JPH09297821A - Character recognition method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To surely detect and correct erroneous recognition at high speed by retrieving a data base from a recognized result through the use of semantic depending relation between items, operating the recognized result with prescribed algorithm, obtaining information which can be recognized at a glance and displaying it with the recognized result. SOLUTION: The filling position of the item is specified from reformat information identified in a format identification part 101 by an item information extraction part 102, and picture information of the position is extracted at every item. A character segment part 103 segments the character from item information at every character. The characters segmented at every character in a character segment part 108 are transmitted to a character recognition part 104 and they are recognized. The recognized result is displayed on the screen of a recognized result display part 105 in the order of the characters, which is decided in the character segment part 103, at every item decided in the item information extraction part 102. An information retrieval part 106 operates and displays information which can be extracted with the recognized result of the character recognition part 104 as a key by retrieving the data base.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition method for recognizing characters written on a form divided into areas for each entry.

[0002]

2. Description of the Related Art Conventionally, a character recognition system for reading a form in which entry items are predetermined for each item extracts a character string for each item based on known format information, and further extracts it for each character. It is divided and sent to the character recognition unit for recognition processing. However, since there is no perfect character recognition so far, the recognition result may include erroneous recognition to some extent.

In order to solve this point, conventionally, an operator checks whether or not each form is read correctly, or if there are many forms to be processed, rejects a form containing ambiguous characters and rejects it. I had to do some work such as reviewing only the submitted form.

On the other hand, in a character recognition system in which the object to be read is numeric data, there is a method using a check digit. In this method, for example, for a numerical string such as a bank account number, a number that is deduced from the numerical string by a certain algorithm from the entry stage is inserted as a check digit at the end of the numerical string or another item, and the numerical value is inserted. It is recognized from the recognition result of the column whether the deduced check digit value and the entered check digit recognition result match. At this time, if they do not match, the operator is prompted to make a correction. If not, the recognition result is regarded as correct and the process proceeds to the recognition of the next form. This allows you to narrow down the items to be reviewed, giving the operator more time,
Accurate review work can be expected.

However, since the check digit is generally a one-digit number, there is a 1/10 chance of overlooking the erroneous recognition when the erroneous recognition overlaps. Further, even in the case of a numerical string, there are some objects that cannot be applied because it is practically impossible for the writer to calculate the check digit like the amount of money, and the problem still remains.

Further, there is the following method. For example, when identifying the transfer account from the account number, recognize both the account number and the holder on the transfer form, and recognize the holder name and the holder column obtained by searching the database from the recognition result of the account number. This is a method of urging correction only when the names of the holders obtained from the above do not match.

[0007] However, in this case, the account number is a number, but in most cases the holder is written in Kanji, so the recognition rate is extremely poor, and even if a correct description is made, correction is urged. More often than not. In addition, when the holder is a corporation, there are various description formats, and even if the holder is correctly described and the recognition format is correct, the description format is often different and correction work is often urged.

In any case, after all, an operator must compare and confirm a list of numbers that is meaningless to humans, and when recognizing such a list of numbers, a human being can quickly and surely. It was necessary to provide an environment to check.

[0009]

In the above character recognition processing, when the operator confirms the recognition result, it is difficult to judge at first glance whether or not it is correct by simply enumerating numbers such as a bank account number. Had to compare character by character. This not only takes a lot of work time, but could lead to an accident by overlooking the misrecognition.

In addition, when the reading target includes kanji, the recognition rate is not only low, but if the character is erroneously recognized, the current recognition method often makes a judgment based on the similarity of the characters. Often, it looks like it looks good, so you need to be very careful when checking.

As described above, when recognizing a sequence of meaningless numbers or a Chinese character string, it is necessary to provide an environment in which a person can confirm the recognition result at high speed and surely. In view of this point, the present invention has an object to provide an environment in which a person recognizes a recognition result at high speed and surely.

[0012]

SUMMARY OF THE INVENTION The present invention, when confirming the recognition result of a character string such as information such as an account number that is meaningless to humans or a similar character, has a semantic dependency between items. By searching the database etc. from the recognition result using the relation, or by deducing the recognition result with a certain algorithm, information that can be confirmed by humans at a glance is also obtained, and by displaying it together with the recognition result, the operator Enables to detect and correct misrecognition at high speed and reliably.

According to the present invention, since information that is difficult for humans to confirm is displayed together with information that can be easily confirmed,
It is faster and more reliable than the conventional method when confirming the confirmation result. Therefore, it is possible to realize a character recognition method for a form with less erroneous recognition than before.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a processing method of the first embodiment. The format identification unit 101 is a portion that identifies the form format to be read. Here, it is identified which existing form the form to be loaded corresponds to. As an identification method, information such as the position of the ruled line is extracted from the form to be read, and the form is identified by comparing the data with the existing form data or by recognizing the format identification number written in advance in the form. There is a method of specifying. The item information extraction unit 102 identifies the entry position of each item from the format identification unit 101 or the format information, and extracts the image information at that position for each item. Further, the character cutout unit 103 cuts out a character from each item information. As for the configuration of the character cutout unit 103, for example, in the case of a form in which characters are written at regular intervals, there is a method of dividing the characters at predetermined intervals. Alternatively, as shown in FIG. 2, there is also a method of expressing the distribution of a black image (mainly pixels of a character portion) in an image as a histogram in the vertical direction and cutting when the histogram becomes 0. In FIG. 2, 201,
202 and 203 are histograms of black pixels corresponding to the respective characters, and 204 and 205 are cutout positions. The characters cut out for each character by the character cutout unit 103 are sent to the character recognition unit 104 where they are subjected to recognition processing. The recognition result here is displayed on the screen of the recognition result display unit 105 in the order of characters determined by the character cutout unit 103 for each item determined by the item information extraction unit 102.
The information search unit 106 searches or deduces a database storing various data for information that can be extracted using the recognition result of the character recognition unit 104 as a key, and the recognition result display unit 105
To display the result. Note that this part of the database will be described in detail later. When the operator notices an error in the recognition result by looking at the recognition result display screen, the recognition result correction unit 1 is operated by a predetermined operation operation.
07 is activated to correct the incorrect recognition result, and the recognition result confirmation unit 108 confirms the correction. The confirmed data is stored in a predetermined memory.

Further, the contents corrected by the recognition result correction unit 107 are fed back to the information retrieval unit 106 and the retrieval result is reflected on the display screen of the recognition result display unit. In addition to the method of activating the recognition result correction unit 107 by the operation as described above, there is also a method of automatically activating the recognition result display unit 105 and the search information display unit 107 at the stage when the processing is completed. This is effective when there is a large possibility that the recognition result will be corrected. However, in this case, it is necessary to confirm the recognition result by the recognition result confirmation unit 108 even when there is no correction.

Now, it is assumed that the character recognition device of the present invention automatically reads a form (also called a form) as shown in FIG. Before the detailed description, the description of FIG. 3 will be given. FIG. 3 is an example of a bank transfer sheet 300, and detailed portions are omitted for clarity of explanation. In FIG. 3, reference numeral 301 is a transfer destination bank name description field, 302 is a branch name description field, and 303 is a transfer destination holder's description field.
304 is the bank number of the bank described in 301, and 305 is the store number of the branch described in 302. 306 is 30
This is the account number of the holder of 3. Here, for simplicity, it is assumed that the transfer destination account is, for example, a savings account. Further, 307 is a column for describing the transfer amount, 308 is an item for describing the name of the transfer person, and 309 is also an item for describing the address.

Generally, the form is often redundant in terms of description, and the transfer form shown in FIG.
The bank name, branch name, and holder are uniquely determined by the set of account numbers. That is, if you enter the bank number, store number, and account number,
You can make a correct transfer without describing the bank name, branch name, and owner. In this sense, the bank name, branch name, and holder name are redundant information. Actually, when reading this transfer paper, only the part such as bank number, store number, account number etc. that is easy to be recognized accurately is read, the operator confirms the number and confirms the existence of the corresponding account and confirms the transfer action. The method of doing it has been taken. With such a method, the number of character types to be recognized in the system such as numbers is small, so it is easy to recognize, but it will be recognized with information that is difficult for humans to understand, and when the operator confirms the recognition result, each character Not only did it take a long time to confirm, but there was a possibility that I would miss it when I got used to it.

Therefore, in the present embodiment, after reading the bank number, store number and account number, the database is searched and the corresponding bank name, branch name and holder are displayed on the recognition result confirmation screen. FIG. 4 is an example of the recognition result confirmation screen. In FIG. 4, 401 is a bank name extracted from the database by searching the bank number, and 404 is an image portion of the bank name extracted from the transfer form. Similarly, 402 is a branch name extracted from the database by searching the bank number and branch number, and 405 is an image portion of the branch name extracted from the transfer paper. Further, 403 is a holder searched from the bank number, store number, and account number, and 406 is an image portion of the holder extracted from the transfer paper. Reference numeral 407 is the recognition result of the bank number, which can be corrected on this screen or by another operation. 410 is an image part of the bank number extracted from the transfer paper. Similarly, 408 is the recognition result of the store number, which can be corrected on this screen or by another operation. Reference numeral 411 is an image portion of the store number extracted from the transfer paper. Further, reference numeral 409 is the recognition result of the account number, which can be corrected on this screen or by another operation. Reference numeral 412 is an image portion of the account number extracted from the transfer paper. Finally, 413 is the recognition result of the amount column, which can be corrected on this screen or by another operation. Reference numeral 410 is an image portion of the amount column extracted from the transfer paper.

In this example, as shown at 412, the last fifth digit of the account number is an extremely vague "0", so 40.
As shown in 9, it is misread as "6". As a result, as indicated by 403, the owner has become "Hinode Sangyo Co., Ltd." as stored in the database. In this way, even if there is little difference in account numbers, displaying the holder often makes the difference clear. As is the case with this example, the error detection is much faster and more accurate than the traditional method of comparing numbers. In addition, it is easier to find an error location than before. Because
In this case, since the bank name and branch name match, the bank number and the store number are considered to match, and only the account number needs to be reviewed. As a result, the correction work can be realized with less than half the effort of the conventional method.

Further, in the case of this embodiment, after correcting the account number from "6" to "0", the correction data is sent to the information retrieving section, the database is re-searched, and then re-displayed on the recognition result display screen. . This not only makes it possible to confirm the correction result, but also makes the correction easier. This means that if the bank number is not recognized correctly, not only will a different bank name be displayed, but even if there is a store number or account number, it will not be possible to search, or it will be as if the branch name or name is irrelevant. People are displayed. This is inconvenient because it forces the operator to review the store number and account number.
Therefore, when the bank number is corrected, the correction contents are automatically sent to the information retrieval unit 106, and the result is recognized by the recognition result display unit 10
If it is redisplayed in 8, the branch name and the holder can be confirmed, and the number of items to be reviewed can be reduced.

As described above, in the present embodiment, the retrieval is automatically performed when one item is corrected. However, when the retrieval takes time, this method takes a lot of work time. In such a case, the operator may arbitrarily activate the information search unit 106 with any function key.

Further, in the past, the display unit was often left-justified regardless of the original image as shown in FIG. In FIG. 5, reference numeral 4101 denotes an original image display window,
Reference numeral 102 denotes a search result display window. Although such a method is convenient in terms of display on the system side, it is not always convenient from the viewpoint of confirmation work by the operator. Although FIG. 6 shows the case of right justification, inconvenience is the same as the case of left justification. This inconvenience is mainly caused by the difference between the original image and the characters, and the method shown in FIG. 7 solves this problem.

In FIG. 7, 4101 is an original image display window and 4305 is a search result display window.
Further, 4301 to 4304 are the center lines of the characters in the original image. In this method, the character display position of the search result is automatically determined so that the center line of the display character of the search result matches the center line of the character of the original image. According to this method, comparison and confirmation can be facilitated as shown in FIG. Of course, the left-justified and right-justified may be easier to perform comparison work depending on the operator and the work content thereof, and it is desirable to be able to switch between them by a simple operation.

Further, if the size of the character font is changed according to the size of the character of the original image as shown in FIG. 8, the comparison work becomes easier. It is desirable to automatically select the character font size that is closest to the character size of each character. However, in the case of handwritten characters, the size of each character may differ. In such a case, the size of each character is changed and the original image is visually matched, or an average character size is obtained, and all the characters are displayed in that size.
In the latter method, the size of the displayed characters is constant, so it is easy to understand the semantic content of the search results. Of course, the size of the font is limited, so there may not be a good font. In this case, it is necessary to enlarge or reduce the original image to bring it closer to the existing character font.

Further, the recognition result may be wrong, and it may be necessary to display a search result different from the original image. In this case, the number of characters is often different, and the above method cannot be used successfully. Therefore, as shown in FIG. 9, if the left end 4501 and the right end 4502 of the character string of the original image are obtained and a character font size that fits exactly between them is selected, comparison work becomes easier compared to the case where the display is shifted to the left and right. . Of course, some of the character size changing methods described above have an error in easiness of work depending on the operator and the contents of the work as in the case described above, so it is desirable to switch between them by a simple operation.

In addition, there is also a small amount of information written like a transfer form in the same branch. In such a case, it is very troublesome and time-consuming to perform the confirmation work on the screen for each form as shown in FIG. Therefore, as shown in FIG. 10, a method is possible in which several forms are simultaneously displayed on the screen and the confirmation / correction work can be performed on the screen. In FIG. 10, reference numerals 4601 to 4603 are information of an account number, a holder, and an amount of money corresponding to one form, respectively. The upper row shows the original image, and the lower row shows the recognition result and the search result. Since it is premised that all transfers are made to accounts in the same branch, there is no need to perform recognition processing for bank numbers and branch numbers, and of course no correction is necessary. However, in this case, the information displayed on the screen may not be enough. In such a case, it is desirable that the detailed information of the form be displayed by selecting the corresponding portion so that detailed confirmation work can be easily performed.
At this time, the correction becomes easier to understand by displaying the list by the contents of the items, in this example, the account numbers obtained as a result of recognition.

Further, as shown in FIG. 4, the recognition result is displayed near the original image on the recognition result display / correction screen, or one or both of them are enlarged or reduced so that the original image and the recognition result can be easily compared. An efficient work environment can be provided.

FIG. 11 is a block diagram showing the overall structure of the second embodiment. In FIG. 11, 507 and 508
The other parts are the same as the corresponding parts in FIG. Reference numeral 507 denotes an information calculation unit, and the recognition result for each item sent from the recognition result display unit or 505 is sent to the information search unit 506 to search for necessary information, and the searched information is stored together with the recognition result. It is sent to the information calculation unit 507. The information calculation unit is a unit that calculates a value deduced by a certain algorithm from the recognition result and the search result. Thereafter, the recognition result, the search result obtained in 506, and the calculation result obtained in 507 are sent to the information consistency determination unit 508, and at the same time, the calculation result is also sent to the recognition result display unit 505 via the information search unit 506 and displayed. To be done. The information consistency determination unit 508 checks and determines the consistency of the transmitted information by a certain algorithm. 5
The information obtained in 08 is sent to the recognition result display unit via 507 and 508 and displayed based on the determination result. It is also possible to display the recognition result of a certain item in a form from a plurality of items and images side by side for each item.

FIG. 12 shows a form which is the object of this embodiment. This is an order form for electrical equipment, and is in a tabular format such as 601. An item 602 describes the total amount of the order. The item to be read here is the "model number"
And "Unit price" and "Quantity" columns. In the same manner as described in the first embodiment, a character string is cut out for each item, and if it is a read item, it is further cut into character units.
recognize.

The recognition result is displayed as shown in FIG.
In FIG. 13, reference numerals 701 to 704 respectively display the recognition result and the original image for each item. First, for 701, the information of the first item "personal computer MZ80" is displayed, and the upper row displays the recognition result and the information retrieved or deduced from it. The lower row shows an image of the letters written on the order form.

The contents of the upper row will be described in detail. As mentioned above, only the "model number", "unit price" and "quantity" are recognized here. Therefore, the other information is obtained by searching or deducing. First, the “product name” is obtained from the recognition result of the “model number” by the information retrieval unit 506 using the database. Further, the "price" is calculated and displayed by multiplying the "unit price" and the "quantity" in the information calculation unit 507 from the recognition result of the "unit price" and the "quantity". Furthermore, "cumulative" is the sum of the previous accumulation, if any, and "price",
Otherwise, the "price" itself is obtained by the information calculation unit 507 and displayed.

The product name "Personal computer M"
As described above, "Z80J" is the information retrieved using the information retrieval unit from the result of reading the model number, but it seems to refer to the same item because it is similar to the item name written by the orderer. Do not match. In this way, the orderer rarely knows the exact product name, and often omits it or describes a slightly different product name. In such a case, it is impossible to determine that the model number is correct even if the recognition process is performed on the product name so that it can be read correctly. In this case as well, it is impossible to automatically determine that the model number is correct from the recognition result of the product name, and a human check is required. At this time, selecting information that has meaning such as a product name rather than selecting information that is meaningless to humans such as a model number as a check target enables not only quick and accurate checking but also a sense of security. Next, 702 displays information on the "extended memory" of the second item. The display format and the search and calculation method are as described for the first item. Here, "C" in the second digit of "Model Number" is replaced with "O"
I have misread it. As a result, the "item" has become "Audio Peaker LO50". This error can be easily found by checking the product name column by the operator, but in this case, the error can be found automatically. First, the unit price can be searched by the information search unit 506 from the recognition result of the “model number”. When the unit price and the recognition result in the “unit price” column are sent to the information consistency determination unit 508 and the information consistency determination unit does not match the two,
A signal indicating that the "model number" or the "unit price" is incorrect is sent to the recognition result display unit 505 through the information calculation unit and the information search unit to issue a warning that the recognition result contains an error. The format of the warning can be realized by thickening the display frame as shown in 702, for example. As a result, the operator can be alerted and a correction can be prompted. In addition to this, the warning format can be changed by changing the display color, blinking the corresponding character (the relevant character is the recognition result part (upper part) of the "model number" and "unit price"), Possible methods include blinking the relevant frame together with the inside of the frame, or moving the cursor to the incorrect location and sounding a warning sound.

The operator who has been warned enters the correction work,
On the same screen, correct "O" in "Model Number" to "C". This correction information is immediately sent to the information search unit, and a search using the new model number is performed. As a result, the correct product name is displayed,
At the same time, the correct unit price is retrieved. These pieces of information are sent to the information calculation section together with other information that has not been corrected, and the "cost" and "accumulation" are calculated by the method as described above.
Further, these pieces of information are sent to the information consistency determination unit. The information consistency determination unit confirms that the unit price information obtained in the information retrieval unit matches the recognition result in the "Unit price" column, and in the reverse flow, the recognition result display unit confirms that the recognition result is consistent. Send a signal. The recognition result display unit that receives this cancels the warning display.

In addition, as shown in FIG. 13, the warning is not issued to all the items in which the inconsistency has occurred or may occur, instead of the "model number" and the "unit price". A method of issuing a warning only to the place causing the inconsistency or only the smallest item including an error may be considered.
In the case of the present embodiment, the minimum items that include an error are “model number” and “unit price”. This is because it can be understood from the information obtained in this embodiment that at least one of the "model number" and the "unit" is incorrect, but it is not known which is incorrect. By narrowing down the items for which warnings are issued in this manner, the number of items to be confirmed by the operator for correction is reduced, and the confirmation work becomes easier.

Reference numeral 703 is the "main body cover" of the third item.
The information of is displayed. There is a misreading that "3" is changed to "8" in the item of "quantity". As a result, the price deducted from the information calculation unit 507 is different from the price entered. The careful operator can notice the error enough by itself, but in this case, the error can be surely noticed from the display of other parts without noticing it. Reference numeral 705 displays total amount information of all orders. Similar to 701 to 704, the upper part is information on the total amount calculated by the information deduction unit 507 from the recognition result, and the lower part is the image itself of the total amount written on the order form.

Since the total amount of money is the most important information in order processing, the operator must check it carefully. Therefore, the operator will surely find an error here. From this, the operator who finds an error confirms that there is a discrepancy in "accumulation" of 704. Next, when we check back 703 and 702, we can see that they are correct up to 702 and they are different from 703. Therefore, the operator seeks the cause of the discrepancy of the “total” of 703 in the information of 703, and first discovers that the “cost” is inconsistent. However, as described above, this item is a value obtained by the information calculation unit 507 from the “unit price” and the “quantity”, and this is not the main cause. That is, even if the operator corrects this, the correction content is not sent to the information search section, or even if it is sent, the "price" is calculated from the recognition result of the "unit price" and "quantity" as described above. The correction result is not reflected. Then, the operator pays attention to "quantity" and notices that "3" is mistakenly read as "8".

When the operator corrects "8" to "3", the "cost" of 703, "accumulation", "accumulation" of 704, and total are corrected in the same flow as when the "model number" of 702 is corrected. And the total matches the total of the original image (lower row). This allows the operator to confirm that the order form is recognized correctly.

FIG. 14 shows a form targeted by the third embodiment. This form is one in which customer (orderer) information is added below the form targeted by the second embodiment. Reference numeral 801 is a customer name entry field, 802 is an entry field for whether or not the customer has previously placed an order with this company, 803 is a customer address entry field, and 804 is a customer telephone number entry field.

FIG. 11 is a block diagram corresponding to this embodiment. The first object to be read in this embodiment is basically a “customer name” column 801 and a “use / non-use” column 802. In the present embodiment, the reading result of these fields and the retrieval / deduction are performed by using the reading result as a key, and the subsequent reading process and retrieval process are changed according to the result. This is the main purpose of this embodiment, and effective operating becomes possible by performing the stepwise reading.

FIG. 15 shows a recognition result display screen. 90
1 is a customer number. This can be obtained by searching the customer database with the recognition result of "customer name" as a key when "use or non-use" is checked from the recognition results of "customer name" and "use or non-use". 902 is a recognition result of "customer name", and 903 is an original image of "customer name". Further, 904 is a "customer address" obtained from the customer database using the customer number or customer name as a key. Furthermore, 905 is the original image. 906 is a "customer telephone number" obtained by the same method as 904, and 907 is its original image.

In the case of FIG. 15, since the recognition of "customer name" is wrong, "customer address" and "customer telephone number" are completely different from the original image. This allows the operator to notice a mistake in recognizing the "customer name". In this way, if it is difficult to find a recognition error with the information of "customer name" alone, by using the dependency between items, the error with a small difference between the correct answers such as "Kida" and "Honda" can be detected in "Kanagawa Prefecture". It is an object of the present invention to make it possible to display a recognition result more easily, such as "." And "Okayama Prefecture".

Further, when the operator corrects the recognition result of "customer name" correctly, the corrected "customer name" is used as a key to search the customer database, and the correct "customer address" is obtained.
And get the "customer phone number".

The processing of this embodiment when the recognition result of "use or not" is "Yes", the recognition of "customer name" is incorrect, and the customer database can be searched by using the incorrect customer name Is. Depending on the recognition situation and search / deduction situation, the following cases are also possible.

First, since the recognition result of "customer name" is incorrect, there are cases in which a search cannot be performed even if "use" is correctly recognized as "present". In such a case, the "customer address" and "customer telephone number" can be left blank to prompt correction, or a warning can be issued by the method described in the second embodiment. Furthermore, in such a case, the "customer telephone number" column is recognized, the customer database is searched from the telephone number, and the obtained information ("customer number", "customer name", "customer address")
Is displayed. This not only saves the operator the trouble of correcting the recognition error, but also absorbs the fluctuation of the notation such as the company name described in the second embodiment.
Trouble such as double registration can be prevented. Also, in this case, by clearly displaying that the information is retrieved from the telephone number, such as changing the color of the telephone number recognition result field, the operator is notified of the situation of retrieval and deduction, and efficient correction work is performed. It can also be done.

Next, even if the "customer name" is correctly recognized and the "use / non-use" is "present", a completely different "customer address" or "customer telephone number" may be retrieved. This mainly happens when the customer changes his address. In such a case, the operator performs certain confirmation means such as confirmation with the customer, then the operator recognizes the "customer address" and "customer telephone number", confirms and corrects the obtained recognition result, and updates the data. As the customer database. By doing so, careless double registration can be avoided and the latest data can always be stored in the database.

In the above case, naturally, a plurality of customer candidates will appear if they have the same surname and name. In this case, a plurality of results obtained by searching a database or the like are displayed and candidates are narrowed down from the results, or the "customer telephone number" column is recognized and the customer is specified.

If the "use / no use" is "absent", information cannot be acquired from the customer database even if the "customer name" column is correctly recognized. But,
Even in such a case, it may be a customer's misunderstanding that "use" is "none", and in the case of a corporation such as a company, the orderer does not know the case that another person used in the past. There are cases. Furthermore, in the case of an individual, it is possible that someone in the family is placing an order even though the ordering person has never used it.

Therefore, in this case, the "customer telephone number" is automatically recognized and the customer database is searched using the recognition result as a key. If there is no corresponding data here, the “customer address” is recognized as a new customer, the operator is informed of the new registration, and the check is urged over all the items. Furthermore, at the time of confirmation / correction, a new customer number is assigned and registered in the database.

If there is a corresponding item, "customer name", "customer address" and "customer number" are displayed. Similar to the above case, the operator is informed that the information is retrieved from the "customer telephone number" by means such as changing the color of the telephone number recognition result column. This allows the operator to check "customer address", "customer name", etc., to see if the information is correct, and if there is an error, perform a certain operation to recognize the part or correct the search result. Give instructions. When the correction is completed, the operator inputs a determination as to whether the information is a new customer information or not, by a certain operation.
If it is determined that the user is new, a new customer number is assigned and registered in the customer database. If not, the obtained information is registered in the database of searched customer numbers.

There is also a method of determining whether or not the user is a new person based on whether or not the "customer name" has been corrected and the "customer address" has been corrected. That is, if any correction is made, it is judged as a new person and new registration is automatically performed.
Of course, at this time, it is possible that the correction content is an address change, so the judgment of new registration based on the presence or absence of correction is made only for "customer name", but it is not made the target of judgment of new registration even if it is corrected. Items may exist.

Finally, in this embodiment, 601 and 602 are the same as those in the second embodiment, and are independent from the processing of this embodiment in the sense that there is no dependency between items, so the second embodiment is executed. It should be noted that the processing of the example and the processing of this embodiment can be performed in parallel.

As described above, in the present embodiment, a character recognition device for reading and recognizing a form in which there is a dependency relationship between items is a recognition result of a predetermined item and retrieval / deduction from the recognition result. Depending on the result obtained, it can be configured to select the subsequent processing. This allows the operator to enter data efficiently,
Database management becomes easier.

[0053]

According to the present invention, an environment is provided in which a character recognition device for reading and recognizing a region divided into items can accurately and easily perform an operator's confirmation work and correction work.

[Brief description of drawings]

FIG. 1 is a system diagram of a first embodiment.

[Fig. 2] Example of character cutout by histogram of black pixels

[FIG. 3] Bank transfer paper to be read in the first embodiment

FIG. 4 is a recognition result display / correction screen of the first embodiment.

[FIG. 5] Excerpt of the recognition result display / correction screen of the first embodiment (when left-justified)

[Fig. 6] Excerpt of the recognition result display / correction screen of the first embodiment (when right-justified)

FIG. 7: Extraction of recognition result display / correction screen of the first embodiment (in the case of character center alignment)

[Fig. 8] Excerpt of the recognition result display / correction screen of the first embodiment (in the case of character size matching)

FIG. 9 is an excerpt of the recognition result display / correction screen of the first embodiment (when the number of characters is different)

FIG. 10: Simultaneously recognizes a plurality of forms according to the first embodiment.
Of the recognition result display / correction screen

FIG. 11 is a block diagram of a second embodiment.

[FIG. 12] Order form to be read in the second embodiment

FIG. 13 is a recognition result display / correction screen of the second embodiment.

FIG. 14: Order form to be read in the third embodiment

FIG. 15: Recognition result display / correction screen of the third embodiment

[Explanation of symbols]

Reference numeral 101 ... Format identification unit 102 ... Item information extraction unit 103 ... Character cutout unit 104 ... Character recognition unit 105 ... Recognition result display unit 106 ... Information retrieval unit 107 ... Recognition result correction unit 108 ... Recognition result confirmation unit 201 ... Character “3” Black pixel histogram of 202 ... Black pixel histogram of character "2" 203 ... Black pixel histogram of character "1" 204 ... Separation point of character "3" and character "2" 205 ... Character "2" and character "1" Separation point 300 ... Bank transfer form 301 ... Bank name entry column 302 ... Branch name entry column 303 ... Nameholder entry column 304 ... Bank number entry column 305 ... Store number entry column 306 ... Account number entry column 307 ... Amount entry column 308 ... Transfer person entry field 309 ... Transferee address 401 ... Display window of bank name retrieved from bank number 402 ... Is it bank number and store number? Display window 403 of searched branch name ... Display window of holder searched from bank number, store number and account number 404 ... Display window of original image of bank name 405 ... Display window of original image of branch name 406 ... Original image display window 407 ... Bank number recognition result display window 408 ... Store number recognition result display window 409 ... Account number recognition result display window 410 ... Bank number original image display window 411 ... Store number original image display window 412 ... Original image display window for account number 413 ... Recognition result display window for transfer amount 414 ... Original image display window for transfer amount 4101 ... Original image display window for holder 4102 ... Holder display window 4201 searched for by account number The owner's email retrieved from the account number Dou 4301 ... Center line of the first character original image of the holder 4302 ... Center line of the second character original image of the holder 4303 ... Center line of the third character original image of the holder 4304 ... Fourth character original image of the holder Center line 4305 ... Display window of holder searched by account number 4401 ... Display window of holder searched by account number 4501 ... Left edge line of original image of holder 4502 ... Right edge line of original image of holder 4601 ... Original image of main portion of first form, recognition result and search result (upper row is original image, lower row is recognition result and search result) 4602 ... Original image of main portion of second form, recognition result and search result ( The upper part is the original image, the lower part is the recognition result and the search result) 4603 ... The original image of the main part of the third form, the recognition result and the search result (the upper part is the original image, the lower part is the recognition result and the search result) 501 ... Forma Recognition unit 502 ... item information extraction unit 503 ... character cutout unit 504 ... character recognition unit 505 ... recognition result display unit 506 ... information retrieval unit 507 ... information calculation unit 508 ... information consistency determination unit 509 ... recognition result correction unit 510 ... Recognition result confirmation part 601 ... Table for entering order details 602 ... Total amount entry column 701 ... Information display window for first item in order table 702 ... Information display window for second item in order table 703 ... Information for third item in order table Display window 704 ... Order display fourth item information display window 705 ... Total amount information display window 801 ... Orderer's name entry field 802 ... Orderer's entry field 803 ... Orderer's name entry field Address entry field 804 ... Orderer's telephone number entry field 901 ... Customer number display window 902 ... Customer name recognition result display window Window 903 ... Customer name original image display window 904 ... Customer address recognition result display window 905 ... Customer address original image display window 906 ... Customer telephone number recognition result display window 907 ... Customer telephone number original image display window

Claims (4)

[Claims] 1. A character recognition method for reading a character written in a form divided into areas for each entry item, for each item, using a semantic dependency between each item, the recognition result or confirmed information. A character recognition method characterized by displaying information on items that can be deduced from both or both. 2. A character recognition device for reading characters written in a form divided into areas for each entry item, for each item, by using the semantic dependency between the items, a recognition result or confirmed information or Deduces information that can be deduced from both of them, and if there is a mismatch between the deduced information, the recognition results, or the deduced information and the recognition result, a warning is issued to the part where there is a mismatch. A character recognition method characterized by the above. 3. The character recognition method according to claim 2, when there is a mismatch between the deduced information, the recognition results, or the deduced information and the recognition result. A character recognition method characterized in that a report is issued only to a location or a location where the number of errors is minimized. 4. In a character recognition method for reading characters written in a form divided into areas for each entry, each item is recognized step by step, and the recognition result up to the previous staircase and the recognition result are recognized. Search using semantic dependencies between items
A character recognition method characterized by performing deduction processing, and determining the items to be subjected to recognition processing or search / deduction processing in the next stage based on the results.