JP2005253005A - Information processing apparatus and method therefor, computer program and computer-readable storage medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technique for embedding the information of both types of characters without hardly causing the sense of incongruity for a document image in which the character types having different character widths such as Japanese or English. <P>SOLUTION: A document analyzer 102 obtains the position and the size of part circumscription rectangle each character image from a document image inputted from an input 101, forms the result of an analysis into a data format for each line, and outputs it to a normalizer 104. In the normalizer 104, a Japanese character image is set between characters that correspond to Japanese characters, and an English character is normalized between characters that correspond to English characters. An information embed part 108 embeds a division by utilizing an embedding technology that exerts an influence upon the blank length between the adjacent characters with respect to the Japanese characters, and embeds information by adjusting the vertical position of a character with respect to the English characters. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a technique for embedding information in a document image and extracting the embedded information.

  In recent years, with the progress of computerization and networking of media such as character data, image data, and audio data, prevention of unauthorized copying of digital data itself and association between digital data and metadata are required.

  On the other hand, in the case of documents / images, the distribution form is still often printed. As described above, since digital data and printed material are used in combination, there is a need for a form of a distribution destination when digital data is distributed as printed material and some means for linking the printed material and digital data.

  Under such circumstances, a technique for embedding watermark information with a digital watermark has been proposed for a multi-valued image such as a still image or a document image. The digital watermark embedding means a technique for embedding watermark information by changing a part of original data so that digital image data, audio data, character data, etc. cannot be perceived by humans.

  As a digital watermark technique for a multi-valued image, various methods using the redundancy of pixel density are generally known. For example, it is a method of embedding by changing each pixel value (Non-Patent Document 1).

  On the other hand, binary images such as document images composed of characters and the like have little redundancy, and it is difficult to realize digital watermark technology. However, there are some known digital watermark methods using characteristics unique to document images. Yes. For example, a method of moving the baseline of a line (Patent Document 1), a method of operating a space length between words (Patent Document 2, Patent Document 3), a method of operating a space length between characters (Non-Patent Document 2), a character And a method of changing the inclination by rotating (Non-Patent Document 3).

  Here, a method of embedding information by manipulating the space length between characters will be described with reference to FIGS.

  FIG. 20 shows a part of a document image before embedding watermark information. FIG. 21 is a diagram showing a part of the document image after watermark information is embedded in a part of the document image of FIG.

  The space lengths P0, S0, P1, and S1 between the characters shown in FIG. 20 become the space lengths P0 ′, S0 ′, P1 ′, and S1 ′ between the characters shown in FIG. 21 when watermark information is embedded. Become. 20 and 21, there are five characters and four spaces between characters. Since two blank lengths are assigned to one bit of information to be embedded, it is possible to embed 2-bit information with four blanks.

  Therefore, when embedding binary (binary) “01”, the character image “child” between P0 and S0 in FIG. 20 is on the left and “ka” between P1 and S1. The character image is shifted (moved) to the right to generate a document image as shown in FIG.

FIG. 22 shows a character image when watermark information is embedded by digital watermarking by rotating the character and changing the inclination. For example, “1” is embedded when rotated clockwise (1 in FIG. 22), and “0” is embedded when rotated counterclockwise (2 in FIG. 22). In FIG. 22, the character image of “through” is rotated clockwise and the character image of “shi” is rotated counterclockwise. In this case, information “10” is embedded in binary. Will be.
Techniques for data hiding W. Bender, D. Gruhl, N. Morimoto, A. Lu IBM Systems Journal, vol. 35, nos. 3 & 4, 1996. ("Electronic document data hiding technique using inter-character space" by King Mongkut University, The 1998 IEEE Asia-Pacific Conf. On Circuits and Systems, 1998, pp.419-422.) “Digital Watermarking with Sealed Images on Japanese Documents” (Yasuhiro Nakamura, Kokoo Matsui), Transactions of Information Processing Society of Japan Vol. Japanese Patent No. 3136061 US Pat. No. 6,086,706 JP-A-9-186603

  As described above, there are several techniques for embedding information in a document image. However, an important point is to make it difficult for human eyes to perceive whether embedding has been performed.

  By the way, focusing on Japanese (a character set such as kanji or flat), although there are some exceptional characters, the sizes of the characters are almost the same, and there is a high tendency to orderly. On the other hand, in the case of English sentences, there are uppercase letters and lowercase letters in English letters, and the size of each letter is not uniform as compared with Japanese letters, and the spacing between the letters constituting the word is narrow. In view of such a difference in characteristics, when embedding information in a document image in which a Japanese character set and an English character set are mixed, it is desired to embed the information so that there is no sense of incongruity in any case.

  The present invention has been made in view of such problems, and a technique for embedding information in both types is unlikely to occur in a document image in which character types having different character widths such as Japanese and English are mixed. It is something to be offered.

In order to solve this problem, for example, the document processing apparatus of the present invention has the following configuration. That is,
An information processing apparatus for embedding information in a document image,
A first embedding unit that affects the blank length between adjacent character images and embeds information;
A second embedding means for influencing the position in the direction orthogonal to the character arrangement direction and embedding information;
Extraction means for extracting information about the position and size of the circumscribed rectangle of each character image in the input document image;
Based on the position and size of each extracted character image, each character image is divided into a first character image type having a first character width and a second character width having a second character width smaller than the first character width. A classification means for classifying into two character image types;
Selecting means for selecting the first embedding means for the character image of the first character image type, and selecting the second embedding means for the character image of the second character image type;

  According to the present invention, by performing embedding processing suitable for each of the first character image type having the first character width and the second character image type narrower than the first character width, In an image in which characters are mixed, a sense of incongruity hardly occurs, and information can be embedded in both types.

  Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

<First Embodiment>
FIG. 1 is a conceptual diagram of the configuration of the digital watermark embedding apparatus according to the first embodiment.

  The digital watermark information embedding device according to the first embodiment performs document analysis in the document image 100 input from the input unit 101 (image scanner or the like) and recognizes character size and position information. A recognition processing unit 103 that recognizes character code information and font information by performing character recognition, a recognition dictionary 105 that is a dictionary used for character recognition in the recognition processing unit 103, and a document image based on a character recognition result. A watermark for embedding information input from a watermark information input unit 107 (information input or registered in advance by the user) using the normalization unit 104 for normalization and the input verification image 100 and the normalized document image 106 It comprises an information embedding unit 108 and an image output unit 109 that generates a digital watermark embedded image 110.

  FIG. 1 is a conceptual diagram of the configuration of the digital watermark embedding apparatus. FIG. 23 shows a specific configuration thereof. The apparatus on the digital watermark extracting side can also be realized with the same configuration.

  In FIG. 23, a computer 2301 is a general-purpose information processing device such as a personal computer that is generally used. The computer 2301 can input an image read from an image input device 2317 such as a scanner, and can edit and store the image. is there. Further, an image obtained by the image input device 2317 can be printed from the printer 2316. Various instructions and the like from the user are performed by input operations from a mouse (registered trademark) 2313 and a keyboard 2314. Inside the computer 2301, each block to be described later is connected by a bus 2307, and various data can be transferred.

  In FIG. 23, the MPU 2302 can control the operation of each block inside the computer 2301, or can execute a program stored therein. A main storage device (configured by a RAM) 2303 is a device that temporarily stores a program and image data to be processed for processing performed in the MPU 2302. A hard disk (HDD) 2304 is an apparatus that can store in advance an OS, application programs, and image data to be transferred to the main storage device 2303 and the like, and can store processed image data.

  A scanner interface (I / F) 2315 is an I / F that is connected to a scanner 2317 that reads an original, a film, and the like and generates image data, and can input image data obtained by the scanner 2317. The printer interface 2308 is connected to a printer 2316 that prints image data, and is an I / F that can transmit image data to be printed to the printer 2316.

  The CD drive 2309 is a device that can read or write data stored in a CD (CD-R / CD-RW) that is one of external storage media. The FDD drive 2311 is a device that can read from the FDD and write to the FDD in the same manner as the CD drive 2309. Similar to the FDD drive 2311, the DVD drive 2310 is a device that can read from and write to a DVD. If an image editing program or printer driver is stored on a CD, FDD, DVD or the like, these programs are installed on the HDD 2304 and transferred to the main storage device 2303 as necessary. It has become.

  An interface (I / F) 2312 is an I / F connected to these in order to receive input instructions from the mouse 2313 and the keyboard 2314. The monitor 2306 is a display device capable of displaying the watermark information extraction processing result and the processing process. Further, the video controller 2305 is a device for transmitting display data to the monitor 2306.

  Note that the apparatus according to the embodiment may be applied to a system including a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.), or a device (for example, a copier, a facsimile apparatus) including a single device. Etc.).

  In the above configuration, by executing a program loaded into the main storage device 2303 in response to an input instruction from the mouse 2313 or the keyboard 2314 using the MPU of 2302, the digital watermark embedding device or the extraction device functions. . At this time, it is possible to monitor the execution status and the result by the monitor 2306.

  Although the apparatus configuration of the embodiment has been described above, the document analysis unit 102, the recognition processing unit 103, the normalization unit 104, the watermark information embedding unit 108, and the like in the configuration of FIG. 1 are realized by a program executed by the MPU 2302. 1 will be described in detail.

  The document analysis unit 102 extracts a circumscribed rectangle (rectangle) from the input document image. The circumscribing rectangle of a character is a rectangle circumscribing the character, and is originally information indicating a region where character recognition is performed, but in a digital watermark, it indicates a character region to be embedded. Each pixel value of the document image is projected onto the vertical coordinate axis, and a blank portion (a portion having no black character) is searched to determine a row, and line division is performed. Thereafter, the document image is projected on the horizontal coordinate axis line by line, and a blank portion is searched and divided into characters. As a result, each character can be cut out by a circumscribed rectangle. The analysis result in the character analysis unit 102 and the document image data input from the input unit 101 are passed to the recognition processing unit 103.

  The analysis result passed from the character analysis unit 102 to the recognition processing unit 103 is in a format as shown in FIG. 24, for example, and is stored in the main storage device 2303 (or HDD 2304) as a storage unit. As shown in the figure, the information is line-by-line, the information indicating the number of lines at the top, the information indicating how many circumscribed rectangles are in the corresponding line, and the number of data related to each circumscribed rectangle. It consists of information (EOL (= End Of Line)) that follows and ends at the end. Information (EOP (End Of Page)) indicating the end of one page is added to the end of the last line, not the end-of-line information. Here, as shown in the figure, the data relating to one circumscribed rectangle includes the position of the circumscribed rectangle (the coordinate position of the lower left corner of the rectangle in the embodiment), the size (height and width) of the circumscribed rectangle, the character code, and the phone type. It consists of a field that stores However, at this stage, the character code and font type are unknown. The data shown in FIG. 24 is hereinafter referred to as document image attribute data.

  FIG. 2 is a block diagram illustrating a detailed configuration of the recognition processing unit 103. In the present embodiment, it is assumed that the recognition processing unit 103 performs character recognition by optical character recognition (OCR). By using the OCR technique, it is possible to identify a character even from a document image subjected to a minute rotation of the character. In addition to character information (character code), multi-font identification is also possible. For the technology related to font identification, see, for example, “Shinichiro Hashimoto,“ Introduction to Character Recognition ”published by the Electronic Communication Association”.

  The feature extraction unit 103a in the recognition processing unit 103 cuts out a corresponding circumscribed rectangle from the input document image based on the document image attribute data (see FIG. 24) as an analysis result from the document analysis unit 102, and cuts out the cut out rectangle. Extracts feature values of character images. The recognizing unit 103b uses the feature amount obtained by the feature extracting unit 103a and each standard font set (font) stored in the recognition dictionary 105 (stored in advance in the HDD 2304 in FIG. 23). Character recognition is performed by comparing the character feature amount, and the font type is also identified. Then, the document image creation data is updated to reflect the character recognition result. That is, a process of storing the recognized result in the character code and font type fields in each “circumscribed rectangle data” in the document image attribute data is performed. At this time, the character code of the Japanese character is represented by 2 bytes, and the English character is represented by 1 byte.

  In the character recognition technology, since a method for identifying whether each character is a full-width character such as Japanese or a half-width character such as English is well known, details are omitted. As shown in FIG. 24, since document image attribute data has been generated, a histogram relating to the width of each circumscribed rectangle is created from this information. When Japanese and English are mixed, two peaks appear in the histogram, and a threshold value (for example, the median value of the two peaks) is obtained from the two peaks. It is also possible to tentatively determine less than English characters and correct the character code of the character recognition result using the tentative determination result to distinguish between English characters and Japanese characters. For example, if the character recognition result is “A” and the circumscribed rectangle width of the corresponding character image is provisionally determined to be an English character, the character recognition result is corrected to “A”.

  FIG. 3 is a block diagram illustrating a detailed configuration of the normalization unit 104. As illustrated, the normalization unit 104 includes a font storage unit 104e (in the embodiment, scalable outline vector font data is stored in the HDD 2304), an image generation unit 104f, an inter-character space acquisition calculation unit 104g, a pitch It is comprised by the determination part 104i.

  The image generation unit 104f reads data for one line from the document image attribute data after the character recognition process. That is, the character code information 104a, font information 104b, and character image (circumscribed rectangle) position and size information of a certain line are input by the number of circumscribed rectangles included in the line. The inter-character space calculation unit 104g distinguishes Japanese characters from English characters based on the character codes stored in the “character code” field, and cumulatively adds the respective blank lengths. In other words, the space length between the character determined to be a Japanese character and the character following it is cumulatively added, and the space book between the character determined to be an English character and the character following it is accumulated. to add.

  The pitch determination unit 104i divides the Japanese character blank book calculated by the inter-character space calculation unit 104g by “the number of Japanese characters−1” included in the line, so that the Japanese average character pitch, that is, The average space length between adjacent characters is calculated. The English average character pitch is similarly calculated.

  Then, based on each calculated average character pitch, the “position” field in all “circumscribed rectangle data” included in the corresponding row in the document image attribute data is updated and normalized. As a result, Japanese characters and English characters are normalized at the character pitch corresponding to each in the target line of the original image, and even if there is a place where the character spacing is extremely narrow, to some extent A blank length of length is assigned.

  In accordance with the updated data, the normalization unit 104 reads out font data corresponding to the character code from the font storage unit 104e, generates a character pattern that matches the circumscribed rectangle size, and generates the generated character according to the normalized position. The pattern is developed in an area secured in the main memory 2303. It should be noted that document images are generated at different character pitches for Japanese characters and English characters (English words). This process is repeated for all characters included in the line of interest. When the character pattern expansion process for one line is completed, the character pattern for the next line is expanded and repeated until the end-of-line character of the last line.

  When the normalized document image for one page is generated in this way, the result and the document image attribute data of FIG. 24 are passed to the watermark information embedding unit 108. The watermark information embedding unit 108 refers to final document image attribute data (see FIG. 24), and prepares two types of prepared in advance according to Japanese characters (2-byte characters) and English characters (1-byte characters). One of the embedding methods is selected, and watermark information is embedded.

  Although details will be described later, for Japanese characters, the height of the circumscribing rectangle is originally uniform, and the embedding that maintains the attributes in view of the fact that adequate space between characters is secured. In order to perform processing, information is embedded by rotating character images. On the other hand, in view of the fact that English characters have a narrow character spacing and are likely to come into contact with adjacent characters when embedded by rotation processing, and the height direction of characters is not uniform, Information is embedded by movement.

  As described above, the operation processing procedure of the digital watermark embedding apparatus according to the embodiment will be described with reference to the flowchart of FIG. This figure is a processing procedure (program loaded to the main storage device 2303) executed by the MPU 2302 shown in FIG.

  First, the document original 100 to be embedded with watermark information is read by the image input unit 101 typified by the scanner 2317 in FIG. 23 and is input to the document analysis unit 102 as document image data (step S601). In this case, the document image data is obtained by inputting a printed matter from the scanner 2317 or the like and converting it into a bitmap. Electronic data created by using a document editing application program, or a hard disk 2304 or a CD drive 2309, various electronic data including application program specific format and text format stored in each storage medium connected to DVD drive 2310, FDD drive 2311, etc. are converted into bitmaps by image processing software etc. Of course, it does not matter.

  Next, the document analysis unit 102 extracts circumscribed rectangles from the input document image data, and generates first-stage document image attribute data (step S602).

  Next, character recognition is performed by the recognition processing unit 103, and the document image attribute data is updated (step S603). At this time, the distinction between Japanese characters and English characters is reflected in the character code.

  FIG. 7 is a flowchart showing the operation procedure of the character recognition process. First, based on the document image attribute data obtained by the document analysis unit 102, a histogram of the width of the circumscribed rectangle of all character images is created, and each character image is either a Japanese full-width character or an English half-width character. A provisional determination is made as to whether or not the character image is present, and the character extraction unit 103a extracts a character feature in accordance with the position and size of the circumscribed rectangle of each character image (step S603a). Here, character feature extraction refers to an operation of extracting a predetermined feature amount included in a character in order to specifically determine the extracted character. As the feature amount in the present embodiment, for example, the circumscribed rectangular region of each character is further divided into small regions, and a histogram of direction components in the small region is taken as a character feature amount, or the distribution of pixel values is uneven. Can be used as a feature amount. In addition, the center of the circumscribed rectangle is used as the position information of the character.

  Then, in the identification unit 103b, the extracted feature value is compared with the feature value of the character or font held in the recognition dictionary 105, and the determination result of whether it is a Japanese character or an English character is obtained. In consideration, character identification (character code generation) and font identification are performed (step S603b). As described above, the document image attribute data is updated with the obtained character code and font type.

  Returning to the description of FIG. Based on the information (position, size, character code, font type) regarding each circumscribed rectangle obtained as described above, the normalization unit 104 generates a normalized document image 106 (step S604).

  The processing contents of the normalization unit 104 in the embodiment will be described with reference to the flowchart of FIG.

  First, attribute data for one line is input from the document image attribute data (step S604a). Then, the character space calculation unit 104g calculates the space length of the Japanese characters in the target line (the sum of the spaces between characters) (step S604b). Thereafter, the pitch determination unit 104i calculates an average pitch (average space length) between characters of each Japanese character by the number of Japanese characters included in the target line (step S604c). In steps S604d and S604e, the same processing is performed for English characters, and the character pitch of the English characters is calculated. Based on the calculated character pitch, the position of the circumscribed rectangle of the target line of the input document image attribute data is updated (step S604f). At this time, the coordinate positions of the circumscribed rectangles in the same line in the direction orthogonal to the character arrangement direction (vertical direction in the case of horizontal writing) are made the same. Then, the character image image of the target line is generated with reference to the font data and the circumscribed rectangle size (step S604g). Then, it is determined whether or not processing has been performed for all rows (step S604h). If it is determined that processing has not been completed for all rows, the processing from step S604a is repeated.

  Returning to the description of FIG. When the normalization process is performed as described above, the generation of the normalized document image and the document image attribute data at that time are obtained, the process proceeds to step S605, and the watermark information (for example, in the PC or the like) is obtained. When applied, it may be a character string (user name, etc.) input from a keyboard, or information stored in advance, but when applied to a copying machine, information specifying the copying machine may be used) Embed. Then, the embedded document image is output (in the embodiment, it is a print output, but it may be stored in a storage medium or transmitted via a network), and the embedding process for one original is finished.

  Here, the embedding process in step S605 in the embodiment will be described in more detail with reference to the flowchart of FIG. Note that watermark information to be embedded is prepared in advance in binary (binary).

  First, one circumscribed rectangle in which information is embedded is determined with reference to the document image attribute data (step S605a).

  Next, it is determined whether the “character code” field of the selected circumscribed rectangle data is a Japanese character (2-byte character code = full-width character) or an English character (1-byte character code = half-width character) (step S605b). If it is determined that the character is a Japanese character, an embedding process by rotation is performed (step S605c). If it is determined that the character is an English character, an embedding process by vertical movement is performed (step S605d).

  When any of the embeddings is performed, it is determined whether or not embedding of the last bit of the embedding information is completed. If not, the processing after step S605a is repeated.

  Here, the process of step S605c will be described with reference to the flowchart of FIG.

  First, 1 bit of information to be embedded is selected (step S401). Then, it is determined whether the bit is “1” or “0” (step S402).

  If it is determined that it is “1”, the circumscribed rectangle of the character image of interest is rotated clockwise by θ around the center position (step S403). On the other hand, when it is determined that the bit to be embedded is “0”, the circumscribed rectangle of the character image of interest is rotated counterclockwise by θ around the center position.

  Next, the process of step S605d, that is, the embedding process for English characters will be described with reference to the flowchart of FIG.

  First, 1 bit of information to be embedded is selected (step S501). Then, it is determined whether the bit is “1” or “0” (step S502).

  If it is determined that it is “1”, the circumscribed rectangle of the character image of interest is moved upward by a predetermined distance, and the result is reflected in the document image attribute data (step S403). On the other hand, if it is determined that the bit to be embedded is “0”, the circumscribed rectangle of the target character image is moved downward by a predetermined distance, and the result is reflected in the document image attribute data.

  In both the Japanese and English embeddings, the character at the beginning of the line is excluded from the embedding target in the embodiment. The reason for this is that, in the case of English characters, it moves up or down as described above, so a reference position is required when moving up and down, and the character at the beginning of the line is used as the reference. is there. This is also because Japanese characters are not rotated. Accordingly, it is sufficient that the reference character is determined, so the end-of-line character may be used, or a predetermined character may be used.

  In some cases, as shown in FIG. 13, when embedding information by vertical movement, the odd-numbered character image in the line is not moved in the vertical direction, and the position of the even-numbered character is moved to the immediately preceding position. You may make it move up and down on the basis of the position of a character (odd-numbered character). That is, when the data to be embedded in the character (2) in FIG. 13 is “0”, the character (2) is moved H1 below the base position of the character (1). Further, when “1” is embedded in the character (4), the character (4) is moved by H2 above the base position of the character (3). In this case, although the number of characters that can be embedded is about half of the number of characters in one line, it is sufficient to use the immediately preceding odd-numbered character as a reference, so that each time a reference position can be determined. Highly accurate embedding can be expected.

  Next, processing in the digital watermark information extraction apparatus according to the embodiment will be described with reference to the flowchart of FIG. The basic configuration is the same as in FIG.

  First, an image to be verified is input via the scanner 2317 or the like (step S1001). However, the image input device is not limited to a scanner, and if the image data is stored in a storage medium, it may be input via a drive that drives the storage medium, or downloaded from a network. It doesn't matter.

  Next, document analysis processing of the input image data is performed (step S1002). This analysis process is the same as that of the digital watermark embedding apparatus described above, and detects the position and size of the circumscribed rectangle of the character image, and generates document image attribute data as shown in FIG.

  In step S1003, character recognition is performed. This character recognition process is also the same as in the first embodiment.

  As a result, information corresponding to the character code and font type in the document image attribute data shown in FIG. 24 is stored.

  Next, based on the circumscribed rectangle, size, character code, and font information of the document image attribute data, embedded information extraction processing is performed according to the character type (step S1004), and the last extracted information is displayed on the monitor 2306. Will be output.

  Although the processing up to the character recognition processing in step S1003 is the same as the digital watermark embedding processing described above, the description thereof will be omitted, and the watermark information extraction processing in step S1004 will be described below.

  FIG. 12 is a flowchart showing the processing content of step S1004.

  First, one of circumscribed rectangles to be extracted is selected with reference to the document image attribute data (step S1004a). Note that the circumscribed rectangle is basically selected from the circumscribed rectangle of the character image at the beginning of the line toward the end of the line, and when one line has been processed, the process moves to the second line at the end of the last line. Do until. However, the circumscribed rectangle of the first letter of each line is excluded from the selection. The reason is that the character at the beginning of the line is not embedded in the embodiment.

  When one circumscribed rectangle is selected, it is determined whether or not it is a Japanese character by examining the “character code” of the circumscribed rectangle. If it is determined that the character is a Japanese character, the process proceeds to step S1004 to extract information by rotation. If it is determined that the character is not a Japanese character, that is, an English character, the process advances to step S1004 to perform an embedded information extraction process based on the vertical movement amount.

  The above process is repeated until it is determined that the process has been performed up to the last line and the last character of the document image.

  Here, the process of step S1004c will be described with reference to the flowchart of FIG.

  First, the position of the center of gravity of the circumscribed rectangle of interest is obtained (step S1004c-1). Next, a standard character pattern is created based on the character code, size, and font type that are the character recognition results of the character image in the circumscribed rectangle, and the standard character pattern and the character image indicated by the circumscribed rectangle are angled. By determining which of the two is rotated by ± θ, the inclination angle of the character image in the circumscribed rectangle in the document image is determined (step S1004c-2).

  If it is determined that the character image of interest is rotated clockwise by θ, the bit embedded in the character of interest is determined to be “1” (step S1004c-3). If it is determined that the rotation is counterclockwise by θ, the embedded data is determined to be “0” (step S1004c-4).

  Next, the process of step S1004d will be described with reference to the flowchart of FIG.

  First, the coordinate position of the circumscribed rectangle of interest (the coordinate in the direction orthogonal to the character arrangement direction) and the coordinate position of the character image at the beginning of the line (reference circumscribed rectangle) are compared. When it is determined that a predetermined amount has moved upward from the reference circumscribed rectangle, the bit embedded in the target character image is “1”, and conversely, it has been determined that the bit has moved downward by a predetermined amount The embedded bit is determined to be “0”.

  As described above, according to this embodiment, when embedding information by changing the position of a character image in a document image in which Japanese and English are mixed, the size of Japanese characters is almost uniform. By embedding information by rotation of the character image based on the characteristic that the space is secured to some extent, it is possible to make it unlikely to cause a sense of incongruity. In the case of English, the space between characters is narrow, but based on the characteristic that the size of individual characters is uneven, embedding information by moving up and down makes it difficult for English characters to feel strange. Is possible. Accordingly, it is possible to embed information about almost all document images in which Japanese and English are mixed (in the embodiment, the characters at the beginning of each line are not to be embedded), and information embedding is also possible when viewed as the entire document image. It is possible to generate a natural document image that is difficult to visually confirm.

<Description of modification>
In the above embodiment, normalization is performed unconditionally when embedding information, and then embedding is performed. However, information may already be embedded in a document image to be embedded with a digital watermark.

  Therefore, first, when embedding a digital watermark, it is desirable to perform a process of extracting embedded information. If the embedded information can be extracted, the information is displayed.

  In particular, if the digital watermark information embedding and extracting apparatus according to the embodiment is applied to, for example, a copying machine, the apparatus configuration may be the configuration shown in FIG. 25 instead of FIG. The difference from FIG. 1 is that a watermark information extraction unit 112 is provided instead of the watermark information input unit 107. That is, a document image to be copied is read, information embedded in the watermark information extraction unit 112 is extracted, and the extracted information is used again as embedded information. Normally, when copying is repeated with a copying machine, the document gradually gets dirty. With the above configuration, the watermark information remains embedded, and the result of copying is character recognition, character Since the pattern is generated and printed, it does not get dirty.

  In the above embodiment, the rotation angle θ of the character image is not particularly limited. However, a value that does not give a sense of incongruity and has a certain degree of embedding extraction accuracy is desirable, so the upper limit will be about 10 degrees.

<Second Embodiment>
In the above embodiment, normalization processing is performed when information is embedded in a character image in a document image. However, an example in which normalization is not required will be described as a second embodiment. That is, it is a method of directly embedding in a digital watermark embedded image. This method imposes certain restrictions on the embedding method, so that processing such as character recognition in the first embodiment is unnecessary, and the load on the device and especially the MPU can be reduced, and the processing speed can be improved. Become.

  FIG. 14 is a conceptual diagram of the configuration of the apparatus according to the second embodiment.

  The watermark information extraction / embedding apparatus according to the second embodiment performs a document analysis in the verification image 200 input from the input unit 201 to recognize the character size and position information, and each character. A pitch type recognition unit 203 that determines the pitch type of the image, a watermark information extraction unit 206 that extracts the watermark information 207, a watermark information embedding unit 208 that embeds the watermark information 207 using the input verification image 200, an electronic The image output unit 209 generates the watermark embedded image 210.

  Hereinafter, differences from the first embodiment will be described. In the second embodiment, the recognition processing unit and the normalization unit are not necessary as described above. Instead, it is different in that the pitch type recognition unit 203 that functions as a part of the normalization unit in the first embodiment is required. Also, in the embodiment, an example of embedding again through the digital watermark extraction process will be described, and it is assumed that Japanese characters are embedded by adjusting the space between characters, and English characters are embedded by moving up and down.

  FIG. 16 is a flowchart showing a processing procedure in the second embodiment.

  When the verification image is input via the input unit 201 (step S1601), the document analysis unit 202 acquires circumscribed rectangle information and creates document image attribute data (step S1602). The steps so far are the same as those in the first embodiment. However, since the character recognition processing is not required in the second embodiment, the two fields of “character code” and “font information” are not required in the document image attribute data shown in FIG. A field for identifying a word character or an English character is provided (hereinafter, this field is referred to as a character type field).

  Next, the pitch type determination unit 203 determines whether the character is a Japanese character or an English character based on the inter-character space length of the document image attribute data (step S1603).

  The details of the pitch type determination unit 203 are shown in FIG. 15 and the processing procedure is shown in FIG.

  First, information about the size and position of each character of data for one page is acquired from the document image attribute information 203a (step S1603a). Next, create a histogram of the inter-character space between each character size (width) of the input character image attribute data and the following characters, to distinguish Japanese full-width characters from English half-width characters, A threshold value of the character size and the space between characters is calculated (step S1603b).

  Next, based on the calculated character size and the inter-character space threshold, it is determined what pitch each character image has, that is, whether it is a Japanese character (step S1603c). Then, as a result of this determination, the data is stored in the “character type” field in each circumscribed rectangle data in the document image attribute data.

  Returning to the flowchart of FIG. When the character type determination process in step S1603 is performed, watermark information is extracted (step S1604). Details of this processing will be described with reference to FIG.

  First, referring to the document image attribute data, one of the circumscribed rectangles is selected (step S1604a). The selection order of circumscribed rectangles is the same as that in the first embodiment, and the characters at the beginning of each line are also excluded.

  In step S1604b, it is determined with reference to the character type field whether the circumscribed rectangular character is a character having a Japanese character pitch. If it is determined that the character pitch is Japanese, the information is extracted based on the character spacing adjustment in step S1604c. If it is determined that the pitch is not a Japanese character pitch, that is, an English character pitch, information extraction processing based on the vertical position determination is performed (step S1604d).

  In step S1604e, this process is repeated until it is determined that the process at the end position of the last line has been completed.

  The information embedding and extracting method using the inter-character space has been described in the prior art. Further, since the information extraction based on the vertical movement amount of the characters has already been described with reference to FIG. 10 of the first embodiment, detailed description thereof is omitted here.

  Returning to the flowchart of FIG. When the information extraction process is completed as described above, the process advances to step S1605 to perform a watermark information update process. In this watermark information update process, the information extracted in the previous step S1604 is left as it is and is newly added to it. The information to be added is intended to facilitate copying history tracking such as identification information unique to the copying machine. As described above, even in a document image having a lower redundancy than that of a multi-valued image, watermark information can be easily added / updated without destroying the original watermark information.

  In step S1606, the added watermark information is reflected in the input document image. The added watermark information is performed from the next line for the next character after the last character for which the presence of the watermark information is confirmed. The processing is as shown in the flowchart of FIG.

  First, in step S1606a, a circumscribed rectangle is selected, and the character type field of the circumscribed rectangle is inspected to determine whether it is a Japanese character pitch. If it is determined that the character image has a Japanese character pitch, information is embedded by adjusting the character spacing in step S1606c. If it is determined that the pitch is not a Japanese character pitch, that is, an English character pitch, information is embedded based on the vertical movement in step S1606d.

  In step S1606e, the above process is repeated until it is determined that all information has been embedded.

  The second embodiment has been described above. According to the second embodiment, although it is not possible to accurately discriminate whether it is a Japanese character or an English character as much as character recognition processing is not performed, an approximate judgment result is obtained based on the Japanese character pitch and the English character pitch. Since the information is embedded on the basis of this, the processing speed is higher than that of the first embodiment, which is particularly effective for a device requiring a copying speed such as a copying machine.

<Other embodiments>
In each of the above-described embodiments, an example of adapting to the inter-character space length, vertical position, and rotation has been described. However, the present invention is not limited to the above-described embodiment. In short, it suffices if a character such as Japanese and a character such as English about half the width are mixed and embedding suitable for each character is performed.

  Furthermore, since it is obvious that most of the functions in the embodiments can be realized by a computer program executed on a personal computer, the present invention naturally includes such a computer program. Further, since the computer program is usually stored in a computer-readable storage medium such as a CDROM and can be executed by being set in a computer and copied or installed in the system, such a computer-readable storage medium is also this book. It is included in the category of the invention.

1 is a conceptual diagram of a configuration of a digital watermark embedding device according to a first embodiment. It is a block block diagram of the recognition process part in FIG. It is a block block diagram of the normalization part 104 in FIG. It is a flowchart which shows the embedding procedure of the watermark information by a character rotation process. It is a flowchart which shows the embedding procedure of the watermark information by the vertical position movement of a character. It is a flowchart which shows the process sequence of the digital watermark embedding apparatus in 1st Embodiment. It is a flowchart which shows the processing content of the recognition process part in FIG. It is a flowchart which shows the process sequence of the normalization part in FIG. It is a flowchart which shows the digital watermark embedding process procedure in FIG. It is a flowchart which shows the information extraction process from a character rotation angle. It is a flowchart which shows the information extraction process from a character up-and-down position. It is a flowchart which shows an information extraction process procedure. It is a flowchart which shows the other example of the information embedding and extraction by a character vertical position. FIG. 6 is a conceptual diagram of a configuration of a digital watermark embedding device according to a second embodiment. It is a figure which shows the structure of the pitch classification recognition part of FIG. It is a flowchart which shows the watermark information embedding processing procedure in the apparatus of 2nd Embodiment. It is a flowchart which shows the process sequence of the pitch classification recognition part of FIG. It is a flowchart which shows the detail of the watermark information extraction process of FIG. It is a flowchart which shows the detail of the watermark information embedding process of FIG. It is a figure which shows the example of the original image in the case of performing digital watermark embedding using the space | interval of a character. It is a figure which shows the image after performing digital watermark embedding using the space | interval of a character. It is a schematic diagram for demonstrating the electronic watermark embedding method using the inclination of a character. It is a concrete block block diagram of the apparatus in embodiment. It is a figure which shows the structure of the document image attribute data produced | generated in 1st Embodiment. It is another structure conceptual diagram of the digital watermark embedding device of the first embodiment.

Claims (14)

An information processing apparatus for embedding information in a document image,
Extraction means for extracting information about the position and size of the circumscribed rectangle of each character image in the input document image;
Based on the position and size of each extracted character image, each character image is divided into a first character image type having a first character width and a second character width having a second character width different from the first character width. Classification means for classifying the character image type of
Embedding means for embedding a division by a first embedding method for a character image of the first character image type, and embedding information by a second embedding method for a character image of the second character image type. A characteristic information processing apparatus. The first embedding method affects the blank length between adjacent character images, embeds information,
The information processing apparatus according to claim 1, wherein the second embedding method embeds information by affecting a position in a direction orthogonal to a character arrangement direction.   The information processing apparatus according to claim 1, wherein the second character width is narrower than the first character width.   4. The apparatus according to claim 1, further comprising extraction means for extracting already-embedded information based on an extraction result of the extraction means, and setting the extracted information as information to be embedded by the embedding means. The information processing apparatus according to any one of claims. Furthermore, according to the first and second character image types classified by the classification means, normalizing means for the interval of the circumscribed rectangle of the input character image;
Character recognition means for recognizing each character image and outputting a character code;
Image generating means for generating a document image according to the recognized character code, the corresponding circumscribed rectangle size, and the normalized position;
5. The information processing apparatus according to claim 1, wherein the information embedding by the embedding unit is performed on the document image data generated by the image generating unit.   6. The information processing apparatus according to claim 1, wherein the first character image type is a full-width character image type, and the second character image type is an alphabet character. The first embedding method embeds 1 bit by rotating the character image at the center of the circumscribed rectangle in a clockwise direction and a counterclockwise direction by a predetermined angle,
The said 2nd embedding method embeds 1 bit by the moving amount | distance to the direction orthogonal to the line direction which is a sequence of the character of a character image, The one of Claim 2 thru | or 6 characterized by the above-mentioned. Information processing device. The first embedding method embeds 1-bit information by moving the target character image to either the immediately preceding character image or the immediately following character image in order to set a difference between the blank lengths on both sides of the character image.
The said 2nd embedding method embeds 1 bit by the moving amount | distance to the direction orthogonal to the line direction which is a sequence of the character of a character image, The one of Claim 2 thru | or 6 characterized by the above-mentioned. Information processing device. An information processing apparatus for extracting information embedded in a document image,
Document image attribute extraction means for extracting attribute information related to the position and size of the circumscribed rectangle of each character image in the input document image;
Based on the position and size of each extracted character image, each character image is divided into a first character image type having a first character width and a second character width having a second character width different from the first character width. Classification means for classifying the character image type of
Extracting means for extracting information about the first character image type character image by the first extracting method, and extracting information about the second character image type character image using the second extracting method; An information processing apparatus comprising: The first extraction method extracts information embedded by an embedding process that affects a blank length between adjacent character images;
The information processing apparatus according to claim 9, wherein the first extraction method extracts information embedded by an embedding process that affects a position in a direction orthogonal to a character arrangement direction. An information processing method for embedding information in a document image,
An extraction step of extracting information about the position and size of the circumscribed rectangle of each character image in the input document image;
Based on the position and size of each extracted character image, each character image is divided into a first character image type having a first character width and a second character width having a second character width different from the first character width. A classification process for classifying the character image types into
An embedding step of embedding information by the first embedding method for the first character image type character image and embedding information by the second embedding method for the character image of the second character image type. A characteristic information processing method. An information processing method for extracting information embedded in a document image,
A document image attribute extraction step for extracting attribute information related to the position and size of the circumscribed rectangle of each character image in the input document image;
Based on the position and size of each extracted character image, each character image is divided into a first character image type having a first character width and a second character width having a second character width different from the first character width. A classification process for classifying the character image types into
An extraction step of extracting information on the first character image type character image by a first extraction method and extracting the second character image type character image by a second extraction method. A characteristic information processing method.   The program for making a computer implement | achieve the function of the information processing apparatus of any one of Claims 1 thru | or 10.   A computer-readable storage medium storing the computer program according to claim 13.

Images