JP5357711B2 - Document processing device - Google Patents

Description

  The present invention relates to a business document processing apparatus and, for example, relates to a technique for automatically collecting character recognition patterns generated when OCR is applied to a paper document.

  In recent years, a large number of paper business documents accumulated in an organization are recognized by scanning and OCR, and the document data is managed by a document management system, thereby improving searchability and safe storage of paper documents. There is a movement to share knowledge.

  Although the recognition accuracy of OCR (Optical Character Reader) has been increasing with the improvement of technology, it is still impossible to completely eliminate misrecognition. Therefore, various countermeasures against misrecognition have been devised. One of them is a method using a replacement dictionary. This is because the correct character string and the misrecognized character string pair are registered in the dictionary for the character string that is likely to be erroneously recognized, and the misrecognized character string registered in the dictionary is included in the OCR target document. If so, it is replaced with the correct character string. This method is effective when the erroneous recognition pattern of the OCR target document is known and the dictionary is comprehensively registered. For example, non-patent documents 1 to 4 are products that use the replacement dictionary, and the replacement dictionary is a general function used for character recognition by OCR.

SEIKO EPSON CORPORATION. EPSON SALES JAPAN CORPORATION 2007. “Read de !! Coco”, [online], [Search June 30, 2009], Internet <URL: http://ai2you.com/OCR/> Media Drive Corporation. “WinReaderPro”, [online], [Search June 30, 2009], Internet <URL: http://mediadrive.jp/products/wrp/index.html> Media Drive Corporation. “E.Typist”, [online], [Search June 30, 2009], Internet <URL: http://mediadrive.jp/products/et/> Panasonic Solution Technologies Co., Ltd. 2009 “Reading Revolution”, [online], [Search June 30, 2009], Internet <URL: http://panasonic.co.jp/pss/pstc/products/yomikaku />

  However, it is common for a user to register an erroneous recognition pattern in a replacement dictionary based on experience. For this reason, the user must sequentially register the erroneous recognition pattern, and the amount of work becomes enormous. Moreover, since the quality of the misrecognition pattern registered according to the skill of the user is fluctuated, there is a problem in that registration of a misrecognition pattern having a high occurrence frequency or over-learning of an erroneous recognition pattern having a low occurrence frequency occurs. Misrecognition pattern registration omissions cause misrecognized character strings to remain in the document after OCR processing, and over-learning results in replacement of the correct character strings in the document. The recognition accuracy is reduced.

  The present invention has been made in view of such a situation, and automatically collects misrecognition patterns registered in a replacement dictionary used when correcting OCR misrecognition, and further selects the collected misrecognition patterns. Applying technology that makes it possible.

  In order to solve the above problems, a document processing apparatus according to the present invention is a document processing apparatus that automatically generates a replacement dictionary for correcting an OCR misrecognized character string, from sample electronic document data obtained by imaging a business document. A matching processing unit that performs misrecognition determination using a segmented correct character string and a post-OCR character string segmented from the post-OCR sample document data obtained as a result of performing OCR on the sample electronic document data. And an analysis processing unit that divides the correct character string into predetermined word units, registers a word including characters determined to be erroneously recognized by the matching processing unit among the divided words, and a storage device. Japanese words stored in the Japanese dictionary data registered with Japanese words and business word dictionary data registered with words used in business And delete the partial or exact matching words from the erroneous recognition pattern candidate filtering comprises a filtering section for storing into said storage device as a mis-recognition pattern misrecognition pattern candidates after the filtering, the.

  Further features of the present invention will become apparent from the best mode for carrying out the present invention and the accompanying drawings.

  According to the present invention, it is possible to automatically collect necessary misrecognition patterns when creating a replacement dictionary for correcting misrecognition that occurs when OCR is applied to a document. Thereby, a user's work amount can be reduced significantly and omission of registration to the replacement dictionary of a misrecognition pattern can be prevented.

  Moreover, it is possible to prevent overlearning that a correct character string is replaced by filtering the collected erroneous recognition patterns.

  As a result, it is possible to create a uniform replacement dictionary regardless of the skill of the user.

It is a functional block diagram which shows roughly the structure of the business document processing apparatus by embodiment of this invention. It is a figure which shows the example of an image obtained by printing and scanning the sample electronic document data 51 memorize | stored in the memory | storage device shown in FIG. It is a figure which shows the example of the sample document data after OCR memorize | stored in the memory | storage device shown in FIG. It is a figure which shows the example of the Japanese dictionary data memorize | stored in the memory | storage device shown in FIG. It is a figure which shows the example of the business word dictionary data memorize | stored in the memory | storage device shown in FIG. It is a flowchart for demonstrating the matching process part in an OCR misrecognition correction program. It is a figure which shows the example of the character string comparison data memorize | stored in the data memory shown in FIG. It is a flowchart for demonstrating the analysis process part in an OCR misrecognition correction program. It is a figure which shows the example of the misrecognition pattern candidate data memorize | stored in the data memory shown in FIG. It is a flowchart for demonstrating the filtering part in an OCR misrecognition correction program. It is a figure which shows the example of the confirmation screen which shows the output result of a misrecognition correction pattern.

  Hereinafter, an embodiment for carrying out an erroneous recognition pattern collection apparatus of the present invention will be described in detail with reference to the accompanying drawings. 1 to 11 are diagrams illustrating an embodiment of the present invention. In these drawings, parts denoted by the same reference numerals represent the same items, and the basic configuration and operation are the same. In addition, in embodiment of this invention, the apparatus, method, etc. which are used are examples, and of course, this invention is not limited to these.

<Configuration of error recognition pattern collection device>
FIG. 1 is a functional block diagram showing a schematic configuration of a misrecognition pattern collection apparatus according to an embodiment of the present invention. This misrecognition pattern collection device includes a storage device 50, an input / output device 30 for inputting and outputting data, a central processing unit 10 for performing necessary arithmetic processing and control processing, and processing in the central processing unit And a data memory 20 for storing data necessary for processing in the central processing unit 10.

  The storage device 50 prints, scans, scans the sample electronic document data 51 such as a Windows (registered trademark) word file created so as to have the same configuration as the business document to be subjected to OCR, and the sample electronic document data 51. Sample document data 52 after OCR obtained as a result of OCR, erroneous recognition pattern data 53 finally output according to the present invention, Japanese dictionary data 54 in which a number of general Japanese words are registered, Business word dictionary data 55 in which a number of words used in documents used in business and the like are registered, and post-OCR sample document data 52 after correction using error recognition pattern data 53 and corrected results obtained Electronic document data 56 is stored.

  The input / output device 30 includes a display device 32 for displaying data, an output unit including a printer (not shown), and a keyboard for performing operations such as selecting a menu for the displayed data. 31 and an input unit including a pointing device 33 such as a mouse, a scanner 34 for capturing a document, and the like.

  The program memory 40 includes a matching processing unit 41 that compares the characters on the sample electronic document data 51 and the corresponding characters in the post-OCR sample document data 52, and an error based on the result output by the matching processing unit. An analysis processing unit 42 that outputs a recognition pattern candidate, and a filtering processing unit 43 that deletes an unnecessary pattern that causes an erroneous correction from the erroneous recognition pattern candidates output by the analysis processing unit. . Each processing unit is stored in the program memory 40 as a program code, and each processing unit is realized by the central processing unit 10 executing each program code.

  The data memory 20 includes character string comparison data 21 obtained from the sample electronic document data 51 and post-OCR sample document data 52, and erroneous recognition pattern candidate data 22 derived from the character string comparison data 21.

  FIG. 2 is a diagram illustrating an example of an image obtained by printing and scanning the sample electronic document data 51 stored in the storage device 50. The document includes three series of character strings. The purpose is to recognize erroneously recognized characters generated after the OCR process by performing character recognition by OCR on the scanned image (FIG. 2) of the sample electronic document data 51 and comparing it with the original sample electronic data 51. (FIG. 7).

  FIG. 3 is a diagram illustrating an example of the post-OCR sample document data 52 included in the storage device 50. 3 shows the result of applying OCR to the image of FIG. Some characters are misrecognized.

  FIG. 4 is a diagram illustrating an example of Japanese dictionary data 54 included in the storage device 50. Many commonly used Japanese words are registered. It includes information such as part of speech, independent word / attachment type, basic form, reading, and type of utilization. These pieces of information are used in an analysis process described later. Specifically, at the time of analysis, whether or not to register in the misrecognition pattern depending on whether any word, including the usage form in the Japanese dictionary data, and the character string after OCR partially match. Used to determine This is because by registering information such as the type of utilization, it is possible to register an erroneous recognition pattern even if any of the utilization patterns matches the character string after OCR.

  FIG. 5 is a diagram illustrating an example of the business word dictionary data 55 included in the storage device 50. Many words (nouns) used in business are registered. However, the word to be registered is not limited to a noun, and information such as part of speech, independent word / attached word type, basic form, reading, and utilization type may be included.

  FIG. 7 is a diagram illustrating an example of the character string comparison data 21 included in the data memory 20, and shows a result of comparing character strings separated from the sample electronic data 51 and the post-OCR sample document data 52. As a result of comparing the characters constituting the character strings, the misrecognition flag is set to “0” if they are the same character, and the misrecognition flag is set to “1” if they are different characters. Yes.

  FIG. 9 is a diagram illustrating an example of the erroneous recognition pattern candidate data 22 included in the data memory 20, and represents the output result of the erroneous recognition pattern for the data in FIGS. The correct character string in the figure is a character string cut from the sample electronic document data 51. The misrecognized character string is a character string corresponding to the correct character string, which is separated from the post-OCR sample document data 52. The misrecognition pattern in the sample document data after OCR is output as a misrecognized character string for each unit representing a collective meaning, and the character string corresponding to the misrecognized character string is output from the sample electronic document data 51 as a correct character string. .

<Processing in wrong recognition pattern collection device>
Next, an outline of processing performed in the central processing unit 10 in the erroneous recognition pattern collection device having the above-described configuration will be described.

  First, the matching processing unit 41 reads the sample electronic document data 51 and the post-OCR sample document data 52, and performs recognition of corresponding characters in each data and determination of erroneous recognition of the characters. The result is stored in the data memory 20 as character string comparison data 21.

  Next, the analysis processing unit 42 reads the sample electronic document data 51 and the word dictionary data 54, and if the word registered in the word dictionary data 54 is included in the sample electronic document data 51, the unit of the word Delimits the string. Next, the character string comparison data 21 is read, a misrecognition pattern is calculated for each divided word unit, and stored in the data memory 20 as misrecognition pattern candidate data 22.

Next, the filtering processing unit 43 reads the erroneous recognition pattern candidate data 22 and deletes unnecessary erroneous recognition patterns. Then, it is stored in the storage device 50 as erroneous recognition pattern data 53.
Each process will be described in detail below.

<Matching process>
Here, a character string is cut out from the sample electronic document data 51 and the post-OCR sample document data 52, and each cut character string is used as a comparison unit, and a recognition error is determined for each character constituting the character string.

FIG. 6 is a flowchart showing an outline of the matching process.
First, the matching processing unit 41 reads the sample electronic document data 51 and the post-OCR sample document data 52, and performs the following processing on the corresponding document pair (step 601).

  The pair of the selected sample electronic document data 51 and post-OCR sample document data is first cut into a group of character strings (step 602). In the example of Fig. 2 and Fig. 3, "Japan △ Software Co., Ltd." and "Eye △ Software Chichishi Co., Ltd.", "Invoice" and "Itouchi Akisho", "June 25, 2009" “June 25, 2009” corresponds to each character string. The character strings can be associated by using, for example, character coordinate information.

  Next, the following processing is performed on all the characters included in all the correct character strings using the character strings cut out from the sample electronic document data 51 as correct character strings (steps 603 and 604).

  The correct character string is compared with the corresponding character string (character string after OCR) of the sample document data after OCR (step 605). In the comparison, for each individual character constituting the correct character string, consistency is confirmed as to whether the individual characters constituting the post-OCR character string correspond correctly. Characters can be associated by a general character string matching method such as DP matching (Dynamic Programming Matching). FIG. 7 shows the comparison results.

  As a result of comparison, if the characters are the same, the misrecognition flag is set to “0” (step 607). If the characters are different, the misrecognition flag is set to “1” (step 608). ). If the character included in the correct character string does not exist in the character string after OCR (for example, “e” in FIG. 7), the erroneous recognition flag of the correct character is set to “1”. Further, when there are a plurality of post-OCR characters corresponding to the correct character (for example, “Ne” in FIG. 7), the erroneous recognition flag is set to “1” for each correct character. Such processing is performed for all characters.

  The finally obtained data is stored in the data memory 20 as the character string comparison data 21 (step 609).

<Analysis processing>
Here, the correct character string is segmented into predetermined word units, and among the segmented words, words including characters whose erroneous recognition flag is determined to be 1 by the above-described matching processing are registered as erroneous recognition pattern candidates.

  FIG. 8 is a flowchart showing an outline of the analysis process.

  The analysis processing unit 42 reads the character string comparison data obtained by the matching processing unit, and performs the following processing for all the character string comparison data (step 801).

  First, the correct character string is cut into predetermined word units (step 802). Specifically, if a word registered in the Japanese dictionary data or the business word dictionary data is included in the correct character string, the correct character string is divided into the word units. If there are more dictionary registered words in the cut words, all the words are cut separately. For example, the “delivery note” which is the correct character string in FIG. 2 is divided into “delivery note” and further “delivery”.

  Next, the following processing is performed on all the cut words (step 803). If a character whose error recognition flag is determined to be “1” is included in the segmented word with reference to the character string comparison data, the word is used as a misrecognition pattern candidate along with the corresponding word in the correct character string. 20 to register. For example, in FIG. 7, the words “delivery” and “delivery note” included in “delivery note” both include the erroneous recognition flag “1”. Accordingly, “delivery” and “in-yarn product”, “delivery note” and “in-yarn product” are registered as erroneous recognition pattern candidates. In this way, the process of step 804 is performed for all words. FIG. 9 is a diagram illustrating examples of erroneous recognition pattern candidates obtained for the data in FIGS. 2 and 3.

<Filtering processing part>
Here, a word that partially or completely matches another word registered in the Japanese dictionary data and the business dictionary data is deleted from the erroneous recognition pattern candidate.

  FIG. 10 is a flowchart showing an outline of the filtering processing unit. The filtering processing unit 43 reads the erroneous recognition pattern candidate data obtained by the analysis processing unit, and performs the following processing for all the erroneous recognition pattern candidates (step 1001).

  The misrecognized character string of the misrecognized pattern candidate registered by the above-described analysis processing is compared with each word of the Japanese dictionary data and the business dictionary data to determine whether or not a partial match or a complete match (step 1002). . For example, a character string “conspicuous” partially matches a Japanese word “conspicuous”. Such a word that partially or completely matches another word must not be registered as a misrecognition pattern. This is because if such a word is registered as a misrecognition pattern, not only a misrecognized word but also another correctly recognized word is judged to be misrecognized, which causes misconversion. is there. For this reason, a word that partially matches or completely matches another word is deleted from the erroneous recognition pattern candidate (step 1003). For example, in a document, if “conspicuous” is uniformly converted to “Hitachi”, a character string such as “conspicuous” is erroneously converted to a character string “conspicuous”. Such erroneous conversion can be prevented by deleting unnecessary patterns by filtering. In this way, the filtering process is performed for all registered erroneous recognition pattern candidates. In the example of FIG. 9, “conspicuous” partially matches “conspicuous” registered in the Japanese dictionary data and business dictionary data, so the pair of “conspicuous” and “Hitachi” is deleted from the erroneous recognition pattern candidates. Is done. “Date / Software” does not partially match “Date / Software”. The partial match mentioned here means that the character string of the misidentification pattern candidate is completely included in the words in the Japanese dictionary data and the business dictionary data. Since “date Δ software” is not completely included in “date Δ software”, “date Δ software” is not deleted from the misidentification pattern candidates. The same applies to “Intra-Inner Crystal” and “Invoice”. On the contrary, since “conspicuous” is completely included in “conspicuous”, “conspicuous” is deleted from the misidentified pattern candidates.

  Next, a confirmation screen for erroneous recognition pattern candidates after filtering is displayed (step 1004). FIG. 11 is a diagram illustrating an example of a confirmation screen. A misrecognition character string / correct answer character string pair, an identification number, and a check box for specifying whether to register are included for each misrecognition pattern candidate. The user can select the erroneous recognition pattern to be finally registered depending on whether or not the check box is checked. After selecting all the erroneous recognition patterns to be registered, the corresponding OK recognition pattern candidate data is stored in the storage device 50 as the erroneous recognition pattern data 53 by pressing the “OK” button. If the user's consent is not obtained, the process can be canceled by pressing “Cancel”.

<Summary>
In the present embodiment, character strings included in the sample electronic document data 51 and the post-OCR sample document data 52 are separated, and a recognition error is determined using the correct character string and the post-OCR character string separated from each other as a comparison unit. Next, the correct character string is segmented into predetermined word units, and among the segmented words, words including characters whose error recognition flag is determined to be 1 by the above-described matching processing are registered as error recognition pattern candidates. Finally, words that partially or completely match another word registered in the Japanese dictionary data and business dictionary data are deleted from the erroneous recognition pattern candidates.

  By executing such processing, it is possible to automatically collect misrecognition patterns when OCR is applied. Moreover, it is possible to prevent overlearning that a correct character string is replaced by filtering the collected erroneous recognition patterns. As a result, a uniform replacement dictionary can be created regardless of the skill of the user, and the cost of creating a replacement dictionary by the user can be greatly reduced.

  Moreover, in this embodiment, a matching process part performs misrecognition determination by setting the misrecognition flag for each character which comprises a correct character string by contrasting the character string after OCR.

  Further, when the character constituting the correct character string does not exist in the character string after OCR, the matching processing unit sets the erroneous recognition flag of the character constituting the correct character string to “1” to correspond to the correct character. If there are a plurality of characters after OCR, the misrecognition flag of one character constituting the correct character string is set to “1”.

  Thereafter, the analysis processing unit corrects the correct character string in units of words included in the Japanese dictionary data in which the Japanese words stored in the storage device are registered and the business word dictionary data in which the words used in the business are registered. Are removed from the erroneous recognition pattern candidates.

  By executing such processing, it is possible to register in the replacement dictionary in units of words and to prevent registration of words that may cause overlearning in the replacement dictionary. Since overlearning frequently occurs when registering in the replacement dictionary in units of characters, in principle, it is impossible to register in the replacement dictionary in units of characters except for some characters. For example, if “day” is misrecognized as “eyes” and it is registered in the replacement dictionary, all the words including “eyes” are replaced with “days”.

  In this embodiment, since the misrecognition pattern is grasped in units of words including the misrecognized character and the misrecognition pattern is registered in the replacement dictionary, overlearning can be suppressed. Moreover, among the recognized misrecognition patterns, misrecognition patterns that may cause overlearning to other words are deleted, and registration to the replacement dictionary is prevented, so that overlearning can be further suppressed. it can.

  The present invention can also be realized by a program code of software that realizes the functions of the embodiments. In this case, a storage medium in which the program code is recorded is provided to the system or apparatus, and the computer (or CPU or MPU) of the system or apparatus reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiments, and the program code itself and the storage medium storing the program code constitute the present invention. As a storage medium for supplying such program code, for example, a flexible disk, CD-ROM, DVD-ROM, hard disk, optical disk, magneto-optical disk, CD-R, magnetic tape, nonvolatile memory card, ROM Etc. are used.

  Also, based on the instruction of the program code, an OS (operating system) running on the computer performs part or all of the actual processing, and the functions of the above-described embodiments are realized by the processing. May be. Further, after the program code read from the storage medium is written in the memory on the computer, the computer CPU or the like performs part or all of the actual processing based on the instruction of the program code. Thus, the functions of the above-described embodiments may be realized.

  Also, by distributing the program code of the software that realizes the functions of the embodiment via a network, the program code is stored in a storage means such as a hard disk or memory of a system or apparatus, or a storage medium such as a CD-RW or CD-R And the computer of the system or apparatus (or CPU or MPU) may read and execute the program code stored in the storage means or the storage medium when used.

DESCRIPTION OF SYMBOLS 10 ... Central processing unit 20 ... Data memory 21 ... Character string comparison data 22 ... False recognition pattern candidate data 30 ... Input / output device 31 ... Keyboard 32 ... Display device 33- ..Pointing device 34 ... Scanner 40 ... Error recognition pattern collection program 41 ... Matching processing unit 42 ... Analysis processing unit 50 ... Storage device 51 ... Sample electronic document data 52 ... Sample document data 53 after OCR ... Misrecognition pattern data 54 ... Japanese dictionary data 55 ... Business word dictionary data 56 ... Corrected electronic document data

Claims (4)

A document processing apparatus that automatically generates a replacement dictionary for correcting an OCR misrecognized character string,
A correct character string cut out from sample electronic document data obtained by imaging a business document and a post-OCR character string cut out from post-OCR sample document data obtained as a result of performing OCR on the sample electronic document data And a matching processing unit for determining misrecognition,
An analysis processing unit that divides the correct character string into predetermined word units and registers a word including a character that is determined to be erroneously recognized by the matching processing unit among the divided words as an erroneous recognition pattern candidate;
A word that partially or completely matches a word included in Japanese dictionary data in which a Japanese word stored in a storage device is registered and a business word dictionary data in which a word used in a business is registered Filtering by deleting from the candidate, filtering, and storing the erroneous recognition pattern candidate after the filtering in the storage device as an erroneous recognition pattern;
A document processing apparatus comprising:   The said matching process part determines the said misrecognition by setting the misrecognition flag for each character which comprises the said correct character string by contrasting the said character string after OCR. Document processing apparatus described in 1.   When the character constituting the correct character string does not exist in the post-OCR character string, the matching processing unit sets a false recognition flag of the character constituting the correct character string to “1”, and the correct character 3. The document processing apparatus according to claim 2, wherein when there are a plurality of post-OCR characters corresponding to the character string, an erroneous recognition flag of one character constituting the correct character string is set to “1”.   The analysis processing unit includes the correct character in units of words included in Japanese dictionary data in which Japanese words stored in the storage device are registered and business word dictionary data in which words used in business are registered. The document processing apparatus according to claim 1, wherein the column is divided.

Images