JPH08263587A - Method and device for document input - Google Patents

Abstract

PURPOSE: To provide the document input device which can correctly recognize an inputted document image in a short time and easily correct the recognition result, and print the result with character size well balanced to a ruled-line frame. CONSTITUTION: A CPU 120 generates a correction result which can grammatically be interpreted from a combination of candidate codes having likelihood above a likelihood level in a likelihood level temporary storage part and displays a 1st candidate for the correction result on a display 106 each time the correction result is generated for a specific number of recognition sequences. The CPU 120 compares the correction result after corrections by an operator with an identical recognition sequence consisting of candidate character codes having the maximum likelihood and outputs the precision of character recognition. When the calculated precision is less than a precision level in a precision level temporary storage part, the reference value of likelihood in the likelihood level temporary storage part is varied.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document input method and a document input device for inputting a document read as an image as a character code by using character recognition under computer control of a word processor, a personal computer or the like.

[0002]

2. Description of the Related Art OCR (O
A physical character reader) and a handwritten character recognition device have been developed and used. In such an OCR or handwritten character recognition device, a character written on a paper surface is read as an image, the image is cut out in character units to extract character characteristics, and the character represented by the image is recognized from the extracted character characteristics. To do. In such character recognition, for example, when recognizing characters such as Kanji, there are a lot of similar characters in Kanji, so there is a limit to the correct answer rate in character recognition processing, and recognition errors occur in the character recognition process. To do. In order to correct this recognition error, the recognition result is compared with the word dictionary, and it is determined whether the sequence of characters obtained as the recognition result is appropriate as Japanese.
Recognition error correction processing is performed. An example of such recognition error correction processing is disclosed in the academic report by Niwa and Kayashima, "Technical Report of the Institute of Electronics, Information and Communication Engineers PRU91-135 character recognition post-processing method and analysis of effect of post-processing". According to the academic conference report, the recognition error generated in the character recognition process is
Reference is made to an example in which correction can be performed with an error correction rate of 0%, and the recognition error correction processing by such a method is useful for increasing the correct answer rate of a Japanese document input as a recognition result. I understand.

[0003]

However, a Japanese form (here, "form" means an office paper, such as a book, a slip, and a notification form, printed with blanks for entering necessary information. .), Reports, pamphlets, etc., are often seen in documents, the characters are arranged at regular intervals in a predetermined frame or in a predetermined interval, that is, Japanese documents in which characters are evenly distributed from an image are displayed. When inputting after converting to a character code, conventional recognition error correction processing
The white space between the characters was regarded as a word delimiter, the collation with the word dictionary was not successful, and the effect of the correction process could not be fully exerted.

Further, when character recognition is performed on the above-mentioned uniformly-allocated character string while leaving layout information indicating at which position on the paper each character is arranged, the character obtained as a recognition result. Since the code is managed separately from the layout information, there is a problem that it takes time and effort to correct the character string of the input result. Specifically, for example, when "notes" are evenly allocated and there is a space between "notes" and "notes", the conventional document input device divides them into "notes" and "notes". The recognized characters will be recognized. In this case, when you want to change the character string "Remarks" to "Other", modify "Bi" to "So" and "Remark" to "Other" for each individual character that is the recognition result, and further It takes a lot of trouble to add a single "no" to the blank space.

Further, when the layout has both vertical and horizontal interpretations, such as characters in the form frame, the direction of the character string to be recognized cannot be determined, and the character string direction cannot be correctly interpreted. Then, there is a problem that the effect of the correction process cannot be exhibited. In addition, when inputting an address display as a character code from an image, there are many place names having similar glyphs in the address display, and thus there is a problem that such place names are likely to be erroneously recognized.

Further, when a copied original or the like is input, a part of the original may be thinned due to uneven copy and the character recognition rate may be significantly lowered. In the conventional recognition error correction processing, uniform processing is performed even in such a case, and therefore, when the character recognition performance is significantly deteriorated due to deterioration of image quality or the like, the optimum candidate is always presented. I couldn't. On the other hand, even when the image quality of the original is good and the recognition rate is high from the beginning, the recognition error correction processing is performed assuming the worst case, and therefore the post-processing requires a long time.

Further, in the conventional document input device, since the characters are printed as they are in the print character size preset in the document input device, the characters obtained as a recognition result are balanced as compared with the set ruled line frame. There was a case that it was printed with a bad character size. In view of the above-mentioned problems, the present invention provides a document type and a document such as characters evenly distributed, characters in a form frame that can be taken in both character string directions, place names in an address display, and characters in a document with poor print quality. Regardless of the status, the document can be recognized correctly and the characters after recognition can be easily corrected, the time required for recognition error correction processing can be shortened, and the size can be balanced with the set ruled line frame. It is an object of the present invention to provide a document input device capable of printing the characters.

[0008]

In view of the above problems, the document input device according to the first aspect of the present invention proposes a plurality of character code candidates as the character recognition result of an image of one character. Is stored in association with the likelihood, the likelihood reference storage means for storing a predetermined likelihood reference value of the likelihood, and the same character string representing the character string in the image of the captured document. A plurality of character string candidates that can be grammatically correctly interpreted from a combination of the character code candidates having a likelihood equal to or higher than the likelihood reference value in the likelihood reference storage means. With respect to the character string candidate generating means to generate and the character string candidates generated corresponding to the character string image, a predetermined candidate is selected from the character string candidates generated for the same character string image. Table to display Means, input means for inputting a correction instruction from the operator to the character string candidates generated corresponding to the character string image, and the character string generated corresponding to the character string image. In accordance with the correction instruction, the correction means corrects the character code that is erroneously recognized to the correct character code, and the corrected character string corresponding to the character string image after the correction by the correction means. And a comparing means for comparing a character string composed of a character code candidate with the highest likelihood corresponding to the same character string image, and after the comparison by the comparing means, based on the comparison result,
A correct answer rate calculating means for calculating a correct answer rate of character recognition from a ratio of the number of different characters between the two and the number of characters included in the one character string, and a predetermined correct answer rate reference value of the correct answer rate is stored. Correctness rate reference storage means, after the correctness rate is calculated, the calculated correctness rate, in comparison with the correctness rate reference value, if a predetermined condition is satisfied, depending on the calculated correctness rate Likelihood reference value changing means for changing the likelihood reference value in the likelihood reference storage means.

According to another aspect of the document input device of the present invention, as a character recognition result of an image of one character, a plurality of character code candidates are stored in association with a likelihood indicating the likelihood of the character code. And a likelihood reference storage unit that stores a predetermined likelihood reference value of the likelihood, and the same character string image that represents a character string in an image of a captured document, in the likelihood reference storage unit. From a combination of the character code candidates having a likelihood equal to or higher than the likelihood reference value,
Character string candidate generating means for generating a plurality of character string candidates that can be grammatically correctly interpreted, and the same character string for the character string candidates generated corresponding to the character string image. A display unit for displaying a predetermined candidate of the character string candidates generated for the image and a correction instruction from the operator for the character string candidate generated corresponding to the character string image. Input means, correction instruction counting means for counting the correction instructions input from the input means, correction instruction reference storage means for storing a predetermined reference value of correction instruction number of the correction instruction number, and corresponding to the character string image Correction means for correcting a character code that is erroneously recognized to a correct character code in accordance with the correction instruction for the candidate of the character string generated by When the counted number of correction instructions is compared with the correction instruction number reference value stored in the correction instruction reference storage means and satisfies a predetermined condition, the likelihood criterion is calculated according to the counted number of correction instructions. Likelihood reference value changing means for changing the likelihood reference value in the storage means.

A document input device according to a third aspect of the present invention is a space detecting means for detecting a space between an image of one character cut out from a character string image representing one line of a character string and an image of one character adjacent thereto. When the detected blank is larger than a predetermined size, the one-character image and the one-character image adjacent thereto are each identified by a word identifying unit that identifies them as characters belonging to different words. A word identification limiting unit that invalidates the identification result of the word identification unit for a character string image in a predetermined area in the image of the document.

According to another aspect of the present invention, there is provided a document input device, which stores character code candidate storage means for storing a plurality of character code candidates as a character recognition result of an image of one character cut out from an image of a captured document. For the same character string image representing a character string in the image of the document, a character string candidate generating unit that combines the character code candidates and generates a plurality of character string candidates that can be grammatically correctly interpreted, A character string candidate storage unit that stores the generated character string candidates and a predetermined character string candidate that corresponds to one character string image are read from the character string candidate storage unit and are stored in the identified ruled line frame. Display means for displaying by equally allocating.

According to a fifth aspect of the present invention, there is provided a document input device, which stores character code candidate storage means for storing a plurality of character code candidates as a character recognition result of an image of one character cut out from an image of a captured document. A character string direction identifying means for identifying whether the direction of the character string in the captured image of the document is vertical or horizontal, based on the distribution of the image area and the blank area in the image of the captured document; From the distribution of the image area and the blank area in the image of the document, when it can be interpreted that the direction of the character string is either vertical or horizontal, in both directions, the same character string image representing the character string, Character string candidate generation means for combining the character code candidates to generate a plurality of character string candidates that can be grammatically correctly interpreted, and for the vertical character string image A predetermined candidate among the character string candidates that have been generated and a predetermined candidate among the character string candidates that have been generated for the horizontal character string image, and compare the predetermined conditions for the predetermined candidate. The character string direction determining means that determines that the character string direction that is more satisfied and the candidate of the character string are valid.

According to another aspect of the document input device of the present invention, a ruled line frame identifying means for identifying a ruled line frame in an image of a captured document and a character recognition result of an image of one character cut out from the image of the captured document. As a character code candidate storage unit that stores a plurality of character code candidates together with the position and character size of the image of the character in the image of the document, the same character representing the character string in the image of the captured document. With respect to the column image, a character string candidate generating unit that generates a plurality of character string candidates that can be grammatically correctly interpreted from the combination of the character code candidates, and a predetermined character string candidate among the generated character string candidates. When printing a candidate in the identified ruled line frame, the character size of the image corresponding to the candidate of the character code in the candidate of the character string can be checked and printed. From the font sizes, a printing character size selecting means for selecting the printing font size closest to the examined character size, the identified ruled line frame is printed, and a predetermined candidate among the generated character string candidates is displayed. A printing unit for printing in a selected print character size is provided in the printed ruled line frame.

A document input device according to claim 7 is the document input device according to claim 6.
In the described document input device, referring to the character code candidate storage means, the position of the image corresponding to the character code candidate in the character string candidates in the image of the document is checked, and the image of the character is displayed. In the image of the document of the image to which the character code candidate in the character string candidate corresponds, by referring to the upper and lower space calculation means for calculating the size of the upper and lower blank areas to be sandwiched and the character code candidate storage means. And the horizontal spacing calculation means for calculating the size of the left and right blank areas sandwiching the image of the character, and the vertical size of the selected print font size is the calculated size of the upper and lower blank areas. In order not to exceed the prescribed size,
A vertical size limiting means for limiting the vertical size of the selected print font size, and a horizontal size of the selected print font size is a predetermined size compared with the calculated size of the left and right blank areas. So as not to exceed the left and right size limiting means for limiting the size of the selected print font size in the left and right direction, and the print character size selecting means includes the upper and lower size limiting means and the left and right size limiting means. From the limited print font size, select the print font size that is closest to the examined character size.

According to another aspect of the document input device of the present invention, a character code candidate storage means for storing a plurality of character code candidates as a character recognition result of an image of one character cut out from the captured image of the document, and a character string. For the same character string image representing, a plurality of character string candidates that can be correctly interpreted as a term by combining the character code candidates,
A character string candidate generating unit that generates with a predetermined priority, and a specific character code is identified from the generated character string candidates, and the character string candidate including the character code and the same character string candidate are identified. Attribute determination means for determining attributes of terms represented by character string candidates generated for a character string image, inter-attribute correspondence term storage means for storing terms having mutually related attributes in association with each other, and predetermined attributes Separately, attribute-based character string candidate storage means for storing the character string candidates whose attributes have been determined together with the priority order,
Regarding the character string candidates stored in the attribute-specific character string candidate storage means, which represent the attribute of the term stored in the inter-attribute correspondence term storage means, the character string candidates are used as keys. A term search means for searching a corresponding term from the attribute correspondence term storage means, and a character string candidate stored in the attribute-specific character string candidate storage means, which is a character of another attribute that is mutually related Priority changing means for increasing the priority of the character string candidates that are common to the search result of the term searching means using the column candidates as keys.

According to a ninth aspect of the present invention, in the document input method, the same character string image representing the character string in the captured image of the document is more than the likelihood reference value stored in the likelihood reference storage unit. A character string candidate generating step of generating a plurality of character string candidates that can be grammatically correctly interpreted from a combination of the character code candidates having a degree; and the character generated corresponding to the character string image. A display step of displaying a predetermined candidate among the character string candidates generated for the same character string image for the column candidate, and the character string candidate generated corresponding to the character string image. , An input step of inputting a correction instruction from the operator, and a character string that is erroneously recognized in accordance with the correction instruction for the character string candidate generated corresponding to the character string image. Correction code for correcting the character code to a correct character code, and after the correction, a corrected character string corresponding to the character string image and a candidate for the most likely character code corresponding to the same character string image. A comparison step of comparing with a character string, and after the comparison, based on the comparison result, from the ratio of the number of characters different between the two and the number of characters included in the one character string, the correct rate of character recognition And a correct answer rate calculation step for calculating the correct answer rate, and the calculated correct answer rate satisfies a predetermined condition by comparing with a correct answer rate reference value stored in the correct answer rate reference storage unit. In this case, a likelihood reference value changing step of changing the likelihood reference value in the likelihood reference storage unit according to the calculated correct answer rate and returning to the character string candidate generating step is provided.

In the document input method according to the tenth aspect, the same character string image representing the character string in the image of the captured document has a likelihood equal to or higher than the likelihood reference value in the likelihood reference storage unit. From the combination of the character code candidates,
A character string candidate generating step of generating a plurality of character string candidates that can be grammatically correctly interpreted; and the same character string for the character string candidates generated corresponding to the character string image. A display step of displaying a predetermined candidate among the character string candidates generated for the image, and a correction instruction from an operator is input to the character string candidate generated corresponding to the character string image. An input step, a correction instruction counting step of counting input correction instructions, and a character that is erroneously recognized according to the correction instruction with respect to the character string candidates generated corresponding to the character string image. A correction step of correcting the code to a correct character code, and after the correction, the counted correction instruction number is compared with a correction instruction number reference value stored in the correction instruction reference storage unit. When satisfying the condition, and a likelihood reference value changing step of changing the likelihood reference value in the likelihood reference storage unit in accordance with the counted number of correction instructions.

In the document input method according to the eleventh aspect of the present invention, a blank detecting step for detecting a blank between an image of one character cut out from a character string image representing one line of a character string and an image of one character adjacent thereto. If the detected space is larger than a predetermined size, the image of the one character and the adjacent 1
The image of characters is a word identification step for identifying characters belonging to different words, and a word identification for invalidating the identification result for a character string image in a predetermined area in an image of a captured document. And a limiting step.

According to a twelfth aspect of the present invention, in the document inputting method, the same character string image representing the character string in the image of the captured document can be combined with the character code candidates and grammatically correctly interpreted. A character string candidate generating step of generating a plurality of column candidates, a character string candidate storing step of storing the generated character string candidates in a character string candidate storage unit, and a predetermined character string corresponding to one character string image. And a display step of displaying the candidates of No. 1 from the character string candidate storage unit and evenly allocating the candidates within the identified ruled line frame for display.

In the document input method according to the thirteenth aspect, the direction of the character string in the image of the captured document is either vertical or horizontal based on the distribution of the image area and the blank area in the image of the captured document. If the direction of the character string can be interpreted as either the vertical or horizontal direction based on the character string direction identification step for identifying whether the character string is in the two directions, and the distribution of the image area and the blank area in the image of the captured document, With respect to the same character string image representing a character string, a character string candidate generating step of generating a plurality of character string candidates that can be grammatically correctly interpreted from the combination of the character code candidates, and the vertical character A predetermined candidate among the character string candidates generated for the directional character string image and a predetermined candidate among the character string candidates generated for the horizontal character string image are compared. To, and a character string direction determining step of determining an effective candidate character string direction and the string towards said given candidate satisfies the conditions set forth previously.

According to a fourteenth aspect of the present invention, there is provided a document input method for a ruled line frame identifying step for identifying a ruled line frame in an image of a captured document and for the same character string image representing a character string in the image of the captured document. A character string candidate generating step of generating a plurality of character string candidates that can be grammatically correctly interpreted from a combination of the character code candidates, and a predetermined candidate among the generated character string candidates, When printing in the identified ruled line frame, the character size of the image corresponding to the character code candidate in the character string candidate is checked, and the character size checked from the printable font sizes that can be printed The step of selecting the print character size that selects the print font size closest to the above, printing the identified ruled line frame, and marking the predetermined candidates of the generated character string candidates. A ruled line frame, and a printing step of printing by the print character size selected.

According to a fifteenth aspect of the present invention, in the document input method according to the fourteenth aspect, after the ruled line frame identifying step, the character code candidate storage section is referred to, and By referring to the character code candidate storage unit, a vertical interval calculation step of checking the position of the image corresponding to the character code candidate in the image of the document and calculating the size of the upper and lower blank areas sandwiching the image of the character. hand,
A left-right interval calculation step of calculating the size of the left and right blank areas sandwiching the image of the character by checking the position in the image of the document of the image corresponding to the character code candidate in the character string candidate, and selecting The vertical size of the selected print font size is limited so that the vertical size of the selected print font size does not exceed the calculated size of the upper and lower blank areas. The size limit step and the size of the selected print font size in the horizontal direction do not exceed the specified size compared with the calculated size of the left and right blank areas. And a left and right size limiting step for limiting the size.

According to a sixteenth aspect of the present invention, in the document inputting method, the same character string image representing a character string is combined with the character code candidates, and a plurality of character string candidates that can be correctly interpreted as a term are determined. A character string candidate generating step that generates together with the priority order, and a specific character code is identified from the generated character string candidates, and a character string candidate including the character code and the same character string as the character string candidate An attribute determination step of determining an attribute of a term represented by a character string candidate generated for an image, and the attribute-determined character string candidates are stored in the attribute-based character string candidate storage unit together with the priority according to a predetermined attribute. An attribute-based character string candidate storage step of storing the character string candidates stored in the attribute-based character string candidate storage unit, wherein A term search step of searching for a corresponding term from the inter-attribute correspondence term storage section using the candidate of each of the character strings as a key for a stored term representing an attribute, and the attribute-based character string candidate storage section Among the character string candidates stored in, the priority change step for increasing the priority of the character string candidates that are common to the search result using the character string candidates of other attributes that are mutually related as a key. ..

[0024]

According to the present invention, in the document input device according to the first aspect, the character code candidate storage means selects a plurality of character code candidates as a character recognition result of an image of one character.
The likelihood is stored in association with the likelihood indicating the certainty. The likelihood reference storage means stores a predetermined likelihood reference value of the likelihood. The character string candidate generation means, for the same character string image representing the character string in the image of the captured document,
A plurality of character string candidates that can be grammatically correctly interpreted are generated from a combination of the character code candidates having a likelihood equal to or higher than the likelihood reference value in the likelihood reference storage unit. The display means displays a predetermined candidate among the character string candidates generated for the same character string image with respect to the character string candidates generated corresponding to the character string image. The input means inputs a correction instruction from an operator to the character string candidates generated corresponding to the character string image. The correction means corrects a character code, which is erroneously recognized, into a correct character code in accordance with the correction instruction with respect to the character string candidates generated corresponding to the character string image. After the correction by the correction unit, the comparison unit compares the corrected character string corresponding to the character string image with a character string including a character code candidate with the highest likelihood corresponding to the same character string image. . The correct answer rate calculation means, after the comparison by the comparison means, based on the comparison result, from the ratio of the number of characters different between the two and the number of characters included in the one character string, the correct answer rate of character recognition, calculate. The correct answer rate reference storage means stores a predetermined correct answer rate reference value of the correct answer rate. Likelihood reference value changing means, after the correct answer rate is calculated, when the calculated correct answer rate satisfies a predetermined condition compared with the correct answer rate reference value, depending on the calculated correct answer rate. Then, the likelihood reference value in the likelihood reference storage means is changed.

Therefore, according to the present invention, the accuracy of character recognition represented by the accuracy rate is calculated for the character string candidates generated corresponding to the character string image, and the calculated character is calculated. Since the number of character code candidates used for the combination by the character string candidate generation means is limited according to the recognition accuracy, it is possible to perform the character string candidate generation processing adapted to the fluctuation of the character recognition accuracy. That is, when the original of the document to be recognized is an original in which the characters are blurred or smeared by being copied many times, and the character recognition accuracy is low, the character code candidates used by the character string candidate generating means are By increasing the number of candidates, the correct answer rate of the generated character string candidates can be increased. Further, when the character recognition accuracy is high, such as when the document of the document to be recognized is obtained as a clear image, the number of character code candidates used for combination by the character string candidate generation means is reduced. The character string candidate generating speed of the character string candidate generating means can be shortened, and as a result, the processing time of the entire document input device can be shortened.

In the document input device according to the present invention, the character code candidate storage means associates a plurality of character code candidates with a likelihood indicating the certainty thereof as a character recognition result of an image of one character. Remember. The likelihood reference storage means stores a predetermined likelihood reference value of the likelihood. The character string candidate generation means is a candidate for the character code having a likelihood equal to or higher than the likelihood reference value in the likelihood reference storage means for the same character string image representing a character string in the image of the captured document. A plurality of candidates for a character string that can be grammatically interpreted are generated from the combination of. The display means displays a predetermined candidate among the character string candidates generated for the same character string image with respect to the character string candidates generated corresponding to the character string image.
The input means inputs a correction instruction from an operator to the character string candidates generated corresponding to the character string image. The correction instruction counting means counts the correction instruction input from the input means. The correction instruction reference storage means stores a predetermined correction instruction number reference value of the correction instruction number. The correction means corrects a character code, which is erroneously recognized, into a correct character code in accordance with the correction instruction with respect to the character string candidates generated corresponding to the character string image. The likelihood reference value changing means, after the correction by the correcting means, compares the counted correction instruction number with a correction instruction number reference value stored in the correction instruction reference storage means and satisfies a predetermined condition. In this case, the likelihood reference value in the likelihood reference storage means is changed according to the counted number of correction instructions.

Therefore, according to the present invention, since the likelihood reference value in the likelihood reference storage means is changed according to the counted number of correction instructions, the trouble of calculating the correct answer rate in claim 1 can be saved, The above effect can be obtained more easily. In the document input device according to the third aspect, the space detecting unit detects a space between the image of one character cut out from the character string image representing the character string of one line and the image of one character adjacent thereto. When the detected blank is larger than a predetermined size, the word identifying means identifies that the one-character image and the one-character image adjacent thereto are characters belonging to different words. The word identification limiting means invalidates the identification result of the word identification means for a character string image in a predetermined area in the image of the captured document.

Therefore, according to the present invention, evenly-assigned characters can be correctly recognized without being divided in the middle of the word by ignoring the word division due to the space between the characters in the ruled line frame. , The recognition error correction effect can be sufficiently exerted. In the document input device according to the fourth aspect, the character code candidate storage means stores a plurality of character code candidates as a character recognition result of an image of one character cut out from the captured image of the document. The ruled line frame identifying means identifies the ruled line frame in the image of the captured document. The character string candidate generation means combines the character code candidates for the same character string image representing the character string in the captured image of the document, and creates a plurality of character string candidates that can be grammatically interpreted. To generate. The character string candidate storage means stores the generated character string candidates. The display means
A predetermined character string candidate corresponding to one character string image is read out from the character string candidate storage means, and is evenly allocated and displayed in the identified ruled line frame.

Therefore, according to the present invention, the predetermined character string candidates that are evenly displayed in the identified ruled line frame are stored as character strings in the character string candidate storage means. When correcting a character string that is evenly allocated and displayed, the character string can be corrected as a unit. Therefore, the operator does not need to input a correction instruction for each character as in the conventional case, and can easily correct the evenly displayed character strings.

In the document input device according to the fifth aspect, the character code candidate storage means stores a plurality of character code candidates as a character recognition result of an image of one character cut out from the image of the captured document. The character string direction identification means is
Based on the distribution of the image area and the blank area in the image of the captured document, it is identified whether the direction of the character string in the image of the captured document is vertical or horizontal. If the character string candidate generation means can interpret that the direction of the character string is either vertical or horizontal from the distribution of the image area and the blank area in the captured image of the document, the character string is generated in each of the two directions. For the same character string image representing, the character code candidates are combined to generate a plurality of character string candidates that can be grammatically interpreted correctly. The character string direction determination means determines a predetermined candidate of the character string candidates generated for the vertical character string image and a predetermined candidate of the character string candidates generated for the horizontal character string image. By comparing with the candidates, it is determined that the character string direction and the candidate of the character string in which the predetermined candidate satisfies a predetermined condition are valid.

Therefore, if it is possible to interpret from the distribution of the image area and the blank area in the image of the captured document that the direction of the character string is either vertical or horizontal, once the character string candidates are set in both directions. In addition, since the character string direction determining means determines the direction of the character string according to another criterion, the direction of the character string can be determined more accurately. Therefore, the effect of the recognition error correction processing can be sufficiently obtained.

In the document input device according to the sixth aspect, the ruled line frame identifying means identifies the ruled line frame in the image of the captured document. The character code candidate storage means selects a plurality of character code candidates as the character recognition result of the image of one character cut out from the captured image of the document, the position of the character image in the image of the document, and the character size. Memorize with. The character string candidate generation means is a plurality of character string candidates that can be grammatically correctly interpreted from the combination of the character code candidates for the same character string image representing the character string in the captured document image. To generate.
The print character size selection means, when printing a predetermined candidate of the generated character string candidates in the identified ruled line frame, an image corresponding to the character code candidate in the character string candidate. Check the character size of and select the print font size that is closest to the checked character size from the print font sizes that can be printed. The printing means prints the identified ruled line frame, and also prints a predetermined candidate of the generated character string candidates in the printed ruled line frame in the selected print character size.

Therefore, according to the present invention, the character size of the image of the character cut out from the image of the captured document is a predetermined candidate of the generated character string candidates in the printed ruled line frame. It is possible to print with the print font size closest to. That is, according to the present invention, a predetermined candidate of the generated character string candidates is printed in the printed ruled line frame with a print font size corresponding to the character size printed in the document to be recognized. Can be printed.

In the document input device according to claim 7, in the document input device according to claim 6, the vertical gap calculating means is
By referring to the character code candidate storage means, the position of the image corresponding to the character code candidate in the character string candidates in the image of the document is checked, and the size of the upper and lower blank areas sandwiching the image of the character is checked. Calculate the The left-right interval calculation means refers to the character code candidate storage means, checks the position in the image of the document of the image corresponding to the character code candidate in the character string candidates, and sandwiches the image of the character. Calculate the size of the left and right blank areas. The vertical size limiting means sets the vertical direction of the selected print font size so that the vertical size of the selected print font size does not exceed a predetermined size compared with the size of the calculated upper and lower blank areas. Limit the size of. The horizontal size limiting means sets the horizontal direction of the selected print font size so that the horizontal size of the selected print font size does not exceed the predetermined size compared with the calculated size of the left and right blank areas. Limit the size of. The print character size selection unit selects a print font size closest to the examined character size from the print font sizes limited by the vertical size limiting unit and the horizontal size limiting unit.

Therefore, according to the present invention, in addition to the above effects, a predetermined candidate of the generated character string candidates is displayed in the printed ruled line frame as a character balanced with the ruled line frame. Can be printed in size. In the document input device according to the eighth aspect, the character code candidate storage means stores a plurality of character code candidates as a character recognition result of an image of one character cut out from the captured image of the document.
The character string candidate generation means combines the character code candidates for the same character string image representing the character string, and generates a plurality of character string candidates that can be correctly interpreted as a term together with a predetermined priority. The attribute determination means identifies a specific character code from the generated character string candidates, and a character string candidate including the character code and a character string generated for the same character string image as the candidate character string. The attribute of the term represented by the candidate of is determined. The attribute correspondence term storage means includes terms having attributes that are related to each other (for example, address name and zip code, employee name and employee number, etc.).
Are stored in association with each other. The attribute-based character string candidate storage means is
The character string candidates whose attributes have been determined are stored together with the priority for each predetermined attribute. The term search means is a character string candidate stored in the attribute-based character string candidate storage means, and represents the attribute of a term stored in the inter-attribute correspondence term storage means Using the candidate (for example, zip code or employee number candidate) as a key, the corresponding term (for example, address name or employee name) is searched from the attribute correspondence term storage means. The priority changing means selects character string candidates (for example, zip code or employee number candidates) of other attributes that are mutually related from the character string candidates stored in the attribute-specific character string candidate storage means. Candidates for character strings that are common to the search results (for example, address names or employee names) of the term search means used as keys (for example, address name or employee name candidates)
Raise the priority of.

Therefore, according to the present invention, among terms expressing the same attribute, for example, "Ota" and "Ota",
When recognizing a term such as an address display or a name, which is a correct term, but contains many character strings that tend to cause recognition errors because it contains many similar glyphs, by comparing with the above search results, An accurate recognition error correction result can be obtained.

In the document input method according to the ninth aspect, a predetermined likelihood reference value and a predetermined correct answer rate reference value are stored in advance in the likelihood reference storage section and the correct answer rate reference storage section, respectively, and characters are stored. The column candidate generation step is performed on the same character string image representing the character string in the image of the captured document,
From a combination of the character code candidates having a likelihood equal to or greater than the likelihood reference value stored in the likelihood reference storage unit,
Generate multiple candidates for a string that can be grammatically interpreted. The displaying step displays a predetermined candidate among the character string candidates generated for the same character string image with respect to the character string candidates generated corresponding to the character string image. The input step inputs a correction instruction from an operator for the character string candidates generated corresponding to the character string image. The correction step corrects a character code that is erroneously recognized to a correct character code for the character string candidate generated corresponding to the character string image according to the correction instruction. In the comparing step, after the correction, the corrected character string corresponding to the character string image is compared with the character string including the character code candidate having the highest likelihood corresponding to the same character string image. In the correct answer rate calculating step, after the comparison, based on the comparison result, the correct answer rate of character recognition is calculated from the ratio of the number of different characters between the two and the number of characters included in the one character string. In the likelihood reference value changing step, after the correct answer rate is calculated, the calculated correct answer rate is compared with the correct answer rate reference value stored in the correct answer rate reference storage unit and satisfies a predetermined condition. To change the likelihood reference value in the likelihood reference storage unit according to the calculated correct answer rate, and return to the character string candidate generating step.

Therefore, according to the present invention, it is possible to obtain the same effect as that of the present invention according to claim 1. The document input method according to claim 10, wherein a predetermined likelihood reference value and a predetermined correction instruction number reference value are stored in a likelihood reference storage unit and a correction instruction reference storage unit, respectively, to generate a character string candidate. In the step, for the same character string image representing the character string in the image of the captured document, from the combination of the candidate of the character code having the likelihood equal to or more than the likelihood reference value in the likelihood reference storage unit, the grammar Generate multiple candidate strings that can be correctly interpreted. The displaying step displays a predetermined candidate among the character string candidates generated for the same character string image with respect to the character string candidates generated corresponding to the character string image. The input step inputs a correction instruction from an operator for the character string candidates generated corresponding to the character string image. The correction instruction counting step counts the input correction instruction. The correction step corrects a character code that is erroneously recognized to a correct character code for the character string candidate generated corresponding to the character string image according to the correction instruction. In the likelihood reference value changing step, after the correction, if the counted correction instruction number is compared with the correction instruction number reference value stored in the correction instruction reference storage unit and satisfies a predetermined condition, The likelihood reference value in the likelihood reference storage unit is changed according to the counted number of correction instructions.

Therefore, according to the present invention, it is possible to obtain the same effect as that of the present invention according to claim 2. In the document input method according to the eleventh aspect, in the white space detecting step, a white space is detected between an image of one character cut out from a character string image representing a character string of one line and an image of one character adjacent thereto. In the word identifying step, when the detected blank is larger than a predetermined size, the one-character image and the adjacent one-character image are identified as characters belonging to different words. The word identification limiting step invalidates the identification result for the character string image in a predetermined area in the captured image of the document.

Therefore, according to the present invention, it is possible to obtain the same effect as that of the present invention according to claim 3. In the document input method according to claim 12, the ruled line frame identifying step identifies a ruled line frame in the image of the captured document. The character string candidate generating step combines a plurality of character string candidates that can be grammatically correctly interpreted by combining the character code candidates for the same character string image representing the character string in the captured document image. To generate. The character string candidate storing step stores the generated character string candidates in the character string candidate storage unit. In the display step, a predetermined character string candidate corresponding to one character string image is read from the character string candidate storage unit, and is equally distributed and displayed in the identified ruled line frame.

Therefore, according to the present invention, it is possible to obtain the same effect as that of the present invention according to claim 4. The document input method according to claim 13, wherein in the character string direction identifying step, the direction of the character string in the image of the captured document is determined based on the distribution of the image area and the blank area in the image of the captured document. Identify which direction is vertical or horizontal. In the character string candidate generating step, when it is possible to interpret that the direction of the character string is either vertical or horizontal from the distribution of the image area and the blank area in the captured image of the document, the character string is extracted in both directions. For the same character string image shown, a plurality of character string candidates that can be grammatically correctly interpreted are generated from the combination of the character code candidates. In the character string direction determination step, a predetermined candidate of the character string candidates generated for the vertical character string image and a predetermined candidate of the character string candidates generated for the horizontal character string image are determined. By comparing with the candidates, it is determined that the character string direction and the candidate of the character string in which the predetermined candidate satisfies a predetermined condition are valid.

Therefore, according to the present invention, it is possible to obtain the same effect as that of the present invention according to claim 5. 15. The document input method according to claim 14, wherein while recognizing a character image, a plurality of character code candidates are stored in advance in a character code candidate storage unit together with a position of the character image in the image of the document and a character size. Then, the ruled line frame identifying step identifies the ruled line frame in the image of the captured document. In the character string candidate generating step, a plurality of character string candidates that can be grammatically correctly interpreted from the combination of the character code candidates for the same character string image representing the character string in the captured image of the document. To generate. The printing character size selecting step, when printing a predetermined candidate of the generated character string candidates in the identified ruled line frame,
The character size of the image corresponding to the character code candidate in the character string candidates is checked, and the print font size closest to the checked character size is selected from the print font sizes that can be printed. In the printing step, the identified ruled line frame is printed, and a predetermined candidate of the generated character string candidates is printed in the printed ruled line frame with the selected print character size.

Therefore, according to the present invention, it is possible to obtain the same effect as that of the present invention described in claim 6. In the document input method according to claim 15, in the document input method according to claim 14, the vertical spacing calculation step refers to the character code candidate storage unit after the ruled line frame identification step, and the character string candidates are referred to. The position of the image corresponding to the character code candidate in the inside of the image of the document is checked, and the sizes of the upper and lower blank areas sandwiching the image of the character are calculated. The left-right interval calculating step refers to the character code candidate storage unit to check the position in the image of the document of the image corresponding to the character code candidate in the character string candidates, and sandwiches the image of the character. Calculate the size of the left and right blank areas.
The vertical size limiting step is performed in the vertical direction of the selected print font size so that the vertical size of the selected print font size does not exceed the predetermined size compared with the size of the calculated upper and lower blank areas. Limit the size of. The right / left size limiting step is performed in the left / right direction of the selected print font size so that the size of the selected print font size in the left / right direction does not exceed the calculated size of the left and right blank areas. Limit the size of.

Therefore, according to the present invention, it is possible to obtain the same effect as that of the present invention described in claim 7. 17. The document input method according to claim 16, wherein terms having mutually related attributes are associated with each other and are stored in advance in the inter-attribute corresponding term storage unit, and the character string candidate generation step is the same character string image representing a character string. With respect to, a combination of the character code candidates is generated, and a plurality of character string candidates that can be correctly interpreted as terms are generated with a predetermined priority. The attribute determining step identifies a specific character code from the generated character string candidates, and detects a character string candidate including the character code and a character string generated for the same character string image as the candidate character string. The attribute of the term represented by the candidate of is determined.
The attribute-specific character string candidate storage step stores, for each predetermined attribute, the character string candidates whose attributes have been determined together with the priority in the attribute-specific character string candidate storage unit. The term search step is a candidate for a character string stored in the attribute-based character string candidate storage unit, and represents the attribute of a term stored in the inter-attribute correspondence term storage unit,
Using the candidate of each character string as a key, the corresponding term is retrieved from the inter-attribute corresponding term storage unit. In the priority changing step, among the character string candidates stored in the attribute-specific character string candidate storage unit, a character string common to the search result is keyed to a character string candidate of another attribute associated with each other. Increase the priority of candidates.

Therefore, according to the present invention, it is possible to obtain the same effect as that of the present invention according to claim 8.

[0046]

1 is a block diagram showing a hardware configuration of a document input device 100 according to a first embodiment of the present invention. The document input device 100 includes a CPU 120, a memory control unit / bus control unit 102, a main memory 101,
Local bus 104, graphics control unit 10
5, display 106, SCSI (Small Co
computer system interface) control unit 107, hard disk device 108, scanner 109, bus bridge circuit 110, external bus 11
1, an input / output control unit 112, a keyboard 113, a printer 114, and a mouse 115. The main part of the document input device 100 is, for example, a personal computer Panacom V24 (CF-V2411RWE).
A general-purpose personal computer such as Matsushita Electric Industrial Co., Ltd., etc., and the individual input / output devices and the like that are the constituents thereof may be standard products that are commercially available. Computing device (C
The PU) 101 is connected to the main memory 101 and the local bus 104 via the memory control unit / bus control unit 102.

The main memory 101 stores a program to be executed by the CPU 120 and data input from each of the input / output disk devices, and provides a work place for the CPU 120 by temporarily storing data during processing. FIG. 2 is an explanatory diagram showing the division of the storage area provided on the main memory 101 for storing the recognition result of the document input device 100 and the storage contents. As shown on the left side of FIG. 2, the image information storage unit 20 is provided on the main memory 101.
1, a block management information storage unit 202, a display data storage unit 203, and a recognition column management unit 204 are provided.

The image information storage unit 201 stores the image of the document to be input, which is read by the scanner 109 or the like, for example. The block management information storage unit 202
The block management information 210 described below, which is created at each stage in the document input process and is shown on the right side of FIG. 2, is stored. The display data storage unit 203 stores, for example, display data such as an image input from the scanner 109 as a recognition target and a recognition result thereof.

The recognition sequence management unit 204 stores each recognition sequence management information 301, which will be described later, created for each recognition sequence. The recognition target block refers to a rectangular area designated as a recognition target in an image input from the scanner 109, and the recognition column is included in the designated recognition target block, for example, corresponds to a single-line character string. Image

The block management information 210 is stored in the scanner 1
In the image input from 09, it is created corresponding to each recognition target block designated as the recognition target text, and each block management information 210 includes image position information 211, ruled line frame information 212, and a character string direction 213. And a recognition column management unit address 214. The image position information 211 is the block identification number of the recognition target block managed by the block management information 210 and the image information in which the recognition target block is stored in the image information storage unit 201, that is, input from the scanner 109. The coordinates located in the image information are shown.

The ruled line frame information 212 indicates the presence or absence of a ruled line frame in the recognition target block. The character string direction 213 indicates whether the direction of the character string is vertical or horizontal in the recognition target block. However, when a character string direction is interpreted in both the vertical direction and the horizontal direction for a certain recognition target block, two pieces of recognition string management information 301, which will be described later, are generated corresponding to each case.

The recognition column management unit address 214 indicates the start address of the storage area in the recognition column management unit 204 in which the recognition column management information 301 for the recognition column included in the recognition target block is stored. FIG. 3 is an explanatory diagram showing a data format of the recognition sequence management information 301 stored in the recognition sequence management unit 204.

The recognition sequence management unit 204 stores the recognition sequence management information 301 created for each recognition sequence included in one recognition target block for each recognition target block. As shown in FIG. 3, each recognition string management information 301 includes recognition character information 302 created for each character included in the recognition string and a correction result 308 obtained for the recognition string.

The recognized character information 302 indicates a recognition result for each character in the recognition string, and includes character position information 303, delimiter information 304, and candidate character information 307. The candidate character information 307 includes a candidate character code 305 and its likelihood 306. The character position information 303 represents the position of the image corresponding to the recognized character information 302 in the recognition target block, for example, a rectangular area on the bitmap with the upper left corner of the recognition target block as coordinates (0, 0). Indicated by position coordinates.

The delimiter information 304 is the character position information 303.
Indicates whether the text image at the position indicated by is the end of a word or phrase. For example, "1" is written if there is a blank of a size equal to or larger than the full-width character immediately after the text image, and "0" is written otherwise. Thus, for example, in a general document, it is possible to eliminate the error of interpreting blank and widely separated columns in a column layout as one line. In addition, here, the size of the space in which the word division is valid is set to be a double-byte character or more, but may be a double-byte character or more, a triple-width character or more, or a quadruple-width character or more.

The candidate character code 305 indicates the character code of the candidate character obtained as a recognition result. Likelihood 306 is
The certainty of the candidate character code 305 obtained as a result of the collation between the character recognition feature extracted from the image at the position indicated by the character position information 303 and the type recognition dictionary separately held by the document input device 100 is shown. . Since the method of calculating the likelihood 306 is the same as the method in the conventional character recognition, description thereof will be omitted.

The correction result 308 is created for each recognition column,
It consists of multiple candidates. The correction result 308 is obtained by combining a candidate character code 305, which is a recognition result for each character, with reference to a word dictionary and generating a plurality of bunsetsu interpretations that can be interpreted as a bunsetsu. The correction result, which is the result of the correction according to the test or the like, is shown. The recognition sequence confirmed by the correction result confirmation unit 213 is stored as the first candidate for the correction result 308.

FIG. 4 is an explanatory diagram showing storage area divisions of the word dictionary storage unit 412 provided on the main memory 101.
The word dictionary storage unit 412 stores various collation dictionaries and tables used for recognition error correction. The word dictionary storage unit 412 includes a dictionary management information storage unit 401, a general dictionary unit 402, a proper noun dictionary unit 403, and a form term dictionary unit 40.
4. A postal code-address matching table unit 405 and a telephone number-address matching table unit 406 are provided.

The dictionary management information storage unit 401 includes a general dictionary address storage unit 407 and a proper noun dictionary address storage unit 4
08, a form term dictionary address storage unit 409, a postal code table address storage unit 410, and a telephone number table address storage unit 411 are provided, and a word dictionary storage in which various collating dictionaries and tables to be described later used for recognition error correction are stored. The addresses in the unit 412 are collectively held.

The general dictionary section 402 stores a general dictionary used for correcting a general document, such as a word dictionary in which independent words and adjunct words are recorded in a trie structure. The proper noun dictionary unit 403 stores proper nouns such as a company name, a department name, and a personal name, and stores a later-described proper noun dictionary used for correcting a business card. The form term dictionary unit 404 stores the form terms that appear frequently in the form and stores a form term dictionary (to be described later) used for correction when inputting the form format.

The postal code / address matching table section 405 is
A zip code-address matching table that stores address name data used for displaying addresses in the area is stored in correspondence with the zip code. The telephone number-address matching table unit 406 is
A telephone number-address matching table in which address name data used for displaying an address in the area is recorded in the area code of the telephone number and the local office number is stored.

The general dictionary address storage unit 407 stores the address of the general dictionary unit 402 in the word dictionary storage unit 412. The proper noun dictionary address storage unit 408 stores the address of the proper noun dictionary unit 403 in the word dictionary storage unit 412. The form term dictionary address storage unit 409 stores the address of the form term dictionary unit 404 in the word dictionary storage unit 412.

Postal code table address storage unit 410
Stores the address of the postal code / address matching table unit 405 in the word dictionary storage unit 412. The telephone number table address storage unit 411 stores the address of the telephone number / address matching table unit 406 in the word dictionary storage unit 412. FIG. 5 is an explanatory diagram showing storage area divisions of the business card data recognition result temporary storage unit 501 provided on the main memory 101.

The business card data recognition result temporary storage unit 501 is
Name storage unit 502, organization name storage unit 503, department name storage unit 504, address storage unit 505, telephone number storage unit 5
06 and a postal code storage unit 507 are provided to temporarily store the recognition result in the business card input process. Name storage unit 502
Stores the correction results 308 of the names of the first to fifth candidates read from the image printed on the business card.

The organization name storage unit 503 stores the correction results 308 of the organization names of the first to fifth candidates read from the image printed on the business card. The department name storage unit 504, read from the image printed on the business card,
Correction results of the department names of the first to fifth candidates 30
Remember 8. The address storage unit 505 stores the correction result 308 of the addresses from the first candidate to the fifth candidate read from the image printed on the business card.

The telephone number storage unit 506 stores the correction results 308 of the telephone numbers of the first to fifth candidates read from the image printed on the business card. Zip code storage unit 5
07, the postcode correction results 308 of the first to fifth candidates read from the image printed on the business card are stored. FIG. 6 is an explanatory diagram showing storage area divisions of the name list information storage unit 601 provided on the main memory 101.

The name list information storage unit 601 stores name list information 608 registered for each individual. Each list information 608 is
A name area 602, an affiliated organization name area 603, an affiliated department name area 604, an address area 605, a telephone number area 606, and a postal code area 607 are provided, and the name list information registered in the document input device 100 is stored. The name area 602 stores the registered personal name.

The affiliated organization name area 603 stores the organization name to which the individual belongs. The department name area 604 stores the department name to which the individual belongs. The address area 605 is
Remember the address of the individual. The telephone number area 606 stores the telephone number of the individual. The postal code area 607 is
Store the postal code of the address.

FIG. 7 is an explanatory diagram showing the storage area division of the work area 700 provided on the main memory 101. The work area 700 includes a document type temporary storage unit 701, a block number temporary storage unit 702, a precision level storage unit 703, an operation level storage unit 704, a likelihood level storage unit 705, a correction dictionary address storage unit 706, and a correction key frequency storage. A unit 707 and a processing buffer 708 are provided to store various variables and intermediate processing results generated at each stage of the recognition processing and the recognition correction processing.

The document type temporary storage unit 701 stores the type of the document which is the current recognition target. The block number temporary storage unit 702 stores the identification number of the recognition target block that is currently the recognition target. Accuracy level storage unit 703
Stores a predetermined accuracy level as a reference value of the accuracy of the character recognition result. The accuracy of the character recognition result is obtained by comparing and collating the character string obtained by arranging the first candidates of the candidate character code 305, which is the character recognition result, with the finally confirmed character string, and the number of matched characters and the number of matched characters. Expressed as a ratio of.

The operation level storage unit 704 is, for example, a back key used for instructing the display of the next candidate when the operator corrects the correction result 308, or a delete key used for rewriting a character recognized by mistake. For the correction instruction keys in the keyboard 113, a predetermined operation level is stored as a reference value of the number of keystrokes. The likelihood level storage unit 705 stores a likelihood level that is predetermined as a lower limit value of the likelihood 306 of the candidate character code 305 that is the character recognition result.

The correction dictionary address storage unit 706 stores up to three addresses of the correction dictionary currently used by the document input device 100. The correction key frequency storage unit 707 is
The number of times of hitting the backward key, the delete key, etc. used for correcting the displayed correction result 308 is sequentially recorded. The processing buffer 708 stores the intermediate processing result generated by the document input device 100 at each stage of the recognition processing and the recognition correction processing.

The memory control unit / bus control unit 102 controls data transmission between the CPU 120 and the main memory 101 and the local bus 104. The local bus 104 is a data transmission path. Local bus 10
4, a display 106 is coupled to the SCSI control unit 10 via a graphics control unit 105.
Via the hard disk device 108, the scanner 10
9 are connected.

The graphics control unit 105 uses the CP
The data such as the processing result by the U120 is converted into display data represented by, for example, bitmap data, and the display 106 is controlled to display the display data on the display surface. The display 106 displays display data represented by bitmap data, for example.

The SCSI control unit 107 is a general-purpose interface for small machines, to which various input / output devices can be connected and which can be shared by a plurality of computers. The hard disk device 108 stores data on a magnetic disk provided inside.

The scanner 109 reads data printed on a paper surface or the like by irradiating a printed matter or the like with light while scanning and detecting the amount of the reflected light. Furthermore, the bus bridge circuit 110 allows the local bus 1
A keyboard 113 and a printer 11 are connected to an external bus 111 connected to the printer 04 via an input / output control unit 112.
4, a mouse 115 and the like are connected. The bus bridge circuit 110 is a connection unit that connects the local bus 104 and the external bus 111 to mutually transmit data.

The external bus 111 is a transmission path for transmitting data from external input / output devices such as the keyboard 113, printer 114, mouse 115 and the like. The input / output control unit 112 controls low-speed input / output devices such as the keyboard 113, the printer 114, and the mouse 115 to connect the input / output devices to the external bus 111.

The keyboard 113 has a plurality of keys,
The key signal corresponding to the pressed key is output. The printer 114 outputs the processing result of the CPU 120 and the like by printing it on a paper surface. The mouse 115 includes a roller and a click key, outputs a cursor control signal corresponding to the rotation of the roller, and outputs a click signal when the click key is pressed.

The CPU 120 exchanges data with the main memory 101 via the memory control unit / bus control unit 102. The CPU 120 executes the procedure program stored in the main memory 101 and controls each of the input / output devices to implement the document input device 100 of this embodiment described later. Hereinafter, the document input device 10
0, that is, the processing procedure of the CPU 120 will be described with reference to flowcharts and specific examples. In the following description, various variables used in the document input device 100 of this embodiment will be described as being initialized in advance.

FIG. 8 is a flowchart showing an outline of the entire processing procedure of the CPU 120. FIG. 9 is an explanatory diagram showing a specific example of the display screen and data in the initial stage of the document input process. When the operator reads the document intended for input from the scanner 109 (step S8)
01), the CPU 120 causes the display 106 to display the menu screen shown in FIG.
The operator is asked whether the document input from is "form form", "business card", or "general document" (step S802).

The CPU 120 branches the subsequent processing according to the response from the operator (step S803). Note that here, the operator inputs the type of the document input from the menu screen, but the scanner 10
The CPU 120 may recognize the document type based on the size of the document read from the document 9, the presence / absence of a ruled line frame, and the like.

In step S803, "general document"
When is selected on the menu screen, the CPU 120
The address of the general dictionary unit 402 that stores the general dictionary for general sentences is set to the dictionary management information storage unit 401.
It is read out from the general dictionary address storage unit 407 in the memory and stored in the correction dictionary address storage unit 706 (step S804).

The CPU 120 performs text recognition input processing and recognition error correction processing for general documents, which will be described later.
In this case, the CPU 120 causes the display 106 to display the first candidate of the correction result 308 together with the text image as the recognition target on the display 106 every time the correction result 308 is obtained for a predetermined number, for example, four recognition columns. The input device 100 activates a text editing program separately held. The operator uses the displayed correction result 308.
With respect to, the recognition error portion is corrected to a correct character using the keyboard 113 or the like, and when the correction is completed, confirmation input is performed. The CPU 120 performs recognition error correction processing for the next character string based on the correction result. In this way, the CPU 120 repeats the recognition error correction processing and the correction result correction processing for each predetermined number of recognition columns, and when the processing is completed for all characters, the recognized and input document is required. If it is printed, the document input is completed (step S805). At this time, the CPU 120 monitors the keystrokes of the backward key and the delete key in the keyboard 113 used by the operator to correct the correction result 308, and increments the value of the corrected key frequency storage unit 707 each time the keystroke is made. .

When "business card" is selected on the menu screen in step S803, the CPU 120 causes the proper noun dictionary section for storing proper noun dictionaries for proper nouns such as business cards and address books. The address of 403 is read from the proper noun dictionary address storage unit 408 in the dictionary management information storage unit 401 and stored in the correction dictionary address storage unit 706 (step S807).

Next, the CPU 120 carries out text recognition input for business cards, which will be described later (step S80).
8), the recognized business card data is collated with the name list information stored in the name list information storage unit 601, and the business card data for which the same individual name is not stored in the name list information storage unit 601 is treated as new name list information. It is newly registered in the name list information storage unit 601. Further, the same personal name is stored in the name list information storage unit 601.
For the business card data stored in the other but different in other parts, the name list information is updated with the information from the business card data (step S809).

In step S803, "form form"
When is selected on the menu screen, the CPU 120
In addition to the general dictionary unit 402 that stores the general dictionary, the address of the form term dictionary unit 404 that stores the form term dictionary in which the form terms that frequently appear in the form are stored is set to the general dictionary in the dictionary management information storage unit 401. It is read from the address storage unit 407 and the form term dictionary address storage unit 409 and stored in the correction dictionary address storage unit 706 (step S810).

Next, the CPU 120 extracts the block surrounded by the ruled line frame and the text area block surrounded by the blank from the form, and the extracted both are
The block management information 210 is individually created as a ruled line frame block and a general block, and is stored in the block management information storage unit 202 (step S811). In addition,
Whether it is a ruled line frame block or a general block is written in the item of ruled line frame information 212 in the block management information 210. In addition, these methods are described in “Technical Report Vol.90 No. 235 (PRU90 68-
80) p33-38 "or" area division of document image using blank area information "Proceedings of IEICE Spring Conference
D-568 p7-410 (1992) ”and the like are used, and since this detailed procedure is not the main subject of the present invention, detailed description thereof will be omitted.

After the CPU 120 recognizes the internal text characters and the character string direction for each of the ruled line frame block and the general block (step S812), the recognized character is recognized in the character string recognized in each block. Equal allocation information corresponding to the column direction is added (step S813). The processing of steps S812 and S813 will be described in detail later.

FIG. 10 shows step S805 shown in FIG.
3 is a flowchart illustrating in detail the content of the text recognition process in FIG. As shown in FIG. 9B, the CPU 120 causes the display 106 to display the document input from the scanner 109. The operator designates the recognition target block by using the mouse 115 while checking the image of the document displayed on the display 106 (step S1001). The shaded area in FIG. 9B indicates an example of the recognition target block designated by the operator in step S1001.

The CPU 120 estimates a character area in the recognition target block and its character direction based on a histogram obtained by mapping the image in the designated recognition target block in the horizontal and vertical directions. Specifically, the CPU 120
Generates a histogram of images that are mapped in the horizontal and vertical directions, and estimates the direction in which there is periodicity in the image and the blank area and the distance between the images is narrow as the character direction. The estimated character string direction and the position of the image of the recognition target block designated in step S1001 are written in the items of the character string direction 213 and the image position information 203 in the block management information 210, respectively. Furthermore, CPU1
20 cuts out an image corresponding to each recognition string included in the recognition target block based on the display data storage unit 203 and the character string direction 213, and at the same time, extracts one character included in each recognition string. Cut out the image of. The position of the image corresponding to each cut out character is written in the item of the character position information 303 of the recognized character information 302 (step S1002).

FIG. 9C shows four recognition columns cut out from the recognition target block shown in FIG. 9B by the processing of step S1002. Therefore, FIG. 9 (b)
4 pieces of recognition sequence management information 301 are created for the image in the recognition target block shown in FIG.
For example, for the recognition string 2, 32 pieces of recognition character information 30
2 is created and stored in the recognition column management unit 204.

Next, the CPU 120 causes the character position information 3
The character identification feature is extracted from the image at the position indicated by 03, and the extracted character identification feature and the document input device 1 are also extracted.
00 is compared and collated with a type recognition dictionary that is separately held, and candidate character codes 305 from the first rank to the tenth rank are generated together with the likelihood 306 in descending order of similarity (step S1003).

The generated candidate character code 305 and likelihood 306 are the candidate character information 30 in the recognized character information 302.
7 is stored in the recognition column management unit 204. This unit character recognition method is based on "National Tehchni".
cal report vol. 35 No. 1 (Fe
b. 1989) ”, etc., and is not the main subject of the present invention, and therefore the description thereof is omitted. As a result, the recognition string management unit 204 causes the recognition character information 30 regarding each recognition character to be recognized.
2 is stored. At this time, the CPU 120 validates the blank area between the character areas as a word or phrase delimiter when the blank area between the character areas is larger than the full width, and otherwise. Invalidate. The CPU 120 determines that the recognized character information 302 corresponding to an image immediately after which is a valid word segmentation by the blank.
“1” is written in the delimiter information 304 of the above, and “0” is written in the delimiter information 304 in the image other than that.

FIG. 11 is an explanatory diagram showing a specific example of the recognized character information 302 generated by the character recognition processing and the display screen displayed after recognition error correction. FIG. 11 (a)
Shows an example of the contents of the recognized character information 302 when the CPU 120 recognizes the image corresponding to the 20th character in the recognition sequence 3 shown in FIG. 9C as “day” in step S1003. ..

Next, in order to increase the matching rate of the character recognition result and reduce unnecessary post-processing collation, the CPU 120 selects the candidate character code 305 of the fifth rank or lower among the candidate character codes 305 in the candidate character information 307. And its likelihood 306
, Which is lower than the likelihood level stored in the likelihood level storage unit 705 is removed from the candidate character information 307 (step S1004). In the example of the recognized character information 302 shown in FIG. 11A, for example, when the likelihood level is “40”, the 8th to 10th candidates whose likelihood 306 is lower than “40” are The recognized character candidates that are removed from the candidate character information 307 and are in the candidate character information 307 are limited to seven candidates from the first rank candidate to the seventh rank candidate. The process of step S1004 will be described in more detail later.

The CPU 120 collates the candidate character code 305 in the candidate character information 307 with the correction dictionary designated by the correction dictionary address storage unit 706 in character string units in recognition string units corresponding to the recognition string management information 301. However, the correction result composed of a sequence of phrases that can be interpreted as a phrase according to a predetermined priority order is written in the correction result 308 (step S1005). The process of step S1005 will be described in detail later.

Each time the CPU 120 writes the correction results 308 corresponding to a predetermined number of consecutive recognition columns, for example, four recognition columns (step S1006), it performs a correction result correction process described later (step S1007), and at this stage. If the recognition remaining block still exists and the recognition input is not completed (step S1008), the process returns to the process of step S1003. If not, the correction result correction process described below is performed as in the process of step S1007. Then, the process ends (step S1009).

FIG. 10A shows a specific processing procedure of the correction result correction processing shown in steps S1007 and S1009. The CPU 120 displays the correction result 3
08 first candidate, together with the image of the recognition target block,
The text is displayed on the display 106 and the text editing function separately stored in the document input device 100 is activated. Figure 11
(B) shows a display example of the display 106 at this time. The upper half of the display screen shown in FIG. 11B shows the image of the recognition target block, and the lower half of the display screen shows the correction result 30.
8 shows the result of correction consisting of 8 first place candidates. When the displayed correction result 308 is incorrect, the operator uses the activated text editing function to use the keyboard 113 to correct the incorrect character in the displayed correction result 308 (step S1011). ). At this time, the CPU 120 monitors the keystrokes of the backward key and the delete key in the keyboard 113 used by the operator to correct the correction result 308, and increments the value of the corrected key frequency storage unit 707 each time the keystroke is made. .

The operator completes the correction, and the correction result 30
When the confirmation instruction of 8 is input from the keyboard 113, the CP
U120 compares and collates the character string obtained by extracting the first candidate of the candidate character code 305 from the candidate character information 307 stored in the recognition string management unit 204 with the final confirmed character string, and The degree of coincidence between columns is calculated. That is, the CPU 120, based on the following formula,
The accuracy r is calculated (step S1012).

R / 100 = (number of matched characters) / (number of collated characters) The CPU 120 compares the calculated accuracy r with a predetermined accuracy level stored in the accuracy level storage unit 703 ( (Step S1013), if the calculated precision r is lower than the precision level, the likelihood level in the likelihood level storage unit 705 is changed to a value 10% lower, and the process ends (step S1014). It should be noted that here, only the case where the calculated precision r is lower than the precision level stored in the precision level storage unit 703 has been described, but conversely, the calculated precision r is the precision level storage unit 703. If the accuracy level is higher than the accuracy level stored therein, the likelihood level in the likelihood level storage unit 705 may be changed to a value 10% higher.

In the above, the accuracy r calculated as described above is compared with the accuracy level stored in the accuracy level storage unit 703, and the likelihood level storage unit 703 is based on the comparison result. The case of changing the likelihood level in 705 has been described, but the likelihood level storage unit 705 is based on another reference value.
The likelihood level within may be changed. Specifically, as described in the processing of step S1011, the CPU 12
For 0, the number of times the backward key and the delete key in the keyboard 113 used by the operator to correct the correction result 308 are counted in the corrected key frequency storage unit 707 each time. As a result, when the operator completes the correction and inputs a confirmation instruction of the correction result 308 from the keyboard 113, the CPU 120 stores the number of keystrokes stored in the correction key frequency storage unit 707 and the operation level storage unit 704. The corrected key frequency storage unit 7 is compared with the operated operation level.
If the number of keystrokes stored in 07 is larger than the operation level, the likelihood level in the likelihood level storage unit 705 is set to 10
The value may be changed to a lower value. Similarly,
When the number of keystrokes stored in the modified key frequency storage unit 707 is smaller than the operation level, the likelihood level storage unit 70
The likelihood level in 5 may be changed to a value 10% higher. This makes it possible to easily determine the accuracy of character recognition without calculating the accuracy r.

FIG. 12 is a flow chart for explaining the candidate character limiting process in step S1004 of FIG. The CPU 120 sets the limited start candidate number N to "5" (step S1201). The CPU 120 is the Nth
The candidate character code 305, which is a position candidate, is extracted together with the likelihood 306 from the candidate character information 307 stored in the recognition string management unit 204 (step S1202), and the extracted likelihood 306 is stored in the likelihood level storage unit 705. A comparison is made with the stored likelihood level to determine whether it is smaller (step S1203). If it is smaller, the value of the likelihood 306 of the Nth or lower candidate in the candidate character information 307 is set to “0” to invalidate this candidate, and the candidate character limiting process is ended (step S1204).

If it is not small, N is incremented (step S1205), the value of N is compared with the maximum number of candidates, "10" (step S1206), and if the value of N is "10" or less, step It returns to the process of S1202. Otherwise it ends. The variable N is stored in the area of the processing buffer 708. 13 is the same as FIG.
It is a flowchart explaining the recognition error correction process in step S1005 of 0.

The CPU 120 sets the interpretation start point to the head position of the recognition sequence stored in the recognition sequence management unit 204 (step S1301). The CPU 120 uses the recognition sequence management unit 2
The candidate character code 305 stored in the candidate character information 307 in 04 is collated with the general dictionary designated by the correction dictionary address storage unit 706 to generate a phrase interpretation that starts from the interpretation start point and is valid as a Japanese grammar. Yes (Step S
1302). The process of step S1302 will be described later in detail.

Next, the CPU 120 selects, in each recognition sequence,
It is determined whether there is a character string left to be processed (step S1303). If there is a character string left unprocessed, the next interpretation start point is set, and then the process returns to step S1302 to obtain the next phrase interpretation from the set interpretation start point (step S1304). If there is no string left to process,
For the recognition sequence to be processed, all bunsetsu interpretation sequences, which are combinations of bunsetsu interpretations that can be interpreted as bunsetsu, have been generated. Arrange the items in order. In addition, when there is a bunsetsu interpretation string having the same number of bunsetsu interpretations in the generated bunsetsu interpretation string, the likelihood 306 of each candidate character code 305 forming each bunsetsu interpretation string.
The larger sum of the values of (hereinafter, referred to as "recognition likelihood score") is arranged first. The CPU 120 prioritizes the respective phrase interpretation strings arranged in this way in the order of arrangement (step S1305).

For example, in the bunsetsu interpretation sequence "present / seven / daily", the number of bunsetsu interpretations is 3; The latter has higher priority. Further, for example, the recognition likelihood score of the phrase interpretation sequence "currently / daily" is "present",
"A", "de", "ha", "day", "constant", "target",
It is the sum of the values of the likelihood 306 of each candidate character code 305 of “NA”. On the other hand, the phrase interpretation sequence "currently / white birch" is "present", "present", "de", "ha",
Candidate character code 3 for "white", "birch", "target", and "na"
It is the sum of the values of the likelihood 306 of 05. Due to the difference in the recognition likelihood scores, the first candidate is “currently / daily”, the second candidate is “currently / birchish” in the generated phrase interpretation sequence. The third candidate is ranked as “current / 7 / daily”. Note that this interpretation start point is also a variable stored in the processing buffer 708.
It is represented by the character position information 303 of the first character of the phrase interpretation.

FIG. 13B is a flow chart for explaining the phrase interpretation generation processing in step S1302 of FIG. 13A in more detail. The CPU 120 refers to the candidate character information 307 corresponding to the character position information 303 specified at the interpretation start point, and corrects the recognized character candidate stored in the candidate character code 305 in the correction dictionary address storage unit 706. All the independent word candidates that can be interpreted as an independent word are generated by collating the characters one by one while collating them with the dictionary for use (step S1311). The generated independent word candidate is stored in the processing buffer 708 secured as an intermediate buffer. The dictionary search method used here is
"Natural language processing in character recognition Information processing Vol.
34 No. 10 pp. 1274-1280 (199
3) ”, etc., and is designed to speed up the dictionary search by digitally searching a trie-structured dictionary by state transitions.

When the retrieval of the independent word is completed, the CPU 120
Sets the target independent word number n to "1" (step S1312), and then sets the n-th word generated in step S1311.
For the th independent word, starting from the next starting point of interpretation,
The subsequent character string is collated with the attached word dictionary (step S1313). Similar to the independent word dictionary, the adjunct word dictionary has adjunct words in a trie structure. Also, CP
Based on the independent word dictionary, the U120 tests the connection between the independent word and the adjunct word and the connection between the adjunct words, and leaves only the one that holds as a Japanese grammar as the adjunct word interpretation. The grammar verification process and the like are methods that are also used in the conventional kana-kanji conversion and the like, and the details thereof are not the main subject of the present invention, and thus the description thereof will be omitted. As a result, the phrase interpretation that can be interpreted as one phrase is the processing buffer 70.
8 generated on.

After incrementing the target independent word number n (step S1314), the CPU 120 determines that the independent word candidate of the target independent word number n is not yet the final candidate among the generated independent word candidates (step S1314). S13
15) Return to the processing of step S1313, and similarly to the above, for the next independent word candidate, perform the auxiliary word dictionary collation of the subsequent character string.

In the dictionary collation in the phrase interpretation generation processing of steps S1311 and S1313, when the word delimiter information of the delimiter information 304 in the recognized character information 302 is valid (that is, "1"),
The CPU 120 enables only the independent word candidate, the adjunct word candidate, and the phrase interpretation that are completed by using the candidate character code 305 having this word division as a word end, and invalidates the others.

FIG. 14 is a flow chart for explaining the business card text recognition processing in step S808 of FIG. The CPU 120, similarly to the recognition target area extraction processing in step S1002 of FIG.
A text area on the business card is extracted from the image of the business card read from (1) based on the histograms that are mapped in the horizontal and vertical directions (step S1401).

The CPU 120 executes step S10 of FIG.
As in the case of the general document described in the recognition candidate generation processing in No. 03, for each image corresponding to one character, a set of candidate character codes 305 up to the tenth place and likelihood 306 is generated, and the recognition sequence management unit It stores in 204 (step S1402). Next, the CPU 120 basically performs the recognition error correction process in the same procedure as the processing procedure shown in FIG. 13A, and stores the correction result in the business card data recognition result temporary storage unit 501 (step S1403). . At this time, the dictionary to be used is the correction dictionary address storage unit 706.
It is a proper noun dictionary whose address is stored in. The process of step S1403 will be described later in detail (step S1403).

The CPU 120 selects the first candidate of each business card data stored in the business card data recognition result temporary storage unit 501,
The display 106 together with the image of the block to be recognized
, And the text editing function separately stored in the document input device 100 is activated. The operator performs the correction result correction process while looking at the display screen (step S14).
04).

FIG. 14B is a flow chart for explaining in more detail the business card recognition error correction processing in step S1403 of FIG. 14A. The CPU 120 performs recognition error correction in the same procedure as the processing procedure shown in FIG. This result is stored in the correction result 308 (step S1411). At this time, the CPU 12
0 is attached to a postal code mark such as “〒” included in the correction result 308, an attribute identification character string such as “TEL” or “FAX”, or a word stored in the proper noun dictionary. Using the part-of-speech information representing the place name, the person's name, and the department name as a clue, the correction result 308 shows the “name”, “affiliation name”, “affiliation department name”, “address”, “telephone number”, “F”.
Which of the “AX numbers” the data represents is estimated, and the first to fifth candidates of the correction result 308 are stored in the corresponding storage area in the business card data recognition result temporary storage unit 501. FIG. 15 is an explanatory diagram showing a specific example of a display screen and data in the business card input process. FIG.
9A shows an example of personal name data stored in the name storage unit 502 of the business card data recognition result temporary storage unit 501 as a result of recognition error correction processing performed by the CPU 120 with reference to the proper noun dictionary. .

Next, the CPU 120 causes the postal code storage unit 5
Using the postal code candidate stored in 07 as a key, the address name data stored in the postal code-address collation table unit 405 corresponding to the postal code is extracted, and the work area 700
It is stored in the internal processing buffer 708 (step S14).
12). Similarly, the CPU 120 controls the telephone number storage unit 50.
Using the telephone number stored in No. 6 as a key, the address name data stored in the telephone number-address matching table unit 406 corresponding to the telephone number is extracted and stored in the processing buffer 708 in the work area 700 ( Step S141
3).

Next, the CPU 120 stores the address name data stored in the telephone number / address collation table unit 406, the address name data stored in the postal code / address collation table unit 405, and the address storage unit 505. Is collated with the address name data stored in (step S141).
4, step S1415). As a result, when there is address name data common to the address name data, it is reset to the address in the address storage unit 505. That is, CPU1
Of the address name data stored in the address storage unit 505, 20 stores only the address name data common to the address name data fetched in the processing buffer 708.
Store in 5. Similarly, of the telephone number data stored in the telephone number storage unit 506, only the telephone number data corresponding to the address name data is stored in the telephone number storage unit 5
Of the zip code data stored in the zip code storage unit 507, the zip code data corresponding to the address name data is stored in the zip code storage unit 507 (step S1416). In step S1416, only the address name data common to the address name data fetched in the processing buffer 708 is stored in the address storage unit 505, and only the telephone number data corresponding to the address name data is stored in the telephone number storage unit 506. In the above, only the corresponding postal code data is stored in the postal code storage unit 507. However, the present invention is not limited to this method, and the priority of each of the corresponding data may be increased by one, or the corresponding postal data may be stored. The address name data, the telephone number data, and the postal code data that do not exist may be invalidated in the address storage unit 505, the telephone number storage unit 506, and the postal code storage unit 507.

If the common address name data does not exist as a result of the collation, the CPU 120 displays on the display 106 that there is an error in the address name data, the postal code data or the telephone number data, and warns ( Step S1417). Specifically, for example, when the first candidate stored in the address storage unit 505 is "Ota Shinmachi, Yao City" and the second candidate is "Ota Shinmachi, Nanao City", the former postal code is " 581 ", that of the latter is" 92 "
6 ", the former area code is" 0729 ", and the latter area code is" 0767 ". When only “926” appears in the postal code candidates stored in the postal code storage unit 507, the first candidate “Ota Shinmachi, Yao-shi” stored in the address storage unit 505 is the address storage unit. It is excluded from the candidates in 505.

In the above description, when the input document is a business card, the address name data is collated based on the relationship between the postal code, the telephone number and the address, but it is not limited to the relationship between the postal code, the telephone number and the address. , Relationship between employee name and employee number,
Other proper nouns having relevance may be associated with each other, stored in advance, and used for collation. Further, not only when the input document is a business card, but also by collating the product name and the model number, the book name and the ISBN number, and the like, the corresponding character strings can be correctly recognized.

FIG. 16 is a flow chart for explaining the procedure of the name list update processing in step S809 of FIG. The CPU 120 extracts the name data from the name storage unit 502, and stores the name area 60 in the name list information storage unit 601.
The name data stored in No. 2 is compared and the matching name data is searched (step S1601).

If there is no corresponding name data in the name area 602 of the name list information storage unit 601 (step S160).
2), the candidate data in the business card data recognition result temporary storage unit 501 is displayed on the display 106 as new name list information.
Displayed in step S1606, and the process proceeds to step S1606 (step S1603). FIG. 15B shows a specific example of the new name list information displayed on the display 106 at this time.
At this time, the operator operates the mouse 115 while looking at the display screen shown in FIG. It is possible to instruct addition of the displayed new name list information or correction of each displayed item.

If the corresponding name data exists in the name area 602 (step S1602), the CPU 120
Compares each business card data in the business card data recognition result temporary storage unit 501 with each name list information in the name list information storage unit 601 corresponding to the retrieved name data. That is, C
The PU 120 includes an affiliated organization name storage unit 503 and an affiliated organization name area 603, and an affiliated department signature storage unit 504 and an affiliated department signature area 6.
04, the address storage unit 505 and the address area 605, the telephone number storage unit 506 and the telephone number area 606, and the postal code storage unit 507 and the postal code area 607, respectively. As a result of the collation, if all of them match, it is determined that the correction is not necessary, and the process ends (step S1604).
If there is any difference between the name card data in the name card data recognition result temporary storage unit 501 and the name list information in the name list information storage unit 601 corresponding to the searched name data, the corresponding name card data recognition Result The business card data in the temporary storage unit 501 and the name list information in the name list information storage unit 601 are displayed on the display 106 in a corresponding manner, and the operator's confirmation is requested (step S1605). FIG. 15 (c)
Shows a specific example of the target display screen displayed on the display 106 at this time. Further, at this time, the operator, like the display screen shown in FIG.
While looking at the display screen shown in FIG.
5 is operated to position the cursor in a predetermined area on the display screen and click the button, so that the CPU 120 can correct the name list information based on the name list data displayed, or correct each displayed item. You can give instructions.

When the operator inputs an instruction to make a correction (step S1606), the CPU 120 activates the text editing function of the document input device 100. The operator uses the text editing function to correct the contents of the displayed business card data. CPU120
Stores the business card data corrected by the operator in the corresponding storage unit in the business card data recognition result temporary storage unit 501 (step S1607). Step S1606
In the case where there is an input from the operator indicating that the contents of the business card data are valid, the business card data stored in the business card data recognition result temporary storage unit 501 is stored in the name book information storage unit 601. The name list information is rewritten (step S1608).

FIG. 17 is a flow chart for explaining the procedure of the form text recognition process in step S812 of FIG. The CPU 120 sets the processing target block number M to "1" (step S1701). Then C
The PU 120 extracts the image information in the Mth recognition target block from the image information storage unit 201 based on the image position information 203 of the block management information 210, and character-recognizes the text included in the image information (step S170).
2). This procedure is performed in the same procedure as the recognition candidate generation process in step S1003 of FIG.

The CPU 120 generates a histogram of the images mapped in the horizontal direction and the vertical direction, and estimates that the direction in which the image and the blank area have periodicity and the interval between the images is narrow is the character direction. FIG. 18 is an explanatory diagram showing a specific example of a text image input in the form format input process and a method of estimating the character string direction. FIG. 18A shows an example of the layout of character images in one recognition target block and their horizontal and vertical mappings. In the vertical direction, the blank area that should be seen in the character spacing is irregular, so in this case, the character string direction is estimated to be the horizontal direction (step S1703). FIG. 18B shows another example of the layout of character images in one recognition target block and the mapping in the horizontal and vertical directions. On the other hand, compared to the example of FIG. 18A, in the case of the form format as shown in FIG. 18B, the character spacing is the same in both directions, so the character string is displayed in both the vertical and horizontal directions. The direction of

The CPU 120 determines whether or not the vertical character string direction can be interpreted (step S1704). If it can be interpreted, the character string direction 2 is stored in the recognition string management unit 204.
13 is set to “vertical”, the block management information 210 corresponding to the vertical recognition sequence is generated (step S1705), and if it cannot be interpreted, the process proceeds to step S1707. C
The PU 120 determines whether or not the horizontal character string direction can be interpreted (step S1706), and if the horizontal character string direction can be interpreted, the character string direction 213 is set to “horizontal” in the recognition string management unit 204, and the horizontal recognition string is recognized. Block management information 210 corresponding to
Is generated (step S1707), and if the horizontal character string direction cannot be interpreted, the process proceeds to step S1708.

Next, since the selected document type is the form format, the CPU 120 ignores the word division due to the blank in the evenly assigned in-table titles. In particular,
The word delimiter information in the delimiter information 304 is set to, for example, "0" and invalidated (step S1708). This allows
It is possible to prevent the character strings in the evenly-allocated titles in the table from being mistakenly divided into words due to the blanks in the evenly-allocated space.

The CPU 120 performs the recognition error correction process in the same manner as the procedure shown in FIG. However, here, the general dictionary in the general dictionary unit 402 stored in the correction dictionary address storage unit 706 and the form term dictionary unit 40 are stored.
Reference is made to the form term dictionary in 4 (step S170).
9). At this stage, when the block management information 210 in which the character string direction is interpreted as the vertical direction and the block management information 210 in which the character string direction is interpreted as the horizontal direction are generated for the image in the same region, the CPU 120 , And the total number of phrase interpretations included in the first candidate of each correction result 308 is compared (step S1710). For example, in the text image shown in FIG. 18B, when the character string direction is interpreted as horizontal, “bank” and “transfer” are obtained as the phrase interpretation,
The total is 2, but when the character string direction is interpreted as vertical, it becomes 3 for "silver", "shake", and "marching" (when "include" is mistakenly corrected as "advance"). There is.

As a result, the CPU 120 adopts the interpretation in the character string direction having the smaller total number of phrase interpretations, and invalidates the block management information 210 of the character string direction interpretation that is not adopted (steps S1711, S171).
2). Finally, the CPU 120 increments the process target block number M (step S1713), and repeats this process until the final recognition target block (step S1714).

FIG. 19 is a flow chart for explaining the print position setting process shown in step S813 of FIG.
In the following, the procedure of processing for setting the print position of the recognized character in each recognition target block after the recognition processing of the form text is completed will be described. The CPU 120 first sets the processing target block number M to "1" (step S190).
1), the recognition sequence management unit 2 corresponding to the Mth recognition target block
The block management information 210 of No. 04 is referred (step S
1902).

Next, the CPU 120 causes the ruled line frame information 21
For the recognition target block in which 2 is “1”, the uniform allocation start control information is added to the delimiter information 304 of each candidate character information 307 corresponding to the first character and the last character of the correction result 308 for each recognition column. The uniform allocation end control information is added. Specifically, for example, the division information 30 is used as the even allocation start control information and the even allocation end control information.
4 is set to "1" (step S190).
3). In addition, in step S1903, the uniform allocation start control information and the uniform allocation end control information are added to each recognition column so that all the character strings in the ruled line frame are evenly allocated. Not limited to. For example, first, from the character position information 303 of characters included in one recognition string, the character spacing between consecutive character images in the recognition string is calculated, and the character strings in the recognition string of the original are evenly allocated. And determine whether or not
According to the determination result, the uniform allocation start control information and the uniform allocation end control information may be added to the recognition string to which the character strings are evenly allocated.

The CPU 120 determines the block number M to be processed.
While incrementing (step S1904), the processing from step S1902 to step S1904 is repeated until the recognition target block ends (step S1904).
1905). FIG. 20 is a flowchart showing a procedure of processing for printing the input form. The following describes the procedure of the form printing process when the recognition result is displayed on the display 106 and the operator inputs an instruction to print the recognition result after the process of setting the print position of the recognition character is completed.

The CPU 120 has a display data storage unit 203.
Is cleared and the block number M to be processed is set to "1" (step S2001). The CPU 120 refers to the image position information 203 corresponding to the Mth recognition target block, the ruled line frame information 204, and the character string direction 213 (step S20).
02), it is determined whether or not there is a ruled line frame (step S2003). If there is a ruled line frame, the ruled line frame is written in the display data storage unit 203 at a position corresponding to the image position information 203 (step S2004).

The CPU 120 sets the processing target recognition column number L to "1" (step S2005). CPU12
0 is the recognized character information 30 corresponding to the character in the Lth recognition string
The printer 114 is referred to by referring to the second character position information 303 information.
From the print font sizes of, select the closest print font size. For example, the print font of the printer 114 is 16 × 16, 24 × 24, 32 × 32, 48.
When the character size calculated from the coordinate value of the character position information 303 is 35 × 34, the CPU 120
32 × 32, which is the print font size having the smallest absolute value of the difference between the print font size of the printer 114 and the calculated character size, is selected (step S200).
6). At the same time, the CPU 120 determines whether the selected character size is within an allowable range defined by a predetermined reference value, and resets it if necessary. Specifically, when the character height of the selected character size is three times or more larger than the line interval or the space between the character frame and the ruled line frame, the height is limited to three times.
When the character width of the selected character size is three times or more larger than the character spacing or the spacing with the ruled line frame, it is limited to three times.

The CPU 120 refers to the recognition sequence management information 301 of the recognition sequence management unit 204 corresponding to the L-th recognition sequence, and delimits information 304 corresponding to the character included in the correction result 308 of the referred L-th recognition sequence. In the case where the uniform allocation control information is added (step S2007), the character position information 303 corresponding to the first character of the correction result 308,
With reference to the last character and the corresponding character position information 303, the remaining character strings of the correction result 308 are arranged between these characters at equal intervals as follows.

That is, when the printing is designated by the input from the operator, the CPU 120 determines that the portion to which the uniform allocation information is added is W, the total number of dots of the recognition target block is W, the number of printing dots of one character is C, When the number of characters is n, the character interval width i is calculated from the following formula and printed. i =
(W-n * C) / (C + 1) For example, when printing 5 characters with a print dot number of 24 dots on the total width dot number of 180 dots in the recognition target block, the character interval width i is 10 as follows. It becomes the dot width.

I = (180-5 * 24) / (5 + 1) = (180-120) / 6 = 60/6 = 10 The CPU 120 repeats this for all the recognition columns in one recognition target block, When finished, the process moves to the next block to be recognized (step S2009, step S2).
010). Further, the CPU 120 causes the above step S2.
The processing from 002 to step S2010 is performed for all the recognition target blocks (steps S2011 and S2012).

As described above, according to the present embodiment, when a general document is input, the character string obtained by extracting the first candidate of the character recognition result from the candidate character information 307 and the final character by the operator's confirmation. By comparing and collating with the confirmed character string and obtaining the accuracy of character recognition, and changing the number of candidate character recognition candidates accordingly, it is possible to perform recognition error correction processing that adapts to variations in character recognition accuracy. .

Specifically, according to the present embodiment, when the original of the document to be recognized is an original which has been copied many times to cause faint or smeared characters, and the character recognition accuracy is low, Candidate character code 305 used for recognition error correction processing
It is possible to increase the accuracy of the correction of the recognition error correction processing by increasing the number of candidates of. Further, according to the present embodiment, when the character recognition accuracy is high, such as when the document of the document to be recognized is obtained as a clear image, the number of candidates of the candidate character code 305 used for the recognition error correction process is set. By reducing the number, the time required for the recognition error correction processing can be shortened, and as a result, the processing time of the entire document input device 100 can be shortened.

Further, by comparing the number of keystrokes of the correction key used by the operator to correct the displayed correction result 308 with a predetermined operation level, the accuracy of the character recognition is roughly determined. Can be judged. Thus, the character string obtained by extracting the first candidate of the character recognition result from the candidate character information 307 and the character string finally confirmed by the operator's confirmation are compared and collated one by one to calculate the accuracy. Therefore, the above effect can be easily obtained.

Further, in the case of inputting in the form of a form, by ignoring the delimiter due to the short space between the characters, evenly assigned characters can be correctly collated without being divided in the middle of the word, and the dictionary collation can be performed. The error correction effect due to can be sufficiently exerted. Also, the document input device 100
Stores the character layout in the manuscript as the character position information 303, and stores the correction result 308 obtained by the recognition error correction processing in units of a recognition sequence composed of a sequence of phrase interpretations. Therefore, even when the character string in the displayed correction result 308 is corrected, the correction can be made in units of one clause interpretation, and the trouble of correcting the characters erroneously recognized one by one can be eliminated. You can However, when correcting the character string that was displayed by evenly allocating it in the ruled line frame in the form input, in order to display the corrected character string evenly in the ruled line frame, the first character of the character string Even distribution information must be added to the last character.

Further, according to this embodiment, the document input device 1
00 is correct by generating a recognition sequence interpretation in both directions for the recognition target block when the interpretation in both the vertical and horizontal character string directions is possible and comparing the results of the recognition error correction processing by dictionary matching. It is possible to solve the problem that the interpretation of the character string direction can be obtained, the direction of the character string is not determined, and therefore the effect of the recognition error correction is not exerted.

Further, according to the present embodiment, in the case of the form format input, the print character size is set based on the character position information 303 on the image and the image position information 211, and the range of the set print character size is set. By restricting, it is possible to solve the problem that the characters in the ruled line frame are printed in a character size that is significantly unbalanced. Also,
According to the present embodiment, in the case of business card input, the correction effect of the place name recognition is enhanced by limiting the address candidates etc. obtained as the correction result by using the postal code or the telephone number as a clue, and a lot of glyph characters are included. It is possible to solve the problem that the recognition accuracy rate of a place name is low. Further, according to the present embodiment, the correct correction result of terms such as proper nouns and general nouns is not limited to the relationship between the postal code and the telephone number and the address name, and is not limited to the case of inputting a business card. 308
Can be obtained.

[0143]

Therefore, according to the present invention as set forth in claim 1, the accuracy of the character recognition represented by the correct answer rate is improved every time the character string candidates for a predetermined number of the character string images are generated. According to the calculated accuracy of character recognition, the number of character code candidates used for the combination by the character string candidate generation means is limited. Therefore, the character string candidate generation processing adapted to the fluctuation of the character recognition accuracy is performed. be able to. That is,
When the original of the document to be recognized is an original in which the characters are blurred or smeared by being copied many times, and the character recognition accuracy is low, the character string candidate generation means selects a character code candidate used for the combination. By increasing the number,
The accuracy rate of the generated character string candidates can be increased. When the character recognition accuracy is high, such as when the document of the document to be recognized is obtained as a clear image, the number of character code candidates used for combination by the character string candidate generation means is reduced. , The character string candidate generation speed of the character string candidate generating means can be shortened,
As a result, the processing time of the entire document input device can be shortened.

Therefore, according to the present invention described in claim 2,
Since the likelihood reference value in the likelihood reference storage means is changed according to the counted number of correction instructions, the time and effort for calculating the correct answer rate in claim 1 can be saved, and the above effect can be obtained more easily. Therefore, according to the present invention as set forth in claim 3, evenly-assigned characters can be correctly recognized without being divided in the middle of a word by ignoring word divisions due to spaces between characters in a ruled line frame. The recognition error correction effect can be sufficiently exerted.

Therefore, according to the present invention of claim 4,
Since the predetermined character string candidates that are evenly displayed in the identified ruled line frame are stored as character strings in the character string candidate storage means, the character string is used as a unit for correction. Can be modified. Therefore, the operator does not need to input a correction instruction for each character as in the conventional case, and can easily correct the evenly displayed character strings.

Therefore, according to the present invention of claim 5,
From the distribution of the image area and the blank area in the image of the captured document, if it can be interpreted that the direction of the character string is either vertical or horizontal, once the character string candidates are generated in both directions, Since the character string direction determining means determines the direction of the character string according to another criterion, it is possible to more accurately determine the direction of the character string. Therefore, the effect of the recognition error correction processing can be sufficiently obtained.

Therefore, according to the present invention of claim 6,
It is possible to print a predetermined candidate of the generated character string candidates in the printed ruled frame with a print font size that is closest to the character size of the character image cut out from the captured document image. it can. That is, according to the present invention, a predetermined candidate of the generated character string candidates is printed in the printed ruled line frame with a print font size corresponding to the character size printed in the document to be recognized. Can be printed.

Therefore, according to the present invention of claim 7,
In addition to the above effect, a predetermined candidate of the generated character string candidates can be printed in the printed ruled line frame in a character size balanced with the ruled line frame. Therefore, according to the present invention as set forth in claim 8, among terms representing the same attribute, for example, "Ota" and "Ota",
When recognizing a term such as an address display or a name, which is a correct term, but contains many character strings that tend to cause recognition errors because it contains many similar glyphs, by comparing with the above search results, An accurate recognition error correction result can be obtained.

Therefore, according to the present invention of claim 9,
It is possible to obtain the same effect as that of the present invention according to claim 1. Therefore, according to the present invention described in claim 10, it is possible to obtain the same effect as that of the present invention described in claim 2. Therefore, according to the present invention described in claim 11, it is possible to obtain the same effect as that of the present invention described in claim 3.

Therefore, according to the present invention of claim 12, it is possible to obtain the same effect as that of the present invention of claim 4. Therefore, according to the present invention of claim 13, it is possible to obtain the same effect as that of the present invention of claim 5. Therefore, according to the present invention of claim 14, it is possible to obtain the same effect as that of the present invention of claim 6.

Therefore, according to the present invention of claim 15, it is possible to obtain the same effect as that of the present invention of claim 7. Therefore, according to the present invention of claim 16, it is possible to obtain the same effect as that of the present invention of claim 8.

[Brief description of drawings]

FIG. 1 is a document input device 100 according to a first embodiment of the present invention.
FIG. 3 is a block diagram showing the hardware configuration of the above.

FIG. 2 is an explanatory diagram showing divisions of a storage area provided on a main memory 101 for storing a recognition result of the document input device 100 and storage contents thereof.

FIG. 3 is an explanatory diagram showing a data format of recognition string management information 301 stored in a recognition string management unit 204.

FIG. 4 is a word dictionary storage unit 4 provided on the main memory 101.
It is explanatory drawing which shows 12 storage area divisions.

5 is an explanatory diagram showing storage area divisions of a business card data recognition result temporary storage unit 501 provided on the main memory 101. FIG.

FIG. 6 is a name list information storage unit 6 provided on the main memory 101.
It is explanatory drawing which shows the memory area division of 01.

7 is an explanatory diagram showing storage area division of a work area provided on the main memory 101. FIG.

FIG. 8 is a flowchart showing an overall outline of the processing procedure of the CPU 120.

FIG. 9 is an explanatory diagram showing a specific example of a display screen and data in an initial stage of document input processing.

FIG. 10 is a flowchart illustrating in detail the content of text recognition processing in step S805 shown in FIG.

FIG. 11 is an explanatory diagram showing a specific example of the recognized character information 302 generated by the character recognition processing and the display screen displayed after recognition error correction.

FIG. 12 is a flowchart illustrating a candidate character limitation process in step S1004 of FIG.

13 is a flowchart illustrating a recognition error correction process in step S1005 of FIG.

FIG. 14 is a flowchart illustrating a business card text recognition process in step S808 of FIG.

FIG. 15 is an explanatory diagram showing a specific example of a display screen and data in a business card input process.

16 is a flowchart illustrating a procedure of name list update processing in step S809 of FIG.

FIG. 17 is a flowchart illustrating a procedure of form text recognition processing in step S812 of FIG.

FIG. 18 is an explanatory diagram showing a specific example of a text image input in a form format input process and a method of estimating the character string direction.

19 is a flowchart illustrating the print position setting process shown in step S813 of FIG.

FIG. 20 is a flowchart illustrating a procedure of a process of printing an input form.

[Explanation of symbols]

 100 Document Input Device 101 Main Memory 102 Bus Control Unit 104 Local Bus 105 Graphics Control Unit 106 Display 107 SCSI Control Unit 108 Hard Disk Device 109 Scanner 110 Bus Bridge Circuit 111 External Bus 112 Input / Output Control Unit 113 Keyboard 114 Printer 115 Mouse 120 CPU

Claims (16)

[Claims] 1. A document input device for recognizing a character image included in an image of a captured document and inputting a document captured as an image as a document having a character code obtained as a recognition result. As a character recognition result of an image of one character, a character code candidate storage unit that stores a plurality of character code candidates in association with a likelihood indicating the likelihood, and stores a likelihood reference value set in advance for the likelihood. Likelihood criterion storage means, and for the same character string image representing the character string in the image of the captured document, of the character code having a likelihood equal to or greater than the likelihood reference value in the likelihood criterion storage means. From a combination of candidates, a character string candidate generation unit that generates a plurality of candidates for a character string that can be grammatically correctly interpreted, and a character string candidate corresponding to the character string image. In contrast, among the character string candidates generated for the same character string image, display means for displaying a predetermined candidate, and for the character string candidates generated corresponding to the character string image, An input unit for inputting a correction instruction from the operator, and the character string candidates generated corresponding to the character string image are corrected in accordance with the correction instruction to correct a character code that is erroneously recognized. A correction unit that corrects to a character code, a character string after correction by the correction unit, which corresponds to the character string image, and a character that is a candidate of a character code with the highest likelihood corresponding to the same character string image. Comparing means for comparing the column, after the comparison by the comparing means, based on the comparison result, from the ratio of the number of characters that differ between the two and the number of characters included in the one character string, character recognition A correct answer rate calculating means for calculating a correct answer rate, a correct answer rate reference storage means for storing a predetermined correct answer rate reference value of the correct answer rate, and a calculated correct answer rate after the correct answer rate is calculated, And a likelihood reference value changing means for changing the likelihood reference value in the likelihood reference storage means according to the calculated correct answer rate when the predetermined condition is satisfied by comparison with the reference value. Characteristic document input device. 2. A document input device for recognizing a character image included in an image of a captured document and inputting a document captured as an image as a document having a character code obtained as a recognition result. As a character recognition result of an image of one character, a character code candidate storage unit that stores a plurality of character code candidates in association with a likelihood indicating the likelihood, and stores a likelihood reference value set in advance for the likelihood. Likelihood criterion storage means, and for the same character string image representing the character string in the image of the captured document, of the character code having a likelihood equal to or greater than the likelihood reference value in the likelihood criterion storage means. From a combination of candidates, a character string candidate generation unit that generates a plurality of candidates for a character string that can be grammatically correctly interpreted, and a character string candidate corresponding to the character string image. In contrast, among the character string candidates generated for the same character string image, display means for displaying a predetermined candidate, and for the character string candidates generated corresponding to the character string image, An input means for inputting a correction instruction from the operator, a correction instruction counting means for counting the correction instruction input from the input means, and a correction instruction standard for storing a predetermined correction instruction number reference value of the correction instruction number. Storage means, correction means for correcting a character code that is erroneously recognized to a correct character code in accordance with the correction instruction for the character string candidates generated corresponding to the character string image, After the correction by the correction means, when the counted correction instruction number is compared with the correction instruction number reference value stored in the correction instruction reference storage means and satisfies a predetermined condition,
And a likelihood reference value changing unit that changes the likelihood reference value in the likelihood reference storage unit according to the counted number of correction instructions. 3. A document input device for recognizing an image of characters included in an image of a captured document and inputting the document captured as an image as a document having a character code obtained as a recognition result, which comprises one line Blank detecting means for detecting a blank between the image of one character cut out from the character string image representing the character string and the image of one character adjacent to the image, and when the detected blank is larger than a predetermined size, 1
The image of a character and the image of one character adjacent thereto are the word identification means for identifying the character as a character belonging to a different word, and the character string image in a predetermined area in the captured image of the document. A document input device, comprising: a word identification limiting unit that invalidates an identification result of the identification unit. 4. A document input device for recognizing a character image included in an image of a captured document and inputting the document captured as the image as a document having a character code obtained as a recognition result. Character code candidate storing means for storing a plurality of character code candidates as a character recognition result of an image of one character cut out from an image of a captured document, and ruled line frame identification for identifying a ruled line frame in the captured image of the document Means and a character string for generating a plurality of character string candidates that can be grammatically interpreted by combining the character code candidates for the same character string image representing the character string in the captured document image Candidate generating means, character string candidate storing means for storing the generated character string candidates, and a predetermined character string candidate corresponding to one character string image, It is read from the unit, a document input device comprising: a display means for displaying to Distribute identified border frame. 5. A document input device for recognizing a character image included in an image of a captured document and inputting a document captured as an image as a document having a character code obtained as a recognition result. Character code candidate storage means for storing a plurality of character code candidates as a character recognition result of an image of one character cut out from an image of a captured document, and distribution of an image area and a blank area in the captured image of the document A character string direction identification means for identifying whether the direction of the character string in the image of the captured document is vertical or horizontal, and the distribution of the image area and the blank area in the image of the captured document. From the above, when it can be interpreted that the direction of the character string is either the vertical or horizontal direction, the character code of the same character string image representing the character string in each of the two directions. A character string candidate generation unit that generates a plurality of character string candidates that can be combined with each other and can be correctly interpreted grammatically, and a predetermined candidate among the character string candidates generated for the vertical character string image. And a predetermined candidate of the character string candidates generated for the horizontal character string image are compared, and the character string direction and the character string of one of the character strings satisfying a predetermined condition by the predetermined candidate. A document input device, comprising: a character string direction determining means for determining a candidate as valid. 6. A document input device for recognizing a character image included in an image of a captured document and inputting the document captured as the image as a document having a character code obtained as a recognition result. Ruled line frame identifying means for identifying the ruled line frame in the image of the captured document and a plurality of character code candidates as the character recognition result of the image of one character cut out from the captured image of the document. Character code candidate storage means for storing together with the position and character size in the image of the document, and the same character string image representing the character string in the captured image of the document, from the combination of the character code candidates, the grammar Character string candidate generation means for generating a plurality of character string candidates that can be correctly interpreted, and a predetermined candidate among the generated character string candidates. When printing in the identified ruled line frame, the character size of the image corresponding to the character code candidate in the character string candidate is checked, and the character size checked from the printable font sizes that can be printed A printing character size selection means for selecting a printing font size closest to the above, and printing the identified ruled line frame, and selecting a predetermined candidate of the generated character string candidates in the printed ruled line frame. And a printing unit that prints in the specified print character size. 7. The document input device according to claim 6, wherein the position in the image of the document of the image corresponding to the character code candidate in the character string candidates is referred to with reference to the character code candidate storage means. By referring to the vertical space calculation means for calculating the size of the upper and lower blank areas sandwiching the image of the character and the character code candidate storage means, and the character code candidates in the character string candidates correspond to each other. The position of the image to be scanned in the image of the document is checked, and the left and right space calculating means for calculating the size of the left and right blank areas sandwiching the image of the character, and the vertical size of the selected print font size are
An upper and lower size limiter that limits the size of the selected print font size in the vertical direction so that it does not exceed the predetermined size compared to the size of the calculated upper and lower blank areas, and the left and right of the selected print font size. The size of the direction is
And a left and right size limiting means for limiting the size of the selected print font size in the left and right direction so as not to exceed a predetermined size as compared with the calculated left and right blank areas. The document input device is characterized in that a print font size closest to the examined character size is selected from the print font sizes limited by the vertical size limiting unit and the horizontal size limiting unit. 8. A document input device for recognizing a character image included in an image of a captured document and inputting the document captured as an image as a document having a character code obtained as a recognition result. Character code candidate storing means for storing a plurality of character code candidates as a result of character recognition of an image of one character cut out from an image of a document, and the character code candidates for the same character string image representing a character string. Character string candidate generating means for generating a plurality of character string candidates that can be correctly interpreted as terms with a predetermined priority, and a specific character code is identified from the generated character string candidates. Then, the attribute of the term represented by the character string candidate including the character code and the character string candidate generated for the same character string image as the character string candidate is determined. Attribute determination means, inter-attribute correspondence term storage means for storing terms having mutually related attributes in association with each other, and storing the character string candidates for which attributes have been determined together with the priority order for each predetermined attribute And a character string candidate stored in the attribute-based character string candidate storage means that represents the attribute of a term stored in the inter-attribute correspondence term storage means. Of the character string candidates stored in the attribute-based character string candidate storage means, a term search means for searching for a corresponding term from the inter-attribute correspondence term storage means using each of the character string candidates as a key , Priority changing means for increasing the priority of the character string candidates that are common to the search result of the term searching means using the character string candidates of other attributes that are mutually related as keys. A document input device characterized by: 9. A predetermined likelihood reference value and a predetermined correct answer rate reference value are stored in a likelihood reference storage section and a correct answer rate reference storage section, respectively, and a character included in an image of a captured document is stored. While recognizing an image as a character, as a character recognition result of an image of one character, a plurality of character code candidates are stored in a character code candidate storage unit in association with a likelihood indicating the certainty, and captured as an image. A document input method for inputting a document as a document including a character code obtained as a recognition result, wherein the same character string image representing a character string in an image of a captured document is stored in the likelihood criterion storage unit. A character string candidate generation step for generating a plurality of character string candidates that can be grammatically correctly interpreted from a combination of the character code candidates having a likelihood equal to or greater than the stored likelihood reference value. A display step of displaying a predetermined candidate among the character string candidates generated for the same character string image with respect to the character string candidate generated corresponding to the character string image, For the character string candidates generated corresponding to the character string image, an input step of inputting a correction instruction from an operator, and for the character string candidates generated corresponding to the character string image A correction step of correcting a character code that is erroneously recognized according to the correction instruction to a correct character code; A comparing step of comparing with a character string composed of the corresponding most likely character code candidates; and after the comparison, based on the comparison result, the number of characters different between the two and the one character string A correct answer rate calculation step of calculating a correct answer rate of character recognition from a ratio with the number of characters included, and a correct answer rate calculated after the correct answer rate is calculated, and the correct answer rate stored in the correct answer rate reference storage unit. When satisfying a predetermined condition compared with the rate reference value, change the likelihood reference value in the likelihood reference storage unit according to the calculated correct answer rate,
And a likelihood reference value changing step of returning to the character string candidate generating step. 10. A predetermined likelihood reference value and a predetermined correction instruction number reference value are stored in a likelihood reference storage unit and a correction instruction reference storage unit, respectively, and characters included in an image of a captured document are stored. While recognizing the image of the image, as a character recognition result of the image of one character, a plurality of candidates for the character code are stored in the character code candidate storage unit in association with the likelihood indicating the certainty, and captured as an image. A document input method for inputting a document as a document consisting of a character code obtained as a recognition result, wherein the same character string image representing the character string in the image of the captured document is stored in the likelihood reference storage unit. The character string candidate generation step for generating a plurality of character string candidates that can be grammatically correctly interpreted from a combination of the character code candidates having a likelihood equal to or higher than the likelihood reference value of A display step of displaying a predetermined candidate among the character string candidates generated for the same character string image with respect to the character string candidate generated corresponding to the character string image, An input step of inputting a correction instruction from an operator for the candidate of the character string generated corresponding to the character string image, a correction instruction counting step of counting the input correction instruction, and the character string image A correction step of correcting a character code that is erroneously recognized into a correct character code according to the correction instruction for the corresponding candidate of the character string generated, and the correction instruction counted after the correction. When the number satisfies the predetermined condition compared with the correction instruction number reference value stored in the correction instruction reference storage unit, the likelihood in the likelihood reference storage unit is calculated according to the counted correction instruction number. Document input method characterized by comprising the likelihood reference value changing step of changing the reference value. 11. A document input method for recognizing a character image included in an image of a captured document and inputting a document captured as an image as a document having a character code obtained as a result of recognition, comprising: A blank detection step of detecting a blank between an image of one character cut out from a character string image representing the character string of and the image of one character adjacent to the image, and when the detected blank is larger than a predetermined size, 1
The character image and the image of one character adjacent to the character image are identified as characters belonging to different words, and a character string image in a predetermined area in the captured image of the document is described above. And a word identification limiting step of invalidating the identification result. 12. While recognizing one character image cut out from the captured image of a document, a plurality of character code candidates are stored in the character code candidate storage unit as a recognition result, and the document captured as the image is stored. A document input method for inputting as a document composed of character codes obtained as a recognition result, the ruled line frame identifying step of identifying a ruled line frame in the image of the captured document, and the characters in the image of the captured document. A character string candidate generating step of generating a plurality of character string candidates that can be grammatically correctly interpreted by combining the character code candidates for the same character string image representing a string, and the generated character string candidates. Is stored in the character string candidate storage section, and a predetermined character string candidate corresponding to one character string image is read from the character string candidate storage section. To, document input method characterized by comprising a display step of displaying to Distribute the identified line border. 13. While recognizing a character image included in an image of a captured document, a plurality of character code candidates are stored in a character code candidate storage unit as a character recognition result of an image of one character and captured as an image. A document input method for inputting a document as a document composed of the character codes obtained as a recognition result, and the document is scanned based on the distribution of the image area and the blank area in the image of the captured document. From the character string direction identification step that identifies whether the direction of the character string in the image is vertical or horizontal, and the distribution of the image area and the blank area in the image of the captured document, the direction of the character string is vertical or horizontal. When it can be interpreted as either direction, the same character string image representing the character string in each of the two directions is grammatically correct from the combination of the character code candidates. A character string candidate generating step of generating a plurality of interpretable character string candidates; a predetermined candidate of character string candidates generated for the vertical character string image; and a horizontal character string. A character string that compares a predetermined candidate among the character string candidates generated for the image, and determines the character string direction and the character string candidate in which the predetermined candidate satisfies a predetermined condition. A document inputting method comprising: a direction determining step. 14. While character-recognizing a character image included in an image of a captured document, a plurality of character code candidates in the image of the character of the character image are detected as a character recognition result of the image of one character. A method for inputting a document stored as an image together with the position and character size in the character code candidate storage unit as a document consisting of the character code obtained as a recognition result. A ruled line frame identifying step of identifying a ruled line frame and a character string that can be grammatically correctly interpreted from the combination of the character code candidates for the same character string image representing the character string in the image of the captured document. And a predetermined candidate of the generated character string candidates is printed in the identified ruled line frame. In doing so, the character size of the image corresponding to the character code candidate in the character string candidate is checked, and the print font size closest to the checked character size is selected from the print font sizes that can be printed. Print character size selection step and printing of the identified ruled line frame, and printing of predetermined candidates of the generated character string candidates in the printed ruled line frame at the selected print character size A method for inputting a document, comprising: 15. The document input method according to claim 14, wherein, after the ruled line frame identifying step, the image corresponding to the character code candidate in the character string candidates is referred to by referring to the character code candidate storage unit. Of the character string candidates by referring to the character code candidate storage unit to check the position of the document in the image, and calculate the size of the upper and lower blank areas sandwiching the image of the character. Of the position of the image corresponding to the character code candidate in the image of the document, the left and right space calculation step of calculating the size of the left and right blank areas sandwiching the image of the character, and the selected print font size The vertical size is
The upper and lower size limiting step that limits the size of the selected print font size in the vertical direction so that it does not exceed the predetermined size compared to the calculated size of the upper and lower blank areas, and the left and right of the selected print font size. The size of the direction is
Document input characterized by comprising a left and right size limiting step of limiting the left and right size of the selected print font size so as not to exceed a predetermined size as compared with the calculated left and right blank areas. Method. 16. An image of one character while preliminarily storing, in an inter-attribute correspondence term storage unit, associating terms having mutually related attributes with each other and recognizing a character image included in an image of a captured document. Is a document input method for storing a plurality of character code candidates as a character recognition result in the character code candidate storage unit and inputting a document captured as an image as a document including the character code obtained as a recognition result. A character string candidate generating step of generating a plurality of character string candidates that can be correctly interpreted as terms by combining the character code candidates for the same character string image representing a column, with a predetermined priority order; A specific character code is identified from among the character string candidates, and a character string candidate containing the character code and the same character string image as the candidate character string are displayed. And an attribute determination step of determining the attribute of the term represented by the generated character string candidate, and storing the character string candidates for which the attribute has been determined together with the priority in the attribute-specific character string candidate storage unit for each predetermined attribute. An attribute-specific character string candidate storage step, which is a candidate for a character string stored in the attribute-specific character string candidate storage unit, and which represents an attribute of a term stored in the inter-attribute correspondence term storage unit Of the character string candidates stored in the attribute-specific character string candidate storage unit, a term search step of searching for a corresponding term from the inter-attribute correspondence term storage unit using the character string candidates as a key A priority order changing step for increasing the priority order of the character string candidates that are common to the search result using the character string candidates of other attributes that are mutually related as a key. Document input method.