JP3383491B2 - Form recognition and classification system - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is a form for accepting a form such as a voucher, recognizing and reading characters written therein, and automatically generating recognition processing information when sorting and storing in a large number of pockets. Recognizing classification system.

[0002]

2. Description of the Related Art For example, in slip processing at supermarkets and convenience stores with stores nationwide, invoice processing for mail order, merchandise delivery and mail delivery processing, a large number of slips are classified, distributed and aggregated. Some work is required. Specifically, a form in which certain items are described is issued, the forms are collected, sorted by destination, and sent to the destination of the form. In the case of a slip or the like, in this case, the work of totaling the amount of money and the like described therein is involved. Such operations typically require large numbers of personnel and are extremely time consuming for processing. Therefore, in general, a large number of specialized personnel have been assigned to the slip processing work.
Further, in order to automate this kind of processing, a system in which a slip itself is eliminated and data is directly input and exchanged using a computer terminal is also widely used.

[0003]

The conventional form recognition and classification system as described above has the following problems to be solved. The present inventors have developed a device that recognizes characters on a form and sorts the form into a large number of pockets according to the contents thereof in order to automatically sort a large amount of forms. In this device, the identification codes of many forms are first read. An image is cut out based on the format information stored corresponding to this form identification code, and characters are recognized. Although there are many types of forms, many format information corresponding to these are registered in advance.

However, the sender of this form may unilaterally partially change the format of the form. In this case, a large number of forms are rejected as unrecognizable, and these are again subjected to manual data input processing by the operator. In this case, it is necessary to inquire the sender of the content of the format change, obtain the original, and create and register the format information again. However, the cost of manually processing a large number of rejected forms is high, and the profit of automation is halved. on the other hand,
Inquiring the sender of the form about the contents of the format change and obtaining the original is time-consuming and extremely complicated.

[0005]

The present invention adopts the following constitution in order to solve the above problems. <Structure 1> The present invention provides a form in which an identification code is attached to each sender.
It is a classification system, and it corresponds to each identification code
Format information indicating the format of each form is stored.
The format information storage section and the form image
The identification code is determined from the data, and the format information storage unit
Import the corresponding format information from
Equipped with a character recognition unit that recognizes characters from the
Character-recognized forms are classified based on the recognition results and
A form recognition classification system that rejects forms that cannot be recognized.
The normal recognition of the character recognition part for each identification code
The number of forms and unrecognizable forms are counted, and
The number of unrecognizable forms for the number of votes exceeds the set threshold.
And the recognition statistics section that instructs the format generation,
When you receive an instruction to generate a matte, you can
Format information from the data
Format that stores information in the format information storage unit
And an information generation unit.

<Description> In order to read a character on a form as image data and cut out the character for recognition processing, format information for displaying a character cut-out position and the like is required. For each sender
Since the formats are different , the format information is registered in correspondence with the form identification code. However, sending source format be pre-formatted information is available
If you change a part of
In this case, the format information is generated using the image data of the unrecognizable form. Therefore, even if the sender changes the format without knowing it, he / she can respond to it.
Wear.

That is, regarding the form of the same identification code,
Equipped with a recognition statistics unit that statistically processes the number of recognized sheets and the number of unrecognizable sheets, the format information generation unit uses the image data of the sheets to generate format information when the number of unrecognizable sheets exceeds a threshold value. And format information
Register in the memory . Threshold set in advance for each identification code
The threshold value is set in advance based on the ratio between the number of recognized sheets and the number of impossible sheets . Accordingly, those unrecognizable rate is abnormally high is determined to have been the format change at sender, and generates the format information by utilizing the image data.

<Structure 2> In the present invention, at least the previous count value of the recognition statistics part
Is further provided with a statistical data storage unit for storing
Is the cumulative value of the current count value and the previous count value or the current count value.
Setting a threshold based on the comparison of the numerical value with the count value
Is characterized by. <Explanation> For example , if the unrecognizable rate is high this time , even though the form of the corresponding identification code was processed normally almost the last time, the author will write the data based on the accumulated count values of this time and the previous time.
Set the threshold value based on the comparison of the current and previous count values.
It

<Structure 3> Furthermore, in the present invention, a character recognition unit for character recognition
A font information storage unit that stores font information,
For changing the font information in the font information storage
It also has a font information re-registration unit, and the character recognition unit
Image data based on the font information stored
Recognize the inability to recognize when disagreement occurs and recognize it
When the statistics department receives a notification, the font information
It is characterized by instructing a change. <Explanation> When the font of a character is changed at the sender, the font dictionary is re-registered instead of automatically registering the format information.

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to specific examples. <Structure of Specific Example> FIG. 1 is a block diagram showing a specific example of the system of the present invention. The system shown in the figure includes a character recognition unit 12 for reading and recognizing the characters written on the form 1 before classifying the form 1. According to the recognition result, the form 1 is classified and sorted in a manner described later. The font information 2 is used for such recognition processing.
1 and format information 22 are stored in the storage unit. In addition to this, a recognition statistics unit 23 for statistically processing the recognition result, an image data storage unit 26, a statistical data storage unit 27, a format information generation unit 28, a font information re-registration unit, and the like are provided. Note that as a result of the recognition processing, the documents that can be classified are transported to the corresponding classification pocket 24 shown in the figure, and if the recognition processing is impossible, the documents are transported to the reject sharing pocket 25.

The font information 21 is a dictionary for character recognition generated in consideration of the character font described in the form. The format information 22 is information for displaying at which position on the form the character to be recognized is written. The recognition statistic unit 23 counts the number of forms that the character recognition unit 12 has performed recognition processing and the number of forms that cannot be recognized,
This is the part that takes predetermined statistics. It should be noted that this statistic is stored in the statistic data storage unit 27, including past statistics as necessary. Further, the image data storage unit 26 stores the image data of the form which has been rejected and which has been determined to be required to generate format information as a result of a certain determination.

[0012] format information generating unit 28, based on image data read from the image data storage unit 26
Based on this, it is the part that generates the format information. In addition,
In this generating method, for example, an image on a form is scanned vertically and horizontally and the number of black dots is counted. As a result, the area in which the ruled lines and the characters written on the form are written is detected. If it is determined from the change in the number of black dots that a character is written on a certain frame or area, that portion is set as an image cutout area for character recognition. Such processing is automatically performed by calculation.
Then, the result is displayed on the display as needed. The operator confirms the displayed contents and judges whether or not the portion set as the image cutout area is acceptable. If there is a problem with the size of the area, correct it. As a result, the format information is confirmed and registered.

The above is a schematic configuration of the system of the present invention. Before explaining the specific operation of the system of the present invention,
A specific method and device of the form classification process will be described. FIG. 2 shows a form processing explanatory diagram suitable for the system of the present invention. In this system, the form is processed as a whole in such a procedure. First, the sender A1 fills in a large number of forms 1 with a destination, a billing amount, and other items. This form 1 is created for various destinations, and this form A is collectively sent from the source A1 to the sorting center. The sorting center is equipped with a form classification processing system. Then, first, in step S1, the form 1 is read one by one, and the data is recognized. Then, in step S2, the forms are classified by destination. In this example, the forms created by the sender A1 are sorted in the order of destinations B1, B2, and B3. These forms are finally sent to the destination as they are, but at this time, the billing amount is calculated for each destination. The billing amount is also written on the form and is recognized by the character recognition unit. Therefore, the following procedure is adopted to collect the recognition results and confirm the results.

First, in step S3, the destination B1
Divide the addressed form into 50 sheets. Then, 50 sheets are taken out and the reading results are totaled (step S3).
Step S4). These are automatically executed by the computer, and the subtotal of the charged amount is printed every 50 sheets (step S).
5).

On the other hand, in addition to the 50 sheets, the read character portions are converted into image data. Then, the necessary portion is printed on the paper as it is (step S
6). That is, the billing amount and the like required for tabulating the form are printed side by side on paper as image data. This is totaled by the operator in step S7. The totaling by the operator may be actually performed using the 50 sheets themselves. However, for example, 50 sheets may include data related to privacy such as personal data. In such a case, a portion unnecessary for such totaling is excluded, and only a necessary portion is once printed on a sheet and used for totaling. This enables privacy protection.

Further, the work of collecting a large number of forms at hand and compiling them is relatively complicated and error-prone. Therefore, if only necessary portions of the respective forms are edited and printed so as to be easily aggregated as image data, the processing becomes efficient. Due to these various advantages, in this example, the main part of the read image is once printed on paper, and the totalization is performed by the operator. Then, the totalization result and step S5
In step S8, the result of totaling is compared. If the two match, the result of recognizing and processing the characters of the form and the data conversion is the same as the result calculated by the operator by actually seeing the form, and it can be seen that there is no error in the calculation or recognition result. In this case, move to step S9,
The billing amount for 50 sheets is fixed. On the other hand, if the matching results do not match, the operator compares the read and recognized data with the image, checks the recognition error, etc., and corrects the data.

FIG. 3 is a schematic block diagram of the form classification processing system. As shown in the figure, this system
Is received by the hopper 2, the image of the form is read by the main body 3, the characters written on the form are recognized, and the sorting is performed based on the result. For this sorting, a large number of pockets 4 for storing the forms according to classification are provided.
Further, a console 5 is provided for controlling this device, and a printer 6 is provided for printing out necessary data.
Is provided.

Some functional blocks provided in the main body 3 are shown on the lower side of this figure. As shown in the figure, the main body section 3 is provided with a reading section 11 for recognizing characters by reading the image of the form 1, and a character recognizing section 12.
Further, in this system, a bar code conversion unit 13 and a bar code printing unit 14 are provided in order to convert the result of such character recognition into a bar code and print the bar code in the margin of the form 1. With this device, the characters written on the form 1 can be recognized and the form can be sorted into the corresponding pockets 4 according to the recognition result. However, the number of pockets 4 is specifically 36, for example, and when the form 1 is classified into about 300, the form 1 may be repeated several times.
It is necessary to set to this system and repeat classification and sorting.

In such a case, if character recognition or the like is performed each time, the recognition process may take a long time and a recognition error is likely to occur. Therefore, in the present invention, a bar code that can be surely read and is less likely to cause a reading error is used. Therefore, for the second and subsequent classification / sorting processing, the barcode is read by the barcode reading unit 15 shown in the figure and the result is used. It should be noted that since the result of character recognition is used for the first time sorting and the output of the bar code reading unit is used for the second and subsequent times, both output signals are input to the sorting storage control unit 16. .

FIG. 4 shows an external view of the form. As shown in this figure, the form 1 is, for example, handwritten or typed on the front side of the form 1, together with the address, the amount charged, and other items. This is recognized by the character recognition unit 12. This back surface is, for example, blank. In this case, the information necessary for repeatedly classifying the form is once bar-coded and printed on the back side. Two barcodes 1 shown
In 8, the prefecture code representing the sender and the address thereof, the recipient, etc. are coded and entered.

In addition to the bar code, the additional information printing unit 17 prints a reference number and a batch serial number so that the operator who handles the form can easily confirm the form. The batch serial number is a number that is assigned to each group when the forms are collected into groups of 50 sheets as described above and are to be aggregated. Note that the reason for setting such a unit of aggregation (which is referred to as a batch) to 50 sheets is that the appropriate amount that can be manually checked is approximately 50 sheets if any mistake is discovered.
Therefore, according to the type, shape or contents of the form,
The number of batches may be arbitrarily selected.

<Operation> FIG. 5 shows a form reading operation flowchart. Here, the specific operation of the system shown in FIG. 1 will be described. First, in the above system, the font information 21 and the format information 22 are downloaded to the storage device in step S1. This is a process for initial setting before the operation of the device is started. Then, in step S2, one form is read and its identification code is read. The format information 22 is registered corresponding to the identification code. Therefore, in step S3, it is determined whether or not the format information corresponding to the identification code is registered. If there is no registration, it is necessary to generate format information, so in step S4, the image data of the form is stored. Further, in step S5, since there is no format information, normal reject processing of the unclassifiable form is performed.

Specifically, the form is conveyed to the reject-use pocket 25 shown in FIG. 1 and necessary recording is performed. Then, the process proceeds to step S8, and it is determined whether all the classification processes have been completed. If the processing has been completed, this operation is exited, but if there are remaining forms, step S2 is executed again.
Then, another form is read. On the other hand, when it is determined in step S3 that the format information is registered, the process proceeds to step S6 and the reject statistical processing according to the present invention is performed. This will be described later in detail with reference to FIG. After such statistical processing is completed, necessary classification processing is performed in step S7. And
Go to step S8.

FIG. 6 shows a reject statistical processing operation flowchart. First, in step S1, it is determined whether the form identification code read last time and the form identification code of this time are the same. If they are not the same, the statistical information is reset in step S2. That is, the statistical information is recorded as a group for a group of forms having the same identification code. Next, in step S3, it is determined whether the form recognition process is possible, that is, whether the form is rejected. If the recognition process is completed, all the following processes are passed, and the process proceeds to the classification process of step S7 shown in FIG.

On the other hand, in the case of the rejected form, in step S4, the reject number is updated by using a counter or the like which counts up the reject number one by one. In step S5, the threshold table corresponding to the form identification code is referred to. That is, it is determined which sender has issued the form, and a threshold value corresponding to the sender is selected. This selection method will be explained later,
For example, in the case where the form of every time about 1000 is issued from the same sender, if the reject number is 10 sheets and 20 sheets, it can be determined that the sender of the entry mistakes, but do not need to register the format information .

On the other hand, if 50 sheets of the form of the sending source, which processes only about 100 sheets each time, are rejected, it can be determined that some format change or the like is performed. In this case, the threshold value is set to, for example, 20 sheets based on the past statistics, and the process proceeds to the determination of the next step S6. That is, it is judged here whether it is the threshold value or more, and if it is the threshold value or more, it is necessary to register the format information, and the process moves to step S7 to store the image data. Next, in step S8, the format information generating unit 28 shown in FIG. 1 is requested to generate the format information. On the other hand, if it is determined that the threshold value is equal to or less than the threshold value, the process proceeds to step S9 and a simple reject process is performed. That is, in this case, a record that this form has been rejected is left, and the form is conveyed to the reject distribution pocket.

FIG. 7 shows an example of a reject log for setting the threshold value described in step S5 of FIG. In this example, every time the form classification process is performed, the recognition statistics unit 23 shown in FIG. 1 records the actual number of read sheets and the number of rejected sheets for each sender. For example, with respect to the sender A, in this case, 5000 sheets are read and 15 sheets are rejected. on the other hand,
For example, looking at the sender C, 40 sheets are rejected for every 2000 sheets read. In this case, send
Fill mistake Tsukemoto A is the cause, the raw format information
No need for composition. On the other hand, the one with the largest number of rejects is sent.
It is said to be due <br/> cause changing fill characters and format filling out such a former C. I slave, there is a format change when it is the number rejected what extent do the threshold is subject to generation of follower over <br/> mat information, the sender of the book <br/> form recognition and Be determined based on unrecognizable data
become. The recognition statistic unit 23 shown in FIG. 1 stores such a reject log in the statistic data storage unit 27 and sets an appropriate threshold value for each form.

In the above examples, all the explanations have been made on the premise that the cause of the reject is the format change of the form. However, for example, even if the character entry areas on the form are the same, there is a case where the character recognition failure rate has increased abnormally because the font of the characters to be entered was changed by the sender . In such a case, if the recognition is performed using an appropriate dictionary corresponding to the character font, the recognition rate is improved. Therefore, it is necessary to re-register the font information instead of the format information. Therefore, as shown in FIG. 1, the format information generation unit 28
Besides, a font information re-registration unit 29 is provided. The format information generation unit 28 or the font information re-registration unit 29 is configured to be activated in consideration of the rejection reason output from the character recognition unit 12 and the past recognition result of the recognition statistics unit 23.

FIG. 8 shows an explanatory diagram of the correspondence according to the reason for rejection. On the left side of FIG. 8, for example, the character recognition unit 12
Listed the reasons for rejects output by etc. The corresponding action is marked on the right. That is, the types of treatment include registration of format information, guidance of the sender, review of recognition logic, and change of registered font information.
That is, in the case of the design change of the form at the sender, the rejection can be eliminated by coping with the registration of the format information. In addition, since the system is not responsible for any entry mistakes made by the sender, the sender is notified of this and the correct entry method is instructed. If the font is suddenly changed at the sender , the registered font information is changed on the system side.

As described above, the form of the sender having an abnormally high character unrecognizable rate is promptly dealt with by automatically generating the format information and re-registering the font information. As a result, it is possible to perform recognition processing and classify the rejected form again. In addition, it is possible to promptly respond without requiring complicated operations such as requesting the original of the form sender and inquiring about the changed contents.

[Brief description of drawings]

FIG. 1 is a block diagram showing a specific example of a system of the present invention.

FIG. 2 is an explanatory diagram of a form process suitable for the system of the present invention.

FIG. 3 is a schematic block diagram of the system of the present invention.

FIG. 4 is an external view explanatory diagram of a form.

FIG. 5 is a flowchart of a form reading operation.

FIG. 6 is a reject statistical processing flowchart.

FIG. 7 is a diagram illustrating an example of a reject log.

FIG. 8 is an explanatory diagram of content of correspondence according to a reason for rejection.

[Explanation of symbols]

1 form 12 Character recognition part 21 Font information 22 Format information 23 Recognition Statistics Department 26 Image data storage 27 Statistical data storage 28 Format information generator 29 Font information re-registration section

Continuation of the front page (56) Reference JP-A-7-114616 (JP, A) JP-A-4-237390 (JP, A) JP-A-7-160807 (JP, A) JP-A-6-162254 (JP , A) JP-A-6-236460 (JP, A) Yoshiaki Yoshida / Hatsukazu Tojo / Shin Sasaki, FAX-OCR system, Oki Electric R & D, Japan, Oki Electric Industry Co., Ltd., April 1, 1993, Vol. 60, No. 2, p. 25-p. 28 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/60 G06F 19/00 G06K 9/00 H04N 1/00

Claims (3)

(57) [Claims] 1. A system for classifying forms to which an identification code is attached for each sender, and format information storage in which format information indicating the format of each form is stored in association with each identification code. Section, and a character recognition section that determines an identification code from the image data of the form, recognizes characters from the image data by capturing the corresponding format information from the format information storage section, and In a form recognition and classification system that rejects forms that are not character recognizable while classifying based on the recognition result, the number of forms normally recognized by the character recognition unit and the number of unrecognizable forms are counted for each identification code, and normal recognition is performed. If the number of unrecognizable forms with respect to the number of forms And a format information generation unit for generating format information from the image data of the unrecognizable form and storing the format information in the format information storage unit when receiving the format generation instruction. Form recognition and classification system. 2. A statistical data storage unit for storing at least a previous count value of the recognition statistic unit, wherein the recognition statistic unit accumulates a present count value and a previous count value or a present count value. The form recognition and classification system according to claim 1, wherein the threshold value is set based on a comparison between the count value and the count value. 3. A font information storage unit storing font information for character recognition by the character recognition unit, and a font information re-registration unit for changing font information in the font information storage unit. When the character recognition unit fails to recognize the image data due to the stored font information, the character recognition unit notifies the recognition statistics unit of the recognition failure, and when the recognition statistics unit receives the notification, the font information is re-registered. 2. The form recognition and classification system according to claim 1, wherein a part is instructed to change the font.