JP6172565B2 - Document processing apparatus and program - Google Patents

Description

  The present invention relates to a document processing apparatus and a program.

  Japanese Patent Laid-Open No. 2004-151867 discloses a development step of expanding a search character string into a corresponding character string image, a filtering step of performing predetermined filtering on the character string image expanded in the expansion step, and a character string filtered in the filtering step Segmentation of images into independent parts, character recognition of each segmented part is performed, and recognition can be performed based on the difference between the recognition character string candidate obtained in the recognition process and the recognition character string candidate obtained in the recognition process A generation process for generating another recognition character string candidate, a predetermined character in a logical sum condition of the other recognition character string candidate that can be combined generated in the generation process and the recognition character string candidate acquired in the recognition process An image search method comprising a character string search step for searching for a column is disclosed.

  Patent Document 2 discloses that an image scanner that inputs image information, OCR software that recognizes an input image, a text information storage unit that stores the recognition result, and that extra characters are inserted into a specified search word. An information processing method and apparatus including document search software that makes assumptions or creates a pattern in which characters are thinned out from a search word and performs a search using the pattern is disclosed.

JP 10-0669494 A Japanese Patent Laid-Open No. 09-016619

  An object of the present invention is to provide a document processing apparatus and program capable of generating a rule capable of sorting documents with high accuracy.

The present invention according to claim 1 is a receiving unit that receives image information of a document, a character information extracting unit that extracts character information including a character string from the image information of the document received by the receiving unit, and the character information extraction Sorting means for sorting the documents received by the receiving means based on the character information extracted by the means, and generating a sorting rule for the sorting means so as to adjust redundancy of character string recognition of the character information extracting means possess a sorting rule generating means, wherein the sorting rule generating means, said character information is set target value of the recognition rate of the character string to be recognized by the extraction means in advance, the character string so that the target value or more A sorting rule with adjusted recognition redundancy is generated, and at least one character constituting the character string extracted by the character information extracting means is set as all target characters. If the recognition rate of the character string extracted by the character information extraction means is predicted not to reach the target value, the characters constituting the character string are matched with all target characters. The document processing apparatus performs processing so that the recognition rate of the character string reaches a target value by gradually replacing the character with the character to be processed .

The present invention according to claim 2, wherein the sorting rule generating means is a document processing apparatus according to claim 1, wherein generating the sorting rule based on the expected recognition rate of each character extracted by the character information extracting section.

According to a third aspect of the present invention, in the document according to the first or second aspect , the sorting rule generating unit generates the sorting rule based on an expected recognition rate for each character size extracted by the character information extracting unit. It is a processing device.

The present invention according to claim 4 is the document processing apparatus according to any one of claims 1 to 3 , wherein the sorting rule generation means generates a new sorting rule when the generated rules overlap.

The present invention according to claim 5, the document processing device according to any one of claims 1 to 4 further comprising a test data generating means for generating test data to sort on the basis of the generated rules by the sorting rule generation means is there.

The present invention according to claim 6 is based on the receiving step for receiving the image information of the document, the character information extracting step for extracting the character information including the character string from the image information of the received document, and the extracted character information. wherein the sorting sorting step documents accepted in the reception step, to execute the rule generating step of generating a sorting rule to adjust the redundancy of the character string recognition, to computer, the sorting rule generation step, the character A target value of the recognition rate of the character string to be recognized in the information extraction step is set in advance, and a sorting rule that adjusts the redundancy of character string recognition so as to be equal to or higher than the target value is generated, and in the character information extraction step Performs processing to replace at least one character composing the extracted character string with a character that matches all target characters When it is predicted that the recognition rate of the character string extracted in the character information extraction step does not reach the target value, the characters constituting the character string are gradually replaced with characters that match all the target characters. This is a program for processing so that the recognition rate of a column reaches a target value .

  According to the first aspect of the present invention, it is possible to provide a document processing apparatus capable of generating a rule that can classify a document with high accuracy as compared with a case without this configuration.

According to the second aspect of the present invention, in addition to the effect obtained by the first aspect of the present invention, the recognition rate can be accurately predicted as compared with the case where the present configuration is not provided.

According to the third aspect of the present invention, in addition to the effect obtained by the first or second aspect of the present invention, it is possible to cope with a case where the character size is different.

According to the present invention of claim 4 , in addition to the effect obtained by the invention of any one of claims 1 to 3, it is possible to cope with the case where the generated sorting rules overlap.

According to the present invention of claim 5 , in addition to the effect obtained by the invention of any one of claims 1 to 4, it can be confirmed whether or not the result predicted by the generated rule is obtained.

According to the sixth aspect of the present invention, it is possible to provide a program that can generate a rule that can classify a document with high accuracy as compared with the case without this configuration.

1 is a schematic diagram illustrating a configuration of a document processing system according to an embodiment of the present invention. It is a block diagram which shows the hardware of the scanner apparatus used for embodiment of this invention. It is a mimetic diagram showing a form classification system concerning an embodiment of the present invention. It is a block diagram which shows the software structure which concerns on embodiment of this invention. It is a top view which shows an example of the form used for embodiment of this invention. It is a schematic diagram which shows the calculation method which calculates the probability of the 2nd example which produces | generates the sorting rule in embodiment of this invention. It is a flowchart for implementing the algorithm of the 1st example which produces | generates the sorting rule in embodiment of this invention. It is a flowchart for implementing the algorithm of the 2nd example which produces | generates the sorting rule in embodiment of this invention. It is a flowchart which shows a part of processing flow in embodiment of this invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a schematic diagram showing an example of the configuration of a document processing system 2 according to an embodiment of the present invention. The image forming system 2 of the present invention includes terminal devices 4-1 to 4-n (n is an integer of 1 or more) configured by a personal computer or the like, and these terminal devices 4-1 to 4-n via a network. The server device 6 connected in this manner, the image forming device 8 connected to the server device 6 via a network, and the scanner device 10 connected to the server device 6 via the network.

  The image forming apparatus 8 includes a scanner unit 8a, and a paper document is read by the scanner unit 8a and converted into an electronic document.

  As shown in FIG. 2, the scanner device 10 transmits and receives data to and from an external device such as a server device 6 and a storage device 16 such as a CPU 12, a memory 14, and a hard disk drive (HDD), and a network. The communication device 18, the scan drive unit 20, and the operation unit 22 are connected through a bus.

  The CPU 12 executes processing based on a program stored in the memory 14 or the storage device 16 and controls the operation of the scanning device 10.

  FIG. 3 shows a form classification system 24 as an example of a document processing apparatus. The form classification system 24 reads a form 26, which is a paper document, by the scanner unit 10 or the scanner unit 8a of the image forming apparatus 8 described above. The read document is recognized by OCR processing or the like and sorted as an electronic document 28. Here, predetermined forms such as “payment receipt”, “payment settlement”, and “force majeure consultation” are classified and stored in each folder 30 prepared in advance in the server device 6 described above, for example.

  FIG. 4 shows the configuration of a document processing program in the scanner device 10. The receiving unit 32 receives an image read by the scanner device 10 as, for example, bitmap data. The document information extraction unit 34 is composed of, for example, OCR software, and extracts character information including a character string from an image received by the reception unit 32 by a predetermined method. For example, as shown in FIG. 5, all the forms handled here have a title, for example, “payment receipt” written at a predetermined position, and a character string at this position is extracted. The character string to be extracted is not limited to being in a specific position, but it is only necessary to capture the attributes of the constituent characters such as color and size.

  The sorting unit 36 sorts (classifies) the form based on the character information extracted by the document information extracting unit 34. Rules for sorting forms in the sorting unit 36 are generated by a sorting rule generating unit 38. The form sorted by the sorting unit 36 is transmitted to the server device 6 as described above, and is filed for each form in a file in the server device 6.

  The test data generation unit 40 generates test data for verifying how correctly the forms to be sorted are sorted. From the test data generated by the test data generation unit 40, the character information is extracted by the character information extraction unit 34, and the data sorted by the sorting unit 36 based on the sorting rules generated by the sorting rule generation unit 38. Are sent to the terminal devices 4-1 to 4-n and the image forming apparatus 8 described above, and the sorting result is output.

  Next, an algorithm for generating a sorting rule in the sorting rule generating unit 38 will be described.

  In this embodiment, it is assumed that the character recognition rate by the character information extraction unit 34 is different for each character. First, the expected recognition rate of the character which comprises the character string which the character information extraction part 34 should extract is set.

As the expected recognition rate, for example, a test chart including various characters is read, and a correct answer rate when the read characters are recognized is obtained in advance. For example, “support” is 98%, “salary” is 90%, “receive” is 85%, “territory” is 80%, and “book” is 95%.
Since the character recognition rate varies depending on the size of the character, when characters having different sizes are used, an expected recognition rate is obtained for each size of each character.

Here, the sorting rules are expressed as follows.
Document title = “Receipt receipt”
This means that all characters must match. Therefore, the expected recognition rate for the character string “payment receipt” is as follows.
Payment * Receiving * Receiving * Area * Book = 98% * 90% * 85% * 80% * 95% = 57%

  As it is, the expected recognition rate is low, so it needs to be increased. To increase the expected recognition rate, the redundancy should be increased. A target recognition rate is set, and the redundancy is adjusted so as to exceed the target recognition rate. For example, in the “payment receipt”, the target recognition rate is set to 90%.

  A first example of increasing redundancy is a method of increasing wildcards. A wild card character is a character that matches all target characters. The wildcard character is represented as “?”. Replace characters with low expected recognition rate with wildcard characters.

In the “payment receipt”, the expected recognition rate of “territory” is the lowest, so “territory” is replaced with a wild card character, and a sorting rule is generated as follows.
Document title = "Supply receipt?"
The expected recognition rate as a character string in this case is as follows.
Payment * salary * receipt *? * Letters = 98% * 90% * 85% * 100% * 95% = 71% (<90%)

Next, since the “estimated” has the lowest expected recognition rate, “accepted” is replaced with a wildcard character, and a sorting rule is generated as follows.
Document title = "Supply ??"
The expected recognition rate as a character string in this case is as follows.
Supply*? *? * Letters = 98% * 90% * 100% * 100% * 95% = 83% (<90%)

Next, since “Sal” has the lowest expected recognition rate, “Salary” is replaced with a wildcard character, and a sorting rule is generated as follows.
Document title = “support”
The expected recognition rate as a character string in this case is as follows.
Support * *? *? * Letter = 98% * 100% * 100% * 100% * 95% = 93% (≥ 90%)
Thus, when the target is not satisfied, the expected recognition rate of 90% can be exceeded by repeatedly increasing the wild card characters.

However, in this first example, there is a possibility that it will overlap with other sorting rules. For example, even in the “payment checkout”, there is a possibility that the document title = “support”.
Also, this method cannot be used when the expected recognition rate of all characters is lower than the target recognition rate.

Next, a second example for increasing the redundancy will be described.
This second example is a method of creating a sorting rule in which one character with a low expected recognition rate is a wild card character and increasing it by logical sum.
In the above example, the first sorting rule created is as follows.
Since the expected recognition rate does not exceed 90% in the document title = “payment receipt?”, The next generated sorting rule is
Document title = "Supply receipt?" + "Supply? Receipt"
It becomes.

The expected recognition rate as a character string in this case is obtained as follows.
Payment * salary * receipt *? * Letters = 98% * 90% * 85% * 100% * 95% = 71%
(<90%) ・ ・ ・ A
Supply*? * Territory * book = 98% * 90% * 100% * 80% * 95% = 67%
(<90%) ・ ・ ・ B
1- (1-A) * (1-B) = 90% (≧ 90%)

    That is, as shown in FIG. 6A, in order to obtain the logical sum probability of the set A and the set B, first, as shown in FIG. The probability (1-B) that is not the set B is obtained. The probability that it is not set A is 29%. Further, the probability that the set is not set B is 33%. Next, as shown in FIG. 6C, a probability (1-A) * (1-B) that is neither set A nor set B is obtained. The probability that it is neither set A nor set B is 10%. Further, the probability of not being set A and not set B, that is, the probability of being set A or set B is 1- (1-A) * (1-B), which is 90%.

    As described above, if the predicted recognition rate does not meet the target even if the lowest character is replaced with a wild card character, the next character with the lowest expected recognition rate is changed to one wild card character, and the logic of the two rules. Find the expected recognition rate when taking the sum. If the goal is not met, repeat until the goal is met. If the expected recognition rate does not meet the target even if the wildcard character is used for each character and the logical sum is taken, two wildcards are used.

FIG. 7 is a flowchart for implementing the algorithm of the first example described above.
First, in step S10, an expected recognition rate of the sorting rule is calculated. In the above example, payment * supply * receipt * area * form = 98% * 90% * 85% * 80% * 95% = 57%.

  When the predicted recognition rate does not exceed the target, n = 1 is set in the next step S12. n is the number of characters in the character string to be recognized. In the next step S14, a sorting rule is generated in which n characters are replaced with wild card characters from characters with a low expected recognition rate. In the above example, the document title = “payment receipt?”.

  In the next step S16, the expected recognition rate of the sorting rule generated in step S14 is calculated. In the above example, payment * payment * receipt *? * Letter = 98% * 90% * 85% * 100% * 95% = 71%.

  In the next step S18, it is determined whether or not the predicted recognition rate calculated in step S16 is equal to or higher than the target recognition rate. If it is determined in step S18 that the predicted recognition rate is greater than or equal to the target recognition rate, the process ends. If the predicted recognition rate is less than the target recognition rate in step S18, the process proceeds to the next step S20. In step S20, n = n + 1. In the above example, since it is less than 90% of the target recognition rate, n = 2 is set in step S20.

  In the next step S22, it is determined whether n exceeds the character string length (5 in the above example). If n exceeds the character string length, no more sorting rules can be generated and the process ends. On the other hand, if it is determined in step S22 that n is less than or equal to the character string length, the process returns to step S14. In the above-described example, the target recognition rate is exceeded when the document title where n = 3 = “support” is not reached.

FIG. 7 is a flowchart for implementing the above-described second example algorithm.
First, in step S30, the expected recognition rate of the sorting rule is calculated. In the above example, payment * supply * receipt * area * form = 98% * 90% * 85% * 80% * 95% = 57%.

  If the predicted recognition rate does not exceed the target, m = 1 and n = 1 are set in the next step S32. m is the order in which the expected recognition rate is low in the character string to be recognized, and n is the number of characters in the character string to be recognized.

  In the next step S34, a sorting rule is generated by replacing n characters with wildcard characters from the mth character with the lowest expected recognition rate. In the above example, the document title = “payment receipt?”.

  In the next step S36, a sorting rule obtained by ORing the created sorting rules is generated. If it has not been created, the sorting rule generated in step S34 is adopted as it is.

  In the next step S38, the expected recognition rate of the sorting rule generated in step S36 is calculated. In the above example, payment * payment * receipt *? * Letter = 98% * 90% * 85% * 100% * 95% = 71%.

  In the next step S40, it is determined whether or not the predicted recognition rate calculated in step S38 is equal to or higher than the target recognition rate. If it is determined in step S40 that the expected recognition rate is equal to or higher than the target recognition rate, the process ends. If the predicted recognition rate is less than the target recognition rate in step S40, the process proceeds to the next step S42. In step S42, m = m + 1. In the above example, since it is less than 90% of the target recognition rate, m = 2 is set in step S42.

  In the next step S44, it is determined whether m exceeds the character string length (5 in the above example). If m exceeds the character string length, the process proceeds to the next step S46, where m = 1 and n = n + 1, and the number of characters to be replaced with wildcard characters is increased by one. On the other hand, if it is determined in step S44 that m is less than or equal to the character string length, the process returns to step S34.

  When the process returns to step S34, a sorting rule is generated by replacing one character with a wild card character from the character with the second lowest expected recognition rate. In the above example, the document title = “Supply?

  In the next step S36, a sorting rule obtained by ORing the created sorting rules is generated. In the above example, the document title = “payment receipt?” + “Payment?

  In the next step S38, the expected recognition rate of the sorting rule generated in step S36 is calculated. The expected recognition rate is 90%, and the processing ends when the target recognition rate is exceeded.

  In the first example described above, for example, the document title = “support” form, and there is a possibility that the sorting rules of “payment receipt” and “payment settlement” overlap. Therefore, as shown in FIG. 9, in step S50, it is determined whether there is an overlapping sorting rule. In step S50, if there is no overlapping sorting rule, the process ends. If there is an overlapping sorting rule, a sorting rule is generated with a new algorithm. For example, a new sorting rule is generated by changing from the algorithm of the first example to the algorithm of the second example.

  The method for increasing the redundancy is not limited to the first example and the second example described above. For example, a character similar to a character with a low expected recognition rate may be included as an OR condition.

2 Document Processing System 8 Image Forming Apparatus 10 Scanner Device 24 Form Classification System 26 Paper Document 28 Electronic Document 32 Reception Unit 34 Character Information Extraction Unit 36 Sorting Unit 38 Sorting Rule Generation Unit 40 Test Data Generation Unit

Claims (6)

Receiving means for receiving image information of the document;
Character information extracting means for extracting character information including a character string from image information of the document received by the receiving means;
Sorting means for sorting the documents received by the receiving means based on the character information extracted by the character information extracting means;
A sorting rule generating means for generating a sorting rule of the sorting means so as to adjust the redundancy of character string recognition of the character information extracting means;
I have a,
The sorting rule generation means includes:
A target value of the recognition rate of the character string to be recognized by the character information extracting means is preset, and a sorting rule that adjusts the redundancy of character string recognition to be equal to or higher than the target value is generated,
A process of replacing at least one character constituting the character string extracted by the character information extracting means with a character that matches all target characters;
When it is predicted that the recognition rate of the character string extracted by the character information extraction means does not reach the target value, the characters constituting the character string are gradually replaced with characters that match all the target characters. A document processing apparatus that performs processing so that the recognition rate of a column reaches a target value . The sorting rule generation means, the character information extracting section based on the expected recognition rate for each character to be extracted by generating a sorting rule claim 1 document processing apparatus according. The sorting rule generation means, the document processing apparatus according to claim 1 or 2 wherein generating the sorting rules based on the expected recognition rate for each size of the characters extracted by the character information extracting section. The sorting rule generation means, if the generated rule is duplicated document processing device according to any one of claims 1 to 3 to produce a new sorting rules. The document processing device according to any one of claims 1 to 4 further comprising a test data generating means for generating test data to sort on the basis of the generated rules by the sorting rule generating means. A reception step for receiving image information of the document;
A character information extraction step of extracting character information including a character string from image information of the accepted document;
A sorting step of sorting the documents received in the receiving step based on the extracted character information;
A rule generation step for generating a sorting rule to adjust redundancy of character string recognition;
To the computer ,
The sorting rule generation step includes:
A target value of the recognition rate of the character string to be recognized in the character information extraction step is set in advance, and a sorting rule that adjusts the redundancy of character string recognition to be equal to or higher than the target value is generated,
A process of replacing at least one character constituting the character string extracted in the character information extraction step with a character suitable for all target characters;
When it is predicted that the recognition rate of the character string extracted in the character information extraction step does not reach the target value, the characters constituting the character string are gradually replaced with characters that match all the target characters. A program that processes the recognition rate of a column to reach a target value .

Images