JP5318742B2 - Character recognition method, character recognition device, and character recognition program - Google Patents

Description

  The present invention relates to a character recognition method, a character recognition device, and a character recognition program, and in particular, a character recognition method, a character recognition device, and a character recognition program for recognizing characters in a continuous form in which data portions in a ruled line form are not regularly arranged. About.

  In a financial institution or the like, a character recognition device that recognizes characters on a form is used to automate the input of characters printed on the form. In this character recognition device, in order to increase the accuracy of character recognition, the heading part and the data part printed on the form are defined in what position and in what order, and the character string is based on the definition information. Recognize.

FIG. 24 is a diagram showing an example of a continuous writing form.
Here, a transfer request form is shown as an example of the continuous entry form. According to this form 100, at the top thereof, this is a “transfer request form”, the transfer person is “A system called“ Co., Ltd. ”, and the designated transfer date is described.

  In the main part of the form 100, details 101 are described in a table format. The description 101 is composed of a heading portion 102, a data portion 103, and a heading footer 104. The heading portion 102 constitutes a heading header of one line, and the data portion 103 is composed of a plurality of detail lines. In the header portion 102, items of “bank name”, “branch name”, “subject”, “account number”, “recipient name”, and “amount” are set. In the data portion 103, character string data corresponding to each item of the heading portion 102 is continuously written for the number of transfer destinations. The heading footer 104 describes the “transfer total” of the heading and its numerical data.

  In the form 100, each item of the heading portion 102 and the character string of the data portion 103 corresponding to the item are aligned and printed in accordance with predetermined definition information. In the definition information, position information and character information about the item of the heading portion 102 and the data portion 103 corresponding thereto are defined, and the character string layout and character string attributes in the form 100 are known. It is like that.

  As described above, in order to recognize the data portion 103 of the form 100, it is necessary to define the position information of the data corresponding to the heading item. Therefore, only the form in which the definition information is registered can recognize characters. . When the format of the form is changed, the definition information of the registered form needs to be changed each time.

  On the other hand, there is known a character recognition device that can flexibly cope with a format change without requiring registration of definition information (see, for example, Patent Document 1). According to the character recognition device described in Patent Document 1, the character information of the form is not created in advance, and the character layout of the form is detected using the regularity of the arrangement of the character strings of the form. The character string of the data part is recognized from the layout, the order of the headings, and the definition information of the character strings corresponding to the headings.

  In addition, in a form in which a character string is printed in a ruled line frame, a heading part and a data part corresponding to the headline part may be described in a plurality of lines in each ruled line frame. In such a case, a character string that is folded back into multiple lines in the middle of each ruled line frame is described, so when recognizing characters, multiple headline items and character strings in each line listed in multiple lines (See, for example, Patent Document 2).

Japanese Patent Laid-Open No. 11-161736 JP 2000-29983 A

  Since the conventional character recognition device recognizes the character string of the form by utilizing the regularity of the arrangement of the character string of the form, the character string of the form includes the heading item and the character string of the corresponding data portion. Are arranged in a line, and data is associated with the headings in the order of the headings. However, in recent years, in addition to the above-mentioned regular forms in which the headline and the character string of the data part are aligned, forms brought in from customers (hereinafter referred to as end users) such as financial institutions (hereinafter referred to as private forms) There is also a need to perform character recognition processing in the same way.

  Since the private form is a form created by the end user with a free concept, the heading and the character string of the data part are not necessarily regularly arranged. For example, in a heading line and a detail line, a heading item and a character string corresponding to the heading item may not be arranged in a line, and the character string may be printed with a minimum interval. Further, when the data corresponding to the headline is, for example, a character string in which the full name is combined with a blank character, the number of items in the headline and the number of character strings do not match in the detail line.

  Furthermore, even if the heading and the character string of the data portion are regularly arranged, one data may be printed as a character string folded in the same column, and only the detail line may be displayed in two lines. As a result, there may be a mixture of one detail line and multiple detail lines.

  Non-standard private forms are the principle of determining the range of a statement based on whether or not the structure of the character strings in two adjacent rows is the same when detecting the character string layout of the form. If the data string for the heading is not aligned, or if the data for the heading does not exist continuously due to wrapping, the layout of the text in the form cannot be accurately detected, and the character recognition of the character string There was a problem that it could not be processed.

  The present invention has been made in view of such a point, and even in a form in which the arrangement of the character string of the data part with respect to the heading part is not regularly arranged, the layout of the character of the form is analyzed, and the form data An object of the present invention is to provide a character recognition method, a character recognition device, and a character recognition program for recognizing a partial character string.

  In the present invention, in order to solve the above problem, in a character recognition method for recognizing a character string of a form having a character string of a heading part and a character string of a data part, the character string extracted from the image data of the form Extract the logical lines in which the character strings are arranged in the horizontal direction from the positional relationship on the form, recognize the character strings in the logical lines, extract the heading lines in which the headline character strings of the details are arranged, and the total number of the headline character strings Detail lines with the same ratio of the number of character strings in the data part are extracted, and the heading item in the heading part is associated with the character string in the data part on a one-to-one basis, and the character string in the data part corresponding to the heading is fixed. Identifies long data and associates it with the heading item. The character string of the data portion corresponding to the heading has a variable length and the first variable length data that does not wrap the character string is identified and associated with the heading item. The headline The heading item is specified by specifying the second variable length data in which the character string of the data portion has a variable length and the character string may be folded, and the heading and the number of character strings of the data portion correspond one-to-one. The character string of the data portion corresponding to the heading has a variable length, and the character string may be folded, and the third variable length in which the number of character strings of the heading and the data portion corresponds 1 to N Identify the data and make it correspond to the item of the headline, combine the character string that spans multiple lines by wrapping as one data, and use the dictionary of the category corresponding to the content of the heading to character string of the data part A character recognition method characterized by recognition is provided.

  In the present invention, a layout analysis processing unit that extracts a character string from image data of a form, extracts a logical line from the positional relationship of the character string, recognizes the character string, and extracts a heading line; An association processing unit for associating a column with a character string of the data part, a character recognition processing unit for recognizing the character string of the data part, a data storage unit storing a dictionary necessary for character string recognition and identification, The association processing unit extracts a line having the same ratio between the total number of headings and the number of character strings of the data in the line, and unconditionally associates the heading part and the data part, and A fixed-length data identification unit that identifies data whose character string in the data part corresponding to the heading has a fixed length, and data that has a variable-length character string in the data part corresponding to the heading and that does not wrap the character string 1st variable to A data specifying unit and a character string of a data part corresponding to a headline have a variable length, and the character string may be folded. The variable-length data specifying unit and the character string of the data part corresponding to the headline have variable length, and the character string may be folded, and the data corresponding to the number of headings and character strings is specified as 1 to N. There is provided a character recognition device including a third variable length data specifying unit and a folded data integrating unit that integrates character strings straddling a plurality of lines by folding as one data.

  Furthermore, in the present invention, there is provided a character recognition program for causing a computer to execute processing for recognizing a character string of a form having a character string of a heading part and a character string of a data part. Extracting logical lines in which the character strings are arranged in a horizontal direction from the positional relationship of the extracted character strings on the form, extracting the heading lines in which the character strings of the details are recognized by recognizing the character strings of the logical lines, Data corresponding to the headline is extracted by extracting details lines having the same ratio between the total number of character strings in the headline and the number of character strings in the data part, and matching the headline item in the headline part with the character string in the data part on a one-to-one basis The character string of the part specifies fixed-length data and corresponds to the heading item, and the character string of the data part corresponding to the heading is variable-length and specifies the first variable-length data that does not wrap the character string. The The character string of the data part corresponding to the heading has a variable length, and the character string may be folded, and the heading and the number of character strings of the data part have a one-to-one correspondence. 2 Specify variable-length data and make it correspond to the heading item. The character string of the data part corresponding to the heading has a variable length and the character string may be folded. Identifies the third variable-length data corresponding to 1 to N and associates it with the item of the headline. By folding, the character strings straddling multiple lines are integrated as one data, and the character string of the data portion There is provided a character recognition program characterized in that a character recognition process is executed using a category dictionary corresponding to the content.

  According to such a character recognition method, character recognition device, and character recognition program, the character string of the data part is analyzed in order from the easily identifiable character string, and the remaining difficult-to-analyze character string is finally deleted by the erasing method. By specifying, the character string of the data part for the heading was specified sequentially. As a result, it is possible to recognize a character string printed in a free layout on a ruled line form.

  The character recognition method, character recognition device, and character recognition program configured as described above are methods of analyzing in order starting from a character string that can be easily specified, so that the character recognition process exceeds the scope of the current standard form creation specifications. There is an advantage that the types of forms that can be handled can be increased.

  In addition, an atypical form in which the heading and the character string of the data part are not aligned in a single line was input by the operator by analyzing the layout. Since a form having many errors in the recognition result can be subjected to character recognition processing, the input load on the operator can be reduced.

It is a block diagram which shows the character recognition apparatus which concerns on 1st Embodiment. It is a figure which shows the example of a form in which the data with respect to a heading are not arranged in a line. It is a figure which shows the example of a form in which the detail line of several lines is mixed. It is a flowchart which shows the flow of a process of a layout analysis process part. It is a flowchart which shows the flow of a process of a matching process part. It is a figure which shows the attribute and characteristic information of the character string acquired from the headline wording dictionary. It is a figure explaining the state with which a header part and a data part respond | correspond one-to-one. It is a figure which shows the form of the result as which the headline part and data part which respond | correspond one-to-one were matched. It is explanatory drawing which measures the character string width | variety of the character string whose data corresponding to a headline are fixed length. It is explanatory drawing which shows the fixed length data in the line where the heading part and the data part do not respond | correspond 1: 1. It is a figure which shows the example of the score showing a coincidence rate. It is a figure which shows the form of the result as which fixed length data was matched. It is explanatory drawing which shows the 1st variable length data in the row | line | column where the header part and the data part do not respond | correspond 1: 1. It is a figure which shows the example of the score showing a coincidence rate. It is a figure which shows the form of the result as which 1st variable length data was matched. It is explanatory drawing which shows the 2nd variable length data in the line where the heading part and the data part do not respond | correspond 1: 1. It is a figure which shows the example of the score showing a coincidence rate. It is a figure which shows the form of a result with which 2nd variable length data was matched. It is explanatory drawing which shows the 3rd variable length data in the line | wire where the heading part and the data part do not respond | correspond 1: 1. It is a figure which shows the example of the score showing a coincidence rate. It is a figure which shows the form of a result associating 3rd variable length data. It is a figure explaining determination whether one specification is a plurality of lines. It is a figure which shows the hardware structural example of the computer used for a character recognition apparatus. It is a figure which shows an example of a continuous writing form.

  Hereinafter, the embodiment of the present invention is applied to a device that uses a transfer request form used in a financial institution or the like as a continuous form form and recognizes a character string printed on the transfer request form. Will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing a character recognition apparatus according to the first embodiment.
The character recognition device 10 includes a layout analysis processing unit 11 to which the output of the scanner 1 is connected, an association processing unit 12 that performs character string association processing, a character recognition processing unit 13 that performs character string recognition, And a data storage unit 14 for storing data necessary for processing.

  The layout analysis processing unit 11 processes the image data of the form acquired by the scanner 1 and analyzes the arrangement of character strings on the form. The association processing unit 12 includes a standard data specifying unit 15, a fixed length data specifying unit 16, a first variable length data specifying unit 17, a second variable length data specifying unit 18, and a third variable length data specifying unit 19. And the folded data integration unit 20, and performs processing for associating the item of the heading part with the data of the corresponding data part from the character string analyzed by the layout analysis processing unit 11.

The character recognition processing unit 13 performs character recognition on the data associated by the association processing unit 12 using an optimum character recognition method corresponding to the content of the headline.
The data storage unit 14 stores the headline wording dictionary 21 and the category / font dictionary 22, and is used for the processing of the layout analysis processing unit 11, the association processing unit 12, and the character recognition processing unit 13. The headline wording dictionary 21 stores a predetermined headline wording, its image data, character string attributes corresponding to the headline, and character string characteristics. The category / font-specific dictionary 22 is prepared for each character recognition category / character font as a dictionary for character recognition of the data portion.

Here, the continuous form form which is processed by the character recognition device 10 will be described.
FIG. 2 is a diagram showing an example of a form in which data for headings is not arranged in a line, and FIG. 3 is a diagram showing an example of a form in which a plurality of detail lines are mixed.

  The form 2 shown in FIG. 2 is composed of a heading part 4, a data part 5 and a heading footer 6, and the character string of the data part 5 corresponding to the heading item of the heading part 4 is aligned in a line. There are no examples, and an example in which the character string of the data part 5 for one heading item includes two or more (hereinafter referred to as N) character strings by blank characters. Such a form 2 correctly associates the character string of the data part 5 with the heading item of the heading part 4 in the process on the premise that the heading item and the corresponding character string are aligned in a line. Can not do it.

  Further, the form 7 shown in FIG. 3 shows a case where one detailed line becomes a plurality of lines in the data portion. This is because although the character string of the data portion corresponding to the heading item of the header portion is aligned in a line, a part of the character string of the data portion is folded halfway. For this reason, the data portion has a form in which one detail line is mixed with a plurality of lines. Such a form 7 cannot correctly associate the character string of the data part with the heading item of the heading part in the process assuming that one detailed line is one line.

Next, the operation of the character recognition device 10 will be described.
FIG. 4 is a flowchart showing a processing flow of the layout analysis processing unit.
The layout analysis processing unit 11 first receives data obtained by scanning with the scanner 1 and converts the form into an image (step S1). Next, the layout analysis processing unit 11 extracts a character string based on the form image data (step S2). The extraction of the character string is performed using an existing technique. For example, the image data of a form is scanned, and a circumscribed rectangle that touches a portion where black pixels are continuous is extracted as a label. Among the extracted labels, labels that are within a threshold distance are extracted as labels constituting one character string.

  Next, the layout analysis processing unit 11 analyzes the positional relationship of the character strings from the extracted label coordinates, and extracts the character strings arranged in the horizontal direction as character strings constituting a logical line (step S3). When this logical line is extracted, the outline of the outer periphery of the form sheet is analyzed. If the image is skewed, the image is rotated so that the character strings in the same logical line are corrected so that they are aligned horizontally. The

  Next, the layout analysis processing unit 11 performs character recognition in all categories for the character string of the entire form (step S4). Here, since it is not known to which category the extracted character string belongs, the layout analysis processing unit 11 is not limited to a specific category, and all category / font-specific dictionaries 22 of the data storage unit 14 are used. Use to recognize the string. This character recognition is performed for all the extracted character strings.

  Then, the layout analysis processing unit 11 matches the result of character recognition with the headline wording dictionary 21 of the data storage unit 14, and extracts the line in which the headings of the details are arranged, that is, the heading header line and the heading footer line (step S5). ). The headline wording dictionary 21 includes “bank name”, “branch name”, “subject”, “account number”, “recipient name”, “amount”, and “transfer” which are heading items used in the transfer request form. "Total" is registered. Of course, in this headline wording dictionary 21, all item names that become other headlines used in business such as financial institutions are also registered.

FIG. 5 is a flowchart showing the processing flow of the association processing unit, and FIG. 6 is a diagram showing the attribute and feature information of the character string acquired from the headline wording dictionary.
First, the association processing unit 12 refers to the headline wording dictionary 21 and acquires the character string attribute and the character string characteristic information of the data portion corresponding to the headline (step S11). As shown in FIG. 6, the character string attributes define the number of characters, the character type, the data format, the character string wrapping, and the ratio of the total number of headings and the number of character strings in the data for one detail line. As the character string characteristics, the order of heading items appearing in the heading part line and the character string width when the data corresponding to the heading has a fixed length are defined.

  FIG. 7 is a diagram for explaining a state in which the heading part and the data part are in one-to-one correspondence, and FIG. 8 is a result form in which the heading part and the data part are in one-to-one correspondence. FIG. 9 is an explanatory diagram for measuring the character string width of a character string whose data corresponding to the heading has a fixed length.

  The standard data specifying unit 15 of the association processing unit 12 sequentially analyzes the detail lines, extracts standard detail lines in which the total number of headings and the number of character strings in the data portion are the same, and data for the headline items Are identified unconditionally (step S12). That is, as shown in FIG. 7, when there are 6 items H1 to H6 in the heading part and 6 pieces of data D1 to D6 in the data part are equal, the heading and the data are in a one-to-one relationship. Because it can respond with, it is made to respond unconditionally. As a result, the bank name is “Iroha Bank”, the branch name is “Head Office”, the subject is “Normal”, the account number is “1234567”, the recipient name is “Aiueo Co., Ltd.”, and the amount is “1,000,000”. ". As a result, as shown in FIG. 8, in the data portion of the form, the first line, the fifth line, and the sixth detailed line enclosed by the square have finished the association process.

  Here, the width of the fixed character column item in the detail line in which the heading and the data are associated one-to-one is calculated (step S13). Since items having fixed character digits in the data have the same character string widths in other detail lines, the widths of the character digit fixed items are calculated here. For the item with fixed character digits, data D3 and D4 corresponding to “subject” and “account number” are selected from the attribute of the character string in FIG. As shown in FIG. 9, the character string widths of the data D3 and D4 are calculated. In step S12, if a detail line is not extracted, the length obtained by multiplying the width of one character by the number of digits from the extracted label of one character sentence and the character string attribute of FIG. 6 is used as the width of the character string. calculate.

  So far, the processing when the ratio between the total number of headings and the number of character strings in the line data is the same has been described, but in the following, the ratio between the total number of headings and the number of character strings in the line data is The processing of different lines will be described.

  FIG. 10 is an explanatory diagram showing fixed-length data in a line where the heading portion and the data portion do not correspond one-to-one, FIG. 11 is a diagram showing an example of a score representing the matching rate, and FIG. It is a figure which shows the form of the attached result.

  The fixed length data specifying unit 16 of the association processing unit 12 sequentially analyzes the detail lines that are not extracted by the standard data specifying unit 15, and the width of the character string, the number of digits of the character string, the attribute of the character string, the character string attribute, The fixed length data for the headline is specified from the position (step S14). In the data part, there are data of subject and account number as data having a fixed character string. For example, when specifying the account number data, first, as shown in FIG. 10, the character string in the line is sequentially compared with the attribute of the character string, and the character of the detail line is determined from the attribute of the character string of the account number data. Data whose column width is 7 digits wide is extracted. In the example of FIG. 10, data D4 and D7 correspond. The comparison between the character string and the attribute of the character string is sequentially performed by evaluation (scoring system) based on the determination element for all the character strings (data D1 to D7) included in one line.

  From FIG. 10 and FIG. 11, there are seven character strings constituting the line, and the attributes and features of the character strings are compared for all seven. The table of FIG. 11 contains the value obtained by converting the comparison coincidence rate according to the determination element into a score, and the higher the coincidence rate, the higher the score. In the example row of FIG. 10, the fourth and seventh character strings (data D4, D7) are scored, but the character string width, the number of characters in the character string, the character type of the character string, and the position of the character string Since the total score of the fourth character string (data D4) is generally high from (order), the fourth character string (data D4) is estimated to be account number data.

  The analysis of the subject data that is fixed length data is performed in the same manner as the analysis of the account number data. Further, the specification of the fixed length data is performed for all the details lines that are not extracted by the standard data specifying unit 15. To analyze. As a result, as shown in FIG. 12, in the data portion of the form, the association processing between the item of the heading portion enclosed by the square and the data of the corresponding data portion is completed. The squares with diagonal lines indicate analyzed character strings.

  FIG. 13 is an explanatory diagram showing the first variable length data in a row where the heading portion and the data portion do not correspond one-to-one, FIG. 14 is a diagram showing an example of the score indicating the matching rate, and FIG. 15 is the first variable It is a figure which shows the form of a result with which long data was matched.

  The first variable length data specifying unit 17 of the association processing unit 12 sequentially analyzes the remaining data of the detail lines that were not extracted by the standard data specifying unit 15, and the character string has a variable length and there is no character string wrapping. First variable-length data whose character type is a single-byte alphanumeric character is specified (step S15). The character string corresponding to the first variable length data is data D7 that is located at the end of the detail line in FIG. 13 and corresponds to the “money amount” in the heading item H6. The table of FIG. 14 includes values obtained by converting the matching rate of the comparison by the determination element into points, excluding the third and fourth data whose association has already been determined. According to the table of FIG. 14, since the score of the seventh character string (data D7) is the highest from the total score, the seventh character string (data D7) is estimated as money data.

  The analysis of the “subject” data, which is the first variable length data, is performed for all the remaining detail lines. As a result, as shown in FIG. 15, in the data portion of the form, the association processing between the item of the heading portion enclosed by the square and the data of the corresponding data portion is completed.

  FIG. 16 is an explanatory diagram showing second variable length data in a line in which the header portion and the data portion do not correspond one-to-one, FIG. 17 is a diagram showing an example of a score representing the matching rate, and FIG. 18 is a second variable It is a figure which shows the form of a result with which long data was matched.

  The second variable length data specifying unit 18 of the association processing unit 12 converts the second variable length data having a character string ratio of 1: 1, a character string having a variable length of 6-digit wrapping, and a character type of characters of a full-width full category. Specify (step S16). In FIG. 16, the character string corresponding to the second variable length data is data D1 and D2 corresponding to the headings “bank name” and “branch name”. The table of FIG. 17 includes values obtained by converting the coincidence rate of the comparison by the determination element into points, excluding the third, fourth, and seventh data that have already been determined to be associated. According to the table of FIG. 17, since the score of the first character string (data D1) is the highest from the total score, the first character string (data D1) is estimated to be bank name data. Next, since the score of the second character string (data D2) is also high, it is estimated that the second character string (data D2) is branch name data.

  The analysis of the data of “bank name” and “branch name” as the second variable length data is performed for all the remaining detail lines. As a result, as shown in FIG. 18, in the data portion of the form, the association processing between the item of the heading portion enclosed by the square and the data of the corresponding data portion is completed.

  FIG. 19 is an explanatory diagram showing third variable length data in a line in which the header portion and the data portion do not correspond one-to-one, FIG. 20 shows an example of a score indicating the matching rate, and FIG. 21 shows the third variable It is a figure which shows the form of a result with which long data was matched.

  The third variable length data specifying unit 19 of the association processing unit 12 generates third variable length data having a character string ratio of 1 to N, a character string having a variable length of 14-digit wrapping, and a character type of characters of all-byte full-category. Specify (step S17). In FIG. 19, the character string corresponding to the third variable length data is data D5 and D6 corresponding to the “header name” of the remaining heading. The table of FIG. 20 includes values obtained by converting the matching rate of the comparison by the determination element into points, excluding the first to fourth and seventh data that have already been determined to be associated. According to the table of FIG. 20, since the score of the fifth character string (data D5) is the highest from the total score, the fifth character string (data D5) is estimated to be the recipient name data. Also, the sixth character string (data D6) has no corresponding heading and has the same character string attribute as the fifth character string, so the sixth character string (data D6) is the fifth character string. It is determined that the data is the recipient name data.

  The third variable length data “recipient name” data is analyzed for all remaining detail lines. As a result, as shown in FIG. 21, in the data portion of the form, the correspondence processing between the item of the heading portion enclosed by the square and the data of the data portion corresponding thereto is completed, and all of the items when one specification is one line Are associated with each other.

  Next, as in the form 7 shown in FIG. 3, the character strings of the data part for the heading item in the heading part are aligned in a line, but there are a case where one detail line in the data part is one line and a plurality of lines. A process when the case is mixed will be described. However, whether or not the character string of the data part with respect to the heading item of the heading part is aligned in one line is not related in the above-described association processing, and here, the data including one detail line including a plurality of detail lines Processing when there is a part will be described.

FIG. 22 is a diagram for explaining whether one specification is a plurality of lines.
First, the loopback data integration unit 20 of the association processing unit 12 pays attention to the fixed-length data (subject or account number data) in which the character string extracted by the fixed-length data specifying unit 16 is adjacent to the line in which the character string exists. From the interval, it is determined whether there is a detail line with character string wrapping, and the number of each detail line is specified (step S18). Referring to FIG. 22, for example, since a certain row of account number data exists continuously in the first and second rows of the data portion, the first row is determined to be one row of details. . In the row with the account number data, the second row is followed by the fifth row, so that there is a vacant space for two rows. Therefore, it is determined that one specification is three rows in the second row. Next to the fifth row in the row with the account number data, the seventh row is the first row, and there is a vacant space for one row. Therefore, it is determined that one specification is two rows in the fifth row.

  Next, when the return data integration unit 20 determines in step S18 that the line is a detail line with character string wrapping, after confirming the position of the account number data, the character string of the data wrapping across multiple lines Are integrated into one item (step S19). That is, in the detail line of the second line, the character string of the bank name data folded over two lines is integrated into one item, and the character string of the recipient name data folded back over three lines Are integrated into one item. Similarly, in the fifth detail line, the character string of the bank name data folded back over two lines is integrated into one item, and the character string of the branch name data folded back over two lines Are integrated into one item.

  As described above, in the association processing unit 12, the association between the headline and the character string of the data part is performed by combining the attribute of the data part of the data part with respect to the headline acquired from the headline wording dictionary 21 and the characteristics of the character string, The attribute of the character string actually recognized in step S4 and the characteristics of the character string are compared, and the character string having a high matching rate is associated. Moreover, the association between the heading and the character string of the data part is analyzed in order from the data that can be easily identified from the attribute and feature information of the character string, and the remaining difficult-to-analyze items are finally identified by the elimination method. I have decided.

  The attributes and characteristics of the character strings associated in this way are known. Therefore, finally, when the character recognition processing unit 13 performs the character recognition process, the character recognition is performed using the dictionary of the category suitable for the character type of the character string. As a result, for example, data whose headline is an account number is recognized using a specific dictionary whose category is narrowed to single-byte numbers, so that it can be recognized with higher accuracy than the first character recognition (step S4). it can.

Finally, the character recognition device 10 notifies the character recognition caller of the result of character recognition by the character recognition processing unit 13.
FIG. 23 is a diagram illustrating a hardware configuration example of a computer used in the character recognition device.

  The entire computer 50 is controlled by a CPU (Central Processing Unit) 51. A random access memory (RAM) 52, a hard disk drive (HDD) 53, a graphic processing device 54, an input interface 55, and a communication interface 56 are connected to the CPU 51 via a bus 57.

  The RAM 52 temporarily stores at least part of an OS (Operating System) program to be executed by the CPU 51 and application programs necessary for character recognition processing. The RAM 52 stores various data necessary for processing by the CPU 51. The HDD 53 stores an OS, an application program for character recognition, a dictionary used for character recognition, and the like.

  A monitor 58 is connected to the graphic processor 54. The graphic processing device 54 displays an image on the screen of the monitor 58 in accordance with a command from the CPU 51. The input interface 55 is connected to the scanner 1, the keyboard 59, and the mouse 60. The input interface 55 transmits signals sent from the scanner 1, the keyboard 59 and the mouse 60 to the CPU 51 via the bus 57.

  The communication interface 56 is connected to a network in the financial institution. The communication interface 56 transmits / receives data to / from other computers via a network.

  With the hardware configuration as described above, the processing function according to the character recognition device 10 of the present embodiment can be realized.

DESCRIPTION OF SYMBOLS 1 Scanner 2,7,100 Form 3,101 Description 4,102 Heading part 5 Data part 6 Heading footer 10 Character recognition apparatus 11 Layout analysis processing part 12 Association processing part 13 Character recognition processing part 14 Data storage part 15 Standard data specification Unit 16 Fixed-length data specifying unit 17 First variable-length data specifying unit 18 Second variable-length data specifying unit 19 Third variable-length data specifying unit 20 Return data integration unit 21 Headline word dictionary 22 Category / font-specific dictionary 50 Computer 51 CPU
52 RAM
53 HDD
54 Graphic Processing Device 55 Input Interface 56 Communication Interface 57 Bus 58 Monitor 59 Keyboard 60 Mouse

Claims (10)

In a character recognition method for recognizing a character string of a form having a character string of a heading part and a character string of a data part,
From the positional relationship of the character string extracted from the image data of the form on the form, a logical line in which the character string is arranged in the horizontal direction is extracted, the character string in the logical line is recognized, and the character string of the headline of the description is arranged. Extract the heading line,
Detail lines having the same ratio between the total number of character strings in the heading and the number of character strings in the data part are extracted, and the heading item in the heading part is associated with the character string in the data part on a one-to-one basis.
Specify data with a fixed-length character string in the data part corresponding to the headline, and make it correspond to the headline item.
The character string of the data portion corresponding to the heading is variable length, and the first variable length data without the character string wrapping is identified and associated with the heading item,
The character string of the data portion corresponding to the headline has a variable length and the character string may be folded, and the second variable length data in which the headline and the number of character strings of the data portion correspond one-to-one is specified. Corresponding to the headline item,
The character string of the data portion corresponding to the headline has a variable length and the character string may be folded, and the third variable length data in which the number of the character string of the heading and the data portion corresponds 1 to N is specified. Corresponding to the headline item,
By concatenating character strings that span multiple lines as a single piece of data,
Recognize the character string of the data part using the dictionary of the category corresponding to the content of the heading,
A character recognition method characterized by the above.   After extracting the heading line, the character string attribute and the character string characteristic of the data part are acquired, and the character string attribute and the character string acquired when analyzing the character string of the detail line corresponding to the heading The character recognition method according to claim 1, wherein a character string corresponding to the headline is specified based on the characteristics of the character string.   Character string attributes include the number of characters, full-width kanji, half-width numbers, half-width alphanumeric characters, full-width full-width categories, data format that indicates whether fixed length or variable length, string wrapping that indicates whether or not the character string is wrapped, and Including a character string ratio indicating the ratio between the total number of headings and the number of data character strings, and the character string features include a character string width indicating the order of headings and the width of fixed-length data, The character recognition method according to claim 2.   After associating the headline item with the character string of the data part on a one-to-one basis, the character string of the character string is compared with the fixed-length character string of the data part for the headline identified from the acquired character string attributes and character string characteristics. The character recognition method according to claim 2, wherein the character recognition method is used for calculating a width and specifying fixed-length data.   The fixed-length data is identified by identifying the character string of the data part, the number of characters in the character string based on the calculated character string width, the obtained character string attribute and character string characteristics, the character type and character of the character string. 5. The character recognition method according to claim 4, wherein comparison is made by a determination factor with respect to the position of the column, a matching rate of each comparison is converted into a score, and the determination is made by a total score.   The identification of the first to third variable length data is performed by identifying the character string of the data portion, the number of characters of the character string, the character type of the character string, and the position of the character string based on the acquired character string attribute and character string characteristics. The character recognition method according to claim 2, wherein comparison is performed using the determination elements, the matching rate of each comparison is converted into a score, and the total score is determined.   For the integration of character strings, the second and third variable values are calculated by calculating the number of one detail line from the interval between adjacent lines where a character string of fixed-length data in the data portion is present, and by folding the detail lines of a plurality of lines. The character recognition method according to claim 1, wherein character strings of long data are integrated. A layout analysis processing unit that extracts a character string from the image data of the form, extracts a logical line from the positional relationship of the character string, recognizes the character string, and extracts a heading line;
An association processing unit for associating the character string of the heading part with the character string of the data part;
A character recognition processing unit for recognizing the character string of the data portion;
A data storage unit storing a dictionary necessary for character recognition and identification;
With
The association processing unit
A standard data specifying unit that extracts lines having the same ratio between the total number of headings and the number of character strings in the line data and unconditionally associates the heading part and the data part;
A fixed-length data specifying unit for specifying data whose character string of the data part corresponding to the heading has a fixed length;
A first variable length data specifying unit for specifying data in which the character string of the data portion corresponding to the heading has a variable length and the character string is not folded;
A second variable-length data specifying unit for specifying data in which the character string of the data portion corresponding to the heading has a variable length and the character string may be folded and the heading and the number of character strings correspond one-to-one When,
A third variable length data specifying unit for specifying data in which the character string of the data portion corresponding to the heading has a variable length and the character string may be folded and the heading and the number of character strings correspond to 1 to N When,
A folded data integration unit that integrates a character string that spans multiple lines as a single data by folding,
A character recognition device characterized by comprising: A character recognition program for causing a computer to execute processing for recognizing a character string of a form having a character string of a heading part and a character string of a data part,
In the computer,
From the positional relationship of the character string extracted from the image data of the form on the form, a logical line in which the character string is arranged in the horizontal direction is extracted, the character string in the logical line is recognized, and the character string of the headline of the description is arranged. Extract the heading line,
Detail lines having the same ratio between the total number of character strings in the heading and the number of character strings in the data part are extracted, and the heading item in the heading part is associated with the character string in the data part on a one-to-one basis.
Specify data with a fixed-length character string in the data part corresponding to the headline, and make it correspond to the headline item.
The character string of the data portion corresponding to the heading is variable length, and the first variable length data without the character string wrapping is identified and associated with the heading item,
The character string of the data portion corresponding to the headline has a variable length and the character string may be folded, and the second variable length data in which the headline and the number of character strings of the data portion correspond one-to-one is specified. Corresponding to the headline item,
The character string of the data portion corresponding to the headline has a variable length and the character string may be folded, and the third variable length data in which the number of the character string of the heading and the data portion corresponds 1 to N is specified. Corresponding to the headline item,
By concatenating character strings that span multiple lines as a single piece of data,
Recognize the character string of the data part using the dictionary of the category corresponding to the content of the heading,
A character recognition program for executing a process.   In the process of integrating the character strings, the number of lines of one detailed line is calculated from the interval between adjacent lines in which the character string of the fixed-length data in the data portion exists, and the second and second lines obtained by wrapping in the multiple detailed lines are returned. The character recognition program according to claim 9, wherein character strings of three variable length data are integrated.

Images