JP2006185342A - Information processor, method and program for classifying character string, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To correctly acquire a character string of each part of a business card, etc., as correct information even when a name, etc., closely resembling a geographic name is included without the needs of designating information indicated by the character string of the each part written in the business card, etc., beforehand or at the time of recognition. <P>SOLUTION: When classifying each character string on the business card read by a scanner 200 by semantic attributes, a CPU of this information processor 100 analyzes the character string with reference to DBs 105 to 107 for the each character string, respectively attaches a score indicating the likelihood of assigning the character string to each of the respective semantic attributes by a plurality of score attaching methods, selects any of combination patterns on the basis of the total values of respective attached scores for each assignment combination pattern of the each character string to each of the semantic attributes and classifies the each character string into each of the semantic attributes with the selected combination pattern. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an information processing apparatus, a character string classification method, a program, and a recording medium that classify semantic information classification represented by character recognition obtained as a result of character recognition and character recognition.

  By specifying the image in a certain format, that is, what kind of meaning character string is described in which position is registered in advance, the user can specify the name at another position and another specific Conventionally, there is a system for recognizing that an address is described at the position of.

Patent Document 1 utilizes the fact that each line of a business card includes a word (a characteristic word such as a person's name or “stock”) serving as an identifier representing a specific classification when performing business card character recognition. Thus, a character recognition device that gives a specific classification to information in a business card is described.
Japanese Patent Laid-Open No. 10-78997

  However, since the conventional system uses the position information of the character string appearing in the business card, it is necessary to register the character string appearance position pattern in advance. If this is not the case, the character string is divided into blocks, and the user is forced to specify which is the name or address, which is very complicated.

  Further, for example, in the case of a horizontal type business card, there is a case in which there is a company name from the upper left to the middle, and a large name is printed in the middle and approximate position information such as a name printed is registered in advance.

  However, in this case, when a business card with a new pattern appears, there is a problem that the business card cannot be recognized unless a new character string appearance position pattern is registered.

  Further, in the character recognition device described in the above-mentioned Patent Document 1, since the semantic attribute is determined for each character string, for example, “Akita Prefecture” or “Yoshinomura Masaru” is used. In the case of a name very similar to the place name, there is a high possibility that the name will be mistakenly recognized as a place name.

  Therefore, even if the character recognition technology of the above-mentioned patent document 1 is used, in order to correctly classify which character string is an address and which other character string is a name, after all, a user check is performed. There was a problem that it was indispensable and very complicated.

  The present invention has been made to solve the above problems, and an object of the present invention is to provide each character in a character string group composed of a plurality of character strings assigned to any one of a plurality of semantic attributes. When classifying a column for each semantic attribute, for each of the character strings, a score indicating the likelihood of assignment of the character string to each semantic attribute is analyzed by a plurality of scoring methods. Assigning and selecting any one of the combination patterns based on the total value of the assigned scores for each combination pattern assigned to each semantic attribute of each character string, and in the selected combination pattern, By classifying each character string into the respective semantic attributes (that is, classifying the semantic attributes of the respective character strings in consideration of the entire character string group), the character strings of the respective parts described in the business card or the like are changed. Information processing apparatus, character string classification method, program, and recording medium that can acquire the character string of each part such as a business card as correct information, without needing to specify in advance whether the information is indicated or during recognition That is.

  The present invention provides an information processing apparatus that classifies each character string in a character string group composed of a plurality of character strings assigned to any one of a plurality of semantic attributes for each semantic attribute, and analyzes the character string. One or a plurality of score assigning means for assigning a score indicating the likelihood of assignment of the character string to each semantic attribute, and each score assigning means for each combination combination pattern of each character string to each semantic attribute And a classifying unit that selects any one of the combination patterns based on the total value of the scores assigned by the method and classifies the character strings into the semantic attributes according to the selected combination pattern. And

  According to the present invention, when each character string in the character string group composed of a plurality of character strings assigned to any one of the plurality of semantic attributes is classified for each semantic attribute, for each character string, Analyzing the character string and assigning a score indicating the likelihood of assignment of the character string to each semantic attribute by a plurality of scoring methods, and assigning each character string to each semantic attribute Since any one of the combination patterns is selected based on the total value of the assigned scores, and the character strings are classified into the respective semantic attributes according to the selected combination pattern, it is described on a business card or the like. It is not necessary to specify what information the character string of each part indicates in advance or at the time of recognition, and the character string of each part such as a business card can be acquired as correct information. An effect such as that.

  Therefore, since the character string is specified by considering the entire character string group, not by the position in the document such as a business card, the character string semantic attribute is specified. For example, “Akita Prefecture” or “Akitagata” Even names that are very similar to place names, such as “Yoshinomura Masaru”, can be classified as names without misrecognizing them as place names.

  As a result, when passing to a subsequent application such as, for example, a business card database, it is possible to save labor by manually checking and specifying or correcting whether a character string is a name or an address.

[First Embodiment]
Hereinafter, details of the present invention will be described with reference to the drawings.

  FIG. 1 is a system configuration diagram showing an example of a system to which the information processing apparatus according to the first embodiment of the present invention can be applied.

  In FIG. 1, reference numeral 100 denotes an information processing apparatus as a character string recognition apparatus. The information processing apparatus 100 OCR recognizes a character string on image data input from an image input apparatus 200 (for example, a scanner or a digital camera). Then, the information processing apparatus 100 determines character string information based on dictionary data stored in the dictionary DB 106 from the OCR recognition result. Further, the information processing apparatus 100 determines the character string information from the character string information determined by the dictionary according to the rules stored in the rule DB 105. Furthermore, the information processing apparatus 100 uses character string information by using establishment calculation such as HMM (Hidden Markov Model) based on appearance probability data stored in the probability information DB 107 from character string information determined by the rule. Determine information.

  FIG. 2 is a block diagram illustrating an example of a hardware configuration of the information processing apparatus illustrated in FIG.

  In FIG. 2, reference numeral 101 denotes a CPU, which is stored in a ROM 103 or a hard disk (HD) 104 (any other storage device such as a flexible disk, CD-ROM, DVD-ROM, etc.) 104. By loading and executing the program according to the present invention on the RAM 102, the character string recognition method of the present invention is realized and the entire computer is controlled. The RAM 102 is used as a work area for the CPU 101.

  Reference numeral 109 denotes a communication interface that enables data transmission / reception with the image input apparatus (for example, a scanner or a digital camera) 200 via a network or other communication medium.

  An input device 110 corresponds to a pointing device such as a keyboard or a mouse. Reference numeral 111 denotes a display device that corresponds to a monitor such as a CRT, LCD, or SED.

  Note that the rule DB 105, the dictionary DB 106, and the probability information DB 107 illustrated in FIG. 1 are constructed in the HD 104 of the information processing apparatus 100. Each function (OCR function, each character string determination function) of the information processing apparatus 100 is materialized by the CPU 101 of the information processing apparatus 100 loading and executing a program stored in the HD 104 on the RAM 102. Is done.

  In addition, the information processing apparatus of the present invention may be configured by a computer, or may be configured by a scanner, a multifunction machine, or the like.

  Hereinafter, a character string recognition method in the information processing apparatus of the present invention will be described with reference to FIGS.

  FIG. 3 is a flowchart showing an example of a first control processing procedure in the information processing apparatus of the present invention, and corresponds to the character string recognition processing. Note that the processing of this flowchart is realized by the CPU 101 of the information processing apparatus 100 loading a program stored in the HD 104 into the RAM 102 and executing it. In the figure, S001 to S008 indicate each step.

  Hereinafter, a case where a character string on a business card is recognized will be described as an example.

  First, image data (an OCR target image) 301 of a document (here, a business card) read by an image input device 200 such as a scanner or a digital camera is input, or a character string of image data stored in the HD 104 When the recognition process is instructed, the CPU 101 starts the process of this flowchart.

  In step S001, the CPU 101 reads character string recognition target image data 301 from a memory (such as HD or RAM), performs OCR processing on the read image data 301, and extracts a character string and its position information (extracted character). Column group 302). For example, a character string is extracted so that a character string in the same row if the business card is horizontal and a character string in the same column if the business card is vertical is treated as one character string. However, if there is a certain distance or more, the character string is divided so that it is handled as a separate character string.

  For example, when the OCR processing of the business card 500 as shown in FIG. 4 is performed, character strings are extracted so that “section manager” and “Yamagata prefecture” (Yamagata Agata (person name)) are treated as different character strings. To do. However, the fact that “section manager” and “Yamagata prefecture”, which are divided into character strings, are on the same line, is stored in the memory for use when applying the rules described later. In the example shown in FIG. 4, seven character strings 501 to 507 are extracted from the business card 500.

  FIG. 4 is a schematic diagram showing an example of a character string recognition target image and a character string extracted from the character string recognition target image in the information processing apparatus according to the present invention, and particularly corresponds to a case where a character string is extracted from a business card. .

  Then, the CPU 101 stores the character string acquired by the OCR process as an extracted character string 302 in a memory (such as HD or RAM).

  In the example shown in FIG. 4, “Yamamoto Electric Co., Ltd.” 501, “XX Business Headquarters” 502, “△△ Research and Development Department First Development Section” 503, “Manager” 504, “Yamagata Prefecture (however,“ Manager ”) Are stored in the same row) ”505”, “1-2-3 Miyoshi 1-2-3, Tokyo, 111-0011” 506, “TEL 03-1234-5678” 507 are stored.

  Next, the CPU 101 performs the processes of steps S002 to S006 on all the character strings extracted in step S001, and determines the semantic attribute for each character string.

  The semantic attribute indicates the classification of information such as “name”, “address”, “telephone number”, “position” (here, the classification of information described in the business card), and is extracted in step S001. In addition, it is determined which of these semantic attributes each character string belongs to. This determination process is executed for all the character strings obtained in step S001. Hereinafter, it demonstrates in detail by step S002-S006.

  In step S002, the CPU 101 reads (focuses on) one of the character strings extracted in step S001, searches the dictionary DB 106 shown in FIG. 5 using the focused character string as a keyword, and based on the search result. A score is assigned to the character string under consideration and stored in the memory.

  FIG. 5 is a diagram showing an example of the dictionary DB in the information processing apparatus of the present invention.

  As shown in FIG. 5, the dictionary DB 106 includes items of a character string 401, a reading 402, a classification 403, and a score (likelihood) 404.

  If this dictionary DB 106 is searched using, for example, “Yamagata Prefecture” as a key, it is found that the classification 403 is “address (part of)” or “person name”. Further, it is understood that the score 404 is “80” in the case of an address and “20” in the case of a person name. For this reason, “Yamagata Prefecture” is regarded as having a high possibility of being an address only by this processing. In this stage, not the final decision but the possibility that “Yamagata Prefecture” is an address and the possibility of a person's name are stored in the memory together with the score.

  The description returns to the flowchart of FIG.

  Next, in step S004, the CPU 101 assigns a score by applying all the rules of the rule DB 103 to the character string under attention, and stores it in the memory.

  FIG. 6 is a diagram showing an example of the rule DB in the information processing apparatus of the present invention.

  As shown in FIG. 6, the rule DB 105 includes items of rule number 601, pattern portion 602, score assigning portion 603, classification 604, and score 605.

  Note that “◯” in FIG. 6 represents a character or a character string. For example, “XXXXX3f” in front of “KK” of “XXXXXX” in the pattern part 602 of Rule 2 is 4 corresponds to “Yamamoto Electric” of “Yamamoto Electric Co., Ltd.” shown in FIG. The rule pattern portion shown in FIG. 6 is assumed to be described using a regular expression, but may be in a description format according to another rule-driven engine technology.

  Also, as in the rule 2 score assigning section 603, “whole character string”, that is, “Yamamoto Denki Co., Ltd.” as a whole, “company name” as in the classification 604 of rule 2, As shown in the score 605, the score “100” is given.

  Since “Yamagata Prefecture” is behind “Manager”, the pattern part and score assigning part of Rule 1 are given a score “100” as the classification “name”.

  Note that it is possible to determine that “section manager” is a title by being assigned as an attribute based on the dictionary search result shown in FIG. However, since the fact that “Yamagata Prefecture” is registered as a person name in the dictionary is not used, the rule of this example is also useful for an unknown person name that is not registered in the dictionary.

  Likewise, “1-2-3 Miyoshi 1-2-3, Tokyo” corresponds to rule 9, and therefore a score “100” is given as “address”.

  In the case of “Yamamoto Electric Co., Ltd.”, the score “100” is given as “company name” because it corresponds to rule 2 as described above.

  Furthermore, since “TEL 03-1234-5678” corresponds to rule 4, a score “100” is assigned as “telephone number”.

  Next, in step S005, the CPU 101 assigns a score using a probability model to the character string of interest and holds it in the memory.

  A probability model (HMM: hidden Markov model) described in Japanese Patent Application Laid-Open No. 2004-46775 can be used as a known technique to acquire the probability that the character string appears.

  For example, if the “semantic attribute” of a character string described on a business card is considered as an internal state, and the “character string” described on the business card is considered as a symbol that can be observed from the outside, the generation process of the character string group described on the business card Can be approximated by a hidden Markov model.

  If this hidden Markov model expresses the format of a business card (appearance pattern of a character string corresponding to each semantic attribute), the probability of transition from one semantic attribute to another (for example, “Name” followed by “Name”) , The probability that “phone number” is written after “address”, etc.), the appearance probability of a character string for each semantic attribute (for example, the appearance probability of “Yamagata Prefecture” in “name”, “location name”) For example, the appearance probability of “Yamagata Prefecture”.

  Thus, by using the hidden Markov model, the naturalness of the connection of the “character string” in the description of the business card can be expressed by the magnitude of the state transition probability.

  In the present embodiment, using such a probability model, the probability of occurrence of the character string of interest is obtained, and the maximum value of the probability becomes a score “100” (for example, if the score is a full score of “100”). Such a coefficient is obtained, and the appearance probability of the character string under consideration is normalized to a score of “100”.

  Note that the probability model applicable to the present invention is not limited to the hidden Markov model, and any probability model may be used as long as it is a model that probabilistically predicts the appearance pattern of the character string corresponding to each semantic attribute. May be.

  Moreover, you may comprise so that the process of the score provision by this probability model may not be performed as an option.

  Next, in step S006, a score calculation process for adding the scores assigned in steps S003 to S005 to the character string under attention is performed. Hereinafter, a description will be given with reference to FIG.

  FIG. 7 is a diagram for explaining score calculation processing in the information processing apparatus of the present invention.

  For example, if the character string under consideration is “Yamagata Prefecture”, score “80” as “address” and score “20” as “name” are given to “Yamagata Prefecture” by the dictionary DB 106. Yes. For “Yamagata Prefecture”, the rule DB 105 assigns a score “100” as “name”. Here, in order to simplify the explanation, it is assumed that no score using a probability model is given.

  In this case, the sum by the score calculation is score “80” as “address” and score “120 (= 20 + 100)” as “name”.

  Note that the information processing apparatus of the present invention does not determine that “Yamagata Prefecture” is a person's name only at this point of time, but applies the optimal allocation process shown in step S008, which will be described later, to the score calculation result to obtain the final result. Character recognition result is determined.

  The description returns to the flowchart of FIG.

  In step S007, the CPU 101 determines whether or not the processing in steps S002 to 006 has been completed for all the character strings in the extracted character string group 302. If it is determined that the processing has not been completed, the process proceeds to step S002. Return processing and process the next character string.

  On the other hand, if the CPU 101 determines in step S007 that the processes in steps S002 to 006 have been completed for all the character strings in the extracted character string group 302, the process proceeds to step S008.

  FIG. 8 shows dictionary-based score assignment results for all character strings in the extracted character string group 302. FIG. Further, FIG. 9 shows a result of assigning a rule-based score to this result and performing score calculation for all the character strings in the extracted character string group 302.

  As shown in FIGS. 8 and 9, the score calculation result for “Yamamoto Electric Co., Ltd.” has a score “20 (dictionary base)” (901 in FIG. 9) as “name” and a score as “company name”. “100 (dictionary base) +100 (rule base)” (903 in FIG. 9).

  For the “section manager”, the score “100 (dictionary base)” is given as the “title”.

  Furthermore, for “Yamagata Prefecture”, the score “20 (dictionary base) +100 (rule base)” (902 in FIG. 9) as “name” and the score “80 (dictionary base)” as “address” (FIG. 9). 904).

  Further, for “〒 111-0011 Tokyo, XX-ku Miyoshi 1-2-3”, the score is “80 (dictionary base) +100 (rule base)” (905 in FIG. 9) as “address”.

  Further, for “TEL 03-1234-5678”, the score is “100 (rule base)” (906 in FIG. 9) as “telephone number”.

  As described above, in this example, “Yamamoto Electric Co., Ltd.” and “Yamagata Prefecture” indicate that the character string may have a plurality of semantic attributes. Any one of the attributes is determined by the optimum allocation process shown in FIG.

  FIG. 10 is a flowchart showing an example of the second control processing procedure in the information processing apparatus of the present invention, and corresponds to the optimum allocation processing shown in step S008 of FIG. Note that the processing of this flowchart is realized by the CPU 101 of the information processing apparatus 100 loading a program stored in the HD 104 into the RAM 102 and executing it. In the figure, S1101 to S1110 indicate each step.

  First, in step S1101, the CPU 101 determines that one character string in the extracted character string group 302 has two or more semantic attributes (characters having a plurality of categories other than the score “0” in the score calculation result shown in FIG. 9). In step S1110, the semantic attribute corresponding to each character string in the extracted character string group 302 (category other than score “0”) is associated in step S1110. Then, the process of this flowchart is terminated using the associated “character string” and “semantic attribute” pair as output information.

  On the other hand, when the CPU 101 determines in step S1101 that one character string in the extracted character string group 302 has two or more semantic attributes (a character string having a plurality of categories other than the score “0”). In step S1102, the CPU 101 calculates a total score based on a combination pattern of character string groups. That is, a pattern having a total score as shown in FIG. 11 is extracted, and the total score of each pattern is calculated.

  FIG. 11 is a diagram showing an example of calculating the total score for each pattern in the information processing apparatus of the present invention.

  As shown in FIG. 11, four patterns A, B, C, and D are extracted from the score calculation result shown in FIG. It should be noted that semantic attributes such as telephone numbers and job titles that are already associated with character strings having only one semantic attribute are omitted.

  In addition, the character strings are arranged so that one character string is not assigned to a plurality of semantic attributes in the same pattern.

  Then, a total score is calculated for each pattern.

  As a result, in the example shown in FIG. 11, the CPU 101 determines in step S1103 that the C pattern having the highest score is most appropriate, and selects the C pattern for the relationship between the semantic attribute and the character string.

  In step S1104, the CPU 101 associates the character string of the pattern selected in step S1103 with the semantic attribute, and uses the associated “character string” and “semantic attribute” as output information. Exit.

  Therefore, in the end, “Yamamoto Electric Co., Ltd.” is “Company Name”, “Manager” is “Title”, “Yamagata Prefecture” is “Name”, “1-2-1 Miyoshi, ○ -ku, Tokyo, 111-0011, Japan” “3” is recognized as “address”, and “TEL 03-1234-5678” is recognized as “phone number”.

  In this way, the business card shown in FIG. 4 is recognized correctly.

  In the example of the business card shown in FIG. 4, since the title “section manager” is listed in front of “Yamagata Prefecture”, the score of the “name” attribute increases when the score is given by the rule DB 105, and “Yamagata” It had a great influence on the recognition of “prefecture” as a “name” attribute.

  Therefore, with reference to FIGS. 12 to 14 shown below, consider the case where the title “section manager” in front of “Yamagata Prefecture” is not described.

  FIG. 12 shows another example of dictionary-based score assignment results for all character strings in the extracted character string group 302. Further, FIG. 13 shows another example of the result of assigning a rule-based score to this result and calculating the score for all the character strings in the extracted character string group 302.

  As shown in FIG. 12 and FIG. 13, the score calculation result for “Yamamoto Electric Co., Ltd.” has a score “20 (dictionary base)” as “name” and a score “100 (dictionary base) as“ company name ”. +100 (rule base) ”.

  The score calculation result for “section manager” is score “100 (dictionary base)” as “title”.

  Furthermore, the score calculation results for “Yamagata Prefecture” are score “20 (dictionary base)” as “name” and score “80 (dictionary base)” as “address” (there is no title, so rule DB 105 Does not apply).

  Also, the score calculation result for “〒 111-0011 1-2-3 Miyoshi 1-2-3 Tokyo” is score “80 (dictionary base) +100 (rule base)” as “address”.

  Furthermore, for “TEL 03-1234-5678”, the score is “100 (rule base)” as “telephone number”.

  The total score pattern is shown in FIG.

  FIG. 14 is a diagram showing another example of calculating the total score for each pattern in the information processing apparatus of the present invention.

  As shown in FIG. 14, four patterns of A, B, C, and D are extracted from the score calculation result shown in FIG.

  As a result of calculating the total score for each pattern, the C pattern has the highest score even in the example shown in FIG.

  Therefore, even if the title “Manager” is not listed before “Yamagata Prefecture”, “Yamamoto Electric Co., Ltd.” is “Company Name”, “Manager” is “Title”, and “Yamagata Prefecture” is “ "Name", "〒 111-0011 Miyoshi 1-2-3, Tokyo, 1-2-3" will be correctly recognized as "address" and "TEL 03-1234-5678" will be correctly recognized as "phone number".

  As described above, a dictionary, a rule, or a probabilistic process for specifying the meaning represented by the character string on the image data is used, and a plurality of ( Taking into account the semantic attributes of the character string (of the entire business card), it is automatically determined whether each part on the image data describes the information corresponding to which semantic attribute (which belongs to which semantic information category) It can be judged (without human intervention).

  Therefore, an excellent character string recognition environment that saves the necessity of manually checking and specifying or correcting whether a character string is a name or address when passing it to a subsequent application such as a business card database. Can be built.

  Note that scoring by rules and scoring by probabilistic processing such as HMM are optional, and it is possible to construct a system that does not use one or both of them.

  For example, when not using a rule, the CPU 101 controls to skip step S004 in FIG. 3 and to skip step S005 in FIG. 3 when not using probabilistic processing.

  Note that the case where only dictionary-based score assignment is performed using the example described above (when the rules and the establishment process are not applied) will be considered below.

  In this case, the same patterns A, B, C, and D as those described above are extracted from the dictionary-based score assignment example shown in FIG.

  However, the total score of pattern A is “100 = 20 + 80”, the total score of pattern B is “100 = 20 + 80”, the total score of pattern C is “200 = 20 + 100 + 80”, and the total score of pattern D is “180 = 100 + 80” Become.

  Therefore, when only the dictionary-based score is given (when the rule and the establishment process are not applied), the pattern C is still determined as the highest score and correctly recognized.

[Second Embodiment]
In the first embodiment described above, a configuration has been described in which OCR recognition is performed from image data that is a character string recognition target, and each semantic attribute of the character string that has been OCR-recognized is determined. However, the processing of steps S002 to 008 in FIG. The execution is not limited to the OCR result of the image data, and any character string data may be used as long as it is a plurality of character string data.

  For example, you may apply to a CSV file as shown in FIG.

  FIG. 15 is a diagram illustrating an example of a CSV file including a character string to be recognized in the information processing apparatus of the present invention.

  As shown in FIG. 15, in the first line, this CSV file has an address in the first column as shown in 1201 and a name in the second column as shown in 1202 (where 1202 is “Akita Prefecture (Agata) ) ") The company name is in the third column and the phone number is in the fourth column.

  In the second row, a telephone number is entered in the third column as indicated by 1203.

  Further, in the third row, as shown in 1205, the name (nomination “XX field × child”) is shown in the first column, and the address is entered in the second column as shown in 1204.

  As described above, even if the column items are different for each row or a name similar to an address is entered, the processing in steps S002 to S008 shown in FIG. Similar to the example of the business card shown in the first embodiment, the data of each column in each row can be classified by semantic attributes (semantic information classification).

  In addition, all the structures which combined any or all of the modifications shown in each embodiment mentioned above are also included in this invention.

  It should be noted that the configuration and contents of the various data described above are not limited to this, and it goes without saying that the various data and configurations are configured according to the application and purpose.

  Although one embodiment has been described above, the present invention can take an embodiment as, for example, a system, apparatus, method, program, or recording medium, and specifically includes a plurality of devices. The present invention may be applied to a system including a single device.

  As described above, when each character string in a character string group composed of a plurality of character strings assigned to any one of a plurality of semantic attributes is classified for each semantic attribute, for each character string, the character A combination of assigning each character string to each semantic attribute by assigning a score indicating the likelihood of assignment of the character string to each semantic attribute by a plurality of methods by one or a plurality of methods. One of the combination patterns is selected based on the total score given for each pattern, and each character string is classified into each semantic attribute based on the selected combination pattern (that is, a character string). By classifying the semantic attributes of each character string in consideration of the entire group), it is possible to confirm in advance or what information the character string of each part described on a business card or the like indicates. Sometimes it is not necessary to specify, it is possible to get a string of each part of the business card as the correct information.

  Therefore, since the character string is specified by considering the entire character string group, not by the position in the document such as a business card, the character string semantic attribute is specified. For example, “Akita Prefecture” or “Akitagata” Even names that are very similar to place names, such as “Yoshinomura Masaru”, can be classified as names without misrecognizing them as place names.

  As a result, when passing to a subsequent application such as a business card database, for example, it is possible to save labor by manually checking and specifying or correcting whether a character string is a name or an address.

  The configuration of a data processing program that can be read by the character recognition device according to the present invention will be described below with reference to the memory map shown in FIG.

  FIG. 16 is a diagram for explaining a memory map of a recording medium (storage medium) that stores various data processing programs that can be read by the character recognition apparatus according to the present invention.

  Although not specifically shown, information for managing a program group stored in the recording medium, for example, version information, creator, etc. is also stored, and information depending on the OS on the program reading side, for example, a program is identified and displayed. Icons may also be stored.

  Further, data depending on various programs is also managed in the directory. In addition, when a program or data to be installed is compressed, a program to be decompressed may be stored.

  The functions shown in FIGS. 3 and 10 in this embodiment may be performed by a host computer by a program installed from the outside. In this case, the present invention is applied even when an information group including a program is supplied to the output device from a recording medium such as a CD-ROM, a flash memory, or an FD, or from an external recording medium via a network. Is.

  As described above, a recording medium in which a program code of software for realizing the functions of the above-described embodiments is recorded is supplied to the system or apparatus, and the computer (or CPU or MPU) of the system or apparatus is stored in the recording medium. It goes without saying that the object of the present invention can also be achieved by reading and executing the program code.

  In this case, the program code itself read from the recording medium realizes the novel function of the present invention, and the recording medium storing the program code constitutes the present invention.

  As a recording medium for supplying the program code, for example, a flexible disk, hard disk, optical disk, magneto-optical disk, CD-ROM, CD-R, DVD-ROM, magnetic tape, nonvolatile memory card, ROM, EEPROM, A silicon disk or the like can be used.

  Further, by executing the program code read by the computer, not only the functions of the above-described embodiments are realized, but also an OS (operating system) or the like running on the computer based on the instruction of the program code. It goes without saying that a case where the function of the above-described embodiment is realized by performing part or all of the actual processing and the processing is included.

  Furthermore, after the program code read from the recording medium is written in a memory provided in a function expansion board inserted in the computer or a function expansion unit connected to the computer, the function expansion is performed based on the instruction of the program code. It goes without saying that the case where the CPU or the like provided in the board or the function expansion unit performs part or all of the actual processing and the functions of the above-described embodiments are realized by the processing.

  In addition, the present invention may be applied to a system composed of a plurality of devices or an apparatus composed of a single device. Needless to say, the present invention can be applied to a case where the present invention is achieved by supplying a program to a system or apparatus. In this case, by reading a recording medium storing a program represented by software for achieving the present invention into the system or apparatus, the system or apparatus can enjoy the effects of the present invention.

  Furthermore, by downloading and reading out a program represented by software for achieving the present invention from a server, database, etc. on a network using a communication program, the system or apparatus can enjoy the effects of the present invention. It becomes.

1 is a system configuration diagram illustrating an example of a system to which an information processing apparatus according to an embodiment of the present invention is applicable. It is a block diagram which shows an example of the hardware constitutions of the information processing apparatus shown in FIG. It is a flowchart which shows an example of the 1st control processing procedure in the information processing apparatus of this invention. It is a schematic diagram which shows an example of the character string extracted from the character string recognition object image in the information processing apparatus of this invention, and this character string recognition object image. It is a figure which shows an example of a structure of dictionary DB in the information processing apparatus of this invention. It is a figure which shows an example of a structure of rule DB in the information processing apparatus of this invention. It is a figure explaining the score calculation process with respect to the character string shown in FIG. It is a figure which shows the score addition result of a dictionary base with respect to all the character strings of an extracted character string group. It is a figure which shows the score calculation result with respect to all the character strings of the extraction character string group. It is a flowchart which shows an example of the 2nd control processing procedure in the information processing apparatus of this invention. It is a figure which shows an example of calculation of the total score for every pattern in the information processing apparatus of this invention. It is a figure which shows the other example of the dictionary-based score provision result with respect to all the character strings of an extracted character string group. It is a figure which shows the other example of the score calculation result with respect to all the character strings of an extraction character string group. It is a figure which shows the other example of the calculation result of the total score for every pattern in the information processing apparatus of this invention. It is a figure which shows an example of the CSV file containing the character string used as recognition object in the information processing apparatus of this invention. It is a figure explaining the memory map of the recording medium (storage medium) which stores the various data processing program which can be read with the character recognition apparatus which concerns on this invention.

Explanation of symbols

200 Image input device (scanner)
100 Information processing device (character string recognition device)
101 CPU
102 RAM
103 ROM
104 HD
105 Dictionary DB
106 Rule DB
107 Probability information DB
109 Communication Interface 110 Input Device 111 Display Device

Claims (10)

In an information processing apparatus for classifying each character string in a character string group composed of a plurality of character strings assigned to any one of a plurality of semantic attributes for each semantic attribute,
One or a plurality of score assigning means for analyzing the character string and assigning a score indicating the likelihood of assignment to each semantic attribute of the character string;
Based on the total value of the scores assigned by each score assigning unit for each combination pattern assigned to each semantic attribute of each character string, select any one of the combination patterns, and the selected combination pattern Classifying means for classifying the character strings into the semantic attributes;
An information processing apparatus comprising: The one or more score assigning means includes:
First storage means for storing dictionary data in which a character string and each semantic attribute are associated with each other, and a score for each semantic attribute of the character string based on the dictionary data A dictionary score assigning means comprising an assigning means,
The information processing apparatus according to claim 1, further comprising: The one or more score assigning means includes:
Second storage means for storing rule data in which a character string rule that associates a character string with each semantic attribute has a score, and a score for each semantic attribute of the character string based on the rule data A dictionary score giving means composed of a second giving means,
The information processing apparatus according to claim 1, further comprising: The one or more score assigning means includes a third assigning means for assigning a score for each semantic attribute of the character string based on a probability model corresponding to a format of the character string group. Granting means
The information processing apparatus according to claim 1, further comprising: The character string classification device includes:
3. The information processing apparatus according to claim 1, wherein a score is calculated when one character string has two or more semantic attributes.   The information processing apparatus according to claim 1, wherein each character string is a character string read from image data. Each of the character strings is a character string read from image data corresponding to a business card,
The information processing apparatus according to claim 1, wherein each semantic attribute includes a name, a title, an address, a telephone number, and a company name. In the character string classification method in the information processing apparatus capable of classifying each character string in a character string group composed of a plurality of character strings assigned to any of a plurality of semantic attributes for each semantic attribute,
For each of the character strings, one or a plurality of scoring steps for analyzing the character string and giving a score indicating the likelihood of assignment to the semantic attributes of the character string;
Based on the total value of the scores assigned in each score step for each combination pattern assigned to each semantic attribute of each character string, select any one of the combination patterns, with the selected combination pattern, A classifying step for classifying the character strings into the semantic attributes;
A character string classification method characterized by comprising:   The program for a computer to perform the character string classification | category method described in Claim 8.   A recording medium storing a computer-readable program for executing the character string classification method according to claim 8.

Images