY (Y) Is to take a large value.
SetX (X) = {i | Xi = X, i∈ [1, s]}
SetY (Y) = {i | Yi = Y, i∈ [1, s]}

FIG. 12 shows an example of a case where such a relationship is not satisfied (a) and a case where the relationship is satisfied (b).
As shown in FIG. 12B, by setting the spectral coordinates, the image position specifying device of the second embodiment can be processed.
In the alignment by the image position specifying device according to the first embodiment, the amount of calculation of O (M ² ) is required on the average by the parallel movement determination unit 6, whereas the image position according to the second embodiment. In the alignment by the specific device, the parallel movement determination unit 6 can process the calculation amount of O (M) on average.

  In the description so far, the frequency spectrum is subjected to inverse Fourier transform to generate an alignment signal pattern, and then the embedding pattern is embedded in the original image by addition on the image. As is apparent, an output image can be obtained by performing an inverse Fourier transform after Fourier transforming the original image and adding a frequency spectrum to the original image. However, the configuration of the third embodiment is excellent in that the intensity adjustment can be performed according to the image position.

In the description so far, the alignment signal embedding device of FIG. 10 has been shown in which the strength adjustment unit 12 is mounted. However, the strength adjustment unit 12 may be omitted.
In this case, the alignment signal pattern becomes an embedded pattern as it is.
Moreover, since it is necessary to adjust an appropriate intensity in advance (for example, when the pixel value takes a value from 0 to 255, the intensity needs to be adjusted in advance to 1 or 2), the reverse When the Fourier transform unit 11 outputs the alignment signal pattern, it is necessary to add a process of multiplying the absolute value of the alignment signal pattern by a real number so that the maximum value or the average value of the alignment signal pattern is set. is there.

In the description so far, the main purpose is to enable alignment when detecting digital watermark data embedded at the same time. Therefore, in many cases, the alignment signal embedding device shown in FIG. 10 is used together with the digital watermark data embedding device or is incorporated into a part of the digital watermark data embedding device.
At that time, the digital watermark data is normally embedded in such a manner that the same embedding is repeatedly applied so that the digital watermark data can be detected even if a part of the image is lost due to clipping.
FIG. 8 shows an example in which the repetition of the alignment signal pattern and the repetition of the digital watermark data are combined. If the alignment can be performed by the image position specifying device of FIG. 1, the digital watermark data can be detected.

  In the above description, when the embedding pattern emb (x, y) is generated from the alignment signal pattern f (x, y), the upper left of the alignment signal pattern is aligned with the upper left of the original image, and is cyclically applied. However, it is not always necessary to match the upper left. However, since the purpose is to enable detection of digital watermark data embedded simultaneously with the alignment signal pattern, the alignment signal pattern and the electronic watermark data must be synchronized in position.

  In the present invention, within the scope of the invention, any combination of the embodiments, or any modification of any component in each embodiment, or omission of any component in each embodiment is possible. .

  DESCRIPTION OF SYMBOLS 1 Image cut-out part (partial image generation means), 2 Fourier transform part (Fourier transform means), 3 Peak detection part (peak coordinate specification means), 4 Spectral correspondence calculation part (correspondence relation specification means), 5 Enlargement determination part (Identification means such as enlargement), 6 Parallel movement determination section, 7 Processing order determination section (Evaluation order determination means), 8 Parallel movement amount candidate selection section (Parallel movement amount identification means), 11 Inverse Fourier transform section (Positioning signal) Pattern generating means), 12 intensity adjusting section (intensity adjusting means), 13 embedding section (alignment signal pattern embedding means).

Claims (10)

  By inputting an image that may have been edited with respect to the original image, and extracting a part of the image from the input image, the same size as the alignment signal pattern embedded in the original image A partial image generating means for generating a partial image, a Fourier transform means for performing a Fourier transform on the partial image generated by the partial image generating means, and outputting a Fourier coefficient as a result of the Fourier transform, and an output from the Fourier transform means. Among the Fourier coefficients, the peak coordinate specifying means for specifying the peak coordinates, which are the coordinates of the Fourier coefficients whose absolute values are prominent, and the spectrum coordinates which are the coordinates on the frequency plane of the frequency spectrum in the alignment signal pattern. Among them, the correspondence for specifying the spectral coordinates corresponding to the peak coordinates specified by the peak coordinate specifying means When executing the evaluation process for each combination of the peak coordinate and the frequency spectrum whose correspondence is specified by the relationship specifying means and the correspondence specifying means, the order of the combination for executing the evaluation process is set to the Fourier coefficient in the peak coordinate. The order determined by the evaluation order determining means, and the order determined by the evaluation order determining means in order from the previous combination, the peak coordinates and the frequency spectrum related to the combination are calculated, and the parallel movement amount of the input image with respect to the original image By substituting into an evaluation expression whose evaluation value changes according to the above, it is determined whether or not the evaluation expression is satisfied, thereby narrowing down the parallel movement amount from all the candidates for the parallel movement amount of the input image. An image position specifying device comprising a parallel movement amount specifying means.   The parallel movement amount specifying means substitutes the peak coordinates and the frequency spectrum relating to the point-symmetrical combination with the correspondence specified by the correspondence specifying means into the evaluation formula to determine whether or not the evaluation formula is satisfied. The image position specifying device according to claim 1, wherein the parallel movement amount is narrowed down from all candidates for the parallel movement amount of the input image.   3. The image position specifying apparatus according to claim 2, wherein the parallel movement amount specifying means determines whether or not there is a point-symmetric movement based on a combination of the peak coordinates and the frequency spectrum related to the finally narrowed parallel movement amount.   2. An identification means for enlarging, for example, enlarging / reducing, rotating and inverting an input image with respect to an original image from a combination of peak coordinates and frequency spectrum whose correspondence is identified by the correspondence identifying means. The image position specifying device according to claim 3.   5. The parallel movement amount specifying means performs the narrowing of the parallel movement amount in the X axis and the narrowing of the parallel movement amount in the Y axis separately. 6. Image location device.   An alignment signal pattern generation unit that generates an alignment signal pattern having a specified pattern size by performing inverse Fourier transform on a plurality of frequency spectra, and an alignment signal pattern generated by the alignment signal pattern generation unit. An alignment signal embedding device comprising alignment signal pattern embedding means for embedding in an image.   7. A plurality of frequency spectra having the same X-coordinate or Y-coordinate relationship are included in the plurality of frequency spectra subjected to inverse Fourier transform by the alignment signal pattern generation means. Alignment signal embedding device. In accordance with the image characteristics of the original image, provided is an intensity adjustment means for adjusting the intensity of the alignment signal pattern generated by the alignment signal pattern generation means,
8. The alignment signal embedding apparatus according to claim 6, wherein the alignment signal pattern embedding unit embeds the alignment signal pattern whose intensity is adjusted by the intensity adjusting unit in an original image.   The partial image generation means inputs an image that may have been edited with respect to the original image, and extracts a part of the image from the input image, thereby aligning the alignment signal embedded in the original image. A partial image generation processing step for generating a partial image having the same size as the pattern, and a Fourier transform unit Fourier-transforms the partial image generated in the partial image generation processing step and outputs a Fourier coefficient as a result of the Fourier transform. A Fourier transform processing step, and a peak coordinate specifying means for specifying a peak coordinate that is a coordinate of a Fourier coefficient with an absolute value protruding from among the Fourier coefficients output in the Fourier transform processing step. And the correspondence specifying means is the coordinates on the frequency plane of the frequency spectrum in the alignment signal pattern. Among the spectrum coordinates, a correspondence relationship identification processing step for identifying a spectrum coordinate having a correspondence relationship with the peak coordinate identified in the peak coordinate identification processing step, and an evaluation order determining means correspond in the correspondence relationship identification processing step. An evaluation order determination processing step for determining the order of the combination for executing the evaluation processing based on the Fourier coefficient in the peak coordinates when executing the evaluation processing for each combination of the peak coordinate and the frequency spectrum for which the identification is specified. The amount specifying means determines the peak coordinate and the frequency spectrum related to the combination in the order determined in the evaluation order determination processing step from the previous combination, and the evaluation value is based on the amount of translation of the input image relative to the original image. Substituting into the changing evaluation formula, whether the above evaluation formula is satisfied By determining, from among all candidate translation of the input image, image localization method and a parallel movement amount determination processing step of performing a narrowing of the parallel movement amount.   The signal pattern generation means performs an inverse Fourier transform on the plurality of frequency spectra to generate an alignment signal pattern having a pattern size designated by the alignment signal pattern generation processing step, and the alignment signal pattern embedding means includes the above An alignment signal embedding method comprising: an alignment signal pattern embedding process step for embedding the alignment signal pattern generated in the alignment signal pattern generation process step in the original image.

Images