JP2013246721A - Character string recognition device, character string recognition program, and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a character string recognition device capable of recognizing a character string more accurately, a character string recognition program, and a storage medium.SOLUTION: The character string recognition device extracts, from an input image, a character string image including a plurality of character images (step 102). The device collates each of the character images included in the extracted character string image with character information previously stored in a character database, and outputs, for each character image, a character indicated by the character information having the highest coincidence with each character image (step 104). The device performs natural language processing based on a character string consisting of the output characters, and estimates a plurality of character string candidates (step 106). For each of the estimated character string candidates, the device collates the extracted character string image with the character information that is stored in the character database and corresponds to each character included in the character string candidate, and calculates the coincidence of each character included in the character string (step 108). Based on the calculated coincidence, the device determines the most reasonable character string (step 112).

Description

  The present invention relates to a character string recognition device that recognizes a character string included in an image, a character string recognition program, and a computer-readable recording medium that records the character string recognition program.

  As a conventional character recognition technique, as shown in Example 1 of FIG. 6, first, an image of a character string area (hereinafter referred to as a character string image) is specified and cut out from an input image, and each character is extracted from the character string image. A technique is generally known in which an image (hereinafter referred to as a character image) is cut out, collated with character information stored in a character database (character DB), and the character with the highest matching score is output as a correct answer (for example, Non-Patent Document 1). Thus, since the conventional character recognition technology recognizes at the level of one character, a character string having no meaning as a word or sentence is often output.

  Therefore, a recognition technique having a function of estimating a character string as a word or a sentence has been proposed. For example, as shown in Example 2 of FIG. 6, in addition to the above-described prior art example 1, a general (natural) character string from a character string made up of the characters with the highest degree of matching recognized in Example 1 above. A technique for estimating a candidate and outputting a character string indicating the highest index among the estimated character string candidates is known (see, for example, Non-Patent Document 2).

Ray Smith, An Overview of the Tesseract OCR Engine, http://tesseract-ocr.googlecode.com/svn/trunk/doc/tesseracticdar2007.pdf Takafumi Yamazoe, Minoru Etoh, Takeshi Yoshimura and Kousuke Tsujino.Hypothesis Preservation Approach to Scene Text Recognition with Weighted Finite-State Transducer, ICDAR, 2011.

  However, in the above prior art, the character string showing the highest index is not always correct. Moreover, when recognizing a character string that is not included in a dictionary or learning data such as proper nouns, a character recognition result without an estimation function may be inferior.

  The present invention has been made to solve the above-described problem, and records a character string recognition device, a character string recognition program, and a character string recognition program that can recognize a character string with higher accuracy than conventional techniques. An object of the present invention is to provide a computer-readable recording medium.

  In order to achieve the above object, a character string recognition device according to the present invention includes an extraction unit that extracts a character string image including a plurality of character images from an input image, and a character included in the character string image extracted by the extraction unit. A first collation unit that collates each image with character information stored in advance in a database, and outputs a character indicated by character information having the highest degree of coincidence with the character image for each character image; Character string estimation means for performing a natural language process based on a character string composed of characters output from the means to estimate a plurality of character string candidates, and a plurality of character string candidates estimated by the character string estimation means, The character information corresponding to each character included in the character string candidate stored in the database is collated with the character string image extracted by the extracting means, and included in the character string candidate. Based on the matching degree calculated by the second matching means for calculating the matching degree of the character and the second matching means, a likely character string is determined from the plurality of character string candidates estimated by the character string estimating means. And a character string determining means.

  Thus, based on the character string obtained by collating each of the character images included in the character string image extracted from the input image with the character information stored in the character database, a plurality of natural language processing is performed. Since character string candidates are estimated and each of the estimated character string candidates is collated with the character string image, the accuracy of the linguistic character string and the degree of coincidence with the character string image by image processing A character string satisfying both of the above is determined as a plausible character string, and the character string can be recognized with higher accuracy than in the prior art. Note that the character string candidates estimated by the character string estimation means are character strings used by humans in daily communication, and are character strings in natural languages such as Japanese and English.

  The second collating unit calculates the degree of matching of each character included in the character string candidate for the estimated plurality of character string candidates, and uses the calculated degree of matching to calculate the character string estimating unit. The degree of coincidence for each character string candidate estimated by the above is calculated, and the character string determining means determines the likely character string based on the degree of coincidence for each character string candidate calculated by the second collating means. be able to.

  In addition, the character string determination unit is more likely to use a character string composed of characters output from the first matching unit when each of the matching degrees for each of the character string candidates is equal to or less than a preset threshold value. It may be determined as a character string.

  Further, the character string determining means has a matching degree of at least one character out of a plurality of character string candidates estimated by the character string estimating means based on the matching degree calculated by the second matching means. A plausible character string may be determined from a character string other than the following character string.

  In addition, the character string determination unit sets in advance each of the degree of coincidence for each character string candidate estimated by the character string estimation unit and the degree of coincidence of character strings made up of characters output from the first matching unit. If the value is equal to or less than the threshold value, the likelihood of the character string is not determined from the image extracted as the character string image by the extraction unit, and the region of the image extracted as the character string image About a region of a character string image that is determined not to be a region of a non-character string image, and is stored in a predetermined storage unit. May store the information related to the area of the character string image in the storage unit separately from the information related to the area of the non-character string image.

  Moreover, the program of this invention is a character string recognition program for functioning a computer as each means of the said character string recognition apparatus.

  The recording medium of the present invention is a computer-readable recording medium on which a character string recognition program for causing a computer to function as each means of the character string recognition device is recorded.

  As described above, according to the character string recognition device, the character string recognition program, and the recording medium of the present invention, it is possible to obtain an effect that a character string can be recognized with higher accuracy than in the related art.

It is a figure which shows an example of a structure of the character string recognition apparatus which concerns on embodiment. It is an example of the flowchart of a character string recognition process routine. It is another example of the flowchart of a character string recognition processing routine. It is explanatory drawing which illustrates the specific example of a character string recognition process typically. It is a figure which shows the other structural example of a character string recognition apparatus. It is explanatory drawing explaining the conventional character recognition technique.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram illustrating an example of a configuration of a character string recognition apparatus according to the present embodiment.

  The character string recognition device 10 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), and a ROM (Read Only Memory) that stores a program for the CPU to execute each processing routine described later. It consists of The computer constituting the character string recognition device 10 may include a storage unit such as a hard disk drive, a communication interface, and the like. A program executed by the CPU may be stored in the hard disk drive. When the CPU reads and executes a program stored in a storage unit such as a ROM or a hard disk, functions described below are realized by cooperating the hardware and the program.

  As shown in FIG. 1, this computer functionally includes a character string image extraction unit 12, a first collation unit 14, a character string estimation unit 16, a second collation unit 18, a character string determination unit 20, a character DB ( Database) 22, language DB 24, and page format storage unit 26.

  In the character DB 22, character information indicating characters of various typefaces (fonts), sizes, and thicknesses is registered in advance for each character. Note that the character information may be, for example, character pattern image information of each character, or may be a feature value indicating a feature of each character. The language DB 24 stores a large amount of natural language data such as Japanese and English that are used by people on a daily basis in a state that can be searched by a computer.

  Although the case where the character string recognition device 10 includes the character DB 22 and the language DB 24 has been described here, for example, the character DB 22 and the language DB 24 are provided in an external device of the character string recognition device 10, and the character string recognition device 10. May refer to the character DB 22 and the language DB 24 by communicating with an external device using communication means.

  The character string image extraction unit 12 acquires an input image as a character string recognition target, and extracts an image of a character string region (hereinafter, a character string image) from the input image by image processing. The character string image includes a plurality of character images.

  The first collation unit 14 collates each character image included in the character string image extracted by the character string image extraction unit 12 with character information stored in the character DB 22 in advance, and for each character image, the character The character indicated by the character information having the highest degree of coincidence with the image is output. Hereinafter, a character string composed of characters output from the first verification unit 14 is referred to as a temporary recognition character string for convenience.

  The character string estimation unit 16 performs known natural language processing using the language DB 24 based on the temporarily recognized character string, and estimates a plurality of character string candidates. The character string candidate estimated here is a character string that a person uses in daily communication, and refers to a character string in a natural language such as Japanese or English. Hereinafter, a set of character string candidates estimated by the character string estimation unit 16 is referred to as a character string candidate group.

  The second collating unit 18 collates each character string candidate in the character string candidate group with the character string image extracted by the character string image extracting unit 12. Here, collation is performed in units of characters included in each of the character string candidates. Specifically, the second collation unit 18 selects character information corresponding to the character included in the character string candidate from the character DB 22, and the selected character information and the character string image extraction unit 12 extract the selected character information. Each character image included in the character string image is collated.

  The character string determination unit 20 determines a likely character string based on the collation result by the second collation unit 18. The character string determined here is output as a character string recognition result.

  Further, the character string determination unit 20 refers to the collation result by the first collation unit 14 and the collation result by the second collation unit 18, and the matching degree of each character string unit of the temporary recognition character string and the character string candidate is all When the threshold value is equal to or lower than a preset threshold, the image extracted by the character string image extraction unit 12 is not a character string region image but an image of a non-character string region other than a character string, such as an illustration or a photograph. The page format storage unit determines and determines the character string from the image, and uses information about the non-character string region in the input image (for example, information indicating the coordinates and size of the non-character string region) as page format information. 26.

  In addition, for an image that is not determined to be an image of a non-character string area, the character string determination unit 20 includes information related to the area (character string area) of the character string image (for example, the character string area) as page format information. Are stored in the page format storage unit 26 separately from the information related to the non-character string region.

  Next, the operation of the character string recognition device 10 will be described in detail with reference to FIG.

  In step 100, the character string image extraction unit 12 acquires an input image as a character string recognition target.

  In step 102, the character string image extraction unit 12 extracts a character string image from the acquired input image.

  In step 104, the first matching unit 14 cuts out individual character images from the character string image extracted by the character string image extraction unit 12. And the 1st collation part 14 collates each of the cut-out character image with the character information previously memorize | stored in character DB22, and character information with the highest matching degree with the said character image for every character image is obtained. Output the indicated character (text data). Here, for example, when the character information is not a character pattern image but a feature value such as a vector indicating the feature of the character, the first collation unit 14 develops the character information into an image and displays the character string image. Match with the character image of. The character image may be converted into the same format as the character information and collated.

  If the character string image extracted by the character string image extraction unit 12 is a horizontally written character string image, the first matching unit 14 moves from left to right (depending on the recognition language, from right to left). Good) Each character image and character information are collated in the order of arrangement, and if the extracted character string image is a vertically written character string image, each character image and character information are arranged in order from the top to the bottom. We will collate. A temporary recognition character string is obtained by arranging characters output after collation in the order of output.

  For the processing from step 100 to step 104, a technique similar to a known conventional method (for example, the method of Example 1 illustrated in FIG. 4) can be used.

  Next, in step 106, the character string estimation unit 16 performs natural language processing based on the temporarily recognized character string to estimate a plurality of character string candidates. In this natural language processing, language data registered in the language DB 24 is referred to. The character string estimation unit 16 outputs a plurality of character string candidates in a probable order without narrowing the estimation result to one character string.

  The language processing techniques used here include word estimation, co-occurrence word estimation, and error correction methods (eg, Non-reference 3: Nagata. Japanese character recognition error correction method using character similarity and statistical language model). Various methods such as Information Processing, 81 (11): 2624-2634, 1998) can be applied.

  In step 108, the second collation unit 18 sequentially collates the plurality of character string candidates with the character string image extracted by the character string image extraction unit 12. In this embodiment, this collation is performed for each character by using the character DB 22 again.

  Hereinafter, the process by the 2nd collation part 18 is explained in full detail. The second matching unit 18 selects a plurality of character information corresponding to each of a plurality of characters included in each character string candidate from the character information stored in the character DB 22. Then, the second collating unit 18 collates each of the selected character information with each of the character images included in the character string image extracted by the character string image extracting unit 12, and is included in each character string candidate. The degree of coincidence with the character image is calculated for each character. In addition, the position in the character string candidate of the character which character information shows, and the position in the character string image of each character image collated with the said character information shall mutually respond | correspond. Also here, as in the processing of the first collation unit 14 in step 104, character information is represented by feature values and the like, and if the character image to be collated and the character information are in a different format, the character information The information is developed into an image and collated with the character image of the character string image. Alternatively, the character image may be converted into a character information format and collated.

  Here, the verification process by the second verification unit 18 will be described in detail. The second collation unit 18 sequentially applies the font registered in the character DB 22 to the estimated character string candidate, or applies a new font that is a combination of a plurality of fonts, and the character of each character constituting the character string candidate After the image is generated, the generated character image is compared with each character image included in the character string image, and the matching degree for each character is calculated in the same manner as the first matching unit 14.

  As a result of the collation, if the character string candidate includes characters with a very low matching score, the matching score for each character string candidate calculated in the next step 110 for the character string candidate is also a low value. It should be noted that a character string candidate that includes a character whose matching degree is smaller than a preset threshold value may be excluded from the character string candidate group regardless of the matching degree of other characters included in the character string candidate. Good.

  In step 110, the second matching unit 18 calculates the degree of coincidence for each character string candidate based on the degree of coincidence for each character calculated in step 108. For example, the sum of the matching degrees for each character included in the character string candidate may be used as the matching degree of the character string candidate, or a value obtained by multiplying the sum by a predetermined coefficient is used as the matching degree of the character string candidate. Also good.

  In step 112, the character string determination unit 20 determines and outputs a plausible character string based on the matching degree for each character string candidate. Specifically, for example, the character string candidate having the highest degree of matching for each character string candidate can be determined as a likely character string.

  Further, the matching degree for each character string candidate may be determined by comparing with a preset threshold value. Specifically, when there is a character string candidate having a matching degree exceeding a preset threshold in the character string candidate group, the character string determining unit 20 selects the character string candidate having the highest matching degree exceeding the threshold. If it is determined as a plausible character string and there is no character string candidate with a matching degree exceeding the preset threshold, a temporarily recognized character string directly extracted from the character string image may be determined as a plausible character string. .

  Furthermore, when there is no character string candidate with a matching degree exceeding the preset threshold, the character string determining unit 20 further determines the matching degree of the character string unit of the temporarily recognized character string from the matching result by the first matching unit 14. When the calculated degree of coincidence does not exceed the preset threshold, the image extracted by the character string image extraction unit 12 is not an image of the character string region, but a non-character string other than a character string, such as an illustration or a photograph. Information indicating the coordinates of the non-character string area in the input image (for example, the coordinates of the upper left vertex of the area) and the size are determined without determining the character string from the image and determining that the image is an area image. Is stored in the page format storage unit 26 as page format information. Note that the matching degree of the temporarily recognized character string in character string units may be calculated in advance by the first matching unit 14.

  For the character string image that is not determined to be a non-character string region image, the character string determination unit 20 uses the coordinates, size, font information, and the like of the character string image region as page format information. The information is stored in the page format storage unit 26 separately from the information related to the non-character string region. For example, the information related to the character string area is distinguished by adding a flag indicating that it is a character string area.

  Conventionally, in character string image extraction, it has been common to extract character string images assuming that all partial regions from the input image are character string regions as much as possible. According to the above method, the character string area and the non-character string area are distinguished and stored as page format information. By using this, further character recognition processing or image processing other than character recognition can be efficiently performed. It can be carried out.

  For example, when a character string is extracted using a read image of a book or document page composed of a plurality of pages having the same page format as an input image, the character string image extraction unit 12 obtains page format information for the first several pages. At this stage, the character string image can be extracted with reference to the page format information stored in the page format storage unit 26 for the subsequent page images. Specifically, the character string image extraction unit 12 can extract a character string image according to the coordinates and size of a character string area other than the non-character string area stored as page format information. This increases the efficiency of the character string image extraction process. Further, in the first verification unit 14, the corresponding character information can be selected from the character DB 22 and used for verification in accordance with the font information stored as the page format information. Thereby, the collation process becomes efficient.

  The information in the page format storage unit 26 can be referred to when extracting an image such as an illustration or a photograph. Thereby, it is possible to improve the efficiency of extracting an image such as an illustration or a photograph instead of a character string.

  For the estimation of the character string by the natural language processing in step 106, a known conventional method (for example, the method of Example 2 shown in FIG. 4) can be used. Further, collation for each character between the character string candidate and the character string image by the second collation unit 18 can be realized by the same technique as the collation of the prior art. That is, character data (image information of various typefaces, sizes, and thicknesses) corresponding to the character string candidate character is extracted from the character DB 22 and compared with each character image of the character string image. The same applies to the calculation of the matching degree for each character.

  In the above description, an example has been described in which the degree of matching for each character string candidate is obtained in advance for all character string candidates, and a plausible character string is determined based on the degree of matching for each character string candidate. It is not limited.

  For example, the character string to be output is determined in consideration of the collation result for each character and the estimation degree of the character string candidate (the probability when the character string candidate is estimated based on the temporarily recognized character string). Also good. For example, character string candidates and character string images are collated in descending order of estimation, and the first character string candidate that does not include characters whose matching degree for each character is lower than a preset threshold is determined as a plausible character string. It can also be output. This will be described below with reference to the flowchart of FIG. In FIG. 3, the same steps as those in FIG. 2 are denoted by the same step numbers and description thereof is omitted.

  In step 120, the second matching unit 18 selects one character string candidate with the highest estimation degree from the character string candidates included in the character string candidate group estimated in step 106. In step 122, the second collating unit 18 collates the selected character string candidate with the character string image. This collation is performed in character units as in step 108. In step 124, the second matching unit 18 calculates the degree of matching for each character for each character included in the character string candidate. In step 126, the character string determination unit 20 determines whether or not a character having a matching degree equal to or less than a preset threshold value exists in the character string candidate. If a negative determination is made in step 126, in step 130, the character string determination unit 20 determines and outputs the currently selected character string candidate as a likely character string.

  On the other hand, if an affirmative determination is made in step 124, the character string determination unit 20 determines in step 128 whether or not collation for all character string candidates in the character string candidate group has been completed. If a negative determination is made in step 128, the currently selected character string candidate is discarded, and the process returns to step 120, where the character string candidate with the next highest estimate is selected and collated. If the determination in step 128 is affirmative, the character string determination unit 20 determines and outputs the temporary recognition character string as a likely character string in step 130.

Next, the operation of the character string recognition device 10 will be described with reference to FIG. 4 by giving a more specific example while comparing with the configuration diagram of FIG. Here, the case where collation is performed in descending order of estimation described with reference to FIG. 3 will be described as an example.
(1) An input image is acquired (FIG. 1A).
(2) A character string image is cut out (FIG. 1B). A portion surrounded by a one-dot chain line is a character string region.
(3) Each character image of the character string image is collated with the character information in the character DB 22 (FIG. 1C), and the character information having the highest matching degree is extracted as an output (FIG. 1D). At this time, a correct character string is not always extracted due to the influence of character segmentation or background.
(4) Natural language processing is performed based on the extracted character string (temporarily recognized character string), and a plurality of character string candidates are estimated (FIG. 1E). Is output (FIG. 1 (F)). As described above, various methods such as word estimation, co-occurrence word estimation, and error correction can be applied to the language processing technique used here. FIG. 3 shows a state in which three character string candidates DRY CLEANER, DRY CLEANING, and DERBY CLOTHING are estimated.
(5) The character string image is collated in order from the character string candidate with the highest estimation. The collation here is performed for each character by extracting the corresponding character information from the character DB (FIG. 1G). From this collation result, a plausible character string is determined (FIG. 1 (H)) and output (FIG. 1 (I)).

  For example, if a character that is equal to or smaller than a preset threshold appears, the character string candidate is discarded, and the process proceeds to the next character string candidate collation. If no character below the threshold appears, this character string candidate is determined as a plausible character string and is output. That is, here, the character string candidate having the highest estimation degree among the character string candidates whose matching degree for each character exceeds all the threshold values is determined as a likely character string. Further, when the matching degree of each character is less than or equal to the threshold value in all character string candidates, the output of (3) is adopted as the final output. Note that, as described with reference to FIG. 2, as a method of calculating the degree of coincidence and character string determination, the sum of the degree of coincidence for each character included in the character string candidate is calculated for each character string candidate. It is also possible to apply a method in which the character string candidate showing the maximum degree of coincidence is determined as a plausible character string and output.

  As described above, the character strings extracted by the image processing (obtained by the collation of the first collation unit 14) are converted into general character string candidates by the natural language processing of the character string estimation unit 16, and again these By matching the character string candidates with the character string image, the final character string recognition result is a character string that satisfies both the accuracy of the linguistic processing character string and the matching degree with the image processing character string image. As a result, a more accurate character string recognition function can be realized.

  The configuration of the character string recognition device is not limited to the above example, and for example, a configuration as shown in FIG. The character string recognition device 11 shown in FIG. 5 includes a character string image extraction unit 12, a first collation unit 14, a character image collation unit 15, a character string estimation unit 16, a character string determination unit 20, a character DB 22, a language DB 24, and a page. It can be expressed by a configuration including the format storage unit 26. Here, the constituent elements having the same reference numerals shown in FIG. 5 and FIG. 1 mean the constituent elements having the same functions, and the description thereof will be omitted.

  In the character string recognition device 11 shown in FIG. 5, the first verification unit 14 and the character image generation unit 15 constitute a second verification unit 19. The character image generation unit 15 sequentially applies the fonts registered in the character DB 22 to the character string candidates estimated by the character string estimation unit 16, or applies a new font that is a combination of a plurality of fonts. A character image of each character constituting the image is generated. When the first collating unit 14 operates as a function of the second collating unit 19, the first collating unit 14 collates the character image generated by the character image generating unit 15 with each character image included in the character string image. Calculate the degree of match. The calculated degree of coincidence is output to the character string determination unit 20. The first collation unit 14 and the character image generation unit 15 realize a function equivalent to that of the second collation unit 18 illustrated with reference to FIG.

  Moreover, although the above-mentioned character string recognition apparatus has a computer system inside, if the computer system is using the WWW system, it shall also include a homepage provision environment (or display environment).

  In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium.

DESCRIPTION OF SYMBOLS 10, 11 Character string recognition apparatus 12 Character string image extraction part 14 1st collation part 16 Character string estimation part 18 2nd collation part 20 Character string determination part 22 Character DB
24 Language DB
26 Page format storage

Claims (7)

Extracting means for extracting a character string image including a plurality of character images from the input image;
Each character image included in the character string image extracted by the extraction means is collated with character information stored in advance in the database, and character information having the highest degree of coincidence with the character image is indicated for each character image. First collation means for outputting characters;
Character string estimation means for performing natural language processing based on a character string composed of characters output from the first collating means, and estimating a plurality of character string candidates;
For a plurality of character string candidates estimated by the character string estimating means, character information corresponding to each character included in the character string candidates stored in the database, a character string image extracted by the extracting means, and Second matching means for calculating the matching degree of each character included in the character string candidate,
A character string determining means for determining a plausible character string from a plurality of character string candidates estimated by the character string estimating means based on the degree of coincidence calculated by the second matching means;
A character string recognition device. The second collating unit calculates a matching degree of each character included in the character string candidate for the estimated plurality of character string candidates, and uses the calculated matching degree to estimate by the character string estimating unit. Calculate the degree of matching for each of the string candidates
The character string recognition apparatus according to claim 1, wherein the character string determination unit determines the likely character string based on a degree of matching for each character string candidate calculated by the second matching unit. The character string determining means determines that the character string consisting of characters output from the first matching means is the likely character string when each matching degree for each character string candidate is equal to or less than a preset threshold value. The character string recognition device according to claim 2. The character string determining means has a preset threshold value for each of the matching degree for each character string candidate estimated by the character string estimating means and the matching degree of the character string composed of characters output from the first matching means. In the following cases, the plausible character string is not determined from the image extracted as the character string image by the extraction means, and the region of the image extracted as the character string image is determined as a non-character string image. It is determined that the area of the character string image, information on the area of the non-character string image is stored in a predetermined storage unit, and the area of the character string image that is not determined to be the area of the non-character string image, The character string recognition device according to claim 2, wherein information relating to the area of the character string image is stored in the storage unit in a manner different from information relating to the area of the non-character string image. The character string determination means has a matching degree of at least one character of a plurality of character string candidates estimated by the character string estimation means based on the matching degree calculated by the second matching means as a threshold value or less. The character string recognition device according to claim 1, wherein a plausible character string is determined from character string candidates other than the character string candidates.   The character string recognition program for functioning a computer as each means of the character string recognition apparatus in any one of Claims 1-5.   A computer-readable recording medium on which is recorded a character string recognition program for causing a computer to function as each means of the character string recognition device according to any one of claims 1 to 5.