JP4863504B2 - Data processing apparatus, data processing method, and data processing program - Google Patents

Description

  The present invention relates to a data processing apparatus that performs time-frequency conversion on target data to execute data processing, and more particularly to a data processing apparatus that performs processing using redundancy of target data.

As one of data processing techniques using temporal frequency conversion, there is a digital watermark technique for embedding information using a specific frequency region that is a redundant portion of target data.
In recent years, with the rapid spread of the Internet, the digital content business has expanded. The digital content business is a business in which literary data such as music data and image data is directly transmitted to a user for sale using the Internet. The digital content business is expected to expand further as a sales system in the future because it can reduce distribution time and costs. However, since digital data can be easily copied, there is a possibility that the user may sell it to a third party without permission. Since the content itself does not include information on who owns the copyright, who sold it, and who purchased it, it is difficult to identify the route even if it is redistributed. If such illegal copying becomes routine, it will not be established as a business, copyright will not be protected, and culture will be hindered.

  Thus, digital watermarking is a technique that has been considered to protect copyrights and prevent unauthorized copying. If there is a watermark, at least it is possible to determine who is the copyright holder, and the fact that there is a watermark can also be a deterrent to people trying to make unauthorized copies.

There are techniques shown in Non-Patent Document 1 and Non-Patent Document 2 as digital watermark techniques.
The technique according to Non-Patent Document 1 is a simple digital watermarking method that uses the hierarchical nature of wavelet transform and the high confidentiality of spectrum spread. First, the target image is subjected to wavelet decomposition into three to four levels, and a signature is embedded in the DC component of each stage in the frequency domain by spectrum spreading. If this combined data is inversely converted, the signature information can be concealed and diffused over the entire image area. Since the spread signal is very weak and does not cause large noise with respect to the image signal, the image including the signature appears to be almost the same as the original image. In addition, since the spread code has the same effect as a cryptographic random number key, it is resistant to signature tampering and the like, and a strong signature for copyright protection can be hidden.

The technique according to Non-Patent Document 2 is a digital watermark method in which an image is converted by wavelet transform and a pseudo random number sequence is embedded as watermark information in the obtained wavelet coefficients. This improves the accuracy of watermark information detection.
Further, as another technique for data processing using time frequency conversion, there is a technique for removing noise from target data.

  It is possible to improve the quality of the target data by specifying the frequency that becomes noise in the target data and removing it. That is, for example, if the target data is sound, the sound can be made clear, and if the target data is image data, the image can be made clear. In addition, when voice recognition processing or image recognition processing is performed by removing noise from the target data, it is possible to recognize a specific voice, shape, or the like more accurately.

There is a spectrum subtraction method (hereinafter referred to as SN method) as a technique related to noise removal. The SN method is a method of removing noise by estimating a noise-only spectrum from a target speech spectrum and subtracting it from the target speech spectrum.
Atsushi Onishi and Kokoo Matsui, “Digital Watermarking Using Multi-resolution Analysis and PN Sequence”, IEICE Transactions, November 1997, Vol. J80-D-II, no. 11, p3020-3028 Akio Miyazaki, Yoshiomi Kajishima, “Improvement of Digital Watermarking Using Multi-resolution Analysis of Images”, IEICE Transactions, January 2002, Vol. J85-A, no. 1, p103-111

  However, since the technique of Non-Patent Document 1 uses a spread spectrum code as a pseudo-random number, high confidentiality is maintained, but it is necessary to know the structure of the random number generator when extracting a watermark. It has the subject that it will become work which requires.

  Further, according to Non-Patent Document 1, since signature data is synthesized by dividing a multiresolution approximate image into blocks, the signature cannot be successfully decrypted when subjected to an attack in which a part of the image is cut off. Has a problem.

Since the technology of Non-Patent Document 2 is a technology that uses the relationship between the multi-resolution representation of an image and human visual characteristics, for example, when embedding a watermark in an image, a location where human visual characteristics can be used (for example, In this case, a contour portion, a shadow portion, etc.) must be found out from the image, and a watermark should be embedded therein. Therefore, there is a problem that the processing becomes complicated.
The SN method related to noise removal has a problem that sufficient performance cannot be obtained in an actual environment where various noise levels occur because noise estimation accuracy varies depending on the noise level.

  Incidentally, an arithmetic method called interval arithmetic is one of arithmetic methods mainly studied in Europe. The interval calculation is an operation method in which calculation is performed using an interval between two numerical values, which is different from normal numerical calculation. Since this calculation method is defined so as to encompass all possible value intervals (error ranges), the interval generally expands as the number of operations increases (this phenomenon is referred to as “interval expansion”). For this reason, many researchers (particularly in Japan) believe that interval arithmetic is almost useless except for numerical methods. However, the inventor has found that "interval expansion" automatically provided by a computer in the calculation process can be regarded as useful information and applied to data processing.

  Therefore, the present invention has been made to solve the above-described problem, and when performing data processing on target data using time-frequency conversion, the redundancy of the target data is reduced using interval arithmetic. An object of the present invention is to provide a data processing device, a data processing method, and a data processing program that expand and use the expanded redundant portion as convenient information for processing.

The data processing apparatus according to the present invention performs a time frequency conversion on the target data by interval calculation in a data processing apparatus that executes data processing for embedding image data using time frequency conversion on the target data. and the target data that is inserted and interval arithmetic means for redundant portion having the target data is generated as an extended expansion target data, the image data to the extended target data the interval arithmetic unit has generated image data Image data insertion means for generating superimposition information; and inverse conversion means for reversely converting the image data superposition information generated by the image data insertion means by the time-frequency conversion to generate image insertion data. To do.

As described above, in the present invention, the time frequency conversion is performed on the target data by the interval calculation, so that the redundant portion of the target data is expanded. ) Information can be revealed. In addition, information can be inserted using the expanded area, or secrecy can be easily maintained without using complicated processing by using the section width and information insertion location as a key. it can.
In addition, since an interval calculation means for performing time frequency conversion by interval calculation is provided, an image can be inserted into a redundant portion by regarding the interval expanded by interval calculation as the redundancy of the target data. In other words, it is not necessary to perform complicated processing such as specifying the place to insert an image in consideration of human visual characteristics, and it is only necessary to insert the image data into the area automatically expanded by the device during the calculation process. Good. Therefore, the processing can be simplified and the processing efficiency can be increased.

Further, since the time frequency conversion is simply replaced with the interval calculation, the conventional apparatus can be easily implemented without drastically changing.
Furthermore, since the image data is inserted into the redundant portion of the input signal, the entire image data can be combined and the resistance to attacks such as tampering and clipping is increased.

(3. Time-frequency conversion by wavelet conversion)
The image processing apparatus according to the present invention is characterized in that the section calculation means performs the time-frequency conversion by wavelet conversion.
As described above, in the present invention, since the time-frequency conversion is performed by the wavelet transform, it can be highly resistant to an attack such as compression using the wavelet transform (for example, JPEG2000).

(4. Image composition by wavelet transform)
The image processing apparatus according to the present invention includes: a component selection unit that selects any one component as selected component data from the four components generated as a result of the image data insertion unit calculating by the section calculation unit; Component changing means for changing a component other than the selected component data selected by the component selecting means into component data obtained as a result of wavelet transform by floating-point arithmetic, and obtaining an arbitrary scalar value from the section of the selected component data A scalar value acquisition unit and an image data superimposing unit that superimposes the image data on the scalar value acquired by the scalar value acquisition unit.
Thus, in the present invention, since one component for synthesizing an image is selected from the four components generated by the wavelet transform, it can be used as one of confidential information.

  Further, by performing wavelet transform by section calculation, the value of the obtained component becomes a section, and an arbitrary value is acquired from the section, so that the value can be used as confidential information. That is, even if an interval can be specified, since it is not possible to extract an image unless it is possible to specify which value was acquired in that interval, high confidentiality can be maintained.

  When changing the three components other than the selected component data to the result of wavelet transformation by floating point arithmetic, change the value by performing wavelet transformation by floating point arithmetic separately from the wavelet transformation by interval arithmetic Alternatively, the result of wavelet transform by interval calculation may be converted to a floating point number. That is, it is only necessary that only the selected component data can obtain the result of the interval operation and the other components can obtain the floating-point result as a result.

(5. Calculate the average deviation)
In the image processing apparatus according to the present invention, the image data insertion unit includes the scalar value acquired by the scalar value acquisition unit, and the component data obtained as a result of wavelet transform by floating point calculation in the selected component data. Average calculating means for calculating an average value of deviations for each pixel in the selected component data, and an average image for generating average image data by multiplying the component data of the image data by the average value calculated by the average calculating means And a data generation means.

  As described above, in the present invention, since the image data is inserted into the redundant part expanded by the interval calculation, the target data is likely to be broken, but the average of the deviations of the selected component data for each pixel is calculated and calculated. By multiplying the image data by the image data and generating the average image data, the component value of the image data to be inserted can be suppressed to the average level of the influence of the section expansion, and the influence on the target data can be minimized. be able to.

(6. Acquisition of maximum or minimum value)
The image processing apparatus according to the present invention is characterized in that the scalar value acquired by the scalar value acquisition means is the maximum value or the minimum value of the section of the selected component data.
As described above, in the present invention, since the scalar value acquired by the scalar value acquisition means is the maximum value or the minimum value of the section of the selected component data, the value acquisition is simplified and the section is maximized. By using it, it is possible to synthesize images firmly.

(7. Addition or subtraction)
The image processing apparatus according to the present invention adds the average image data to the maximum value when the scalar value acquired by the scalar value acquisition unit is the maximum value of the section of the selected component data. When the scalar value acquired by the value acquisition means is the minimum value of the section of the selected component data, the average image data is subtracted from the minimum value.

  Thus, in the present invention, when the scalar value acquired by the scalar value acquisition unit is the maximum value of the section of the selected component data, the average image data is added to the maximum value, and the scalar value acquired by the scalar value acquisition unit If the value is the minimum value of the selected component data interval, the average image data is subtracted from the minimum value and the image is inserted. Images can be combined.

(8. Section setting)
The image processing apparatus according to the present invention is characterized by comprising section setting means for setting a section calculated by the section calculating means within a range designated by a user.
Thus, in the present invention, since the section calculated by the section calculation means is set in the range specified by the user, the user can arbitrarily set the section and use the information as confidential information. Can do. That is, confidentiality can be easily maintained without using pseudo-random numbers or the like.

In addition, since the degree of influence on the target data by the synthesized image varies depending on the set value of the section, it is possible to perform processing suitable for each target data by setting the section value according to the quality of the target data. it can. For example, if the target data is an image and the image quality of the original image is rough, it is possible to synthesize the image firmly by expanding the range of the section. The effect is not so much of a concern because of the roughness. Conversely, if the original image is clear, the influence on the original image can be reduced by narrowing the range of the section.
Note that the section may be set for each pixel of the selected component, or may be set uniformly for the selected component.

(9. Image extraction)
The image processing apparatus according to the present invention is obtained as a result of calculation by floating point arithmetic means for performing wavelet transform by floating point arithmetic on the image insertion data into which the image data has been inserted, and by calculation by the floating point arithmetic means. Among the two components, difference calculation means for calculating a difference between the component data of the component in which the image data is inserted and the scalar value acquired by the scalar value acquisition means, and the difference calculated by the difference calculation means Image data extracting means for extracting the image data by dividing by the average value calculated by the calculating means.
As described above, in the present invention, since the synthesized image can be easily extracted simply by using the value acquired by the scalar value acquisition unit and the average of the deviation for each pixel calculated by the average calculation unit, the processing is simplified. . These values can also be used as information for maintaining confidentiality.

  Further, since the image data is inserted, the extracted result can be confirmed by human eyes. In other words, just as the watermark on the banknote can be seen through light, if the extraction process is performed on the image, the inserted image data can be confirmed by visual observation, so that the presence of the image data can be known at a glance. It becomes possible.

The data processing apparatus according to the present invention performs a time frequency conversion on the target data by interval calculation in the data processing apparatus that executes data processing for removing noise using time frequency conversion on the target data. it is, the target data, noise which the target data has to detect the interval arithmetic means for generating as an extended expansion target data, said target data, the noise of the target data from the difference between the expansion target data It is characterized by comprising noise detecting means and noise removing means for removing at least part of the noise detected by the noise detecting means from the target data.
As described above, in the present invention, since the noise of the target data is detected from the difference between the target data and the extended target data expanded by the redundant portion by the interval calculation, it is complicated to estimate the noise or specify the noise. There is no need to perform complicated calculations. Further, since noise can be easily detected, the detected noise can be easily removed.

  Although the present invention has been described as an apparatus, the present invention can be understood as a system, method, and program as will be apparent to those skilled in the art. These outlines of the invention do not enumerate the features essential to the present invention, and a sub-combination of these features can also be an invention.

Embodiments of the present invention will be described below. The present invention can be implemented in many different forms. Therefore, the present invention should not be construed based only on the description of the present embodiment. Also, the same reference numerals are given to the same elements throughout the present embodiment.
In this embodiment, the apparatus will be mainly described. However, as will be apparent to those skilled in the art, the present invention can be implemented as a system, a method, and a program for operating a computer. In addition, the present invention can be implemented in hardware, software, or hardware and software embodiments. The program can be recorded on any computer-readable medium such as a hard disk, CD-ROM, DVD-ROM, optical storage device, or magnetic storage device. Furthermore, the program can be recorded on another computer via a network.

(First embodiment of the present invention)
(1. Configuration)
(1-1 Hardware configuration of data processing apparatus)
FIG. 1 is a schematic diagram illustrating a hardware configuration of a data processing apparatus 100 according to the present embodiment.
The data processing apparatus 100 according to the present embodiment includes a CPU (Central Processing Unit) 101, a main memory 102, a motherboard chip set 103, a video card 104, an HDD (Hard Disk Drive) 111, a bridge circuit 112, An optical drive 121, a keyboard 122, and a mouse 123 are provided.

  The main memory 102 is connected to the CPU 101 via the CPU bus and the motherboard chip set 103. The video card 104 is connected to the CPU 101 via an AGB (Accelerated Graphics Port) and a motherboard chip set 103. The HDD 111 is connected to the CPU 101 via a PCI (Peripheral Component Interconnect) bus and the motherboard chip set 103.

  The optical drive 121 is connected to the CPU 101 via a low-speed bus, a bridge circuit 112 of the low-speed bus and the PCI bus, the PCI bus, and the motherboard chip set 103. A keyboard 122 and a mouse 123 are also connected to the CPU 101 with the same connection configuration. The optical drive 121 is a drive that reads (or reads and writes) data by irradiating a laser beam on an optical disk, and corresponds to, for example, a CD-ROM drive and a DVD drive.

FIG. 1 is merely an example schematically showing a hardware configuration of the data processing apparatus 100 according to the present embodiment, and various other configurations can be adopted as long as the present embodiment is applicable.
The data processing apparatus 100 is constructed by copying the data processing program to the HDD 111 and performing a so-called installation (installation shown here is merely an example) that allows the copied data processing program to be loaded into the main memory 102. The data processing program is loaded into the main memory 102 and activated when the user instructs the OS (Operating System) that controls the computer to activate the data processing apparatus 100.
The data processing program may be provided from a recording medium such as a CD-ROM, or may be provided from another computer connected to the network via the network interface 114.

(1-2 Overall module configuration of data processing device)
FIG. 2 is a module configuration diagram of the entire data processing apparatus according to the present embodiment.
The data processing apparatus 100 includes an input processing unit 205, a synthesis processing unit 201, an extraction processing unit 1401, a noise processing unit 1601, and an output processing unit 270.
The input processing unit 205 performs processing for inputting target data to be processed. The target data to be processed is data that can be decomposed into time frequencies, such as sound and images.

The synthesis processing unit 201 performs processing for synthesizing image data (watermark image) with target data.
The extraction processing unit 1401 performs processing for extracting the watermark image synthesized by the synthesis processing unit 201.
The noise processing unit 1601 performs processing for removing noise of target data.
The output processing unit 270 performs processing for outputting the target data in which the watermark image is inserted and the target data from which noise is removed to a screen or paper.

(1-3 Module configuration of synthesis processing unit)
FIG. 3 is a module configuration diagram of the synthesis processing unit of the data processing apparatus according to the present embodiment.
The composition processing unit 201 includes an interval arithmetic processing unit 210, a floating point arithmetic processing unit 215, an image data insertion processing unit 220, and an inverse conversion processing unit 260.
The interval calculation processing unit 210 is a processing unit that performs wavelet transform on input target data by interval calculation.
The floating point arithmetic processing unit 215 is a processing unit that performs wavelet transform on input target data by a normal floating point arithmetic.
The image data insertion processing unit 220 performs processing for inserting a watermark image into wavelet transformed data.
The inverse transform processing unit 260 performs a process of returning the target data by making the four components into one by performing the inverse operation of the wavelet transform.

(1-4 Module configuration of image data insertion processing unit)
FIG. 4 is a module configuration diagram of an image data insertion processing unit of the data processing apparatus according to the present embodiment.
The image data insertion processing unit 220 includes a component selection processing unit 225, a component change processing unit 230, a scalar value acquisition processing unit 235, an average calculation processing unit 240, an average image data generation processing unit 245, and an image data superimposition processing unit 250. .

  The component selection processing unit 225 performs a process of selecting an arbitrary one from the four components obtained by the two-dimensional wavelet transform performed by the section calculation processing unit 210. A watermark image is inserted into the component selected here. Further, information indicating which component has been selected can be used as confidential information.

  The scalar value acquisition processing unit 235 performs processing for acquiring an arbitrary scalar value from the section of the component data selected by the component selection processing unit 225. Arbitrary scalar values can make maximum use of the width of the section by obtaining the maximum or minimum value of the section. Also, this arbitrary scalar value can be used as confidential information in the same manner as described above.

  The average calculation processing unit 240 performs a process of calculating an average of deviation for each pixel of the selected component data selected by the component selection processing unit 225. The deviation referred to here is a difference between the value acquired by the scalar value acquisition processing unit 235 and the component data in the calculation result by the normal floating point calculation. In other words, this is a deviation from the result of the floating point calculation, which is caused by the interval calculation.

The average image data generation processing unit 245 performs processing for multiplying the component value of the watermark image to be inserted by the average value calculated by the average calculation processing unit 240. Therefore, the maximum of the component values of the watermark image can be suppressed to the average of the influence due to the section expansion.
The component change processing unit 230 performs processing to replace the component data of three components other than the component selected by the component selection processing unit 225 with the component data calculated by the floating point arithmetic processing unit 215.

  The image data superimposition processing unit 250 performs a process of superimposing the average image data generated by the average image data generation processing unit 245 on the scalar value acquired by the scalar value acquisition processing unit 235. When the scalar value acquisition processing unit 235 acquires the maximum value of the section, the component value of the average image data is added to the scalar value. When the scalar value acquisition processing unit 235 acquires the minimum value of the section, the component value of the average image data is subtracted from the scalar value. That is, the image data is superimposed outside the section to make the insertion of the image data strong.

(2. Operation)
(2-1 interval calculation)
Here, the section calculation will be described in detail. The interval calculation is not calculated between numerical values performed by numerical calculation, but is calculated using an interval between two numerical values.
Section A is defined as follows.

R is a real number. The lower limit and the upper limit of the range of the section A are set to inf (A) = a1 and sup (A) = a2, and the width of the section A whose section width is not 0 is set to w (A) = a2-a1. The four arithmetic operations of two sections A = [a1, a2] and B = [b1, b2] are defined by the following equations.

  According to the basic calculation of the equation (2), generally, the width of the section is expanded as the number of operations increases. This is because, when an operation is performed in two sections, it is defined to include all of the sections that can be taken by the calculation. For example, if A = [− 1, 2], B = [1, 4], and C = [− 2, 2], then A + B = [0, 6] and A + B−C = [− 2, 8]. . The widths are 3 for A and B and 4 for C. In contrast, the widths of A + B and A + B + C are 6 and 10, respectively. The phenomenon that the interval expands when the calculation is performed in this way is often called “interval expansion”. The above is the detailed description regarding the interval calculation.

(2-2 Wavelet transform)
The wavelet transform is one of the typical time-frequency conversion methods and is a well-known method, and therefore, the basic description is omitted. However, the Dovecy wavelet used in this embodiment will be described in detail here. explain.

(2-2-1 Dovecy wavelet transform)
Dovecy wavelets are complex in shape of wavelets and scaling functions and cannot be represented by known functions. However, the natural number N is used to characterize the value. Here, a sequence of Dovecy wavelets p _k and q _k is given. Table 1 below is the Dovecy sequence p _k .

q _k is obtained by the following equation by inverting p _k in the time axis direction and alternately inverting the sign of the coefficient.

(2-2-2 One-dimensional discrete wavelet transform)
In the Dovecy wavelet transform, since the scaling function φ _{0, k} (t) is not given, the scaling coefficient s _k ⁽⁰⁾ cannot be calculated. However, it is generally known that there is no problem even if the sequence f (n) obtained by sampling a signal is set to s _k ⁽⁰⁾ . Approximate function of arbitrary signal f (t)

The following equation holds.

Using a two-scale function, φ _{j-1, n} (t)

And

Therefore, when replacement is performed from equation (5),

It becomes. The wavelet expansion coefficient w _k ^(j) is calculated using a two-scale function.

And below the integral symbol is s _n ^(j-1) .

It becomes. FIG. 5A shows how the signal is decomposed from the equations (8) and (10). FIG. 5A shows how the scaling coefficient is decomposed to a low accuracy level.
The reconstruction is obtained by the following equation.

From this equation, the signal is reconstructed as shown in FIG. FIG. 5B shows a state of reconstruction from s _k ^(j) and w _k ^(j) with low accuracy.

(2-2-3 2D discrete wavelet transform)
In the two-dimensional discrete wavelet transform, first, the discrete wavelet transform is performed in the horizontal axis direction, and then the discrete wavelet transform is performed in the vertical axis direction. The image data f (m, n ⁾ is regarded as a level 0 scaling function s _{m, n} ⁽⁰⁾ . The horizontal wavelet transform is

It becomes. s _{m, n} ^{(j + 1, x)} and w _{m, n} ^{(j + 1, x)} represent a scaling factor and a wavelet expansion series in the horizontal axis direction, respectively.
Next, when wavelet transform in the vertical axis direction is performed

And when we put together the formula,

It becomes. This is illustrated in FIG. 5C. When s (0) is decomposed, each element of s (1), w (1, h), w (1, v), and w (1, d) can be obtained. s (1) is a low frequency component of s (0). w (1, h) is a high frequency component in the horizontal direction, w (1, v) is a vertical direction, and w (1, d) is a diagonal direction. The reconstruction is performed by the following equation as in the case of one-dimensional conversion.

As a result, image decomposition and reconstruction can be performed by wavelet transform. The above is the detailed description regarding the wavelet transform.

(3. Algorithm)
Next, an algorithm of the data processing apparatus 100 according to the present embodiment will be described.
The image data to be subjected to C _0. When wavelet transform is performed on C ₀ , as shown in FIG. 5C, low frequency and high frequency components (C ₁ , D ₁ , E ₁ , F ₁ ) in the horizontal, vertical, and diagonal directions, respectively. )obtain. Since normal wavelet transform is performed by floating point arithmetic, the components (C ₁ , D ₁ , E ₁ , F ₁ ) obtained therefrom are floating point numbers. If interval arithmetic is used instead of floating point arithmetic, IC ₁ , ID ₁ , IE ₁ , and IF ₁ are obtained instead of C ₁ , D ₁ , E ₁ , and F ₁ . In the wavelet transform, when interval calculation is performed using δ, the signal expression is replaced with equation (4).

It becomes. Here, δ _k is a small positive real number. If the watermark image data to be binary inserted is X, the insertion process is as follows.

(1) Select one of the components of the four sections (IC ₁ , ID ₁ , IE ₁ , IF ₁ ) and calculate the maximum value or the minimum value. For example, when IF ₁ is selected, the maximum value is set to F ₁ ′ = sup (IF ₁ ).
(2) Replace other components with floating point numbers. For example, IC ₁ , ID ₁ , and IE ₁ are replaced with C ₁ , D ₁ , and E ₁ .
(3) Perform the following calculation,

Obtain αX. At this time, N _F1 is the number of pixels, and F ₁ (i) is the i-th pixel value in F ₁ .

(4) When the minimum value is acquired in (1), F ₁ '-αX is performed, and when the maximum value is acquired, F ₁ ' + αX is calculated.
(5) Replace F ₁ ← F ₁ '-αX or F ₁ ← F ₁ ' + αX.
(6) C _{1 to} F ₁ are restored using inverse wavelet transform to obtain image insertion data jC ₀ into which a watermark image has been inserted.
The above is the algorithm related to image data insertion of the data processing apparatus.

(4. Operation of data processing device)
FIG. 6 is a flowchart showing processing of the data processing apparatus 100 according to the present embodiment. Processing of the data processing apparatus 100 for realizing the above algorithm will be described.
First, image data to be processed is input (step S601). Two-dimensional wavelet transform is performed on the input image data. In the present embodiment, wavelet transformation (step S602) by interval computation and wavelet transformation (step S603) by floating point computation are performed. One component is selected from the four components obtained by the wavelet transform by the interval calculation (step S604). A watermark image is inserted into the component selected here.

FIG. 7 is a diagram illustrating a state in which wavelet transformation is performed by floating point arithmetic or interval arithmetic. When the two-dimensional wavelet transform is performed on the image data C _{0 as the} target data by floating point calculation, four components C ₁ , D ₁ , E ₁ , and F ₁ are obtained. Naturally, the component data values of C ₁ , D ₁ , E ₁ , and F ₁ are floating point numbers. Further, when two-dimensional wavelet transformation is performed on C ₀ by interval calculation, four components IC ₁ , ID ₁ , IE ₁ , and IF ₁ are obtained. Of course, the component data values of IC ₁ , ID ₁ , IE ₁ , and IF ₁ are intervals. One component for inserting the watermark image is selected from the four components obtained by the wavelet transform by the interval calculation. Here, it is assumed that a watermark image is inserted into IF ₁ .

Returning to FIG. 6, for the three components IC ₁ , ID ₁ , and IE ₁ other than the component selected in step S604 (here, IF ₁ ), the components C ₁ , D ₁ , and E ₁ obtained by the floating point calculation are changed to Replace (step S605). The maximum value or the minimum value of the section of the component IF ₁ selected in step S604 is acquired (step S606).

FIG. 8 is a diagram showing the state of component replacement and value acquisition. As can be seen, replaces the IF ₁ 3 one component IC ₁ except, ID _1, IE ₁ to _{C 1, D 1, E 1} , for IF ₁ is here the maximum value F ₁ '= sup of section ( IF ₁ ). At this time, the values of the four component data are all values rather than sections. By acquiring the maximum value or the minimum value of the section, it is possible to take a value away from the midpoint of the section, and it is possible to insert an image firmly using the section to the maximum. In addition, since the maximum value and the minimum value are values recognized by the apparatus, it is easy to obtain the values and the processing can be simplified.

  It is not always necessary to acquire the maximum value or the initial value. Any value may be taken as long as it is within the section, but if it is the midpoint, the original image is approached, and the merit of the section calculation is reduced.

Returning to FIG. 6, the average value of the deviation for each pixel between the value (F ₁ ) obtained by the floating point calculation and the value (F ₁ ′) obtained in step S606 is calculated (step S607). The calculated average value is multiplied by the component value of the watermark image to be inserted to generate average image data whose maximum value is within the average value (step S608). The component value of the average image data is superimposed on the value acquired in step S606 (step S609). At this time, if the maximum value of the section is acquired in step S606, the average image data is added. If the minimum value of the section is acquired in step S606, the average image data is subtracted. That is, since the maximum value is acquired with F ₁ ′ = sup (IF ₁ ), the addition process is performed here.

FIG. 9 is a diagram illustrating a state in which the average image data is superimposed on the component data of the target image data. When the average value of the deviations of F ₁ and F ₁ ′ for each pixel is α, the watermark image X to be inserted is multiplied by α. If F ₁ ′ is the maximum value in the section, F ₁ ′ + αX is performed, and if F ₁ ′ is the minimum value in the section, F ₁ ′ −αX is performed. By multiplying the watermark image X by the average value of the deviation of each pixel of F ₁ and F ₁ ′ by α, the component value of the watermark image to be inserted can be suppressed to the average degree of influence by the section expansion, and the target data can be obtained. Can be minimized. Further, if F ₁ '+ αX is performed for the maximum value of the section, and F ₁ ' -αX is performed for the minimum value of the section, an image can be inserted at a point farther from the middle point of the section. Become. That is, if αX is subtracted from the maximum value of the section (or if αX is added to the minimum value of the section), the value approaches the midpoint of the section, and the watermark image to be inserted is the original image data. It approaches. In this case, since the merit of using the interval calculation is reduced, it is necessary to perform the above processing in order to make the most of the advantage of the interval calculation.

Returning to FIG. 6, the component data F ₁ ′ + αX on which the average image data αX is superimposed is replaced with F ₁ , wavelet inverse transformation is performed (step S 610), and the image insertion data in which the watermark image is inserted is output. (Step S611) The process ends.

FIG. 10 is a diagram illustrating a state where the wavelet inverse transformation is performed. The average image data F ₁ ′ + αX is replaced with F _1, and the inverse operation of the wavelet transform by a normal floating point operation is performed to generate the image insertion data jC ₀ .

In the present embodiment, the target data is the image data C ₀ , but is not limited to image data, and may be one-dimensional data such as sound. In that case, the input signal can be considered as a two-dimensional signal by cutting it by a certain length and stacking it vertically. That is, the same processing as in the case of an image can be performed. It may also be video data.

Further, the average calculation processing unit 240 and the average image data 245 in FIG. 4 are not indispensable processing means, but the presence of them makes it possible to guarantee the quality of the target image.
Furthermore, in the present embodiment, the processing for inserting the image data as a watermark image has been described. However, the data processing according to the present embodiment is not a watermark image but a technique for inserting the image so that the image is not visually recognized. The device is available for anything.

(Second embodiment of the present invention)
(1. Configuration)
(1-1 Module configuration of synthesis processing unit)
FIG. 11 is a module configuration diagram of the synthesis processing unit of the data processing apparatus according to the present embodiment. The difference from the synthesis processing unit 201 of the data processing apparatus 100 according to the first embodiment is that the floating point arithmetic processing unit 215 is not provided.
Therefore, the data processing apparatus 100 according to the present embodiment does not perform normal wavelet transform by floating point arithmetic.

(1-2 Image Data Insertion Module Module Configuration)
FIG. 12 is a module configuration diagram of an image data insertion processing unit of the data processing apparatus according to the present embodiment. The difference from the image data insertion processing unit 220 of the data processing apparatus 100 according to the first embodiment is that a conversion processing unit 226 is provided.
The conversion processing unit 226 performs processing for converting components other than the component selected by the component selection processing unit 225 from section data to floating-point numbers. That is, as a result, as in the case of the first embodiment, only the component data selected by the component selection processing unit 225 becomes interval data, and the other component data becomes floating-point numbers, but the processing steps are different. .

(2. Operation of data processing device)
FIG. 13 is a flowchart showing processing of the data processing apparatus 100 according to the present embodiment.
First, image data to be processed is input (step S1301). Two-dimensional wavelet transform is performed on the input image data by interval calculation (step S1302). One component is selected from the four components obtained by the wavelet transform (step S1303). A watermark image is inserted into the component selected here. Since the wavelet transform is performed by the section calculation, the four component data are section data. Here, the three component data other than the component selected in step S1303 are converted from section data to floating point data (step S1304).
In the case of conversion to floating point data, calculation may be performed based on the interval data, or the midpoint of the interval may be obtained and converted to the midpoint.

FIG. 14 is a diagram illustrating a state in which component data other than the selected component is replaced from the interval data to the floating point data. When the two-dimensional wavelet transform is performed on the image data C _{0 as the} target data by the interval calculation, four components IC ₁ , ID ₁ , IE ₁ , and IF ₁ are obtained. Of course, the component data values of IC ₁ , ID ₁ , IE ₁ , and IF ₁ are intervals. One component for inserting the watermark image is selected from the four components obtained by the interval calculation. Here, it is assumed that a watermark image is inserted into IF ₁ . The component data (IC ₁ , ID ₁ , IE ₁ ) other than the selected component is converted into values (C ₁ , D ₁ , E ₁ ) when wavelet transform is performed by floating point arithmetic by conversion processing. Since IF ₁ remains the section data, processing for obtaining the maximum value or the minimum value is performed as in the case of the first embodiment.
Returning to FIG. 13, after step S <b> 1305, a watermark image is inserted by performing the same processing as in the first embodiment.

(Third embodiment of the present invention)
(1. Module configuration of the synthesis processing unit)
FIG. 15 is a module configuration diagram of the synthesis processing unit of the data processing apparatus according to the present embodiment. The difference from the first embodiment is that a section setting processing unit 206 is provided.
Interval setting unit 206 performs setting of the section designated by the user (the value of [delta] _k in Equation (16)). That is, the user can arbitrarily set the interval calculation interval. The set section information can be used as confidential information. If the section is set wide, it is easy to insert a watermark image, but the original image is easily broken. Conversely, if the section is set to be narrow, it is difficult to insert the watermark image, but the original image can be inserted without breaking. Specifically, it is preferably set to about 0.005 to 0.01.

(2. Operation of data processing device)
FIG. 16 is a flowchart showing processing of the data processing apparatus 100 according to the present embodiment.
First, image data to be processed is input (step S1601). The user designates a section for the section calculation, and sets the designated section to δ _k in Expression (16) (step S1602). Based on the set section, wavelet transform is performed by section calculation (step S1603). Subsequent processing is the same as in the first embodiment.

  As described above, since the user can arbitrarily set the section, the processing according to the image data to be processed can be performed. For example, when the target image data is originally a rough image, even if the section is widened and the influence of the watermark image appears on the target image data, it is not so noticeable. Conversely, when the target image data is clear, the interval can be narrowed so that the target image data is not affected by the watermark image. In addition, the section information can be used as a key for extracting a watermark image.

(Fourth embodiment of the present invention)
(1. Module configuration of the extraction processing unit)
FIG. 17 is a module configuration diagram of the extraction processing unit of the data processing apparatus according to the present embodiment. Here, processing for extracting the watermark image inserted in each of the above embodiments will be described.
The extraction processing unit 1401 includes a floating point arithmetic unit 215, a difference calculation processing unit 1410, and an image extraction processing unit 1420.

  The floating point arithmetic unit 215 is a processing unit that performs wavelet transform on input target data by a normal floating point arithmetic. Again, the data to be processed is data in which a watermark image has already been inserted.

  The difference calculation processing unit 1410 includes the component value of the component in which the watermark image is inserted among the four components obtained by the wavelet transform performed by the floating point arithmetic unit 215, and the scalar value in the section acquired when the watermark image is inserted. The process which calculates the difference with is performed. Since the scalar value in the section acquired when the watermark image is inserted is information known only to the inserted person, it can be used as a key.

The image extraction processing unit 1420 performs processing for extracting the watermark image X. Specifically, a process of dividing the difference calculated by the difference calculation processing unit 1410 by α described above is performed (details will be described later in the description of the algorithm). The difference calculation processing unit 1410 and the image extraction processing unit 1420 extract the watermark image X by performing a reverse operation to the insertion process.
The output processing unit 270 performs processing for outputting the extracted watermark image.

(2. Algorithm)
The watermark image extraction algorithm will be described below. In order to extract the watermark image, information about the component in which the watermark image is inserted and information on the scalar value acquired from the section are necessary. These pieces of information are the key for extracting the watermark image.

(1) performs a wavelet transform in a normal floating-point operations on the image data jC ₀ watermark image is embedded, is decomposed into jC _1, jD _1, jE _1, jF _1.
(2) When a watermark image is inserted into the IF ₁ component, | jF ₁ −F ′ ₁ | is calculated.
(3) The watermark image X is extracted by the following equation.

The above is the algorithm for watermark image extraction. As a result, it can be seen that the inverse operation is performed in the case of insertion.

(3. Operation)
FIG. 18 is a flowchart showing processing of the data processing apparatus 100 according to the present embodiment.
First, image data into which a watermark image has been inserted is input (step S1801). The input image is subjected to wavelet transformation by normal floating point arithmetic (step S1802). The difference between the component value of the component in which the watermark image is inserted among the obtained four components and the scalar value of the section acquired when the watermark image is inserted is calculated (step S1803). The difference is divided by an average value α of deviations calculated at the time of watermark image insertion (step S1804). The watermark image obtained thereby is output (step S1805), and the process is terminated.
In this way, by performing the inverse operation of the insertion, the watermark image can be extracted by a simple process. Since the information cannot be extracted without using the information at the time of inserting the watermark image, the confidentiality can be maintained.

(Other embodiments)
(1. Module configuration of data processing device)
FIG. 19 is a module configuration diagram of the noise processing unit of the data processing apparatus according to the present embodiment. Here, the redundant part by the interval calculation is not used for insertion of the watermark image but for noise removal. The noise processing unit 1601 includes a section calculation processing unit 210, an actualization processing unit 1605, a noise detection processing unit 1610, and a noise removal processing unit 1620.

The section calculation processing unit 210 is a processing unit that performs wavelet transform on target data input in the same manner as described above by section calculation.
The revealing processing unit 1605 performs a process of intentionally revealing the redundant part expanded by the interval calculation. That is, a process of emphasizing the noise portion by making use of the error due to the section is performed.
The noise detection processing unit 1610 performs processing for detecting noise from a difference between an image in which noise has been revealed by the manifestation processing unit 1605 and an input target image.
The noise removal processing unit 1620 performs a process of removing at least a part of the noise by subtracting the noise detected by the noise detection processing unit 1610 from the input target data.

(2. Operation of data processing device)
FIG. 20 is a flowchart showing processing of the data processing apparatus 100 according to the present embodiment.
First, data for which noise is to be removed is input (step S2001). Wavelet transform is performed on the input target data by interval calculation (step S2002). Using the section expanded by the section calculation, noise is made obvious by the error and emphasized (step S2003). The difference between the input target data and the data with enhanced noise is calculated (step S2004). The calculated difference is subtracted from the target data to remove noise (step S2005), and the data from which the noise has been removed is output (step S2006), and the process is terminated.

  FIG. 21 is a schematic diagram illustrating a process of removing noise. FIG. 21A shows target input data. This is image data or audio data. FIG. 21B shows the target data subjected to wavelet transform by interval calculation to reveal the noise part and emphasize it. Since the redundant part is expanded by the interval calculation, the target data is destroyed. In other words, the noise part is emphasized. FIG. 21 (c) shows the difference between FIG. 21 (b) and FIG. 21 (c), which is a redundant part expanded by section calculation. FIG. 21D shows the result of subtracting the noise remaining as the redundant portion from the target data. Noise is removed and clear data is obtained.

  As described above, by using the interval expansion of the interval calculation, it is possible to emphasize noise and easily remove the noise. Further, since it is not necessary to perform processing such as noise estimation, the processing is simplified and the accuracy of noise removal can be improved.

  In addition, as a method of removing noise other than the above, first, time frequency conversion is performed by interval calculation, and the portion where the interval width exceeds a predetermined value in the high frequency region is regarded as noise, and the portion is regarded as noise. The noise can be easily removed by replacing it with 0 and performing an inverse transformation. In any case, since the noise can be revealed by using the interval calculation, the noise can be easily removed.

  According to each of the embodiments described above, data redundancy can be expanded by combining interval calculation and time-frequency conversion. And it can be applied to various technologies by not taking the redundancy as an error negatively but as a positive extension of the available area.

  The data processing apparatus according to the present invention is not limited to the wavelet transform, and can be applied to any time-frequency transform of orthogonal transform. For example, Fourier transform, discrete cosine transform (DCT), modified discrete cosine transform (MDCT), discrete sine transform (DST), Hartley transform (HLT), Walsh Hadamard transform (WHT), Haar transform (HT), Karhunen label It can be applied to conversion (KLT) and the like.

  In addition to the digital watermark technique and the noise removal technique, a characteristic frequency can also be specified by using interval calculation. For example, the present invention can be applied to a technique in which a portion emphasized by combining a lifting wavelet having a learning function and interval calculation is regarded as a signal feature.

  Although the present invention has been described with the above embodiments, the technical scope of the present invention is not limited to the scope described in the embodiments, and various modifications or improvements can be added to these embodiments. . And embodiment which added such a change or improvement is also contained in the technical scope of the present invention. This is apparent from the claims and the means for solving the problems.

(Inserting and extracting watermark images)
(1. Conditions)
A 256 gray scale image of 256 × 256 bits size is used as an image into which a watermark image is to be inserted (FIG. 22A). This is registered in the Image Standard Library (SIDBA: Standard Image DataBAse) for image processing researchers, and is an image used by many researchers. The watermark image is binary and has a size of 128 × 128 bits (FIG. 22B). The target image is an image turned by a woman, and the watermark image is an image in which 16 “transparent” characters are arranged in a row, four vertically and four horizontally.

In this example, IF ₁ was selected from four components (IC ₁ , ID ₁ , IE ₁ , IF ₁ ) obtained by wavelet transform, and δ _k in equation (16) was uniformly set to 0.01. In addition, INTLAB, which is one of the MATLAB toolboxes, supports interval calculation. As the scale function, a three-stage wavelet transform was performed using a Dovecy function.

  Here, it supplements about MATLAB and INTLAB. MATLAB is an abbreviation for MATRIx LABoratory and has a library for matrix calculation, vector operation, graph display, and the like. Since programs can be created more easily than conventional programming languages, scientific and technical calculations can be performed in a short time. In addition, many libraries are provided as standard, but the function of MATLAB can be expanded by installing an extension package called a toolbox as needed. The INTLAB used this time is an abbreviation for INTerval LABoratory, and it is possible to easily implement the interval calculation by using such a tool box.

(1. Results)
FIG. 22C shows image data with a watermark image inserted, and FIG. 22D shows the watermark image extracted from FIG. 22C. From FIG. 22C, it is impossible to recognize at a glance that the watermark image is inserted. That is, the watermark image can be inserted with little influence on the target image. Also, from FIG. 22D, the watermark image could be extracted in almost the same state as before insertion.
Further, the noise ratio of the image data in which the watermark image in this case is inserted was calculated by the following equation.

N is the number of pixels, and T is the number of gradations -1. That is, in this embodiment, N = 65536 and T = 255. The calculation result was PSNR = 35.3. Generally, if the PSNR value is 35 or more, it can be said that the image quality is generally good, so that it can be said that the image quality of the image data in which the watermark image is inserted maintains high quality.

(1. Conditions)
The experimental environment is the same as in the case of Example 1, but different experiments are used, and the experiment is performed separately when δ _k in equation (16) is set to 0.01 and when set to 0.005. It was. FIG. 23A shows the target image, and the watermark image is the same as in the first embodiment.

(2. Results)
23 (b) is an image data in which the watermark image is inserted in a case where [delta] _k is set to 0.01. FIG. 23C is a watermark image extracted from FIG. At this time, the noise ratio was PSNR = 35.47. Further, FIG. 23 (d) is an image data in which the watermark image is inserted in a case where [delta] _k is set to 0.005. FIG. 23E shows a watermark image extracted from FIG. At this time, the noise ratio was PSNR = 41.53. In any case, the image quality is generally good because the PSNR value is 35 or more. In particular, when δ _k is set to 0.005, PSNR = 41.53, which is a very good image quality.

(1. Conditions)
The experimental environment was the same as in Example 2, but the experiment was performed using different target images. FIG. 24A shows a target image.

(2. Results)
Figure 24 (b) is a picture data in which the watermark image is inserted in a case where [delta] _k is set to 0.01. FIG. 24C is a watermark image extracted from FIG. At this time, the noise ratio was PSNR = 38.17. Further, FIG. 24 (d) is an image data in which the watermark image is inserted in a case where [delta] _k is set to 0.005. FIG. 24E is a watermark image extracted from FIG. At this time, the noise ratio was PSNR = 44.08. Here again, it can be said that the image quality is generally good because the PSNR value is 35 or more, and particularly when δ _k is set to 0.005, the image quality is very good at PSNR = 44.08.

(1. Conditions)
The experimental environment was the same as in Example 2, but the experiment was performed using different target images.

(2. Results)
Figure 25 (b) is an image data in which the watermark image is inserted in a case where [delta] _k is set to 0.01. FIG. 25C is a watermark image extracted from FIG. At this time, the noise ratio was PSNR = 34.73. Although the image is somewhat degraded numerically, it is difficult to recognize the degradation of the image visually, and there is not much difference from the target image. Further, FIG. 25 (d) is a picture data in which the watermark image is inserted in a case where [delta] _k is set to 0.005. FIG. 25E is a watermark image extracted from FIG. At this time, the noise ratio was PSNR = 40.88. In either case, the extracted watermark image is somewhat degraded, but can be recognized as a watermark image.

From the results of Examples 1 to 3, it can be said that by using the data processing apparatus of the present invention, it is possible to obtain image data with good image quality even when a watermark image is inserted. In addition, better to set the _k also δ in each case to 0.005, a high image quality than setting it to 0.01. However, even when set to 0.01, for a sufficiently good quality, the value of [delta] _k can be said that it is desirable to set 0.005 to about 0.01.

(Watermark image extraction when image is damaged)
(1. Conditions)
The experimental environment is the same as that in the first embodiment, but the image is damaged in this embodiment. In other words, this is an experiment in the case where image data into which a watermark image has been inserted has been tampered with.

(2. Results)
FIG. 26A shows image data in which a watermark image is inserted. FIG. 26B shows an image damaged (falsified) with respect to the image data. Here, there are scratches like black graffiti. The watermark image extracted from FIG. 26B is FIG. Referring to FIG. 26C, the watermark image is also damaged at the portion where the target image is damaged. However, since the watermark can be uniformly inserted as an image as a whole, the watermark image can be sufficiently confirmed in portions other than the scratches. In other words, when inserting a watermark into a part of an image or searching for an inconspicuous part and inserting it, the watermark may disappear due to such scratches, or extraction may not be possible. Since a watermark image can be inserted, such a problem does not occur. In addition, if the extracted watermark and the inserted watermark are not identical according to the judgment by the machine, the presence or absence of the watermark may not be determined. However, by inserting it as a watermark image as in this embodiment, the watermark can be visually confirmed by human eyes. The presence or absence can be confirmed.

(Watermark image extraction when the image is compressed)
(1. Conditions)
The experimental environment is the same as that in the first embodiment, but the image is compressed in this embodiment. The compression method is JPEG2000 or JPEG.

(2. Results)
FIG. 27A shows image data in which a watermark image is inserted. FIG. 27B shows a watermark image obtained by compressing the image data using JPEG 2000 and extracting the compressed image. The compression rate at this time is 14.67%. That is, the watermark image is extracted from the image data in which about 85% of the component is lost. Compared with the first and second embodiments, the extracted watermark image is broken, but the presence of the watermark image can be easily confirmed.

  Further, FIG. 27C shows a watermark image compressed by JPEG and extracted from the compressed image. The compression rate at this time is 28.95%. That is, the watermark image is extracted from the image data in which about 70% of the component is lost. In this case as well, the watermark image is somewhat broken, but the presence of the watermark image can be easily confirmed. In the digital watermark technology, it is important to extract a clear watermark, but it is most important that the presence of the watermark can be confirmed with certainty. From this point of view, in this embodiment, the fact that the presence of a watermark image can be visually confirmed even when most of the components are lost can be said to be very effective for digital watermarking.

(Watermark image extraction when image is cut out)
(1. Conditions)
Although the experimental environment is the same as that in the first embodiment, a part of the image (128 × 128 pixels) is cut out in this embodiment.

(2. Results)
FIG. 28A shows image data in which a watermark image is inserted. FIG. 28B shows a part of the image data cut out by 128 × 128 pixels. A watermark image extracted from the image of FIG. 28B is FIG. Compared to the original watermark image, the image is considerably broken, but it can be confirmed that the watermark is included.
The above experimental results show that the digital watermark technique is robust against various attacks, and a high-quality watermark image can be obtained.

It is the schematic diagram which showed the hardware constitutions of the data processor which concerns on 1st Embodiment. It is a module block diagram of the data processor which concerns on 1st Embodiment. It is a module block diagram of the synthetic | combination process part of the data processor which concerns on 1st Embodiment. It is a module block diagram of the image data insertion process part of the data processor which concerns on 1st Embodiment. It is a figure for demonstrating wavelet transformation. It is a flowchart which shows the process of the data processor which concerns on 1st Embodiment. It is a figure which shows a mode that wavelet transformation was performed by the floating point calculation or the interval calculation. It is the figure which showed the mode of component replacement and value acquisition. It is the figure which showed a mode that average image data was superimposed on the component data of the image data used as object. It is the figure which showed a mode that wavelet reverse transformation was performed. It is a module block diagram of the synthetic | combination process part of the data processor which concerns on 2nd Embodiment. It is a module block diagram of the image data insertion process part of the data processor which concerns on 2nd Embodiment. It is a flowchart which shows the process of the data processor which concerns on 2nd Embodiment. It is the figure which showed a mode that component data other than the selected component are replaced from area data to floating point data. It is a module block diagram of the synthetic | combination process part of the data processor which concerns on 3rd Embodiment. It is a flowchart which shows the process of the data processor which concerns on 3rd Embodiment. It is a module block diagram of the extraction process part of the data processor which concerns on 4th Embodiment. It is a flowchart which shows the process of the data processor which concerns on 4th Embodiment. It is a module block diagram of the noise processing part of the data processor which concerns on other embodiment. It is a flowchart which shows the process of the data processor which concerns on other embodiment. It is the schematic diagram which showed the process in which noise is removed. It is a figure which shows the result of Example 1. It is a figure which shows the result of Example 2. It is a figure which shows the result of Example 3. It is a figure which shows the result of Example 4. It is a figure which shows the result of Example 5. It is a figure which shows the result of Example 6. It is a figure which shows the result of Example 7.

Explanation of symbols

100 Data processing device 101 CPU
102 Main memory 103 MB chipset 104 Video card 111 HDD
DESCRIPTION OF SYMBOLS 112 Bridge circuit 114 Network interface 121 Optical drive 122 Keyboard 123 Mouse 201 Composition processing part 205 Input processing part 206 Section setting processing part 210 Section arithmetic processing part 215 Floating point arithmetic processing part 220 Image data insertion processing part 225 Component selection processing part 226 Conversion Processing unit 230 Component change processing unit 235 Scalar value acquisition processing unit 240 Average calculation processing unit 245 Average image data generation processing unit 250 Image data superimposition processing unit 260 Inverse conversion processing unit 270 Output processing unit 1401 Extraction processing unit 1410 Difference calculation processing unit 1420 Image extraction processing unit 1601 Noise processing unit 1605 Realization processing unit 1610 Noise detection processing unit 1620 Noise removal processing unit

Claims (11)

In a data processing apparatus that executes data processing for embedding image data using time-frequency conversion for target data,
By performing by the section calculates the time-frequency transform on the target data, the target data, and the interval arithmetic means for redundant portion in which the target data has to generate as an extended expansion target data,
Image data insertion means for generating image data superimposition information by inserting image data into the extension target data generated by the section calculation means;
A data processing apparatus comprising: inverse conversion means for inversely transforming the image data superposition information generated by the image data insertion means by the time-frequency conversion to generate image insertion data . The data processing apparatus according to claim 1,
The data processing apparatus , wherein the section calculation means performs the time-frequency conversion by wavelet conversion . The data processing apparatus according to claim 2, wherein
The image data inserting means is
Component selection means for selecting any one component as selected component data from the four generated components as a result of calculation by the section calculation means;
A component changing unit that changes a component other than the selected component data selected by the component selecting unit to component data obtained as a result of wavelet transform by floating-point arithmetic;
Scalar value acquisition means for acquiring an arbitrary scalar value from the section of the selected component data;
An image data superimposing unit that superimposes the image data on the scalar value acquired by the scalar value acquiring unit . The data processing apparatus according to claim 3, wherein
The image data inserting means is
An average calculation for calculating an average value of deviations for each pixel in the selected component data from the scalar value acquired by the scalar value acquisition means and the component data obtained as a result of wavelet transform by floating point calculation in the selected component data Means,
A data processing apparatus comprising: average image data generation means for generating average image data by multiplying the component data of the image data by the average value calculated by the average calculation means . The data processing apparatus according to claim 3 or 4,
The data processing apparatus according to claim 1, wherein the scalar value acquired by the scalar value acquiring means is a maximum value or a minimum value of a section of the selected component data . The data processing apparatus according to claim 5 , wherein
When the scalar value acquired by the scalar value acquisition unit is the maximum value of the section of the selected component data, the average image data is added to the maximum value, and the scalar value acquired by the scalar value acquisition unit A data processing apparatus characterized by subtracting the average image data from the minimum value when the value is the minimum value of the section of the selected component data . The data processing apparatus according to any one of claims 1 to 6,
A data processing apparatus comprising section setting means for setting a section calculated by the section calculation means in a range designated by a user . The data processing apparatus according to any one of claims 4 to 7,
Floating point arithmetic means for performing wavelet transform by floating point arithmetic on the image insertion data into which the image data has been inserted;
Difference calculation means for calculating the difference between the component data of the component into which the image data is inserted and the scalar value acquired by the scalar value acquisition means among the four components obtained as a result of the calculation by the floating point calculation means When,
A data processing apparatus comprising: an image data extraction unit that extracts the image data by dividing the difference calculated by the difference calculation unit by the average value calculated by the average calculation unit . In a data processing apparatus that executes data processing for removing noise using time-frequency conversion for target data,
By performing time frequency conversion on the target data by section calculation, the section data is generated as expanded target data in which the noise of the target data is expanded, and
Noise detection means for detecting noise of the target data from the difference between the target data and the expansion target data;
A data processing apparatus comprising: noise removal means for removing at least part of noise detected by the noise detection means from the target data . In a data processing method in which a computer executes data processing for embedding image data using time-frequency conversion for target data,
Computer
By performing time-frequency conversion on the target data by section calculation, the section data is generated as expanded target data in which the redundant portion of the target data is expanded; and
An image data insertion step for generating image data superimposition information by inserting image data into the extension target data generated by the section calculation step;
A data processing method , comprising: performing an inverse transformation step of inversely transforming the image data superimposition information generated in the image data insertion step by the time-frequency transformation to generate image insertion data . In a data processing program for causing a computer to perform data processing for embedding image data using time-frequency conversion for target data,
Wherein by performing the interval calculating a time-frequency conversion on the target data, the target data, interval arithmetic means redundant portion in which the target data has to generate as an extended expansion target data,
Image data insertion means for generating image data superimposition information by inserting image data into the extension target data generated by the section calculation means;
A data processing program for causing a computer to function as inverse conversion means for inversely converting the image data superimposition information generated by the image data insertion means by the time-frequency conversion to generate image insertion data .