JP5508896B2 - Image processing method - Google Patents

Description

  The present invention embeds information as a latent image in an image such as a still image or a moving image so that it is difficult for the human eye to understand, and detects and processes the embedded information as a latent image when some processing is applied to the image. Technology and a method of embedding information.

There are roughly the following three types of techniques for adding information to images.
・ A method to keep information separately from the image and handle them together ・ A method to superimpose information in the image in a visible state ・ A method to hide information in an invisible state in an image Information to be invisible in an image As an example of the concealment method, the digital watermark technique is a technique for expressing embedded identification information by using a redundant part of content. The embedded identification information is invisible to the person viewing the content, but can be extracted from the content using a special detection method.

  The following technologies are disclosed as latent image technologies that can also be used for digital contents.

JP-A-8-197828 (summary) JP 2003-118200 A (paragraph 0009) JP 2000-76418 A (paragraph 0011)

Conventional latent image technology has been developed mainly for printed materials and is difficult to use for digital content. Hereinafter, the term “latent image” is used in a broad sense, and a technique for enhancing security and services by hiding the latent image is referred to as a “latent image technique”.
Here, in order to avoid the confusion of the term “printed material technology”, when there is an image B in which the latent image is hidden with respect to the original image A, the former A is referred to as a cover image and the latter B is referred to as a stego image. Since the hidden latent image is superimposed on the cover image, it is referred to as a superimposed image. Images refer to still images, moving images, 3D moving images, and the like.

  In addition to conventional printing applications, the present invention is also intended for cases where image processing is applied to a stego image when the stego image is digital content.

  According to Patent Document 1, a fine line added on a stego image is unlikely to change greatly due to digital image processing. The same applies to the minute points in Patent Document 2. For example, a noise reduction filter process may be performed in the compression encoding process, and a superimposed image expressed by minute points is likely to disappear. When the cover image is a pattern composed of thin lines, the entire pattern of the cover image is easily smoothed by the scale reduction process, and when both the thin lines and the thick lines are smoothed with the average value of the density in a large area, the density difference Does not occur.

  The technique of Patent Document 3 is invisible on the cover image, and the superimposed image cannot be directly recognized by human eyes.

  The present invention also provides a method and a method for creating a stego image in which hidden information is visualized (passive visual detection is possible) when analog / digital image processing is performed on a stego image of digital content. Provide methods and operational methods. In particular, the present invention provides a latent image technique that focuses on components emphasized after processing in image processing accompanying reduction processing and recompression encoding processing as digital image processing.

  The present invention also provides a method for detecting consciously hidden information.

  Of the techniques disclosed in this specification, the configuration and operation according to one aspect are as follows.

  In the following description, an image may be referred to as data.

  The superimposed image is hidden in the cover data so as not to be visible to the human eye, stego data is created, and the superimposed image is detected from the stego data that has undergone image processing. The superimposition process is performed by changing the cover data slightly. The superimposed image is given a characteristic that is visualized by reduction (referred to as a geometric deformation characteristic) and a characteristic that is visualized by recompression coding (referred to as a recompressed coding characteristic), and the superimposed image is used as a latent image. Lightly superimpose on the cover data. Since it is difficult for human eyes to detect superimposed images, stego data can be sold and distributed normally.

  However, when stego data has undergone image processing. Since the signal component of the superimposed image is enhanced by the geometric deformation characteristic and the recompression coding characteristic, the superimposed image is visualized and made visible to the human eye. In addition, when a character for checking fraud is written on the superimposed image or an image generation time to be proved is displayed, information such as a character hidden in the stego data as a latent image can be detected as necessary. Furthermore, the detection result can be reflected in other devices and system operations.

  An image superimposing device stores a program (or module) for realizing an image superimposing method on a computer equipped with arithmetic means, input means, output means, memory (primary storage means), disk (secondary storage means), and communication. This can be realized by operating above.

  Image superposition method includes superposition position map alignment specification processing, scramble processing, time-space domain division processing, buffer, geometric deformation characteristic setting processing, recompression characteristic setting processing, superimposition data creation processing, superimposition processing, compression coding parameter adjustment Processing, stego data output processing, auxiliary information output processing, electronic signature processing, encryption processing, image quality analysis processing, conversion coefficient / quantization table extraction processing. Further, the cover data and the superimposed image are aligned using the superimposed position map, and the intensity of the superimposed image is adjusted using the intensity map table.

  In the geometric deformation characteristic setting process, the superimposed image is processed based on the following technical principle. Attention is paid to a partial area of the cover image that becomes the size of the superimposed image when the cover image is reduced by a certain ratio R. In the superimposition processing, attention is paid to a location A that corresponds to the pattern of the superimposed image and a location B that does not correspond in the partial area on the cover image. For the location A, the average value of the pixel values of the partial areas on the cover image is slightly changed. Place B is not changed. When such a change is given to the cover image, when the cover image is reduced, places where the average value is intentionally changed are connected to each other and perceived by humans, and the superimposed image appears to float.

  In the recompression characteristic setting process, the superimposed image is processed based on the following technical principle. Set a cutoff frequency according to the recompression encoding conditions (bit rate, etc.), remove the frequency component higher than the cutoff frequency from the superimposed image, and have a frequency within the predetermined range that is lower than the cutoff frequency. Increase the signal power of the component. By this processing, mosquito noise corresponding to the shape feature is emphasized in the superimposed image. If the cover image contains many intermediate frequency components and high frequency components when the stego data on which the superimposed image is superimposed is recompressed, the intermediate frequency components and high frequency components of the superimposed image are processed after the recompression encoding process. Many remain. The recompression encoding process is performed for each partial area (macroblock) of interest. For this reason, the continuity between the pixel value of the attention area and the pixel value of another area adjacent to the attention area is not ensured. Therefore, in human vision, block noise is recognized in addition to the emphasized mosquito noise at the position of the pixels constituting the superimposed image. Since these visually recognized noises appear to be connected across a plurality of attention areas, it becomes easier for a person to perceive a superimposed image.

The effects obtained by the above embodiment according to one aspect are as follows.
• Stego images do not interfere with normal viewing unless image processing is added.
-The superimposed image is less disturbing than the visible mark.
-Compared with invisible digital watermarks, it can be seen directly from counterfeit content and illegal copies (pirated versions), so it has a greater anti-fraud effect.
-Anyone can detect the superimposed image.
-Superimposed images can also be visualized by taking pictures of stego images with a mobile phone or video camera.
-Since one-way processing of the entire image is possible, it is difficult to delete the superimposed image (check pattern). Even if the image is forcibly erased, the trace of the superimposed image (check pattern) remains, and the value of the original picture is greatly reduced and the check effect continues.
-When creating a stego image, it is possible to set the required image quality by adjusting the intensity.
-Suppressing an increase in the amount of stego image data due to the addition of superimposed images
-Even if the detection algorithm (visual) is disclosed, the safety and reliability of the content will not be reduced as a system. Conventional digital watermarks could not guarantee safety unless the algorithm was kept secret. The present invention provides a digital watermark that is safe even if the algorithm is disclosed.
-By superimposing the superimposed image itself and delivering the scramble key safely, only the person with the appropriate key can confirm the superimposed image, and the security of the superimposed image (check pattern) itself can be established. A new configuration of key exchangeable digital watermark is provided.
-A conventional digital watermark equivalent function can be provided by using a machine-readable superimposed image instead of a visually identifiable superimposed image (such as a logo or check pattern).

Furthermore, the effects obtained by a typical system to which the latent image technology disclosed in this specification is applied are as follows.
・ Content authentication:
By superimposing authentication information, authority contact information, etc. as a character image, it is possible to provide a content authentication means that can be easily verified.
・ Falsification detection:
By superimposing the authenticity information as a character image, it is possible to visually detect whether the image is a forgery or an original. A special device is not required to check the authenticity, and the verification operation is facilitated by making the check pattern visible when the viewing screen size is changed by the player operation.
・ Copyright protection:
By superimposing the video viewer ID and distribution route information as a check pattern. The content value in terms of image quality is greatly reduced when an illegal copy is created such that the monitor output screen is re-captured using a video camera or the like. In addition, the user can be explicitly controlled, which contributes to the prevention of unauthorized copying.
・ Information leakage suppression:
By superimposing the accessor ID, the management organization name, etc. as a check pattern, the unauthorized user can be explicitly checked, thus preventing information leakage due to an operation error. It also has a deterrent effect on malicious information leakage.
-Copy generation management:
By using it together with other security technologies such as digital watermarking, the security of each application system can be enhanced. If the same information as the superimposed image cannot be detected as a digital watermark, the image can be determined to be fake. By embedding a digital watermark when creating a stego image, if a digital watermark is detected from a cover image, the cover image is determined to be an image in which the superimposed image has already been written, and the superimposed image is prevented from being written twice. To do.
・ Information hiding:
By making the superimposed image a bar-code-like pattern expression, the machine reading accuracy can be improved instead of visual confirmation.
Software protection:
Image processing software that outputs a stego image in which a product serial number is superimposed on a cover image makes it easy to prove the identity of the software processed image.
・ Distribution management:
By displaying the route, administrator, server name, etc. as the superimposed image, it becomes easy to prove the identity of the route through which the image has passed.
・ Billing:
By displaying a charged / settled (non-completed) mark as a superimposed image on the stego image, the user can easily check the status of the content such as the charging status.

  The latent image technology disclosed in the present specification can be applied to software, devices, services, systems, contents, media, and application applications for hiding information in images and detecting and processing the hidden information. In particular, it can be used for content authentication, falsification detection, copyright protection, information leakage suppression, copy generation management, record management, information hiding, software protection, distribution management, billing management, and the like.

  According to the present invention, there is provided a stego image creation method, a detection method, and an operation method in which hidden information is visualized when image processing is performed on the stego image with hidden information.

It is a conceptual diagram of a latent image technique. It is a system implementation example of latent image technology. It is an example of the detection method of a latent image. It is an example of composition of an image superposition device. It is a conceptual diagram of the principle of giving geometric deformation characteristics to a superimposed image. It is a figure which illustrates the process sequence which provides the geometric deformation characteristic to a superimposition image. It is a figure which illustrates in detail the creation method of an intensity map. It is explanatory drawing of an intensity map table. It is a figure which illustrates the superimposition processing procedure of a cover image at the time of illustrating a moving image, and a superimposed image. It is a figure which illustrates the concept of the provision process procedure of the recompression encoding characteristic to a superimposition image. FIG. 10 is an explanatory diagram of a basic principle that a superposed image is visualized when a stego image undergoes compression-encoded image processing in giving re-compression encoding characteristics. It is a figure which illustrates the detailed procedure which provides the recompression encoding characteristic with respect to a stego image. It is an image figure of the superimposition image superimposed according to the superposition position map. It is a figure which illustrates the process sequence of a superimposition apparatus, when a machine determines automatically the detection of a superimposition image irrespective of recognition of a human eye. It is a figure which illustrates the process sequence of a detection apparatus, when a machine automatically determines the detection of a superimposed image irrespective of recognition of a human eye. It is a figure which illustrates the process sequence of the electronic watermark detection embedded in the superimposed image itself.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that components having the same function are denoted by the same reference symbols throughout the drawings for describing the embodiment, and the repetitive description thereof will be omitted.

  FIG. 1 shows a conceptual diagram of the latent image technology.

  The latent image technique in this embodiment is to hide the superimposed image 102 from the cover data 101 so as not to be seen by human eyes, create stego data 104, and detect the superimposed image 102 from the stego data 106 that has undergone image processing 105. Refers to technology. The superimposing process 103 is performed by minutely changing the cover data. Since the superimposed image is not visible to the human eye, the stego data 104 can be sold and distributed normally. However, when the stego data 104 has undergone the image processing 105. The superimposed image is visualized and becomes visible to the human eye. For example, it can be used for fraud deterrence or authenticity proof by writing a character for checking fraud on the superimposed image 102 or displaying an image generation time to be proved.

  FIG. 2 shows a system implementation example of the latent image technology.

  The user inputs the cover data 101 to the input device of the image superimposing device 203. Further, the user designates characters, sounds, images, moving images, database (DB) search results, etc., and the user inputs them as the superimposed image 102. Alternatively, a position time generation source using a GPS (Global Positioning System) or the like, or an output result of another system such as a DB may be input to the image superimposing device 203 via a network. The image superimposing device 203 superimposes the superimposed image 102 on the cover data 101 and outputs the stego data 104. In the example of FIG. 2, the ID of the stego data distribution destination user and a character string indicating that image copying is prohibited are designated as the superimposed image 102 or the check pattern. When the stego data 104 is input to the decoder 205 (player or player), it is output to the monitor 206 and looks to the eyes of the person 207 the same as the cover data 101.

  If the stego data 104 is illegally copied and the illegal copy stego data 208 passes through an image processing device 209 such as a transcoder, a processed image 210 subjected to the image processing is output. When the processed image 210 is input to the decoder 211 (player or player), the processed image 210 is output to the monitor 212, and the person 213 sees the superimposed image superimposed on the cover data. Alternatively, by inputting the stego data 208 to the detection device 214, the superimposed image 102 or information 215 such as characters, sounds, and images that are the basis of the superimposed image 102, which is designated by the user in advance, is output. .

  The output of the detection device 214 may be connected to another system or service via a communication device such as a network, and the system operation and service content may be controlled in correspondence with the detection information. If the detection information 215 is copy control information corresponding to 4 bits (prohibited copying, copying permitted, etc.), the device control such as copy control, retransmission control, management record control corresponding to the detection information 215 is converted into stego data in another system. Can be done against.

  FIG. 3 shows an example of a latent image detection method.

  Although the latent image can be detected through an image processing apparatus such as a transcoder, there are other methods for easily detecting the latent image. Even if the reproduced image is reduced and displayed by adjusting the decoder 302, the latent image can be detected by the eyes of the person 304. Alternatively, when the decoder 302 is normally played back and the output to the monitor 303 of the decoder 302 (when it is invisible to humans at this stage) is captured by the video camera 305, the digital camera 306, and the camera-equipped mobile phone 307, the stego data 301 Are subjected to DA-AD conversion (analog conversion), geometric deformation (reduction, etc.), and recompression coding processing. The processed image after such recompression encoding processing can be detected by the eyes of the person 308 even by these methods in order to visualize the latent image in the same manner as the transcoder. A screen may be printed using the video printer 307.

  FIG. 4 shows a configuration example of a processing unit that realizes the image superimposing apparatus.

  The image superimposing device 203 is a program that realizes an image superimposing method on a computer that includes an arithmetic unit (CPU), an input device, an output device, a memory (primary storage device), a disk (secondary storage device), and a communication device ( Or a module) on a memory.

  The program may be stored in advance in a memory or a disk in the computer, or may be introduced from another device via a medium that can be used by the computer when necessary. The medium refers to, for example, a storage medium that can be attached to and detached from the computer, or a communication medium (that is, a wired, wireless, optical network, or a carrier wave or digital signal that propagates through the network).

  Each processing unit shown in FIG. 4 is realized by the CPU executing a program (or module) that realizes each processing unit stored in a memory or a disk.

  As input reception processing realized by the program, certification key generation / input processing 401, work key generation / input processing 402, cover data input processing 403, superimposition information input processing 404, superimposition data input / editing / imaging processing 405, scramble key There is a generation / input process 406 and an intensity input / intensity map setting process 407. The input device includes a multimodal user interface, and in each of the above processes, an input by a user and an input from another system are accepted.

  A conversion coefficient / quantization table detection process 408 extracts a conversion coefficient and a quantization table as image features from the cover data received by the cover data input process 403. When the cover data is compression-encoded data, the frequency conversion is often performed. For example, in the case of MPEG, the conversion coefficient indicates a DCT (Discrete Cosign Transformation) coefficient. When the cover data is uncompressed data, frequency conversion is performed using a predetermined quantization table to obtain a conversion coefficient.

  The information received by the superimposition information input process 404 is information that is the basis for positioning the superimposition data with respect to the cover data. The upper right, the center, the whole, etc. are designated, and a superposition position map is created based on these. See FIG. 13 for the concept of alignment. If the user designation information received by the superimposition data input / editing / imaging processing 405 is a character, it is converted into a superimposition data by converting it into a character image. In the case of voice, it is converted into superimposition data by converting it into a character and a character image by voice recognition. You may edit text, audio, images, video, etc. Specify the parameters such as the original size of the superimposed data, the size of the cover data, and the scaling factor by which geometric deformation method is used to float the superimposed data. A predetermined value such as a reaction at an area ratio of ½ may be used, or a plurality of parameters may be determined separately for each partial region of the image.

  The superimposition information input process 404 accepts input of user designation information related to the superposition position map. The superposition position map alignment designation processing 409 generates a superposition position map. The superimposition position map shows the size of the superimposition image (or its reduced / enlarged image), the position of the superimposition image, the frame time, number, visualization characteristics, type, memory storage address of the original data, etc. in the image coordinate system of the cover data. This data is determined (see also FIG. 13).

  In the superimposition information input process 404, information may be embedded in the superimposed image using a digital watermark technique. For example, Japanese Patent Application Publication No. 2007-013357 discloses a digital watermark technique that can be applied to binary images such as check characters. It may be information specified by the input device or automatically generated information such as time.

  Using the scramble key generated or received by the scramble key generation / input process 406, the scramble process 411 encrypts the superimposed data. This is performed when it is designated in the superimposition information input process 404 that the superimposition data is encrypted. When the superimposed image is scrambled by the scramble processing 411, the intensity map is scrambled by the scramble processing 416 using the same scramble key as the key used for the superimposed image, whereby the intensity map and the pixel of the superimposed image are Take position correspondence. Thereafter, the superimposition data creation processing 417 creates superimposition data reflecting the intensity map.

  The time-space area dividing process 412 divides time and space into partial areas. The time-space region division processing 412 follows the designation of the superposition position map that is accepted / generated by the superposition information input processing 404 and the superposition position map position designation processing 409. In the case of a moving image, a buffer 413 for processing may be used depending on the frame type. Further, the space-time domain division processing 412 performs geometric deformation characteristic setting processing 414, recompression characteristic setting processing on the superimposed image (or scrambled superimposed image) output from the superimposed data input / edit / image processing 405. After performing special processing, which will be described later, using 415, a superimposed data creation process 417 creates superimposed data.

  The intensity input / intensity map setting process 407 inputs / sets / generates an intensity map designated by the user, and later adjusts the intensity of the superimposed data to assist the work of creating the superimposed data. The intensity map is determined in advance by setting the upper and lower limits of the amount of change in the pixel value of the image and the amount of change in the frequency component, or may be specified by the user or calculated from the relationship with the brightness value of the cover data. Also good. The data of the created or input or predetermined intensity map is stored in an intensity map table on the memory or disk (see FIG. 8 for the concept of the intensity map table).

  The superimposition process 418 performs superimposition of the superimposition data and the cover data along the superimposition position map designated by the superimposition position map alignment designation process 409. The image may be superimposed by adding the frequency components of the image or by adding the pixel values of the luminance / color difference signal components. At the time of addition, the image quality analysis processing 424 may adjust the strength of the superimposed data component in advance with respect to the cover data or the superimposed data so that the image quality of the cover data does not deteriorate significantly.

  The compression coding parameter adjustment processing 419 performs compression coding on an image that has undergone the superimposition processing if it is uncompressed. In addition, if the image that has undergone the superimposition process has been compression-encoded and has been generated by adding transform coefficients, the data change (for example, data size, upper limit value, changed quantization table replacement, etc.) associated with the operation Adjust parameters such as header information and file management information.

  The stego data output process 420 outputs stego data including a latent image. When the user gives an instruction to encrypt the stego data, the encryption process 421 encrypts and outputs the stego data using the work key generated / input accepted by the work key generation / input process 402. . For the stego data or the encrypted stego data, the electronic signature processing 422 may add the electronic signature data and certificate data by using the signature key generated / input accepted by the certification key generation / input processing 401.

  The auxiliary information output processing 423 outputs a scramble key, work key, electronic signature, certificate, signature key, time stamp, time information, management information, user information, and the like. The output destination of the stego data output processing 420 and the output destination of the auxiliary information output processing 423 may be a disk, a monitor, a portable medium, or another system connected to the network. Also good.

  FIG. 5 is a conceptual diagram showing the principle of imparting geometric deformation characteristics to a superimposed image.

  Attention is paid to a partial area yo502 of the cover image yo501 that becomes the size of the superimposed image when the cover image 501 is reduced by a certain ratio R. The superimposed image ym is set to 503, a width wd, and a height hd. The partial area of the cover image yo501 has a width ws and a height hs. A function for obtaining the least common multiple is LCM ().

  In the virtual rectangular area 505, an area having a width LCM (ws, wd) and a height LCM (hs, hd) is equivalent to an integral multiple of the width and height of the superimposed image 503. Similarly, the same rectangular area 505 is equivalent to a value obtained by multiplying the width and height of the attention area 502 of the cover image 501 by another integer. Paying attention to these geometrical relationships, one pixel 509 in the superimposed image 503 corresponds to the (ws / wd) * (hs / hd) region 508 in the cover image 501. In FIG. 5, the black small block 503 indicating the superimposed image corresponds to one pixel of the superimposed image, but the black small block 508 of the cover image and the black small block 506 of the virtual rectangular area are directly connected to the respective pixels. It shows the relative position and size in the whole image without corresponding.

  In the cover image yo501, the (ws / wd) * (hs / hd) attention area 508 is a pixel value reflected as a pixel of the superimposed image size at the time of reduction. For example, the average pixel value of the attention area 508 may become the pixel value of one pixel of the reduced cover image after the cover image is reduced. Alternatively, the pixel value of one specific pixel in the attention area 508 may become the pixel value of one pixel of the reduced cover image after the cover image is reduced.

  In the superimposition processing, attention is paid to a location A that corresponds to the pattern of the superimposed image and a location B that does not correspond in the partial area on the cover image. For the location A, the average value of the pixel values of the partial areas on the cover image is slightly changed. Place B is not changed. When such a change is given to the cover image, when the cover image is reduced, places where the average value is intentionally changed are connected to each other and perceived by humans, and the superimposed image appears to float. If the partial area of the cover image is completely reduced to the size of the superimposed image, the superimposed image appears most clearly. It is known from the research results of cognitive psychology that human beings have a psychology that tries to connect distant figures. For this reason, the superimposed image appears to appear even in another reduction.

  FIG. 6 is a diagram showing a processing procedure for imparting a geometric deformation characteristic to the superimposed image in the geometric deformation characteristic setting process 414.

  In the superimposed data input / edited image processing 405, the superimposed image that has been accepted / generated is referred to as a superimposed image sA. Further, in the cover data input process 403, an input accepted / generated image is used as the cover image 501. The superposition position map generated by the superposition position map alignment designation processing 409 is the size of the superposition image (or its reduced / enlarged image), the position of the superposition image, the frame time, the number, and the visualization characteristics in the image coordinate system of the cover data. , Type, data for determining the memory storage address of the original data, etc. (see also FIG. 13). Attention is paid to a partial area 507 of the cover image yo that becomes the size of the superimposed image 503 when the cover image yo 501 is reduced by a certain ratio R. The superimposed image ym503 has a width wd and a height hd (see FIG. 5). As the attention area 507 of the cover image 501, one is selected from a plurality of overlapping position candidates indicated by the overlapping position map. As pre-processing of step 601, these processes are repeated / sequentially processed.

  In step 601, management information is embedded in the superimposed image sA using a digital watermark to create a superimposed image sB. As described above, the management information may be input by a user, may be determined in advance, or may be input from another system (such as a time generation source) from the outside. Good.

  In step 602, the superimposed image sB is enlarged to xs1 times and ys1 times in the vertical and horizontal directions, and when the superimposed image sB is a binary image (1 bit per pixel), it is converted into a multivalued (eg, 8 bits per pixel) A superimposed image sC of the value image is obtained.

xs1 = LCM (ws, wd) / wd
ys1 = LCM (ws, wd) / wd
In step 603, a low-pass filter (LPF) is applied to the enlarged superimposed image sC to obtain a superimposed image sD. Note that the size of the superimposed image sD is LCM (ws, wd) * LCM (hs, hd).

  In step 604, the superimposed image sD after LPF is reduced to 1 / xs2, 1 / ys2 in the vertical and horizontal directions to obtain a superimposed image sE.

xs2 = LCM (ws, wd) / ws
ys2 = LCM (hs, hd) / hs
In step 605, the intensity input / intensity map setting process 407 reads and sets the intensity map (see also FIGS. 7 and 8). A superimposed image sF is created by superimposing uniformly / selectively so that the average pixel value of the block of interest changes during superimposition according to the intensity map. Alternatively, they are superimposed so as to maintain the image quality of the cover image. For example, in the case of selectively overlapping so that the average pixel value of the block of interest does not change, the following is performed. Even if the average pixel value of the block of interest does not change, if you create a state in which a specific pixel value protrudes, image processing in which only the specific pixel value is selectively reflected on the processed image at the time of reduction (the nearest neighbor) In the case of reduction by the law), it is possible to hide the latent image while suppressing deterioration of the image quality of the stego image. An intensity map serving as a judgment criterion for such various processes is prepared in advance or dynamically generated and selected and used by the user.

  In step 606, the cover image and the superimposed image sF are superimposed to obtain a stego image. Superimposition processing 418 is performed. The superimposition target of the cover image may be a luminance component, a color difference component, or an RGB component. Superimposed images may also be interpreted as luminance shading, and the upper 4 bits of 8-bit information are red components, and the inversion information of lower 4 bits (evaluated as a larger value closer to 0) is a different color such as a blue component. It may be separated.

  FIG. 7 is a diagram showing in detail the method for creating the intensity map described in step 605.

  In step 701, the superimposed image sE709 is read.

  In step 702, management information designated by the user is set.

  In step 703, the entire superimposed image is divided into small partial areas in the entire superimposed image, and the following steps 704 to 707 are repeated.

  In step 704, the target partial area of the superimposed image is set.

  In step 705, the intensity map table (see FIG. 8) storing the intensity map is referred to.

  In step 706, the superimposed image is analyzed to obtain a feature amount of the superimposed image. The line segment direction (eight directions), the degree of unevenness of the object (black and white density), area, center of gravity, etc. are determined. The determination of the line segment direction is performed by calculating the correlation between the line segment image directed in the eight directions of up, down, left and right and the pixel value of the superimposed image, and the direction having the minimum correlation value is set as the line segment direction. The unevenness can be achieved by the size of the frequency component of the target partial region.

  In step 707, intensity map data corresponding to the feature amount of the partial area is selected. Various selection methods are possible. For example, select a map that distributes large intensity values parallel to the line segment direction, or select a map that distributes large intensity values in a direction orthogonal to the line segment direction, or has a large degree of unevenness / If small, select a homogeneous / distributed intensity map. The user may select it.

  In step 707, an intensity map 710 is created by adding or multiplying the intensity information (intensity parameter) specified in the selected intensity map by a user-specified intensity setting value. If the intensity map is reduced at an appropriate ratio, as shown in the reduced intensity map 711, distant points are connected and the latent image is easily visualized more clearly.

  In step 708, the superimposed image sF is output.

  FIG. 8 is an explanatory diagram showing an intensity map table.

  The intensity map determined based on the intensity map creation method of FIG. 7 is stored in an intensity map table 801 on a disk or memory. Each record of the intensity map table 801 includes a partial region pixel array 802, ID 803, line segment direction 804, object unevenness (black and white density) 805, management information 806, and intensity setting value 807. Related information such as area and center of gravity may be stored. Homogeneous / selective superposition so that the average pixel value changes during superposition. The management information 806 is a flag specified in step 605 and step 606 when the user specification such as overlaying so as to maintain the image quality of the cover image is not reflected. When this flag is off, priority is given to user designation.

  In step 707, intensity map data corresponding to the feature amount of the partial area is selected. The intensity setting value 807 is selected according to the arrangement of pixel values of the partial area of interest (illustrated as a 4 × 4 area of black and white binary in FIG. 8). Alternatively, the strength setting value 807 may be obtained using the line segment direction 804, the degree of unevenness (monochrome density) 805 of the object, etc. as a table search key.

  FIG. 9 is an explanatory diagram illustrating a procedure for superimposing a cover image and a superimposed image when a moving image is exemplified.

  In step 901, a superimposed position map is created.

  In step 902, the superimposed image is subjected to frequency conversion and quantization.

  In step 903, cover image reading, decoding, and color space adjustment are performed.

  In step 904, iterative control is performed in GOP (Group Of Picture) units or frame units of the cover image.

  In step 905, the frame data is stored in the frame buffer.

  In step 914, the luminance information and chrominance information included in a certain frame are repeatedly controlled in units of slices or macroblocks, and attention is paid sequentially to one macroblock. The size of the macroblock is 16 × 16 or the like.

  In step 906, the superimposed position map is referred to.

  In step 907, it is determined whether or not the position of the macro block of interest is a position specified in the superimposition position map (a position where the superimposition image is superimposed in the cover image). If it is not the superposition position, the process returns to step 914 to refer to the next macroblock. If it is determined in step 907 that the position of the macro block of interest is a superposition position, the processing from step 908 to step 913 is performed.

  In step 908, it is determined whether the target macroblock is an intra macroblock. When the macro block is not an intra macro block, information on the target macro block exists in a form in which a specific macro block in the frame information at another time is used as an entity and an address is referenced using a motion vector.

  In step 909, the macro buffer information (corresponding to a picture) existing in another frame in time is obtained from the macro block information corresponding to the target macro block with reference to the frame buffer.

  In step 910, the noticed macroblock information is replaced with the actual macroblock information existing in another frame in time as necessary. That is, the compressed expression by the motion vector is locally canceled and converted to the frequency component of the pixel value set.

  By referring to the superposition position map and replacing the macroblock with a smaller macroblock and performing the minimum necessary cancel operation, an increase in the amount of output data due to motion vector cancellation can be suppressed.

  In addition, when macroblock information existing in another frame in time has already superimposed a superimposed image, the motion vector may not be locally canceled. For example, this is a case in which macroblock information existing in another frame in time is referred to in a background portion having no motion in a compressed expression using a motion vector. When a certain macroblock information is rewritten, the rewrite result is reflected as it is in other frames referring to the macroblock.

  In step 911, the macro blocks of the superimposed image and the cover image are superimposed. In the case of data representation by frequency transformation, the pixel value obtained by inversely transforming the value added in the frequency domain into the spatial domain is added to the cover image and the superimposed image as the spatial (ie, picture) pixel value due to the linearity of the transformation function. The resulting pixel value is technically equivalent.

  In step 912, compression coding parameter adjustment is performed along with rewriting of the GOP structure, frame information, slice information, and macroblock information in the above processing. For example, since the data amount changes, the data amount after the change is recorded and calculated.

  In step 913, header information to be output is set.

  In step 915, encoding processing such as transform coefficients is performed.

  In step 916, if processing in units of GOP has been completed, GOP information is read from the buffer and output. If it is in the middle of GOP or frame processing, the process returns to step 905, the intermediate information is stored in the buffer, and the process proceeds to the next process (step 914). In step 904, if all GOPs have been processed, the process ends.

  FIG. 10 is an explanatory diagram showing the concept of a re-compression encoding characteristic imparting processing procedure for a superimposed image.

  The addition of the recompression coding characteristic is a process in the recompression characteristic setting process 415. Attention is paid to a partial area 1002 in the cover image 101. This partial area data 1003 corresponds to a macroblock such as JPEG or MPEG, and is a rectangular area such as 16 × 16. FIG. 10 shows a 4 × 4 example. Attention is paid to the partial area 1005 in the superimposed image 102, and the geometric deformation characteristics shown in FIG. The data group 1006 after giving the geometric deformation characteristics has the same picture size as the data group 1003. In addition, the setting of the superimposed position map and the intensity map is reflected, and the pattern is in a state that is technically equivalent to the pattern 710 in FIG. 7, and it is difficult for human eyes to see the pattern of the superimposed image.

  In compression encoding processing such as MPEG and JPEG, DCT conversion is used for frequency conversion. In the compression encoding process, after frequency conversion, quantization is performed so that many bits are allocated to low frequency components, and data compression is performed. The numerical value of the frequency component becomes irreversible after quantization, and the pixel value obtained by inversely transforming the quantized frequency component is slightly different from the original pixel value. A natural image uses the characteristic that signal power concentrates on low frequency components. However, the pattern of the superimposed image may not be a natural image, such as when checking characters are written, and not only low frequency components but also high frequency components are important. By appropriately maintaining and processing the high-frequency component, it is possible to impart recompression coding characteristics to the superimposed image. However, if the frequency component is divided into low frequency, intermediate frequency, and high frequency, the high frequency component is easily lost by the compression encoding process. The selection and processing level of the intermediate frequency is important.

  The quantization transform coefficient table 1007 is a set of values obtained by frequency transforming and quantizing the data group 1003. In the conventional notation of DCT transform coefficients, the effective value is biased to the upper left low frequency component, and the lower right high frequency component is often zero. Here, the lowest frequency among the frequencies at which the value of the quantized transform coefficient is zero is referred to as a cutoff frequency.

  FIG. 10A is a graph showing frequency components obtained by frequency-converting and quantizing the partial region data 1003 of the cover image 101. In the graph, the horizontal axis represents the frequency of the image, and the vertical axis represents the spectral value in the conversion region.

  FIG. 10B is a graph showing frequency components obtained by frequency conversion of the data group 1006 after the geometric deformation characteristics of the superimposed image 1004 are given. In FIG. 10B, a frequency portion exceeding the cutoff frequency is referred to as a cutoff region 1008.

  FIG. 10C is a graph obtained by applying a cutoff frequency to the frequency component shown in FIG. 10B and removing the high frequency component. The frequency calculated by the quantization transformation coefficient table 1007 is used as this cutoff frequency. Alternatively, a value obtained by minutely changing the cutoff frequency within a predetermined range may be used.

  If a set of frequency components from which high-frequency components have been removed is returned to the spatial domain by inverse frequency transformation (IDCT or the like), a pattern that is close to the original pattern is reproduced. However, many high-frequency components to be removed are included near the edge portion of the original superimposed image 1005. When only high-frequency components are selectively removed in the frequency domain, after returning to the spatial domain by inverse frequency transformation (such as IDCT), the edge portion of the superimposed image may be disturbed and mosquito noise may appear. The part that appears to be mosquito noise has a cognitive effect on the frequency band near the cutoff frequency, such that the wave of the frequency part above the cutoff frequency is reflected by the cut-off frequency as a wall. It becomes easy to perceive.

  In FIG. 10D, after removing the high frequency component higher than the cutoff frequency from the frequency component shown in FIG. 10B, the frequency band (bandwidth fd) below the cutoff frequency is processed. Is. Constants may be determined in advance for the maximum and minimum values of the bandwidth fd. In the case of DCT conversion, the frequency axis can be expressed by a multiple kθ (k = 1 to n) of the base frequency θ.

  When the value of k as the cutoff frequency is kc, the frequency component of (kco) × θ indicates a frequency one unit lower than the cutoff frequency on the horizontal axis in FIG. The above processing may be performed by concentrating on the frequency component of (kc-1) × θ. Alternatively, using the predetermined two constants α and β, the frequency band of (kc−α) × θ to (kc−β) × θ may be set as the bandwidth fd and may be processed.

  For the frequency band (bandwidth fd) lower than the cut-off frequency, a value corresponding to the power of the spectrum value in the region portion above the cut-off region is added. The power is obtained as a waveform area calculation on the graph and is the sum of the spectrum values in the frequency band within the calculation range. Alternatively, a value obtained by slightly changing the power of the spectrum value within a predetermined range may be used.

  FIG. 10E is a graph showing frequency components of the output stego image. A frequency component obtained by frequency conversion / quantization of the data 1003 of the partial region of the cover image shown in FIG. A frequency component obtained by performing the processing shown in FIG. 10D on the frequency component obtained by frequency-converting the data group 1006 after the geometric deformation characteristics of the superimposed image are defined as B. When the parameter α obtained by multiplying the intensity setting value 807 determined in FIG. 8 by a predetermined constant is used, the frequency component C of the stego image can be calculated as a linear sum of A and B (C = A + αB).

  The cut-off frequency also changes depending on the setting during frequency conversion. When a JPEG or MPEG compression setting is high compression and a small file is output, the cutoff frequency is set to a low frequency. In the case of low compression, the frequency is high. Conversely, if the cut-off frequency is intentionally set low, it is possible to superimpose a superimposed image that is easily visualized at high compression, and conversely, a superimposed image that is easily visualized at low compression. In this manner, a superimposed image having a recompression characteristic corresponding to the setting such as the bit rate and the image quality when the stego image is compressed can be created.

  FIG. 11 is an explanatory diagram showing the basic principle that a superimposed image is visualized when a stego image undergoes compression-encoded image processing in the provision of recompression encoding characteristics.

  FIG.11 (e) is a figure equivalent to FIG.10 (e), and is a graph which shows the frequency component of the stego image output. In the graph, the horizontal axis represents the frequency of the image, and the vertical axis represents the spectral value in the conversion region. The vertical axis may be considered as a DCT coefficient after quantization.

  In FIG. 11E, the spatial component (pixel value) 1101 of the stego image obtained by inversely transforming the frequency component of the stego image includes the region of interest 1002 of the cover image 101, the geometric deformation characteristics, and the recompression coding characteristics. The superimposed image X to which is added is a picture superimposed. More specifically, a process of adding a recompression coding characteristic to the data 1106 of the area to which the geometric deformation characteristic is given is added to the data 1005 of the attention area of the superimposed image 102, and the transform component shown in FIG. Is the superimposed image X. When a person visually observes the stego image of the stego image 1101 on which the superimposition image X is superimposed and the entire stego image 104 including the spatial component 1101 in a part of the region 1102, the design of the original superimposition image 102 is difficult to see.

  FIG. 11F is a graph showing frequency components obtained by converting a partial region 1102 of the stego image 104 by recompression encoding processing. When the frequency component shown in FIG. 11 (f) is further inversely transformed, it appears as a pattern close to the pattern in which the superimposed image 102 is superimposed on the attention area 1002 of the cover image, and the shape and characters of the superimposed image are easily visible.

  In the recompression coding process, image processing such as a smoothing filter works so as to cancel out the mosquito noise-like noise emphasized by the process of FIG. Further, when a component of the superimposed image in the attention area 1104 in the stego image 104 is extracted, it is spatially spread so as to blur after the recompression encoding process due to the influence of the superimposed cover image. In addition, when the cover image includes many intermediate frequency components and high frequency components, many intermediate frequency components and high frequency components of the superimposed image remain after the recompression coding process. However, the recompression encoding process is performed for each partial area (macroblock) of interest. For this reason, continuity of pixel values with another macro block adjacent to the attention area 1104 is not ensured. Therefore, in human vision, block noise is recognized in addition to the emphasized mosquito noise at the position of the pixels constituting the superimposed image. When these visually recognized noises appear to be connected across a plurality of attention areas 1104, a person perceives a superimposed image.

  FIG. 11D ′ shows frequency characteristics obtained by frequency-converting the attention area 1104 of the stego image 104 again. The transform component quantization value and the cut-off frequency change, and the frequency characteristic (transform band characteristic) of the superimposed image is approximately restored.

  FIG. 12 is an explanatory diagram showing a detailed procedure for giving a recompression coding characteristic to a stego image.

  In step 1201, the transform coefficient of the encoded cover image is restored, and the macroblock and quantization table information are extracted.

  In step 1202, the zero value of the macroblock quantization transform coefficient information 1007 is scanned to set a cutoff frequency.

  In step 1203, the superimposed image data is read and divided into regions. According to the superposition position map, the position corresponding to the cover image is determined, frequency conversion and quantization are performed for each partial region, and a conversion coefficient is calculated.

  In step 1204, the frequency components larger than the cut-off frequency of the cover image are further divided out of the conversion coefficients of the superimposed image, and the processing in steps 1205 and 1206 is performed.

  In step 1205, a cutoff frequency is set from the conversion coefficient of the superimposed image as shown in FIG. A recompression condition (such as a bit rate) may be designated by a user input instruction, and the corresponding cut-off frequency may be set lower or higher.

  In step 1206, the cut-off area shown in FIG. 10C is removed by applying an LPF that passes a conversion coefficient lower than the cut-off frequency to the conversion coefficient of the superimposed image.

  In step 1210, the signal power (total conversion coefficient) of the superimposed image is calculated. The appropriateness of the signal power is determined, and if the signal power is smaller than a predetermined constant value Emax, the processing of step 1207, step 1208, and step 1209 is performed.

  In step 1207, the energy of the cut-off area of the superimposed image is calculated. Three parameters: signal power (total conversion coefficient, graph area), spectrum shape (from left, right, homogeneous, etc.), and cutoff frequency band (frequency range of which frequencies are deleted) in cutoff region 1008 Get.

  In step 1208, the energy of the cut-off area of the superimposed image is adjusted. If the signal power in the cutoff region is smaller than a predetermined threshold value, the signal power is raised. If it is larger, the signal power is reduced.

  In step 1209, the energy of the non-cut-off region of the superimposed image is adjusted. Processing is performed on the energy in the non-cut-off region of the superimposed image according to the width (frequency difference) of the cut-off frequency band. A value corresponding to the energy of the cut-off area of the superimposed image is added or deleted. Alternatively, a value obtained by multiplying a predetermined constant may be used. As shown in the graph of FIG. 10 (d), the signal power is adjusted by the width of the cutoff frequency band for the spectrum value that is in the non-cutoff region and is in the frequency component lower than the cutoff frequency.

  In step 1211, the conversion coefficient of the cover image and the conversion coefficient of the superimposed image are calculated. Operations are added. Subtraction or multiplication may be used. If the sum is simply added, the cover image becomes brighter as a whole. Therefore, a method may be used in which an image processing filter is applied to the cover image so that the signal power of the cover image is lowered / increased in advance. This image processing filter may be a noise removal filter.

  In step 1211, the macroblock information of the stego image is adjusted and encoding is performed. At the time of superposition, the quantization table of the cover image can be changed to another table. In that case, the macro block information of the stego image is adjusted so as to output the used table.

  FIG. 13 is an image diagram showing a superimposed image superimposed according to the superimposed position map.

  In a stego image, a superimposed image is difficult to be visually recognized unless geometric deformation or recompression coding is performed. However, since it is difficult to explain if a diagram cannot be visually recognized, a visible superimposed image is drawn for explanation.

  FIG. 13A is a conceptual explanatory diagram when a cover image is divided into a plurality of partial regions and a plurality of superimposed images having different visible characteristics (geometric deformation characteristics, recompression encoding characteristics, etc.) are superimposed. The partial area 1302, the partial area 1303, the partial area 1304, and the partial area 1304 have different geometric deformation characteristics. For example, when the stego image is reduced to an area ratio of 70%, the partial area 1302 is visualized, and when the area ratio is reduced to 50%, the partial area 1303 and the partial area 1304 are visualized. The partial areas may overlap each other or may be in different areas. The secure area 1305 is obtained by encrypting “prohibited copy ID: 0011”, encoding a signature, verification information, secret information, 1-bit information indicating the presence of a check character, and the like in a bar code form. If 1-bit information indicating the presence of a check character is detected, but the check character appears to disappear, it can be seen that the check character has been deleted by some means, contributing to applications such as tampering detection. . The partial region 1302, the partial region 1303, the partial region 1304, and the partial region 1304 may have recompression encoding characteristics with different compression rates.

  FIG. 13B shows an example in which superimposed images having different visual characteristics are superimposed depending on the moving image frame time. The moving image includes a plurality of frames 1306. A plurality of superimposed images having different visibility characteristics (geometric deformation characteristics, recompression encoding characteristics, etc.) are superimposed on a plurality of frames. The superimposed image itself may be moving image data with movement. When the entire moving image undergoes image processing, the location of the superimposed image having visible characteristics (geometric deformation characteristics, recompression coding characteristics, etc.) changes, and the superimposed image is easily recognized by human eyes.

  FIG. 14 is an explanatory diagram showing a processing procedure of the image superimposing device 203 when the machine automatically determines the detection of the superimposed image regardless of recognition of human eyes.

  In step 1401, the cover data input process 403 receives an input of a cover image.

  In step 1402, the superimposition position map alignment designation process 409 sets a secure area.

  In step 1403, the superimposed data input / edited image processing 405 encodes the superimposed information. The electronic signature processing 422 further adds an electronic signature and a public key certificate to basic information such as a user ID as superimposition information. Superimposition data input / edited image processing 405 encodes these bits to form a bit string having a length N of 0110001,. For encoding, an existing code system such as a JIS code may be used. Further, an error correction code is added to the bit string.

  The superimposition information encoded as a bit string in this way is encoded into a superimposed image data format. An area of n * n cells (n * n is equal to or less than N) is defined with four squares of a pixels as one cell. In accordance with the value of the superimposition information bit string, black if 0, white if 1, and white or black colored cells are sequentially arranged from the upper left to the lower right. When n squares are arranged, the operation moves to the lower row. In this way, encoded superimposed image data is created. The superimposed image data becomes a two-dimensional barcode-like pattern 1305. It may be converted into a multi-value image or may be arranged at random. Information indicating a randomly arranged arrangement may be held outside as a key and used at the time of detection.

  In step 1404, the superimposition process 418 performs superimposition of the superimposed image.

  In step 1405, the stego data output process 420 performs stego image output.

  FIG. 15 is an explanatory diagram illustrating the processing procedure of the detection device 214 when the machine determines whether to detect a superimposed image regardless of human eye recognition.

  The detection device 214 is a program (or a program for realizing the detection device 214) on a computer including an arithmetic unit (CPU), an input device, an output device, a memory (primary storage device), a disk (secondary storage device), and a communication device (or This can be realized by operating the module on the memory.

  The program may be stored in advance in a memory or a disk in the computer, or may be introduced from another device via a medium that can be used by the computer when necessary. The medium refers to, for example, a storage medium that can be attached to and detached from the computer, or a communication medium (that is, a wired, wireless, optical network, or a carrier wave or digital signal that propagates through the network).

  Each process shown in FIG. 15 is realized as a process by the CPU executing a program (or module) that realizes each process stored in the memory or the disk.

  In step 1501, an input of a stego image is accepted.

  In step 1502, superimposed image separation of the secure area 1305 is performed from the stego image. Since the secure area is in the form of a two-dimensional bar code, after converting the frequency of a stego image, a transform component having a multiple frequency component with one square corresponding to one bit as the base of the frequency is extracted, and inversely transformed to reconstruct the pattern. do it.

  In step 1503, the superimposition information is decoded. The reconstructed secure area is an image rectangular area with one side having a length of (a * n) and includes n * n cells. The average value of the pixel values of one square set in step 1403 is calculated and compared with a threshold value to determine whether the square is 0 or 1. A bit string of 1-bit information of n * n is extracted. A user ID or the like is decoded as superimposition information from the obtained bit string.

  In step 1504, the validity of the superimposed information is verified. Decode with error correction code. In addition, the electronic signature and public key certificate are verified. The validity can be verified when the error correction code is decoded or the signature is verified.

  In step 1505, superimposition information is output. Basic information such as a user ID is output as superimposition information.

  FIG. 16 is an explanatory diagram showing a processing procedure for detecting the digital watermark embedded in the superimposed image itself by the detection device 214.

  Each process shown in FIG. 16 is realized as a process by the CPU executing a program (or module) that realizes each process stored in the memory or the disk of the detection device 214 described above.

  The digital watermark is embedded in the superimposed image itself by the superimposed information input process 404.

  In step 1601, an input of a stego image is accepted.

  In step 1602, the stego image is analyzed.

  A superimposed image region is extracted according to the set superimposed position map. The superimposed image constituent pixels are extracted according to the set intensity map. The extracted pixel information is reduced according to the geometric deformation characteristics.

  In step 1603, the superimposed image is separated.

  In step 1604, a digital watermark is detected from the superimposed image. It can be detected by calculating the correlation with the digital watermark pattern. Japanese Patent Application Publication No. 2007-013357 discloses a digital watermark technique that can be applied to binary images such as check characters. Extract superimposition information.

  In step 1605, the validity of the embedded information is verified. For the verification, an error correction code or an electronic signature can be used.

  In step 1606, the embedded information is output. The result may be output to a monitor, which is one of output devices connected to the detection device, so that the person can visually recognize the detection result.

  In step 1607, the corresponding device is controlled.

  The output device of the detection device is equipped with a communication device through a network or device connection, and is connected to other devices and systems. Corresponding devices connected to the detection device include a still image viewing device, a still image recording device, a printing device, a digital camera, a moving image reproducing device, a moving image recording device, a video printer, and a video camera. Digital signage, photo frames, etc. may be connected. Function information that specifies the operation method is defined corresponding to each function such as rendering, encoding, decoding, saving, and printing of these devices. Also, state information is defined for each device state such as recording or playback. The function information and status information may be a serial number system. Also, the control information may mean “double speed recording OK”.

  The corresponding device includes an input device, an output device, a disk, a memory, an arithmetic device, and a control device, and the input device of the corresponding device is connected to the detection device. When function information and status information is received from the input device of the corresponding device, the control device of the corresponding device performs control corresponding to the function information and status information. In advance, function information, state information, and the like are expressed as digital watermark embedding information, and embedded as a digital watermark in a superimposed image in a stego image. In step 1607, information such as function information and status information detected as a digital watermark is sent to the response crisis, and the corresponding device is controlled.

  As an application thereof, for example, an integrated video camera including a detection device and a moving image recording device can be configured. The stego image taken via the camera lens is input to the detection device, and when the recording is detected as function information, the function information is sent to the moving image recording device, and the moving image recording device performs video recording. When recording NG is detected, video recording is not performed or stopped.

101: Cover data, 102: Superimposed image, 104: Stego data, 203: Image superimposing device, 214: Detection device, 801: Intensity map table.

Claims (10)

An image processing method using a computer to generate a stego image obtained by superimposing a superimposed image representing the information as a latent image on a cover image before embedding information,
A geometric deformation characteristic setting step for processing the superimposed image so that the superimposed image is visualized when the stego image is reduced;
And a recompression characteristic setting step for processing the superimposed image so that the superimposed image is visualized when the stego image is reduced. The image processing method according to claim 1,
In the geometric deformation characteristic setting step , when the cover image is reduced by the ratio R, the size of the superimposed image becomes the size of the superimposed image. An image processing method comprising: changing an average value of pixel values of the partial areas on the cover image, and maintaining the pixel values without changing the partial areas that do not correspond. The image processing method according to claim 2,
The calculator includes an intensity map that defines a position of a pixel set and an overlap intensity;
A superimposed position map that associates the pixels of the superimposed image with the positions of the pixel set included in the partial area of the cover image,
The geometric deformation characteristic setting step selectively changes the pixel value of the partial region according to the pixel value of the superimposed image and the intensity map with respect to the partial region of the cover image;
And a step of outputting a stego image. The image processing method according to any one of claims 1 to 3,
The recompression characteristic setting step includes:
Calculating a cutoff frequency of the cover image according to re-compression encoding conditions;
Setting a cutoff frequency of the superimposed image according to the recompression encoding condition;
And a step of processing a frequency component equal to or lower than a cutoff frequency of the superimposed image. The image processing method according to claim 4,
The recompression characteristic setting step includes:
Removing a frequency component equal to or higher than the cutoff frequency from the superimposed image;
Adding a value corresponding to power of a spectrum value equal to or higher than the cutoff frequency to a predetermined range in a frequency band lower than the cutoff frequency of the superimposed image. An image processing method. In the computer, a latent image detection method for detecting the superimposed image from the stego image created by the image processing method according to any one of claims 1 to 5,
Display or print the result of geometric deformation of the stego image,
Alternatively, a latent image detection method for visualizing the superimposition information by either displaying the result of photographing the displayed or printed stego image. The image processing method according to claim 3,
The superposition position map includes the position, size, and image processing characteristics of one or more superposition areas,
The intensity map includes arrangement information of pixel values of partial areas, an ID, a line segment direction, an unevenness coefficient, management information, and an intensity setting value, and the intensity in a direction orthogonal to the line segment direction. Contains records that contain strongly set strength settings,
Superimposing a superimposed image on one or more positions of the cover image according to the superimposed position map;
Searching for a record in which the arrangement information of the pixel values of the superimposition position map matches using the arrangement of the pixel values of the partial area of the cover image as a key;
Reflecting the intensity setting value of the record in changing the partial area of the cover image;
When a plurality of records are searched for the strength setting value of the record,
Calculating a line segment direction A of the superimposed image;
Searching for the intensity setting value information B for the plurality of records searched using the line segment direction A as a key;
Reflecting the intensity set value information B in the change of the partial area of the cover image. The image processing method according to claim 7, comprising:
The cover image and the superimposed image are represented by frequency components,
Imparting a reduction processing characteristic or a recompression encoding processing characteristic to the superimposed image;
Performing a process of superimposing the cover image and the superimposed image on a frequency space;
And a step of converting the motion vector into a frequency component of a pixel value set if the cover image data includes a motion vector expression. In the latent image detection method for detecting the superimposed image from the stego image created by the image processing method according to any one of claims 1 to 8,
Extracting the superimposed image from the stego image;
Detecting information from the superimposed image;
And outputting the extracted information. A latent image detection method comprising: Computer input / output control in a system in which a computer A having memory, arithmetic means, communication means, and input / output means and a computer B having memory, arithmetic means, communication means, and input / output means are connected by a network. A method,
The computer A is
Executing the image processing method according to any one of claims 1 to 8,
Transmitting the information extracted from the stego image to the computer B,
The computer B includes a step of receiving the information and performing input / output control of the computer B based on the information.

Images