JP4294456B2 - Specific character string search device, specific character string search method, specific character string search program, and recording medium - Google Patents

Description

The present invention is a specific character string search capable of performing a high-speed search for a specific character string from a bitmap document image even when a low-performance CPU is mounted or when only a small memory resource is mounted. The present invention relates to a device, a specific character string search method, a specific character string search program, and a recording medium .

  In recent years, a document image of a bitmap image is applied to a character recognition device, and a character string to be searched is extracted from character strings obtained as a result of recognition.

  For example, in a copying machine, in order to internally process an image, complicated image processing is possible. Since dedicated hardware (DSP or the like) is used for this processing, the CPU does not require so high performance, and a low-priced CPU is used. Also, a large capacity is not required for the storage device mounted on the copying machine.

  By the way, when an OCR is installed in a copying machine, a facsimile machine, a portable phone, a PDA or the like, if there are many character types to be recognized, such as Japanese OCR, a long time is required for the character recognition processing. That is, in Japanese OCR, character recognition processing for converting a character image into a character code takes longer time than character and line segmentation processing. For this reason, real-time processing of OCR is not possible unless a high-performance CPU is used in a copying machine or the like. Further, the capacity of the pattern dictionary is enormous, and a character recognition function cannot be installed in a copying machine or the like with a small memory resource.

  For this reason, in a copying machine, when the “confidential” character is included in the document, the copying function is stopped, and therefore, a high-performance CPU and a large-capacity storage device must be installed. Will soar.

The present invention has been made in view of such circumstances, and even when a low-performance CPU or a small memory resource is installed, a specific character string is searched from a bitmap document image. It is an object of the present invention to provide a specific character string search device, a specific character string search method, a specific character string search program, and a recording medium.

In order to achieve the above object, the present invention provides a specific character string search device that performs character recognition on a search target image and extracts a specific character string in the image, from a pattern dictionary in which a large number of characters are registered. A partial dictionary is generated by extracting only one or more specific characters included in the specific character string and a plurality of small difference character dictionary records having a small difference from each of the one or more specific characters. Character recognition means for performing character recognition processing between the specific character and the small difference character for the character in the search target image using the dictionary generation means, the partial dictionary generated by the dictionary generation means, and the character recognition For the plurality of character strings obtained by the means, the key with the specific character is used. Performing Rakuta recognition processing, in which a character string search means for extracting a character string having the specific character string and identity.
The character recognizing means may determine whether the character in the search target image is the same / non-identical with each character constituting the specific character string and the small difference character.
Further, in a specific character string search method for performing character recognition on a search target image and extracting a specific character string in the image, one of the character strings included in the specific character string from a pattern dictionary in which a large number of characters are registered. A dictionary generation step of extracting only dictionary records of the above-mentioned specific character and a plurality of small difference characters having a small difference from each of the one or more specific characters to generate a partial dictionary; and A character recognition step for performing character recognition processing between the specific character and the small difference character for the character in the search target image using the generated partial dictionary, and a plurality of character strings obtained by the character recognition step , Character recognition process with the specific character What is obtained by a character string search step of extracting a character string having the specific character string and identity.
Moreover, the said character recognition step is good to determine the same / non-identity with respect to the character in the said search object image with each character and small difference character which comprise the said specific character row | line | column.
Also provided is a specific character string search program that causes a computer to execute each step of the character string search method.
Also provided is a computer-readable recording medium storing the above-described specific character string search program.

With the above-described configuration of the present invention, it is possible to obtain a specific character string from a bitmap document image at high speed.

  FIG. 1 is a functional block diagram showing a first embodiment of a specific character / graphics detection apparatus of the present invention. In FIG. 1, the specific character / graphic detection device 1 includes a control unit 11, a storage unit 12, an image input unit 13, and an operation / display unit 14.

  The control means 11 controls the specific character / graphics detection apparatus 1 as a whole. The storage unit 12 includes a RAM, a ROM, a magnetic storage device, and the like.

  The storage means 12 stores a program 121 (character / graphic recognition program P1, dictionary generation program P2, character string / graphic search program P3, search result output program P4), a partial dictionary 122, and a pattern dictionary 123. A storage area 124 is provided.

  The programs P1, P2, P3 and P4 are executed by the control means 11, respectively. The program P1 and the control means 11 constitute a character / graphic recognition means M1, and the program P2 and the control means 11 constitute a dictionary generation means M2. The program P3 and the control means 11 constitute a character string / graphics search means M3, and the search result output program P4 and the control means 11 constitute a search result output means M4.

  The image input unit 13 may be an image input device such as an image scanner or a digital camera, may be a communication unit (not shown) that can acquire an image by communicating with an external device, or an image from an application. Image acquisition means (program) (not shown) that can be acquired may be used.

The operation / display unit 14 is a user interface, and typically includes an operation unit such as a keyboard and a mouse, and a display unit such as a display.
[First operation example]
Hereinafter, the first operation example of the specific character / graphics detection apparatus 1 of FIG. 1 will be described with reference to the flowchart of FIG. In this operation example, first, the user uses the operation / display unit 14 to specify a search target image TG (in this operation example, the image shown in FIG. 3) and a search character string / graphic LS (here, (Designation of character string “confidential”) is performed (S101).

  When designating the search target image TG, an image file input in advance from the image input unit 13 and stored in the storage unit 12 can be used as the search target image TG, and a certain page in such an image file, Alternatively, the image area in this page can be designated, and the search character string / graphic LS (character string “confidential” designation) can be performed by inputting a character string from a keyboard or the like, or in a certain file using a mouse or the like. This can be done by selecting a character string / graphic.

  The dictionary generation means M2 extracts from the pattern dictionary 123 the dictionary records (the recorded data is a feature amount) of the characters “machine” and “dense” that constitute the designated character string “confidential”. It is stored as the partial dictionary 122 (S102). That is, in this operation example, the partial dictionary 122 is generated as a dictionary including only “machine” and “dense” words as dictionary records.

  On the other hand, the character / graphic recognition unit M1 takes the search target image TG into a predetermined work area (usually secured as a work area in the storage unit 12), and uses the partial dictionary 122 generated by the dictionary generation unit M2. Character recognition is performed (S103).

  As shown in FIG. 4, the recognition result of each character included in the search target image TG by the character / graphic recognition means M1 is the character “machine”, “dense”, unrecognizable character ( Reject character that is not determined as any character). In FIG. 4, the recognized “machine” and “dense” are indicated by white characters, and the reject character is indicated by “□”.

  The character string / figure search means M3 receives the recognition result by the character / figure recognition means M1, and searches for the “confidential” character string from the recognition result (S104). FIG. 5 shows a “confidential” portion (first line 11th, 12th, second line 2, 2nd, 3rd, etc.) in the recognition result (character string) by the character / figure recognition means M1 shown in FIG. (11th, 12th, 5th row, 3rd, 4th, 6th row, 5th, 6th) are indicated by black characters.

  The search result output means M4 receives the coordinate information of the character string portion on the search target image TG searched as “confidential” by the character string / graphic search means M3, and emphasizes this portion as shown in FIG. Then, it is output to the display unit (display) of the operation / display means 14 (S105) (the search target image TG after the enhancement is indicated by the search result image RG). In FIG. 6, “Report” in the 11th and 12th characters in the first line, “Report” in the 2nd and 3rd characters in the 2nd line, “Secret” in the 11th and 12th characters in the 5th line 3, “Current” of the fourth character, and “Trend” portion of the sixth and fifth character are highlighted. This emphasis method is performed by changing the color of the corresponding character string (for example, changing to red if the color on the search target image TG is black), shading the corresponding character string, and underlining the corresponding character string. This can be done by adding lines.

The user simply looks at the emphasized characters in the search result image RG as shown in FIG. 6 displayed on the display unit (display) of the operation / display unit 14 and displays the search character string / figure in the search target image TG. It is possible to easily find out whether or not “confidential” which is LS is included, or when it is included.
[Second operation example]
Hereinafter, the second operation example of the specific character / figure detection apparatus 1 in FIG. 1 will be described with reference to the flowchart in FIG. 7 by taking as an example the case of searching for the character string “confidential” as in the first operation example. In this operation example, the processing in S201 and S202 is the same as the processing in S101 and S102 in the first operation example (see FIG. 2), and the user uses the operation / display unit 14 to search for an image TG (see FIG. 3). And the search character string / graphic LS (character string “confidential”) are designated (S201), and the dictionary generating means M2 determines the characters “machine” and “ The “dense” dictionary record is extracted from the pattern dictionary 123 and stored as the partial dictionary 122 (S202).

  On the other hand, the character / figure recognition means M1 takes the search target image TG into a predetermined work area (usually secured as a work area in the storage means 12), as in the process of S103 of the first operation example, and creates a dictionary. Character recognition is performed using the partial dictionary 122 generated by M2 (S203). At this time, in this operation example, the recognition result of each character included in the search target image TG by the character / graphic recognition unit M1 is the character “machine” included in the partial dictionary 122 as shown in FIG. , “Dense”, or an unrecognizable character (a reject character that is not determined as any character). However, in this operation example, the character / figure recognition unit M1 stores the difference DD for each character included in the search target image TG. The degree of difference DD can be obtained by an appropriate well-known method from, for example, a plurality of feature amounts of the dictionary record for each character.

  The character string / figure search means M3 receives the recognition result including the degree of difference DD by the character / figure recognition means M1, searches for a “confidential” character string from the recognition result (S204), and further each “confidential” character. An average value of the dissimilarity DD of the characters (“machine”, “fine”) constituting the column is calculated, and this is compared with a predetermined threshold value SH, and the “secret” value lower than the threshold value SH is calculated. The character string is extracted as what is actually displayed as “confidential” (S205). FIG. 8 shows a “confidential” portion (first line 11th, 12th, second line 2, 2nd, 3rd, etc.) in the recognition result (character string) by the character / graphic recognition means M1 shown in FIG. The eleventh, twelfth, fifth row, third, fourth, sixth row, fifth and sixth characters are indicated by black characters, and each character (“machine”, “secret” is indicated in the “confidential” portion. )) Is indicated by a numerical value.

In this operation example, the threshold value SH is set to “20”. In FIG. 8, the average value of the 11th and 12th “secret” in the first row in the recognition result (character string) by the character / graphic recognition means M1 shown in FIG. 3 is (89 + 76) / 2 = 82. 5. The average value of the second and second “secret” in the second row is (68 + 78) / 2 = 73, and the average value of the second and eleventh and twelfth “secret” is (13 + 8) / 2 = 10.5, the average value of the third and fourth “secret” in the fifth row is (48 + 97) /2=72.5, and the average value of the fifth and sixth “secret” in the sixth row is (53 + 64). ) /2=58.5 der is, takes a value smaller than "20" is a threshold SH "confidential" in the second row 11, only the 12-th "confidential". Therefore, the character / figure recognition means M1 can determine that only the “secret” in the second and eleventh and twelfth rows is actually displayed as “confidential” on the search target image TG.

  The search result output means M4 receives the coordinate information of the character string portion on the search target image TG extracted as “confidential” in S205 by the character string / figure search means M3, and receives the coordinate information. As shown, it is emphasized and output to the display unit (display) of the operation / display unit 14 (S105) (the search target image TG after the enhancement is indicated by a search result image RG). In FIG. 9, the “confidential” portion of the 11th and 12th characters in the second line is highlighted. In this operation example, as in the first operation example, this emphasis method changes the color of the corresponding character string (for example, changes to red if the color on the search target image TG is black), and the corresponding character. This can be done by shading a column or adding an underline to the corresponding character string.

In this operation example, since the “confidential” portion is highlighted on the display unit (display) of the operation / display unit 14 as shown in FIG. 9, “confidential” is included in the search target image TG. Whether or not it is included, the description can be further recognized.
[Third example of operation]
Hereinafter, the third operation example of the specific character / graphics detection apparatus 1 in FIG. 1 will be described with reference to the flowchart in FIG. In this operation example, the process of S301 is the same as the process of S101 in the first operation example (see FIG. 2), and the user designates the search target image TG (see FIG. 3) using the operation / display unit 14, and The search character string / graphic LS is designated (designation of the character string “confidential”) (S301).

  Next, the dictionary generating means M2 has a plurality of small differences DD with respect to the characters “machine”, “secret”, and “machine” constituting the designated character string “confidential” (in this operation example, five). ) And a plurality of (5 in this operation example) character dictionary records whose degree of difference DD is smaller than “dense” are extracted from the pattern dictionary 123, and six dictionary records for each character are stored in the partial dictionary. It is stored as 122 (S302). That is, when extracting the dictionary record for “machine” in S302, the degree of difference DD is extracted from the other five dictionary records having a small degree of difference DD from the dictionary record of “machine”. In general, the feature amount of a character constituting a dictionary record is represented by a value of an n-dimensional (n is an integer of 2 or more) feature amount space SS. In this operation example, this feature amount space is represented for convenience of explanation. The two dimensions shown in FIG. In FIG. 11A, the feature coordinates are indicated by “◯”, and a 6-character dictionary record of “char”, “磯”, “some”, “foundation”, “楚”, and “machine” is displayed. It is stored as a partial dictionary 122. Similarly, when extracting a dictionary record for “dense” in S302, a 6-character dictionary record of “firefly”, “honey”, “sakae”, “secret”, “house”, and “dense” is the partial dictionary 122. Saved as

  On the other hand, the character / figure recognition means M1 takes the search target image TG into a predetermined work area (usually secured as a work area in the storage means 12), as in the process of S103 of the first operation example, and creates a dictionary. Character recognition is performed using the partial dictionary 122 generated by M2 (S303). At this time, in this operation example, the recognition result for each character included in the search target image TG by the character / graphic recognition means M1 is “char”, “character” included in the partial dictionary 122 as shown in FIG. Twelve characters such as 磯, 幾, “foundation”, “，” and “machine”, “firefly”, “honey”, “sakae”, “secret”, “house” and “dense”, and unrecognizable characters (Reject character that is not determined as any character).

  The character string / figure search means M3 receives the recognition result by the character / figure recognition means M1, and searches the character string of “confidentiality” (second character 10th and 11th characters: indicated by symbol P in FIG. 12). (S304).

  The search result output means M4 emphasizes the character string portion on the search target image TG detected in S304 by the character string / graphic search means M3, for example, as shown in FIG. (S305) (the search target image TG after enhancement is indicated by a search result image RG). In FIG. 9, the “confidential” portion of the 11th and 12th characters in the second line is highlighted. In this operation example, as in the first and second operation examples, this emphasis method changes the color of the corresponding character string (for example, changes to red if the color on the search target image TG is black). This can be done by shading the corresponding character string, adding an underline to the corresponding character string, or the like.

  In this operation example, since the “confidential” portion is highlighted on the display unit (display) of the operation / display unit 14 as shown in FIG. 9, “confidential” is included in the search target image TG. Whether or not it is included, the description can be further recognized. In particular, in this operation example, since a plurality of (here, five) character dictionary records that are highly likely to be recognized erroneously are extracted, the recognition accuracy is increased.

In the above operation example, when the dictionary generation unit M2 in S302 creates the partial dictionary 122, as shown in FIG. 11B, the search character string / figure LS is configured in the two-dimensional feature amount space SS. Characters within a predetermined value (“30” in FIG. 11B) centering on each character (“machine” in FIG. 11B) can be extracted as a character dictionary record. In FIG. 11B, the dictionary generation means M2 extracts three characters “磯”, “楚”, and “machine” as a character dictionary record.
<< Second Embodiment >>
FIG. 13 is a functional block diagram showing a second embodiment of the specific character / graphics detection apparatus of the present invention. In FIG. 13, a copying machine 2 (corresponding to the specific character / graphic detection device in FIG. 1) includes a central processing unit (CPU) 21, a memory 221, a magnetic storage device 222, an image scanner 23, an operation panel 24, and the like. The printer 25 is provided.

  The central processing unit 21 corresponds to the control means 11 of FIG. 1, the memory 221 and the magnetic storage device 222 correspond to the storage means 12 of FIG. 1, the image scanner 23 corresponds to the image input means 13 of FIG. Reference numeral 24 includes various keys and a display and corresponds to the operation / display means 24 of FIG. Although not shown, the printer 25 includes a paper feeding mechanism for copying paper, a paper discharging mechanism for printed paper, and the like.

  The memory 221 stores a program 2211 (a character / figure recognition program P1, a dictionary generation program P2, a character string / figure search program P3, a search result output program P4), and a partial dictionary 2212. A pattern dictionary 2221 is stored in the magnetic storage device 222. The programs P1, P2, P3, and P4 are respectively executed by the central processing unit 21, and the program P1 and the central processing unit 21 constitute character / graphic recognition means M1, and the program P2 and the central processing unit 21 are dictionary generation means M2. The program P3 and the central processing unit 21 constitute a character string / figure search means M3, and the search result output program P4 and the central processing unit 21 constitute a search result output means M4. The memory 221 has a storage area for storing software such as a printer driver for controlling the copier, a storage area for temporarily storing the read original, and a setting for storing the copier settings. A storage area or the like is secured.

  The copying machine 2 according to the present embodiment is configured not to print (do not copy) when a specific character string “confidential” is described in a copy document.

  In general, in OCR for Japanese and Chinese, the amount of character recognition processing for converting character images into character codes is enormous compared to character and line segmentation, and there is also a pattern dictionary used for character recognition. Large capacity. Therefore, when performing OCR processing with scarce resources, it is necessary to reduce the number of characters to be recognized, reduce the amount of calculation of character recognition processing, and reduce the capacity of the pattern dictionary.

  Since hard disks have recently become inexpensive, the capacity of the magnetic storage device 222 mounted in the copying machine 2 of the present invention can be configured to be sufficiently large. The pattern dictionary 2221 can be stored in the memory 221, but the pattern dictionary 2221 is preferably stored in the magnetic storage device 222 because the dictionary capacity is large. In addition, it is impractical for the central processing unit 21 to perform character recognition processing (matching processing) for all characters in the pattern dictionary 2221 in view of the performance of the processing device mounted on the copying machine 2.

  For this reason, in this embodiment, the central processing unit 21 having a particularly low processing capability performs character recognition processing (matching processing) for a limited number of characters in the partial dictionary 2212 stored in the memory 221. I made it.

  When the user inputs a character string that restricts copying using the operation panel 24, for example, the dictionary generation means M2 acquires this character string and generates a partial dictionary 2212. In addition, the user may not always be comfortable using the operation panel 24 for input. In such a case, the user can designate a character string from an external device such as a computer and transfer it to the dictionary generation means M2 of the copying machine 2 via the network.

  The operation of the copying machine 2 when the character string that restricts copying is "confidential" will be described below with reference to the flowchart of FIG. In this embodiment, the user first designates the search character string / graphic LS (designation of the character string “confidential”) using the operation panel 24 (S401).

  Next, the dictionary generating means M2 has a plurality of small differences DD with respect to the characters “machine”, “secret”, and “machine” constituting the designated character string “confidential” (in this operation example, five). ) And a plurality of (5 in this operation example) character dictionary records having a small difference DD with respect to “dense” are extracted from the pattern dictionary 2221, and six dictionary records for each character are stored in the partial dictionary. It is stored as 2212 (S402). This process is the same as S302 in the third operation example of the first embodiment (see FIG. 10).

  On the other hand, when the copying machine 2 reads a document on the contact glass, the character / graphic recognition unit M1 uses the read image as a search target image TG and a predetermined work area (usually secured as a work area in the memory 221). (S403), and character recognition is performed using the partial dictionary 2212 generated by the dictionary generation means M2 (S404). Note that this character recognition can be performed only on the upper part (for example, upper ３) of the search target image TG. As a result, it is possible to avoid the application of the copy restriction when the character string “confidential” appears in a normal sentence (when there is no need to restrict the copy).

  In the present embodiment, the recognition result for each character included in the search target image TG by the character / graphic recognition means M1 is the same as S303 in the third operation example of the first embodiment. The character string / figure search means M3 receives the recognition result by the character / figure recognition means M1, and searches for a “confidential” character string from the recognition result (S405).

Next, when the character string / graphic search means M3 does not extract the character string “confidential” (“NO” in S406), the search target image TG stored in the memory 221 is printed out (S407). When it is determined (“YES” in S406), the user is notified that the print output is restricted without printing the search target image TG stored in the memory (S408). For example, the notification may be displayed on the display of the operation panel 24 by characters or the like, the warning lamp of the operation panel 24 may be turned on, or a message or warning sound may be generated from the speaker of the operation panel 24. Good.
<< Third Embodiment >>
FIG. 15 is a functional block diagram showing a third embodiment of the specific character / graphic detection apparatus of the present invention. In FIG. 15, the specific character / graphic detection device 3 is connected to the dedicated server 4 via the network 500. In the present embodiment, the pattern dictionary and the partial dictionary are held / managed by different devices (specific character / graphic detection device 3 and dedicated server 4).

  The specific character / graphic detection device 3 includes a control unit 31, a storage unit 32, an image input unit 33, an operation / display unit 34, an image output unit 35, and a communication unit 36. The control means 31 controls the specific character / graphic detection device 3 as a whole.

  The storage means 32 stores a program 321 (character / graphic recognition program P1, character string / graphic search program P3, search result output program P4) and a partial dictionary 322.

  The programs P1, P3, and P4 are respectively executed by the control means 31, and the program P1 and the control means 31 constitute a character / graphic recognition means M1, and the program P3 and the control means 31 constitute a character string / graphics search means M3. The search result output program P4 and the control means 31 constitute search result output means M4.

  The image input unit 33 may be an image input device such as an image scanner or a digital camera, or may be an image acquisition unit (program) (not shown) that can acquire an image from a certain application.

  The operation / display unit 34 is a user interface, and typically includes an operation unit such as a keyboard and a mouse, and a display unit such as a display. The image output means 35 is a printer, and the communication means 36 is a means for connecting to the network 500.

  The dedicated server 4 includes a control unit 41, a storage unit 42, and a communication unit 43. The control means 41 controls the entire dedicated server 4.

  The storage means 42 stores a program 421 (dictionary generation program P2) and a pattern dictionary 422. The program P2 is executed by the control means 41, and the program P2 and the control means 41 constitute a dictionary generation means M2. The communication unit 43 is a unit for connecting to the network 500.

  The capacity of the pattern dictionary increases as the character type, font, and the like to be recognized are expanded. In addition, maintenance may be required as appropriate according to trends in the world (new font announcements, font trends, etc.). Therefore, in the present embodiment, the pattern dictionary 422 is centrally managed by the dedicated server 4 and the partial dictionary 322 is cut out and used as necessary, so that the maintainability and recognition performance of the pattern dictionary 422 can be expected.

  In the present embodiment, the character string to be detected and the partial dictionary 322 have a small amount of data, and therefore the dictionary generation means M2 and the pattern dictionary 422 are arranged in the dedicated server 4 here. The specific character / graphic detection device 3 can be accessed via the network 500. For example, when the load on the network 500 does not increase extremely, only the pattern dictionary 422 can be arranged in the dedicated server 4.

  Hereinafter, an example of the operation of the specific character / graphics detection device 3 of FIG. 15 will be described with reference to the flowchart of FIG. The user inputs a character string to be searched from the operation / display unit 34. Here, when the character string “confidential” is input, this character string is sent to the dictionary generation means M2 on the dedicated server 4 via the network 500 (S501). The dictionary generation means M2 extracts a specific dictionary record from the pattern dictionary 422 based on the received character string “confidential”, and transmits it to the specific character / figure detecting device 3 via the network 500.

When the specific character / figure detecting device 3 receives the dictionary record, it stores it as a partial dictionary 322 (S502). Thereafter, as in S303 to S305 in the third operation example of the first embodiment, character recognition is performed on the search target image TG using the partial dictionary 322 (S503), the character recognition result is received, and a “confidential” character string is searched. Then, the character string portion on the search target image TG searched as “confidential” is emphasized and displayed on the display of the operation / display unit 34 (S505).
<< 4th Embodiment >>
FIG. 17 is a functional block diagram showing a fourth embodiment of the specific character / graphic detection apparatus of the present invention. In FIG. 17, the specific character / graphic detection device 6 is substantially the same as the configuration of the specific character / graphic detection device 1 (see FIG. 1) of the first embodiment, and the control means 61, storage means 62, Image input means 63, operation / display means 64, program 621, partial dictionary 622, pattern dictionary 623, processing target image storage area 624, character / graphic recognition means M1 ′, dictionary generation means M2, character string / graphic search means M3 The search result output means M4 is composed of the control means 11, storage means 12, image input means 13, operation / display means 14, program 121, partial dictionary 122, pattern dictionary 123, processing target image storage area 124, character in FIG. The configuration is almost the same as that of the figure recognition means M1, the dictionary generation means M2, the character string / graphic search means M3, and the search result output means M4. However, in this embodiment, the matching method by the character / graphic recognition means M1 ′ is different from the matching method by the character / graphic recognition means M1 in FIG.

  In addition, the pattern dictionary 623 includes information on the data distribution in the learning data, whereas the pattern dictionary 123 in FIG. 1 does not include information on the data distribution. The information included in the pattern dictionary 623 is indispensable when a general statistical method such as a Bayes discriminant function or a modified secondary discriminant function is used, and it is not uncommon for it to be included in the pattern dictionary. . In this embodiment, it is assumed that the feature amount standard deviation value in the category is included in the record of the pattern dictionary 623. FIG. 18 shows a state where “dictionary” and “dense” dictionary records are extracted by the dictionary generation device. FIG. 18 shows the feature amount space SS, and the inside of the ellipse indicates the standard deviation within the category of the learning data.

  Hereinafter, an operation example of the specific character / graphics detection device 6 of FIG. 17 will be described with reference to the flowchart of FIG. In FIG. 19, the processing of S601 and S602 is the same as S101 and S102 in the first operation example of the first embodiment. That is, the user uses the operation / display unit 64 to specify the search target image TG and the search character string / graphic LS (here, the character string “confidential”) (S601).

  The dictionary generation means M2 extracts from the pattern dictionary 623 the dictionary records of the characters “machine” and “dense” that make up the designated character string “confidential” (the recorded data is a feature quantity), and this is extracted. It is stored as the partial dictionary 122 (S602).

  Next, a feature amount is extracted from the character image cut out from the search target image TG (S603), and the feature amount of the cut out character image and the feature amount of the characters (“machine”, “fine”) in the partial dictionary 622 are obtained. Each difference is obtained (S604).

The character / graphic recognition means M1 ′ recognizes all the corresponding characters (in this case, “machine”, “fine”) for character images whose feature amounts are within the standard deviation range, and the result is It outputs (S605). In FIG. 18, the points indicated by black circles indicate the feature amounts of the character images in the search target image TG, and the character images corresponding to the points included in each ellipse are recognized as “machine” and “dense”. For the points included in both ellipses, the two characters “machine” and “dense” are recognized characters. FIG. 19 shows an example of the recognition result. “(Confidential)”
Indicates that the character part is recognized as "machine" and "dense".

  Further, the character recognition result is received, and the “confidential” character string is searched by the character string / graphic search means M3 (S606). For example, in the case of “(confidential) secret”, the part is “confidential” and “confidential”, so in the case of “(confidential) secret”, when the search string is “confidential” Hit (see the part enclosed by an ellipse in FIG. 20). On the other hand, in the case of “secret (confidential)”, since the portion is “confidential” and “confidential”, no hit occurs when the search character string is “confidential”.

  The character string portion on the search target image extracted as “confidential” is emphasized and output to the display by the search result output means M4.

  In the first to fourth embodiments described above, the case of searching for the character string “confidential” has been described as an example. However, it is possible to search for a graphic such as a logo that can be recognized by the character / graphic recognition means M1. For example, it is possible to detect only a manuscript of a specific company by using a logo or the like attached to a letterhead of a specific company as a search target graphic.

It is a functional block diagram which shows 1st Embodiment of the specific character and figure detection apparatus of this invention. 3 is a flowchart illustrating a first operation example of the specific character / figure detecting apparatus of FIG. 1. It is a figure which shows the example of the search object image applied to the specific character and figure detection apparatus of FIG. It is a figure which shows the recognition result about each character contained in the search object image by the specific character and figure detection apparatus of FIG. FIG. 5 is a diagram showing a “confidential” portion in the recognition result by the character / figure recognition unit shown in FIG. It is a figure which shows the example which emphasized the character string part on the search object image concerning a search result in the 1st operation example, and output it to the display part of the operation / display means. It is a flowchart which shows the 2nd operation example of the specific character and figure detection apparatus of FIG. The “confidential” part in the recognition result (character string) by the character / graphic recognition means shown in FIG. 3 is indicated by black characters, and each character (“machine”, “secret”) is displayed in the “confidential” part. It is the figure which showed the difference degree numerically. It is a figure which shows the example which highlighted the character string part on the search object image concerning a search result in the 2nd operation example, and output it to the display part of the operation / display means. 6 is a flowchart illustrating a third operation example of the specific character / figure detecting apparatus in FIG. 1. It is the figure which represented the feature-value of the character which comprises a dictionary record in two dimensions. It is a figure which shows the recognition result about each character contained in the search object image by a character and figure recognition means in a 3rd operation example. It is a functional block diagram which shows 2nd Embodiment of the specific character and figure detection apparatus of this invention. 6 is a flowchart showing the operation of the copying machine when the character string that restricts copying is “confidential”. It is a functional block diagram which shows 3rd Embodiment of the specific character and figure detection apparatus of this invention. It is a flowchart which shows the operation example of the specific character and figure detection apparatus of FIG. It is a functional block diagram which shows 4th Embodiment of the specific character and figure detection apparatus of this invention. It is a figure which shows a mode that the dictionary record of "machine" and "dense" was extracted by the dictionary production | generation apparatus of FIG. It is a flowchart which shows the operation example of the specific character and figure detection apparatus of FIG. It is a figure for demonstrating operation | movement of the specific character and figure detection apparatus of FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Specific character and figure detection apparatus 2 Copier 3,6 Specific character and figure detection apparatus 4 Dedicated server 11, 31, 41, 61 Control means 12, 32, 42, 62 Storage means 13, 33, 63 Image input means 14, 34, 64 Operation / display means 21 Central processing unit 23 Image scanner 24 Operation panel 25 Printer 35 Image output means 36, 43 Communication means 121, 321, 621, 421, 2211 Program 122, 322, 622, 2212 Partial dictionary 123, 623 , 422, 2221 Pattern dictionary 124, 624 Processing target image storage area 221 Memory 222 Magnetic storage device 500 Network M1 Character / graphic recognition means M2 Dictionary generation means M3 Character string / graphics search means M4 Search result output means

Claims (6)

In a specific character string search device that performs character recognition on a search target image and extracts a specific character string in the image,
A dictionary record of one or more specific characters included in the specific character string and a plurality of small difference characters having a small difference from each of the one or more specific characters from a pattern dictionary in which a large number of characters are registered A dictionary generating means for extracting only the data and generating a partial dictionary;
Using the generated partial dictionary by the dictionary generation means for a character in the search target image, and character recognition means for performing character recognition processing of the specific character and the small degree of difference character,
For the character recognition plurality of character string obtained by means that, the performing character recognition processing of a particular character, and a character string search means for extracting a character string with a specified character string and identity The specific character string search device characterized by the above. 2. The specification according to claim 1, wherein the character recognition unit determines whether the characters in the search target image are the same / non-identical with each character constituting the specific character string and a small difference character. Character string search device. In a specific character string search method for performing character recognition on a search target image and extracting a specific character string in the image,
A dictionary record of one or more specific characters included in the specific character string and a plurality of small difference characters having a small difference from each of the one or more specific characters from a pattern dictionary in which a large number of characters are registered A dictionary generation step of generating only a partial dictionary by extracting only
Using the generated partial dictionary by said dictionary generating step, for the character in the search target image, and character recognition step of performing character recognition processing of the specific character and the small degree of difference character,
For a plurality of character string obtained by the character recognition step to the performing character recognition processing of a particular character, and a character string search step of extracting a character string with a specified character string and identity The specific character string search method characterized by the above.   4. The specification according to claim 3, wherein the character recognition step determines whether the characters in the search target image are the same / non-identical with each character constituting the specific character string and the small difference character. Character string search method. A specific character string search program that causes a computer to execute each step of the character string search method according to claim 3 or 4 . A computer-readable recording medium storing the specific character string search program according to claim 5.

Images