JP3159087B2 - Document collation device and method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for collating documents between documents, and more particularly, to the method of identifying documents in electronic data format created by a word processor or the like with documents printed on paper sheets by a printer or the like. The present invention relates to a document matching method and apparatus for supporting a task of checking a document.

[0001]

2. Description of the Related Art In many cases, it is necessary to confirm the identity of two documents. A typical example is the identification of documents of high importance such as a contract. Hereinafter, in the present invention, the operation of performing the same check between a plurality of documents is referred to as document collation.
Since such a document is usually written on a paper, the document collation has hitherto been performed by arranging two papers side by side and visually confirming each one. Since this work mostly depends on humans, the burden on the operator is heavy.

On the other hand, the spread of personal computers and communication networks in recent years has promoted the conversion of documents into electronic data. Document collation between documents in electronic data format is usually O
It is built into the S command and can be performed very easily. For example, in MS-DOS (registered trademark of Microsoft Corporation), FC command is used, and UNIX (X / Op
en Company Ltd. Registered trademark) in cm
The p command is a typical example. For details such as the functions of these commands, refer to each OS (Operati
on System) manual or "Introductory MS-DOS ((ASCII Learning System Introductory Course)""written by Koji Murase, pp123, (published by ASCII Publishing Company) or S
"UNIX SVR4" by tephen Coffin
pp150, (published by SoftBank Corp.) and the like.

[0003]

However, there is no proposal in the above-mentioned known documents regarding document collation between a document in electronic data format and a document described on paper. In the future, as the spread of electronic data format documents spreads widely, the situation where paper sheets and electronic data documents are mixed will occur, and it is thought that the need for such confirmation will increase. , Can not respond to this.

[0004] The collation between the electronic data format and the paper document is as follows.
Compared with the above-described document comparison between paper documents, the burden on humans is even greater. The main reasons are as follows:
The following 3) can be considered. 1) It is difficult to arrange the documents side by side and compare them within one field of view. 2) Generally, the layout of the document is different between the electronic data format and the paper. 3) Generally, the character shape used is different. An object of the present invention is to automate such a document collation operation and reduce a burden on an operator.

[0005] In addition, when preparing and submitting important documents, it is necessary to confirm the contents of the documents once before submitting them with caution. However, the present invention also requires such work. It is intended to help.

[0006]

In order to solve the above-mentioned problems, a document collating apparatus according to the present invention is provided with a recognition section for optically reading a character written on a sheet, and is stored in advance as electronic data. A collating unit for collating the character code and the recognition result by the recognizing unit one character at a time, and switching a display method so as to indicate a position where the operator pays particular attention to look according to the result of the collation by the collating unit, A confirmation unit that displays a character corresponding to the character code of the electronic data,
The recognition result by the recognition unit includes a character code obtained based on a character image obtained by optically scanning a character written on the sheet and a reliability of the recognition result of the character code, and the confirmation unit includes: Displayed so as to indicate a position where an operator pays special attention to the character code of the electronic data and the character code of the recognition result by the collating unit and the position of the operator who pays special attention to the character code of the recognition result. A method is switched to display the characters of the electronic data, and the confirmation unit displays a character code of the electronic data.
Match with the character code of the recognition result
And the reliability for the character code is set in advance.
A first case higher than the first value, wherein the matching result is a match
And the second reliability is lower than the first value.
A case and a third case in which the collation results do not match
The electronic data in a different display method.
Is displayed .

Further, the document collation method of the present invention stores
The character corresponding to the character code of the electronic data
Printed on, to get the recognition result Ri printing characters on the sheet optically <br/> reading the electronic data character code
And said recognition result and de acquired collation to collation result one by one, paying particular attention by the operator in response to the comparison result
Switching the display method to indicate the viewing position, the electronic
Display character corresponding to the character code data, confirming visually the comparison result.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, a first embodiment of a document matching method and apparatus according to the present invention will be described with reference to the drawings.

In the document collation method of this embodiment, character recognition processing is used as means for automatically collating electronic character data with character data on paper. However, at the current level of character recognition technology, not all characters can be read completely correctly. In addition to handwriting, especially in the case of typographical characters, such as kanji, there are many types of characters, and when the fonts are diversified, the accuracy decreases. Therefore, the purpose of the present embodiment is to reduce the burden on humans by performing character recognition on portions that can be collated by character recognition and performing the other portions on humans. For this purpose, it is also important to devise a method for facilitating the manual document collation method itself.

Nevertheless, it is particularly important to devise ways to improve the accuracy of character recognition as much as possible, since this leads directly to reducing the burden on humans. Here, the reasons why the accuracy of character recognition is generally insufficient are summarized as follows (1) to (3). (1) Incorrect (automatic) analysis of the reading area. As a result, different reading areas are read. (2) Incorrect character extraction. This error is likely to occur in sentences that contain full-width and half-width characters, and where the text size changes,
In particular, if attention is paid to the character type, it is likely to occur at a position including a specific character such as a separated character (a character such as kan) or a punctuation mark. (3) An error occurs in individual character recognition. In general, as the character type increases and the character shape becomes more diversified, it relatively decreases.

In document collation, electronic data can be regarded as correct answer data in character recognition processing. In this embodiment, the above-mentioned (1) to (1) to (4) are used by utilizing this property.
An attempt is made to reduce the problem (3) and improve the accuracy of character recognition.

FIG. 1 is a block diagram showing the configuration of the first embodiment of the present invention. The sheet 11 is a sheet on which one data to be collated with a document is described. The electronic data 12 is the other data to be compared, and is usually stored as a character code on a storage medium such as a hard disk or a memory. FIG. 2 shows an example of the sheet 11. The character code of the electronic data 12 is O
JIS code, shift JIS code, EU according to S etc.
C code or the like is used.

The sheet 11 is optically scanned by a character recognizing unit 21, and the written contents are converted into character codes and output. The character recognition unit 21 is a general character recognition device (hereinafter referred to as OCR (OCR) that has been commercialized in various forms.
optical sensor reader) can be applied, and it is a known device that does not require a special technique.

The character recognition unit 21 will be described in more detail. FIG. 3 shows the structure of a typical character recognition unit. First, the optical scanning unit 211 optically scans the sheet 11,
Obtain a digital image via A / D conversion. The digital image is processed by the subsequent recognition processing method, required processing speed, H /
The image is acquired as a binary image, a grayscale image, a color image, or the like according to the W scale. At this time, a dropout image obtained by optically dropping out a specific color may be used. Also, even if you place the sheet by hand like a flatbed scanner,
It may be automatically conveyed. Further, in the automatic transport type, it is possible to have two stackers, such as a form OCR, and to discharge the sheets into an accept stacker and a reject stacker according to the coincidence / mismatch of the document collation.

The area extracting section 212 extracts a reading area (character area) from the obtained digital image. At this time, the reading area may be automatically extracted based on information such as the projection and the arrangement of the black connected areas as in the document OCR.
The reading area may be extracted using a format definition given in advance as used in CR.

The character extracting section 213 extracts individual characters from the reading area. At this time, the document may be automatically extracted based on the features used in the region extraction of the document OCR, or may be extracted using the format definition.

The character recognizing unit 214 extracts features from the cut out individual characters and performs character recognition using a character dictionary. Although various methods have been developed for character recognition, any method can be used. For example, a method of finding the minimum distance or the maximum similarity between the characteristics of the input character and the characteristics of the individual character categories registered in the dictionary in advance during collation, and a leaf (leaf) using a tree structure ) May be used as the determination result, but any method may be used.

As described above, the character recognition unit 21 may use any method as long as the characters to be read described on the sheet 11 can be finally converted into the character code format. What is different is that there is no need to correct the character recognition result based on linguistic information, which is usually called knowledge processing or character recognition post-processing. The reason is that correcting character recognition results using linguistic information is a process that is not suitable for document collation as in the present invention, because there is a risk of falsification of the character recognition results in accordance with the knowledge possessed. It depends on the reason. Further, since the electronic data 12 itself is considerably rich knowledge information, it is better not to use other knowledge. The method of using the electronic data 12 will be described later.

Next, the collating unit 22 verifies the character code of the output result of the character recognizing unit 21 with the electronic data 12 and confirms a match / mismatch. The operation of the matching unit 22 will be described with reference to FIG. FIG. 4 shows an example of a character recognition result for the sheet of FIG. Although the character recognition result 31 and the electronic data 32 in FIG. 4 are character codes, they are expressed as characters here for easy understanding.

In the character recognition result 31, the ninth character katakana "te" is read as the symbol "@", and the eleventh character katakana "to" is read as the kanji "to" (I). The collating unit 23 compares the character recognition result 31 and the character code at the corresponding position of the electronic data 32 one by one in order from the front in order to distinguish whether or not the corresponding character codes match, and compares the result with each individual character. The character is stored in the document comparison result 33. In the document comparison result 33, 0 indicates a match and 1 indicates a mismatch.

FIG. 4 shows an example of the collation result. In the present invention, it is only necessary to know whether the character codes match or not, and the order of collation and the method of expressing the collation result are not essential.

Finally, the verification result is displayed by the visual confirmation unit 23, and the operator performs visual confirmation. At this time, electronic data 3
2 is displayed on the display, but in order to indicate the position where the operator pays particular attention, the mismatched portion is emphasized and displayed based on the matching / mismatching of the matching result 33. This procedure will be described with reference to FIG. First, the electronic data 32 is read out character by character in order from the beginning (step 231) (step 232), and the collation result at the corresponding position of the collation result 33 is referred to (step 233). ) Are not matched (1), highlighting is performed (step 235). This procedure is repeated until the last character of the electronic data (step 236).

As highlighting, reverse display, blink,
Changing the color, changing the character shape (for example, italic), changing the size, and the like. FIG. 6 shows a display example in which the size is changed as an example.

By the way, the character recognizing section 21 can also obtain the layout information of the characters described on the paper sheet in the recognition process. Therefore, if the visual confirmation unit 23 displays the collation result by using the layout analysis result by the character recognition unit 21, the layout on the paper and the layout of the electronic data can be made similar.
Confirmation work becomes easier. In general, character recognition is performed in units of lines. Therefore, the simplest layout information can be obtained by adding a line feed code (usually 0x0a) to a character recognition result after recognition of one line is completed. Collation unit 2
In step 2, the collation is performed by ignoring the line feed code included in the character recognition result. In the visual check unit 23, if the electronic data is displayed by referring to the line feed code position of the character recognition result, such a layout display is performed. Can be adjusted.

As the display means, there is a printing means such as a printer in addition to the display described in this embodiment, and the same effect can be obtained.

Based on the display data indicating the collation result from the collation unit 22, the operator focuses on the sheet 11 and the emphasized portion of the display data, and confirms the coincidence / mismatch of the documents. Document collation is now complete, but the result can be entered using a keyboard, mouse, etc., if there are multiple destinations for the sheet, the destination for the sheet can be specified, and the result of the document collation can be reflected in electronic data. The processing to be performed is appropriately set according to the operation situation.

Such an invention relates to a personal computer and a scanner or a personal computer and an OCR
It can be easily realized by a combination of the above.

Next, a second embodiment of the present invention will be described. This embodiment differs from the above-described first embodiment only in two places, that is, the output format of the recognition result of the character recognition unit 21 and the display method of the visual confirmation unit 23. Therefore, this part will be particularly described. The character recognition unit 214 shown in FIG.
Recognizes each character image and outputs a character code as a result. At this time, the character image itself used for recognition is passed to the collating unit 22 together with the recognition result (character code). The character image is in a format that makes it possible to know the character number corresponding to the read result from the head and to access the character image at random. For example, an individual character image is referred to as CIXXX. One method is to create a file with the name YYY. Here, XXX represents the number of the image of the character from the beginning. For example, XXX = 002 for the second character. The extension YYY indicates a storage format of a commonly used image. The leading CI is a name indicating a character image, and can be arbitrarily selected.

In the collating unit 22, the character code as the recognition result is used for the same processing as described above, and the character image is passed to the visual checking unit 23 without being used for any processing. When displaying the character code of the electronic data indicating the collation result, the visual check unit 23 displays the corresponding character image above or below the character code. At this time, the size of the character image becomes easier to see if it is enlarged or reduced according to the size of the screen or the display size of the code of the electronic data. Further, since similar sentences may be bothersome in two rows, only the character image corresponding to the electronic code at the cursor position in the display data or only the mismatched portion may be displayed. Since the image display and the enlargement / reduction processing are well-known techniques, detailed description will be omitted.

In this embodiment, since the character image to be displayed by the visual confirmation unit 23 need only be known, the interface from the character recognition unit is free within that range. Although the present embodiment has been described using an image interface based on a file format, an image stored continuously with an access table as shown in FIG. 7 may be used as an interface.
As shown in (2), the visual confirmation unit 23 may receive only the cutout position as information, and request the character recognition unit 21 again only a required character image.

Next, a third embodiment of the present invention will be described.

This embodiment is also different from the first and second embodiments in the output format of the recognition result of the character recognition unit 21 and thereafter.

First, the character recognition unit 21 outputs the recognition reliability of each character in addition to the character code as the recognition result.
In character recognition using similarity or distance as a criterion, similarity, reciprocal of distance / code conversion (conversion of ±), or the magnitude of the difference between the first and second candidates are used as recognition reliability. If the character recognition is a method using a tree structure, the reliability may be defined in advance for each node or leaf, and the accumulated value of the nodes or leaves passed at the time of determination may be used as the reliability.

Next, the processing of the collating unit 22 will be described with reference to FIG. Similar to FIG. 4, reference numeral 31 denotes a character recognition result;
Indicates electronic data. 34 is the reliability, and Sx represents the reliability of the X-th character. Here, when the reliability is S0 or more, the character recognition result is assumed to have high reliability. The threshold value S0 may be determined experimentally. The collation result 35 is determined as follows.

The recognition result matches the electronic data and the reliability is S0 or more... 0 The recognition result matches the electronic data and the reliability is less than S0... 2 The recognition result does not match the electronic data... 1 In FIG. Among the reliability S1 to S15 of the recognition result, only the reliability S6 is less than S0, and the rest is the collation result 35 in the case of S0 or more.

In the visual checking section 23, the portion for switching the display method according to the collation result 35 is different. For example, as shown in FIG.
The double-width, 2 portion is displayed as a vertical double-width of 0.

Next, a fourth embodiment of the present invention will be described.

A list of character types to be read by the character recognizing unit 21 is given to the collating unit 22 in advance as a sequence of character codes. Accordingly, the electronic data has a structure in which a character type that does not produce a correct result by character recognition, that is, a character type not to be read can be checked.

The processing of the matching unit 22 at this time will be described with reference to FIG. In FIG. 11, reference numeral 41 denotes a character recognition result;
Represents the reliability, 43 represents the electronic data, and 44 represents the collation result. The eighth character of the electronic data 43 is the Roman numeral I, the character type to be read for character recognition, and the character recognition result 41.
The 8th character is the alphabet I, and the reliability is 43
Is S6 <S0, and the rest are larger than the threshold value S0. The collating unit 22 obtains a collation result in the following procedure.

A list of character types to be read is searched for each character code in the electronic data 43. In the case of a character not to be read, 3 is set as the collation result. In the case of a character not to be read, the collation result is set according to the procedure described in the third embodiment. According to this procedure, a collation result 44 is obtained based on the character recognition result 41, the reliability 42, and the electronic data 43. The technique of searching for a specific character code from a list of character codes is a known technique, and a detailed description thereof will be omitted.

The visual confirmation unit 23 differs in the part for switching the display depending on the collation result 44. For example, as shown in FIG. 12, the part where the collation result 44 is 1 is a quadruple angle of 0 and the part where the collation result 44 is 2 is 0. The portion that is 3 times the vertical double width of 3 can be displayed with the horizontal double width of 0.

Next, a fifth embodiment of the present invention will be described. In the present embodiment, in particular,
There is a characteristic in the character cutout part.

As shown in FIG. 13, the electronic data 12 can be referred to from the character extracting section 213. Of course, even if the electronic data 12 cannot be referred to directly, the contents may be referred to via the collation unit 22 by data transfer or the like.

The operation of the character extracting section 213 will be described below with reference to FIG. In FIG. 14, reference numeral 51 denotes an example of an image from which an area has been extracted, and reference numeral 52 denotes a block figure representing a circumscribed rectangle of a single black connection area of the image. In the image 51, hiragana is printed in full-width and katakana is printed in half-width. In normal character extraction, the extraction of the "was test" part is often mistaken. This is because the character pitch is disturbed in this portion.

In this embodiment, electronic data is used to stably perform the character extraction. Normal 2-byte characters are 2 by
In the te code, a half-width character is represented by a 1-byte code, and character extraction is performed using this property. For example, in the example shown in FIG. 14, by separating the electronic data into a 2-byte code and a 1-byte code, it can be seen that the first three characters are full-width, the next three characters are half-width, and the following two characters are full-width. .

Since the width of the full-width character can be inferred from the height of the line, the block figure 52 is cut out with reference to the code of the electronic data 12. Since the first character is full-width, 521 and 522 are integrated to form the first character. Similarly, the second character is 523, and the third character is 524 because it is too small.
24 and 525 are integrated to form a third character. Since the next three characters are half-width characters, 526 to 528 are cut out as single characters, and 529, 530 and 531 are integrated as full-width characters.

In this manner, stable and accurate character extraction can be performed. Note that extraction of a single black connected region and bottom-up block integration of a mixture of full-width and half-width characters are well-known technologies, and therefore, only a method of using electronic data will be described here.

In this description, the single black connected area is used as a basic feature of character extraction, but it goes without saying that the same processing can be performed by character extraction by projection.

Next, a sixth embodiment of the present invention will be described. This embodiment also has a feature in the character cutout unit 213 as compared with the other embodiments as in the above-described fifth embodiment.
The character recognition unit 21 holds shape information of each character in a form that can be referred to from each character code. The shape information is
Given as the circumscribed frame shape of the character, a square frame (represented by code 1 in the following description), a vertically long rectangular frame (2), a horizontally long rectangular frame (3), a small rectangular frame (4), an indefinite / blank 0) etc. FIG. 15 shows an example in which a shape code is referred to from a character code using a table. In FIG. 15, the JIS code is described as an example, and the shape information corresponding to the character indicated by the character code is 1 byte at a position advanced by the hexadecimal code byte corresponding to the character code (YYYY) from the top of the table. 0 to 4 are described. Therefore, if the character code (YYYY) is known,
Then, the code (0, 1, 2, 3, or 4) of the corresponding shape information can be referred to. This table can be easily and automatically created using a learning character database used when creating a character dictionary.

A processing procedure using the shape information based on this table will be described with reference to FIG. 16 in FIG.
Reference numeral 1 denotes an example of a character string image obtained by the region extraction 212. The first four characters (up to “′”) are full-width characters, and the next three characters
The characters (test) are printed in half-width and the remaining four characters are printed in full-width. Reference numeral 62 in FIG. 16 is a diagram in which a single connected area of a black point is extracted from the image 61 in FIG. 16 and each is represented by its circumscribed frame. According to the conventional method, a cutting error easily occurs between 625 and 633, and in particular, 626, 630,
Errors are likely to occur in the integration and separation of 632 and 633.

In this embodiment, first, electronic data is referred to. Referring to a table of a shape (not shown) using the electronic data in the method of FIG.
Shape information 114 (square,..., Small) is obtained. When the character is cut out from 62 based on this information, the first character is composed of 621 and 622,
3 is one character, and the position of 624 is a square whose shape information is a square. Therefore, 624 and 625 are integrated into one character, 626 to 630.
Since there is shape information that each character is 631, the position of 631 and the next character are squares, 631 to 633 are integrated, and one character is output to 634 and 635.
3 can be obtained. Since a method of integrating blocks is a known technique, a detailed description thereof will be omitted. Further, here, the processing based on the connected area of the black pixels has been described. However, the same processing can be easily realized by projecting a character.

In this manner, stable character extraction is realized by reducing character extraction errors.

Next, a seventh embodiment of the present invention will be described. This embodiment is applied to a case where the electronic data 12 is compared with a part on the sheet 11.

Aside from the range in which character extraction is to be performed, by specifying a portion on the sheet 11 where document collation is desired in the format definition, the electronic data 12 can be collated with a portion on the sheet 11. This method uses the form OCR
Since this is a basic method, it can be easily analogized, and a detailed description is omitted.

As another method, there is a method in which an area analysis result is presented to the operator during the execution of the processing in the area extracting unit 22 and the operator specifies an area to be compared with the document. This method is also similar to the method realized by the document OCR, and can be easily realized by a person skilled in the art, so that the detailed description is omitted.

The necessity of such a function is, for example, when a comparison is to be performed from the third line of the text described on the sheet.
The position of the third line cannot be determined until after the character is extracted from the beginning, but since it is necessary to specify only a part of the position, it is useful to specify the document collation position separately from the character extraction. This method can also be specified in the middle of a line.

Next, an eighth embodiment of the present invention will be described.

This embodiment is characterized by a method of dividing a character dictionary used by the character recognition unit 214 and a control method at the time of reading.

The purpose of dividing the dictionary is to improve the accuracy of character recognition. In character recognition, it is a general fact that the reading accuracy decreases as the number of categories to be identified (the number of character types to be read) increases. In particular, at the time of inputting (creating) electronic data, it is difficult to restrict the input target character type to a creator who is not aware of the character type that can be read by the OCR.
The range is up to the IS second level, which is more than 6,500 characters. Therefore, the reading accuracy is improved by using a divided dictionary for a certain read character and reducing the number of read target character types in a pseudo manner. Further, the effect that the processing speed is improved by reducing the read character type is also obtained.

Here, a specific method of dividing a character dictionary will be described. First of all, symbols, hiragana, katakana, uppercase letters, lowercase letters, JIS first level, JIS second
There is a method of dividing the dictionary according to the level. Since this dividing method conforms to the JIS code, it can be realized very easily. As a second division method, there is a method of dividing a kanji for each radical. Classification of radicals with respect to JIS codes is easy to prepare because they are commercially available. For example, such data can be found in, for example, Riki Kamigaki, "PC Word Processor Kanji Dictionary" (published by Natsume Co., Ltd.). As another division method, there is a method of dividing two or more characters having a high rate of misreading into different dictionaries in a character recognition test at the time of creating a dictionary. In this method, for example, a pair such as “Yu (Kanji)” and “Ta (Katakana)” is divided into different dictionaries. Such a division method is based on the result of the character recognition experiment only if there is only an experiment result.
It is relatively easy to program by taking out combinations of characters that are equal to or more than% and by dividing them into a plurality of dictionaries so that these characters do not overlap each other. As the latter segmentation technique, such a combination of characters with a high rate of misreading is used as a penalty evaluation function,
An automatic clustering method often used in pattern recognition can be used.

As shown in FIG. 17, the character dictionary is arranged so that each element (dictionary of individual characters) of the divided dictionary is continuous as a unit. In the dictionary division table, the head storage positions of the division dictionaries are described in the storage order of the division dictionaries. Therefore, if the Z-th divided dictionary is accessed, the start position from the beginning of the dictionary is stored at the Z-th position in the dictionary division table, and the end position (start position of the divided dictionary Z + 1) is stored at the Z + 1-th position in the table. It will be.

Next, an example of such a divided dictionary access method will be described with reference to FIGS. 17 and 18. FIG. First, a character code at a corresponding character position is read from the electronic data 12 (step 81 in FIG. 18). Based on this code, the number (Z in FIG. 17) of the divided dictionary containing the correct answer of this code is acquired using the dictionary number table (step 82 in FIG. 18). The procedure for acquiring the number of the divided dictionary from the character code is exactly the same as the procedure for acquiring the shape code described with reference to FIG. The numbers of the divided dictionaries are given in the storage order of the dictionaries as described above.

Next, the dictionary division table is accessed from the number of the division dictionary (step 83). This access method is as described above.

When the storage positions of the divided dictionaries are obtained in this manner, the dictionaries are sequentially read (step 84), collation is performed (step 85), and the read position of the next character dictionary is set (step 86). At this time, if the next read position has reached the end position of the divided dictionary, the process ends, otherwise, steps 84 to 86 are executed (step 8).
7).

Here, the method of accessing the divided dictionary has been described. After collating the entire dictionary, based on the divided information of the dictionary, the character codes having the highest performance among the character codes contained in the divided dictionary are selected from the character recognition result candidates. The procedure is also possible.

Next, a ninth embodiment of the present invention will be described.

This embodiment is different from the other embodiments in the procedure of the collating unit 22.

The matching unit 22 lists pairs of inhomogeneous characters that cannot be distinguished from each other as a list. Homomorphic characters have the same character shape but different character codes. Some commonly used characters include “I (Kanji) and # (symbol)”, “H” (hi) and “Iwaku”, ( To (Katakana) and To (kanji) "," C
There are a considerable number of such homomorphic pairs represented by "(uppercase) and c (lowercase)". Strictly speaking, “day and day” are different from homomorphic characters, but may be treated as homomorphic characters because it is extremely difficult to distinguish them in character recognition. A list of such character code pairs that are consecutively arranged is a pair of homomorphic characters.

In FIG. 19, the character recognition result 31, the electronic data 32, and the collation result 33 are the same as those in FIG. 4 described above, and the derivation procedure is the same as the method already described.
The matching result 34 is derived from the matching result 33 with reference to a list of pairs of homomorphic characters.

More specifically, a list of homomorphic characters is searched using the character code of the electronic data 32 at the position where the collation result 33 does not match, that is, 1 in FIG. As a result, if the character code is found in the list, the character code of the other party having the same variant is read from the list, and the code is compared with the character code of the character recognition result 31 at the corresponding position. As a result, if they match, the matching result 33 is switched from non-matching to matching and stored in the matching result 34. In the comparison result 33, when there is no character code corresponding to a list other than the mismatch (0 in FIG. 19) and the homomorphic character pair, the comparison results 33 and 34 have the same value.

In FIG. 19, there is a pair of “to” (Katakana) and “to” (Kanji) in the list of homomorphic characters.
Because there are no "〒" and "te", "to" (Katakana) and "to"
Only the matching result corresponding to (Kanji) is switched from 1 to 0.

The tenth embodiment of the present invention is applied when confirming a once created document. In other words, a document with high importance is created after writing a document with ordinary characters, reconfirming the contents, and then submitting or sending the document. The present embodiment supports such work. .

Referring to FIG. 20, operation terminal 91 includes:
Stores electronic data created by a word processor or sent using e-mail or the like. This electronic data corresponds to the electronic data 12 shown in FIG. As the operation terminal 91, a personal computer, a server, or the like is usually used.

Next, the electronic data is printed from a printer 92 connected to the operation terminal 91 to create a sheet. Next, the created paper is applied to a character recognition device 93 to recognize its contents. The character recognition device 93 corresponds to the character recognition unit 21 shown in FIG. This recognition result is stored in the operation terminal 91.
Then, the electronic data stored in the operation terminal is compared with the character recognition result, and the mismatched portion is confirmed using the display. The operation terminal 91 is the collating unit 22 shown in FIG.
And the visual confirmation unit 23. As a result, if a match is confirmed, the document collation of the paper is terminated.

In this embodiment, the configuration of the operation terminal, the printer, and the character recognition device has been described. However, a configuration in which the character recognition device is replaced with a scanner and the character recognition itself is executed on the operation terminal side is also possible. At present, there is a product in which a printer and a scanner are integrated, so that it is possible to easily integrate the printer and the scanner.

At this time, if the printing font, printing size, and printing layout of the printing unit are fixed, the area analysis is dedicated to the layout, and the character dictionary is dedicated to the printing, it is quite comparable to a normal character recognition device. It is possible to obtain a highly accurate character recognition accuracy.

Next, an eleventh embodiment of the present invention will be described. In the present embodiment, in the configuration shown in the above-described tenth embodiment (FIG. 20), it is possible to relatively freely select a font for printing and to keep character recognition accuracy high. .

In the configuration shown in FIG. 20, the character font, size, and layout to be printed on paper can be controlled on the operation terminal 91 side. Therefore, it is possible to notify the character recognition device 93 of the font / size / layout information to be printed in advance. These information are stored in the printer 92.
Immediately before the print data is sent to the character recognition device 93 (character recognition unit), the notification is made in the format shown in FIG.

FIG. 21 shows the format of each character of the electronic data, and size information of the size of this format, which is the number of characters, is transferred to the character recognition device. The character code 101 represents the character code of the electronic data, the print position (X, Y) 102 represents the upper left position of the circumscribed rectangle of the character, and the coordinate system is an XY coordinate system having the origin at the upper left of the form. The font type 103 represents a print font by a predetermined number. The font size (X, Y) 104 is represented by a horizontal (X) direction size and a vertical (Y) direction size of a circumscribed rectangle of the print font. In addition, as the unit of the coordinates and the size, 1/10 mm or the resolution size of the optical system is used.

In the character recognition device 93, the printer 9
2 has an individual character dictionary for each font size used, and allows divided access. The storage order of the divisions is a numerical order indicating the font type used in FIG. Since the divided dictionaries and a method for accessing the divided dictionaries have already been described in the eighth embodiment, the description will be omitted.

The operation of the character recognition device 92 will now be described with reference to FIG.
This will be described with reference to FIG. Reference numeral 111 denotes an example of a character image to be read after the area is extracted. Since the character image is printed using a proportional font, the printing pitch is not constant. Extract a single black connected area from the character image 111,
What is expressed by the circumscribed frame is 112 in FIG.

There is a method for extracting an area by using the print position 102 shown in FIG. 21 as information. However, the printing accuracy of a printer for a personal computer is occasionally shifted by 1 to 2 mm in the vertical and horizontal directions, so that it is difficult to use the information as highly accurate information. Therefore,
The region extraction is first performed by the method described in the first embodiment, and the print position 102 is used as auxiliary information.
For example, when there are two small points and it is not clear which one to select, a final decision is made using the actual position information 102.

Character extraction is performed by repeating integration as necessary while comparing with the circumscribed frame using the font size. For example, in the case of FIG. 22, when 1121 and 1122 are integrated, they substantially match the font size 104 of “ko” received from the operation terminal 91.
122 is integrated to be the first character. The size of 1123 is approximately the same as the font size 104 of
Since the characters 124 and 1125 are integrated and substantially coincide with the font size 104 of “ha”, this is set as one character. Similarly, each of 1126 to 1130 has a size corresponding to “for test” and is cut out as one character.
1131 to 1133 are integrated to correspond to the size of “de”, and 1134 and 1135 are also cut out as “su”.

The dictionary used for character recognition also switches the dictionary to be collated based on the information of the font type 103 for each character and uses a necessary divided portion. For example, for a character in font A, the character dictionary uses only the font A portion.
As described above, since the font type 103 in FIG. 21 matches the division number of the character dictionary, the method of using the same can be performed by the same processing as that described in the eighth embodiment.

In this embodiment, it is not necessary to use all of the information shown in FIG. 21, and it is possible to apply only some of the information.

As described above, it is possible to stably extract characters even when printing a proportional font or the like in which the print pitch is not constant, which has conventionally been considered difficult to extract characters stably. Also, by switching to a dedicated dictionary for each character in accordance with the font, character recognition can be performed with high accuracy.

[0087]

As described above, according to the present invention,
Most of the document collation between the document in the electronic data format and the document described on the paper can be automated, which has the effect of reducing the work load due to the conventional manual operation. Furthermore, since the amount of manual collation is reduced and much is automated, the processing speed is improved, leading to more efficient business and customer service.

Further, since all the visual check operations can be performed on the display on the terminal or on a single sheet of paper, it is possible to further reduce the number of manual operations and increase the efficiency of operations and services.

Further, by making a portion having low reliability of character recognition a target of visual confirmation, it is possible to prevent a decrease in document collation reliability due to an error in character recognition, thereby realizing highly accurate document collation.

Further, by pointing out a place where a collation error occurs, it is possible to reduce a human error in confirming the visual confirmation part. Therefore, more accurate document matching can be realized.

Further, by improving the character extraction accuracy of a character string in which full-width characters and half-width characters are mixed, the accuracy of character recognition can be improved, and as a result, visual confirmation work can be reduced.

Further, in addition to a character string in which full-width characters and half-width characters are mixed, the character recognition accuracy can be improved by improving the character extraction accuracy of a symbol or the like which has been apt to be erroneous in the past, and as a result, the visual confirmation work can be performed. Can be reduced.

Further, it is possible to collate a part of the document described on the paper sheet with the electronic data, which has an effect of expanding the applicable range of the document collation. In particular, there is an effect that the document can be collated from the middle of the document.

Further, the accuracy of character recognition can be improved, and the visual confirmation work can be reduced. For example, if katakana and kanji are used as separate divided dictionaries, misreading such as katakana “ta” and kanji “evening” is eliminated, and the character recognition accuracy is improved.

Further, it is possible to cope with the same and different characters which are difficult to completely cope with in character recognition, and there is an effect that the visual check operation can be further reduced.

Further, this is for supporting document collation after creation of an important document, and the work efficiency can be improved.

Further, according to the tenth aspect of the present invention, it is possible to particularly improve the character recognition accuracy, and it is possible to realize more efficient work. Also, it is possible to handle various types of documents such as proportional pitch printing and documents in which the printing font is switched for each character, which is difficult to achieve high reading accuracy with a normal character recognition device. Connect.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a diagram illustrating an example of a sheet used for document matching.

FIG. 3 is a diagram illustrating a configuration example of a character recognition unit in FIG. 1;

FIG. 4 is a diagram illustrating a processing example of a matching unit in FIG. 1;

FIG. 5 is a diagram showing a flow of processing of a visual confirmation unit in FIG. 1;

FIG. 6 is a diagram showing a display example of a visual confirmation unit in FIG. 1;

FIG. 7 is a diagram illustrating an example of a character recognition unit-collation unit image interface by file transfer.

FIG. 8 is a diagram illustrating an example of a character recognition unit-collation unit image interface based on cutout position passing.

FIG. 9 is a diagram illustrating a processing example of a matching unit according to a third embodiment of the present invention.

FIG. 10 is a diagram illustrating a display example of a visual confirmation unit according to the third embodiment.

FIG. 11 is a diagram illustrating a processing example of a matching unit according to a fourth embodiment of the present invention.

FIG. 12 is a diagram illustrating a display example of a visual confirmation unit according to a fourth embodiment.

FIG. 13 is a block diagram showing a configuration of a fifth embodiment of the present invention.

FIG. 14 is a diagram illustrating an operation of extracting a character according to the fifth embodiment.

FIG. 15 is a diagram illustrating a method for acquiring shape information according to a sixth embodiment of the present invention.

FIG. 16 is a diagram for explaining the character extracting operation in the sixth embodiment.

FIG. 17 is a diagram illustrating a method of referring to a character dictionary according to the eighth embodiment of the present invention.

FIG. 18 is a diagram illustrating a character recognition process according to the eighth embodiment.

FIG. 19 is a diagram illustrating a processing example of a matching unit according to a ninth embodiment of the present invention.

FIG. 20 is a diagram showing a configuration of a tenth embodiment of the present invention.

FIG. 21 is a diagram illustrating an example of print character information according to an eleventh embodiment of the present invention.

FIG. 22 is a diagram for explaining the character cutting operation in the eleventh embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 Sheet 12 Electronic data 21,129 Character recognition part 22 Collation part 23 Visual confirmation part 91,130 Operation terminal 92 Printer 93 Character recognition device

Claims (11)

(57) [Claims] 1. A recognition unit for optically reading characters written on a sheet, a collation unit for collating character codes stored as electronic data in advance with a result of recognition by the recognition unit, and a collation unit. A confirmation unit that switches a display method so as to indicate a position where the operator pays particular attention to view in accordance with a result of the collation by the unit, and displays a character corresponding to a character code of the electronic data; The recognition result includes a character code obtained based on a character image obtained by optically scanning a character written on the sheet and a degree of reliability of the recognition result of the character code. The operator pays special attention to the character code of the electronic data and the character code of the recognition result according to the verification result and the reliability of the character code of the recognition result. Switching the display method to indicate the position to pay and view, and displaying the characters of the electronic data, the checking unit is that the collation result between the character code of the electronic data and the character code of the recognition result is the same, A first case in which the reliability of the character code is higher than a first value set in advance, a second case in which the matching result is matched, and the second case in which the reliability is lower than the first value. In a third case where the collation results do not match, the characters of the electronic data are displayed by different display methods. 2. A recognizing unit for optically reading characters written on a sheet, a collating unit for collating a character code stored as electronic data in advance with a result of recognition by the recognizing unit, and a collating unit. A confirmation unit that switches a display method so as to indicate a position where the operator pays particular attention to view in accordance with a result of the collation by the unit, and displays a character corresponding to a character code of the electronic data; A first storage unit that stores in advance a character type that can be read by the computer; and the collation unit determines a display method according to whether a character of the electronic data is a character type stored in the first storage unit. A document collation device that switches and displays characters of the electronic data. 3. A recognizing unit for optically reading characters written on a sheet, a collating unit for collating a character code stored in advance as electronic data with a result of recognition by the recognizing unit, and a collating unit. A confirmation unit that switches a display method so as to indicate a position where the operator pays particular attention to view in accordance with a result of the collation by the unit, and displays a character corresponding to a character code of the electronic data; A scanning unit that optically scans a character written on the sheet, an area extracting unit that extracts a character area from image data obtained by scanning, and a character image in the extracted area as one character. And a character recognition unit that recognizes a character image cut out for each character by using a character dictionary, wherein the character cutout unit stores character code data of the electronic data. A document collating device, wherein a cutout position of the character image corresponding to the character code is controlled according to a length. 4. A recognizing unit for optically reading characters written on a sheet, a collating unit for collating a character code stored in advance as electronic data with a result of recognition by the recognizing unit, character by character, and the collating unit. A confirmation unit that switches a display method so as to indicate a position where the operator pays particular attention to view in accordance with a result of the collation by the unit, and displays a character corresponding to a character code of the electronic data; A scanning unit that optically scans a character written on the sheet, an area extracting unit that extracts a character area from image data obtained by scanning, and a character image in the extracted area as one character. A character extracting unit that extracts a character image extracted for each character by using a character dictionary, and a character extracting unit that extracts a character image corresponding to each character code of the electronic data. A second storage unit for storing shape information indicating a tangent frame, wherein the character cutout unit sets a cutout position of the character image based on shape information obtained by referring to the second storage unit. A document matching device characterized by controlling. 5. A recognizing unit for optically reading a character written on a sheet, a collating unit for collating a character code stored as electronic data in advance with a result of recognition by the recognizing unit, and the collating unit. A confirmation unit that switches a display method so as to indicate a position where the operator pays particular attention to view in accordance with a result of the collation by the unit, and displays a character corresponding to a character code of the electronic data; Includes a plurality of character dictionaries that are divided based on a predetermined condition and can be specified by a character code of the electronic data, and are located at positions corresponding to the characters when recognizing characters described on the sheet. A document collating apparatus for acquiring a character code of the electronic data, determining the character dictionary to be used in accordance with the character code, and recognizing the character using the character dictionary. 6. A recognizing unit for optically reading characters written on a sheet, a collating unit for collating a character code stored in advance as electronic data and a result of recognition by the recognizing unit one character at a time, A display unit that switches a display method so as to indicate a position where the operator pays particular attention to the position to be viewed in accordance with a result of the comparison by the unit, and displays a character corresponding to a character code of the electronic data. A third storage unit that stores a combination of a plurality of characters in the form of a character code in advance, wherein the collation unit determines that the character code of the electronic data does not match the character code of the recognition result in the collation result When the character code of the electronic data is stored in the third storage unit, a character code of the other character having a similar shape is acquired, and the acquired character code and the recognition result are obtained. Document matching apparatus characterized by collating the character code again. 7. A printing unit that prints a character corresponding to a character code of electronic data stored in advance on a sheet, a recognition unit that optically reads the character printed on the sheet, and a character of the electronic data. A collating unit that collates a code and a recognition result by the recognizing unit one character at a time, and switches a display method so as to indicate a position where an operator pays special attention to look according to a result of the collation by the collating unit,
A document verification device for displaying a character corresponding to a character code of the electronic data. 8. The electronic data holds at least one piece of information of a character font, a character size, or layout information together with the character code in the electronic data unit or the character code unit. A printing process is performed based on the information held by the electronic data, and the recognition unit controls a cutout position of a character image obtained from printed characters on the sheet based on the information, and a character dictionary to be used. 8. The document matching apparatus according to claim 7 , wherein the specification is performed. 9. When the character of the recognition result and the character of the electronic data corresponding to the character do not match in the collation result by the collating unit, the confirmation unit determines that the electronic data is determined to be mismatched. document matching device according to any one of claims 1 to 8, wherein the emphasizing and displaying the data of the character. 10. The recognition result by the recognition unit includes a character image obtained by optically scanning a character written on the sheet and a character code obtained by performing a recognition process on the character image. a document matching device according to any one of claims 1 to 8, wherein the displaying side by side with the character image corresponding to the character and the character of the electronic data. 11. A character code corresponding to a character code of electronic data stored in advance is printed on a sheet, a character printed on the sheet is optically read to obtain a recognition result, and a character code of the electronic data is obtained. And the recognition result are collated one character at a time to obtain a collation result.The display method is switched so as to indicate a position where the operator pays special attention to look according to the collation result, and the character code of the electronic data is A document collating method comprising displaying a corresponding character and visually confirming the collation result.