JPH1063773A - Recognition result editing method, pattern recognition system, and information recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently perform edition such as the confirmation, the correction or the like of a recognition result even if the amount of documents in large. SOLUTION: In the case where the editing processing function of a first editing processing part 31 is a function for performing matching check on the recognition result with its original documents over the whole sentences, a second editing processing part 32 does not check the recognition result over the whole sentences (i.e., in this case, the editing processing result) but picks out the recognition result (editing processing result) at the prescribed sampling rate from the editing processing result over the whole sentences, and the editing processing such as confirmation, correction or the like is performed on only the recognition result (editing processing result) picked out.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a pattern recognition process.
The present invention relates to a recognition result editing method for editing a recognition result obtained by performing (for example, character recognition processing), a pattern recognition system, and an information recording medium.

[0002]

2. Description of the Related Art In general, in a character recognition system, a document such as a manuscript is read by a scanner or the like to form a document image, and then a character image is cut out from the document image to perform a character recognition process to obtain a recognition result. ing. In such a character recognition system, the character recognition process itself does not take much time, but it takes considerable time to confirm and correct the recognition result. In other words, confirmation of the recognition result, editing such as correction, etc., are performed by matching the original document (or document image) with the recognition result over the entire text (all pages, all lines), and usually, the accuracy is corrected. In order to achieve this, the matching check of such a whole sentence was repeated twice.

[0003]

As described above, conventionally, when editing such as confirmation and correction of a recognition result, the recognition result and a document (or a document image) based on the recognition result are conventionally used.
The matching check over the whole sentence was repeated twice, so it took a considerable amount of time to confirm the recognition result and edit the correction, etc., especially when checking the recognition result of a large number of documents and editing the correction etc. In addition, there is a problem that the working efficiency is significantly reduced.

[0004] The present invention provides a recognition result editing method, a pattern recognition system, and an information recording medium capable of efficiently performing confirmation such as confirmation and correction of a recognition result even when a large number of documents are present. It is intended to be.

[0005]

In order to achieve the above object, according to the present invention, there is provided a recognition result editing apparatus for performing an editing process on a recognition result obtained by performing a pattern recognition process on a predetermined document image. In the method, the recognition result is subjected to at least two editing processes, and in each editing process, the recognition result sampling rates are made different from each other, and in each editing process, the recognition result sampling is performed at a predetermined sampling rate. When performed, only the extracted recognition result is edited.

According to a second aspect of the present invention, in the recognition result editing method of the first aspect, when the certainty factor of the recognition result is calculated together with the recognition result in the pattern recognition process.
The third editing process is characterized in that the recognition result with the certainty factor equal to or higher than the predetermined threshold value is skipped, and only the recognition result with the low certainty factor is corrected.

According to a third aspect of the present invention, in the recognition result editing method of the first aspect, in each editing process, a correction history before and after editing is managed, and a character having a high frequency of correction is managed by the correction history. It is characterized by outputting information.

According to a fourth aspect of the present invention, in the recognition result editing method according to the first aspect, a ratio of correction performed in the second editing process is output as a correction ratio.

According to a fifth aspect of the present invention, there is provided a dictionary for recognition, recognition processing means for performing pattern recognition processing on a document image using the dictionary, and editing processing for a recognition result by the recognition processing means. Editing processing means for performing at least two times of editing processing on the recognition result. At this time, in each editing processing, the recognition result sampling rates are made different from each other, and in each editing processing, When a recognition result is extracted at a predetermined extraction rate, an editing process is performed only on the extracted recognition result.

According to a sixth aspect of the present invention, in the pattern recognition system of the fifth aspect, the sampling rate is set to be updatable.

According to a seventh aspect of the present invention, in the pattern recognition system according to the fifth aspect, when the recognition processing means calculates the reliability of the recognition result together with the recognition result, the editing processing means performs the first editing. The processing is characterized in that the recognition result with a certainty factor equal to or higher than a predetermined threshold value is skipped, and only the recognition result with a low certainty factor is modified.

According to an eighth aspect of the present invention, in the pattern recognition system according to the seventh aspect, the editing processing means comprises:
Furthermore, it has a function of outputting the rate of correction made in the second editing process as a correction rate, and the threshold of certainty can be updated according to the correction rate output from the editing processing means. It is characterized by having.

According to a ninth aspect of the present invention, in the pattern recognition system of the fifth aspect, dictionary management means for managing a dictionary is further provided. When information on a high character is output as modified character information, the dictionary has a function of updating the dictionary based on the modified character information.

The invention described in claim 10 is the same as the claim 9.
The described pattern recognition system is characterized in that the sampling rate is reduced each time a dictionary is updated.

According to an eleventh aspect of the present invention, a recognition result obtained by performing a pattern recognition process on a predetermined document image is subjected to at least two editing processes. In each editing process, when a recognition result is extracted at a predetermined sampling rate, a program for performing an editing process only on the extracted recognition result is recorded. It is characterized by.

According to a twelfth aspect of the present invention, in the pattern recognition processing, the confidence of the recognition result is calculated together with the recognition result. It is characterized in that a program for correction is recorded.

According to the first to twelfth aspects of the present invention, the editing work efficiency can be significantly improved in editing such as confirmation and correction of the recognition result.

In particular, the invention according to claim 1, claim 5, claim 6, or claim 11 provides at least two
In each editing process, the sampling rate of the recognition result is made different from each other, and in each editing process, when the recognition result is sampled at a predetermined sampling rate, the recognition result is extracted. Since the editing process is performed only on the recognition result, editing such as confirmation and correction of the recognition result can be efficiently performed.

According to the second, seventh and twelfth aspects of the present invention, when the certainty factor of the recognition result is calculated together with the recognition result in the pattern recognition process, the first editing process requires a certain threshold or more. Is skipped, and only the recognition result with low confidence is corrected. This eliminates the need for the user to look through the entire document at the time of confirming or correcting the recognition result, and can increase the work efficiency.

Further, the invention according to claim 3 or claim 9 is as follows:
When information on a character having a high frequency of correction is output as corrected character information in the editing processing means, the recognition rate can be improved by updating the dictionary based on the corrected character information.

The invention according to claim 4 or claim 8 is
By outputting the rate of correction in the second editing process as a correction rate, and updating the certainty threshold in accordance with the correction rate, the recognition rate can be improved.

Further, the invention according to claim 10 reduces the sampling rate each time the dictionary is updated. This makes it possible to gradually increase the work efficiency each time it is used.

[0023]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a configuration example of a pattern recognition system according to the present invention. In the example of FIG. 1,
The pattern recognition system is configured as a character recognition system that recognizes characters in a document.

Referring to FIG. 1, the character recognition system includes an image input unit 1 for inputting a document such as a document as a document image, a document image storage unit 2 for storing the document image read by the image input unit 1, For example, a character image to be recognized is cut out from a recognition dictionary 5 in which a standard pattern and a standard feature amount of each character to be recognized are registered and a document image input from the image input unit 1. A recognition processing unit 4 for performing a character recognition process of extracting the characteristics of the character image and comparing it with the dictionary 5; a recognition result storage unit 6 for storing the recognition result from the recognition processing unit 4;
An edit processing unit 8 for editing such as correction, and an edit processing unit 8
Edit processing result storage unit 9 for storing the edit processing result from
A dictionary management unit 10 that manages updating of the dictionary 5 and the like;
A condition setting unit 40 for setting conditions such as an editing process for the recognition result.

FIG. 2 is a diagram showing an example of a hardware configuration of the character recognition system of FIG. Referring to FIG. 2, this character recognition system is realized by, for example, a personal computer or the like, and controls a CPU 11 that controls the entire system, a ROM 12 that stores a control program of the CPU 11, and the like.
RAM 13 used as a work area of U11
And a scanner 14 for reading a document such as a manuscript as a document image
A document image file 15 in which a document image read by the scanner 14 (or a character image cut out from the document image for recognition) is compressed and stored, for example, in page units, and a dictionary for character recognition is stored. A dictionary file 16, a text file 17 in which recognition results obtained by performing a character recognition process on a document image (character image) are stored as text (coded information), and a recognition result editing process. Editing result file 4 in which results are stored
1, a temporary file 42 used as a temporary storage area at the time of editing processing, and used to set conditions for editing processing for the recognition result, and to perform editing such as confirmation and correction on the recognition result. It has a display device 18 and an input device 19.

Here, the scanner 14, the document image file 15, the dictionary 16, the text file 17, and the editing result file 41 are the image input unit 1, the document image storage unit 2, the dictionary 5, and the recognition result storage unit 6 of FIG. , Edit processing result storage unit 9
Respectively. The CPU 11 has the functions of the recognition processing unit 4 and the editing processing unit 8 in FIG. Further, as will be described later, the CPU 11 stores the dictionary management unit 1 of FIG.
It can also have a function of zero.

The display device 18 and the input device 1 shown in FIG.
9, the CPU 11, the RAM 13, and the like also have a function as the condition setting unit 40 in FIG. That is, conditions such as editing processing are input from the input device 19 or the like, and the CPU 1
1 can be set in the RAM 13.

The functions of the recognition processing unit, the editing processing unit, the dictionary management unit, the condition setting unit, and the like in the CPU 11 are, for example, a software package (specifically, an information recording medium such as a CD-ROM). 2), and therefore, in the example of FIG. 2, when the information recording medium 20 is set, a medium driving device 21 that drives the information recording medium 20 is provided.

For the input device 19, for example, a keyboard, a mouse or the like can be used.
A display such as a CRT or LCD can be used. In this case, an icon or the like displayed on the screen of the display 18 is pointed by a mouse to select a process, to start or end a predetermined process, or to move a cursor on the screen by a keyboard or a mouse. Or scrolling can be performed.

In other words, the pattern recognition system (character recognition system) of the present invention causes a general-purpose computer system having an image scanner, a display, and the like to read a program code recorded on an information recording medium such as a CD-ROM. The present invention can also be implemented in an apparatus configuration in which a microprocessor of this general-purpose computer system executes pattern recognition processing (character recognition processing). In this case, the information storage medium for storing the pattern recognition processing program, the editing processing program, the dictionary management program, and the like according to the present invention is not limited to a CD-ROM, but may be an RO.
M, RAM, FD, etc. may be used. The input of the document image may be given as an image file without using a scanner.

In the character recognition system shown in FIGS. 1 and 2, the recognition processing section 4 may obtain the recognition result and also obtain the certainty (reliability) of the recognition result. The process of calculating the certainty factor can be performed by, for example, a method disclosed in Japanese Patent Application Laid-Open No. 4-212883.

That is, the degree of certainty indicates how likely the character of the final recognition result is to be.
It can be represented by a numerical value of 00%, or can be represented by quantizing the numerical value in several steps. For example, it can be represented by the following three ranks A, B, and C. A rank: The recognition result is correct. B rank: The recognition result is suspicious (I do not know whether it is correct or wrong). C rank: The recognition result is wrong.

In this case, the recognition processing section 4 comprehensively determines a certainty factor based on information obtained by a plurality of stages of processing for obtaining a final recognition result. For example, from the pattern matching processing, the evaluation value of the first candidate or the difference between the evaluation values of the first candidate and the second candidate, the path selection processing, the evaluation value at the time of determining the path, and the rule processing, which rule is applied and modified. Gathering information indicating whether or not it was done, and information indicating the result of language correction from language processing, and using this information as evidence, for example, Dempster & Shafer
Comprehensively evaluates the confidence using the probability theory of

The determination of such a certainty factor may be performed in the final processing stage by collecting information obtained in the previous processing stages and collectively, or may be performed based on the information obtained in each processing stage. The operation of obtaining the certainty factor and updating the certainty factor obtained by the pre-processing stage by this operation is repeated until the final processing stage.

As described above, when certainty is required in the recognition processing unit 4, as described later, the editing processing unit 8
In editing processing such as confirmation and correction of recognition results in
By changing visual conditions such as color and brightness when displaying the recognition result or displaying characters or symbols in accordance with the recognition result in accordance with the certainty, the apparatus user can be sure of the recognition result. It is possible to easily recognize the degree, find the character that needs correction quickly and accurately, and perform the correction work efficiently.

The editing unit 8 performs a first editing process on the recognition result from the recognition unit 4 (more specifically, the recognition result stored in the recognition result storage unit 6). And a first edit processing unit 31 that stores the result of the edit processing in a first result storage unit 9a (first result file 41a in FIG. 2) of the edit processing result storage unit 9. The second and subsequent editing processes are performed on the recognition result from the recognition processing unit 4 (more specifically, the result is stored in the first result storage unit 9a (the first result file 41a in FIG. 2). The first edit processing result is further edited), and the second edit stored in the second result storage unit 9b of the edit processing result storage unit 9 (the second result file 41b in FIG. 2). And a processing unit 32.

Here, the first edit processing section 31 and the second
When performing the editing process on the recognition result, the editing processing unit 32 displays, for example, a reference line for displaying the recognition result or the editing process result on the screen of the display device 18 and one side of the reference line. Displays a recognition result or an edit processing result which is a current edit target along the reference line, and displays the current edit target displayed on one side of the reference line on the other side of the reference line. A document image (character image) corresponding to the recognition result or the edit processing result can be displayed along the reference line.

FIG. 3 shows an example of a document for a certain page. FIGS. 4A and 4B show the results of recognition or editing of the document shown in FIG. The display examples on the display device 18 for performing the operations are shown.

In the display example of FIG. 4A, a vertical reference line L ₁ is displayed on the screen, and along the reference line L ₁ , immediately next to the reference line L ₁ , a recognition result sequence or a recognition result sequence to be edited is displayed. The edit result column “Today is fine weather” is displayed in the row direction (= vertical),
Moreover, its is immediately to the left, the editing subject to recognition result string or editing result document image corresponding to the columns (character image) string "today becomes fine weather," row direction along the longitudinal reference line L ₁ (= Vertical).

In the display example of FIG. 4B, a horizontal reference line H ₁ on the screen is displayed, and a recognition result sequence to be edited is displayed immediately above the reference line H _1. Alternatively, the edit processing result column “Today is fine weather” is displayed in the row direction (= horizontal), and the document image corresponding to the recognition result sequence to be edited or the edit processing result sequence is immediately below. (character image) string "today becomes fine weather," are displayed in the row direction (= horizontal) along the side of the reference line H _1.

Note that, on the display screen, whether the row direction is vertical as shown in FIG. 4A or the row direction is horizontal as shown in FIG. For example,
For example, the condition can be set by designating the row direction as “vertical” or “horizontal” in the condition setting unit 40 (input device 19 in FIG. 2) in FIG.

The first edit processing unit 31 and the second edit processing unit 32 display the recognition result (text) displayed on one side of the reference line on the display screen of the display device 18 and the reference line. It has a function of constantly scrolling or moving the document image (character image) displayed on the other side in conjunction with it. Specifically, for example, in the state of FIG.
Moving one row ₁ of the display has been that the recognition result to the right (the text), for example, in the right direction, as shown in FIG. 5, in conjunction with this, the document that is displayed on the left side of the reference line L ₁ Display control can be performed such that the image (character image) is also moved rightward by one line. Further, for example, in the state of FIG. 4 (a), the right to view has been that the recognition result of the reference line L ₁ is moved two characters (text), for example, in the upward direction, as shown in FIG. 6, in conjunction with this is allowed, the document image (character image) displayed on the left side of the reference line L ₁ also can be displayed control to move two characters upward.

As described above, on the display screen of the display device 18, the recognition result or the edit processing result displayed on one side of the reference line and the document image displayed on the other side of the reference line.
(Character image) and the function of always displaying them (corresponding to the reference line and displaying them in parallel with each other across the reference line). Second
When the editing processing unit 32 has the recognition result or the editing processing result and the corresponding document image (character image)
The comparison with the column is facilitated, and the operator can easily confirm and correct the recognition result or the edit processing result.

When the recognition result or the edit processing result is confirmed or corrected, a character currently being corrected may be displayed on the screen of the display device 18 so as to be identifiable that the character is currently being corrected. it can. Specifically, FIG.
In the example shown in FIG. 7, when "heaven" is erroneously recognized as "husband" and the character currently being corrected in the recognition result or the editing processing result is "husband" as shown in FIG. 7, the recognition result is displayed. In the area, the character of "husband" may be surrounded by, for example, a rectangle, and in the display area of the document image, the corresponding character image of "heaven" may be surrounded by, for example, a red rectangle.

In the display examples shown in FIGS. 4A and 4B, when a document image (character image) is displayed, for example, a document stored in the document image file 15 in a compressed state as described later. It is also possible to enlarge the image (character image) and display it at, for example, the same size as the document image input at the time of the scanner.

In the display examples shown in FIGS. 4A and 4B,
On the screen, only the recognition result (text) and the document image (character image) are displayed, but on this screen, for example, the entire image of the page to be edited is further displayed, and the correction is performed on the page to be edited. The first function to display information such as characters, the number of characters with low confidence, the total number of characters, etc.
It is also possible to provide the editing processing unit 31 and the second editing processing unit 32 of the above.

As described above, the first edit processing unit 31 and the second
With regard to the basic display control function within one display screen as described above, these can be the same.

However, in order to efficiently perform editing such as confirmation and correction of the recognition result, in the present invention, the first editing processing unit 31 and the second editing processing unit 32 have different editing processing functions. I have.

That is, when the editing processing function of the first editing processing section 31 and the editing processing function of the second editing processing section 32 are to be the same, the editing processing is repeated twice or more in the same manner. If the editing process is to perform a full-text matching check between the recognition result and the source document, as described above,
Work efficiency is significantly reduced.

In the present invention, in order to solve such a problem in the related art, the editing processing function of the first editing processing unit 31 performs, for example, a matching check over the entire text between the recognition result and the original document. If it is, the second editing unit 32 does not check the recognition result over the entire sentence (that is, in this case, the editing result), but instead of checking the recognition result over the whole sentence, at a predetermined sampling rate, A recognition result (edit processing result) is extracted, and editing processing such as confirmation and correction is performed only on the extracted recognition result (edit processing result).

More specifically, for example, n pages of document images of a document such as a meeting minutes document are stored in the document image storage unit 2 in units of pages, and the recognition processing unit 4 stores the page images for the n pages of document images. When the recognition processing is performed for each unit and the recognition results for n pages are stored in the recognition result storage unit 6 (text file 17), the first editing processing unit 31 stores the recognition results for n pages in all pages, for example. (n pages), while the second edit processing unit 32 functions to perform edit processing (confirmation and correction) in page units, instead of performing edit processing for all pages, for example, for all pages (n pages) ) Minutes recognition result
From the (editing result), a predetermined extraction rate w (0 <w <1)
Has a function to extract recognition results (editing results) for n × w pages in page units, and to edit only the extracted recognition results (editing results) for n × w pages in page units .

Here, as a method for extracting the recognition result (editing result) for n × w pages at a predetermined sampling rate w from the recognition result (editing result) for n pages, for example, 1
Generate n × w random numbers in the range from to
With this random number, n × w pages can be randomly extracted (extracted) from n pages.

When the recognition result (editing result) for n × w pages is extracted from the recognition result (editing result) for n pages in the manner described above, the extracted n
Recognition results (editing process results) for xw pages are temporarily stored in a temporary file 42 before the editing process is performed. In the editing process,
When the display control processing as shown in FIGS. 4A and 4B is performed, the original document image corresponding to the recognition result (edit processing result) for n × w pages is also temporarily stored in the temporary file 42. It has become.

The sampling rate w can be arbitrarily set by the user, for example, by the condition setting section 40 (the display device 18 and the input device 19), and the sampling rate w is updated. (Change) is possible. The sampling rate w is updated once it is set.
Until (change) is made, for example, the current extraction rate is held in the RAM 13.

Specifically, the setting of the sampling rate w can be performed in the following manner. First, in a state where the initial screen as shown in FIG. 8 is displayed, when the user points at the icon (%) of “sampling rate setting”, a sampling rate setting screen as shown in FIG. 9 is displayed. Here, the current sampling rate w is displayed in the sampling rate setting window of the sampling rate setting screen.
(%) Is displayed. In the initial state of this system (immediately after setup), a default value “100” is set as the sampling rate w, and therefore, “100” is displayed in the sampling display window at this stage. When the user wants to set (change) this to a desired sampling rate, the user inputs, for example, a numerical value (%) of the desired sampling rate, that is, a numerical value in the range of 1 to 100 from the input device 19.
For example, if the total number of pages of the conference copy manuscript to be edited is 10
When there are 0 pages and 20 pages are to be extracted, the extraction rate w is 20/100 = 20 (%).
Therefore, the user inputs "20".

Thus, the numerical value of the sampling rate setting window is
It changes from “100” to “20”. The sampling rate w is 20
If (%) is sufficient, the user confirms that the numerical value of the sampling rate setting window is “20”, and points to the icon (OK) on the sampling rate setting screen. As a result, the sampling rate w is determined to be 20 (%), and is held as a parameter in, for example, the RAM 13 unless it is updated (changed). When the user wants to cancel this setting, for example, to change the numerical value from “20”, the user points to the icon (cancel) on the sampling rate setting screen. As a result, the screen returns to the initial screen, and performs the above operation again to display the sampling rate setting screen,
Enter the value of the desired extraction rate.

In this manner, setting, updating (changing), and the like of the sampling rate w can be performed.

FIG. 10 is a flowchart for explaining a processing operation example and an operation example of the first editing processing section 31. Referring to FIG. 10, the first edit processing unit 31
By displaying a screen as shown in (a) and (b), the user can be made to perform editing processing such as confirmation and correction in page units. That is, the first editing processing unit 31 first determines whether the end button, the next page button, or the confirm button has been selected (steps S1 and S1).
2, S3).

As a result, when the next page button is selected, the recognition result and the document for n pages stored in the recognition result storage unit 6 (text file 17) and the document image storage unit 2 (document image file 15) are displayed. The first page of the unprocessed image is displayed on the screen as shown in FIGS. 4A and 4B (step S4). Thus, the user can confirm and correct the recognition result of this page as described above (step S5). When the editing process for one page is completed in this way, the process returns to step S1, and the same process as described above is repeated. That is, in this repetition processing, when the user sequentially selects the next page button (step S2), the recognition result storage unit 6 (text file 17) and the document image storage unit 2
The recognition results and document images for n pages stored in the (document image file 15) are sequentially displayed on the screen in page units, and the user sequentially confirms and corrects the recognition results for n pages in page units. be able to.

In such a series of processing, when the user selects the confirm button (step S3), the first editing processing section 31 checks whether all the recognition results for n pages have been edited (confirmation, correction, etc.). Is checked (step S6).
As a result, edit (confirmation, correction, etc.) the recognition results for n pages
Are performed, the editing of the recognition result (editing for n pages) is determined, and the result of the editing process is stored in the second result storage unit 9b (the second result file 41b).
(Step S7).

In the above series of processing, when the user selects the end button (step S1), the execution of the editing processing (program) in the first editing processing section 31 ends.

In the above-described processing example, only the processing of sequentially selecting the next page by the next page button is shown. Can also be added.

FIGS. 11 and 12 are flow charts for explaining an example of processing operation and an example of operation of the second editing processing section 32. Referring to FIGS. 11 and 12, for example, FIG.
Is displayed (step S11), it is determined whether or not the desired number of pages have been extracted and stored in the temporary file 42 (step S12).

If the desired number of pages have not been extracted yet, the pages corresponding to the desired number of pages are extracted from the total pages, and the processing until these are stored in the temporary file 42 (steps S13 to S18). Perform

That is, it is determined whether or not the sampling rate w is set to a desired value (step S13). When the sampling rate w is not set to the desired value, the user sets the sampling rate setting screen as described above. It is displayed and a desired extraction rate w is set (step S14). In this way, for example, when the desired extraction rate w is set in the RAM 13, the second editing processing unit 32 sets the random number value at this extraction rate w within the range of the total number of pages n. A page number for n × w (%) is randomly extracted from the total page number n by using the generated random number (step S15). Next, the recognition result of each page corresponding to the page number of the extracted n × w (%) pages
(Editing result) and the document image are stored in the first result storage unit 9a.
(First result file 41a) and document image storage unit 2
(Document image file 15) and store them in the temporary file 42 (step S1).
7).

After the recognition results (editing process results) and document images for the n × w (%) pages extracted in this way are stored in the temporary file 42, the second editing unit 3
2 displays a list of information on the extracted pages (for example, page numbers for n × w pages) on the screen (step S18).

At this stage, the second editing processing unit 32 displays a screen as shown in FIGS. 4A and 4B and allows the user to perform editing processing such as confirmation and correction in page units. Can be done. That is, the second editing unit 32 determines whether the end button, the extraction button, or the confirmation button has been selected (steps S19, S20, S20).
21).

As a result, when the extraction button is selected, among the recognition results (editing results) and document images of n × w (%) pages extracted and stored in the temporary file 42, For example, the top page of
The screen is displayed as shown in (a) and (b) (step S2
2). Thus, the user can confirm and correct the recognition result (editing result) of this page as described above (step S23). When the editing process for one page is completed in this way, the process returns to step S12, and the same process as described above is repeated. That is,
In this repetition processing, when the user sequentially selects the extraction button (step S20), the recognition results (editing processing results) and document images for the n × w (%) pages extracted and stored in the temporary file 42 are displayed. Are sequentially displayed on the screen in page units, and the user can sequentially confirm and correct the recognition results for n × w (%) pages in page units.

In such a series of processing, when the user selects the confirm button (step S21), the second editing processing unit 32 edits the recognition result (editing processing result) for n × w (%) pages. It is checked whether all (confirmation, correction, etc.) have been performed (step S24). As a result, when all the edits (confirmation, correction, etc.) of the recognition results (edit processing results) for n × w (%) pages are performed, the sampling edit (n × w (%)) for the recognition results (edit processing results) is performed. Confirm (Edit for page),
The result of this editing process is stored in the second result storage unit 9b (second result file 41b) (step S25).

In the above series of processes, when the user selects the end button (step S19), the execution of the sampling and editing process (program) in the second editing unit 32 is ended.

When the recognition processing unit 4 obtains a certainty factor together with the recognition result, the first editing processing unit 31 and the second editing processing unit 32 check and correct the recognition result or the editing processing result. The low confidence character in the recognition result or the edit processing result can be displayed so as to be distinguishable from other characters. For example, in the display area of the recognition result or the editing processing result, the low confidence character in the recognition result or the editing processing result is reversely displayed, and in the display area of the document image (character image), the low confidence character is reversely displayed. You can also.

When the confidence is calculated together with the recognition result, the first edit processing unit 31 and the second
The editing processing unit 32 of the present embodiment displays the recognition result or the editing process result and / or the document image (or the document image) displayed on the screen of the display device 18 when editing such as confirmation or correction of the recognition result or the editing process result. Characters with a certainty factor (for example, 90%) or more in character images) are skipped.
Only characters with low confidence may be displayed in order. In other words, a character whose certainty factor is equal to or less than a predetermined threshold (for example, 90%) is extracted from the recognition result or the editing process result and / or the document image (character image), and only the character is recognized as the recognition result (editing result ) And / or character images can be displayed one after another.

In this case, the certainty threshold is set to be updatable by, for example, the condition setting unit 40 (or the input device 19 in FIG. 2).

In addition, such a skip display of the low certainty character is performed by, for example, as shown in FIGS. 4A and 4B, one side of the reference line is displayed along with the current edit target along the reference line. Is displayed on the other side of the reference line, and the recognition result sequence or the edit processing result column to be currently edited displayed on one side of the reference line is displayed on the other side of the reference line. A corresponding document image (character image) sequence can be displayed on the display screen along the reference line, or a display screen of only the recognition result or the edit processing result, or a display screen of only the document image (character image) can be displayed. You can do it.

As described above, the first editing processing unit 31 and the second editing processing unit 32 skip (skip) characters having a certainty factor equal to or higher than a predetermined threshold value,
If a function of displaying only characters with low confidence is provided, the user does not need to look through the entire document when checking or correcting the recognition result or the editing result, thereby improving work efficiency. .

In the above-described example, the skip display function (skip correction function) based on the certainty factor is provided in both the first edit processing unit 31 and the second edit processing unit 32. Such a skip correction function is provided only to the first edit processing unit 31 that edits all pages, and the second edit processing unit 32 that edits only a predetermined page extracted from all pages is skipped. It is also possible not to have the correction function.

That is, by providing the first edit processing section 31 with the skip correction function, the correction work can be efficiently performed by the skip correction function even when the entire page is targeted. On the other hand, even if this skip correction function is not provided to the second edit processing unit 32, the edit target in the second edit processing unit 32 is the first edit processing unit 3.
1 is the result of the editing process resulting from the skip correction, and the second editing processing unit 32 edits only a part of all pages. Even in the case where the confirmation and the correction are sequentially performed every time, the correction work efficiency does not decrease so much, and the confirmation accuracy and the correction accuracy can be further improved by not performing the skip correction.

According to the user's request, the skip correction function is provided only in the second edit processing section 32 which edits only a predetermined page extracted from all pages. , The first to edit all pages
It is also possible to prevent the edit processing unit 31 from having the skip correction function. In addition, it is also possible that neither the first edit processing unit 31 nor the second edit processing unit 32 has the skip correction function.

Further, in the present invention, the first edit processing section 31 and the second edit processing section 32 create and manage histories (correction histories) before and after editing (confirmation, correction, etc.). Based on such a correction history, information on characters having a high frequency of correction is managed as corrected character information. When the user operates the corrected character notification button in FIG. 8, for example, as shown in FIG. Is displayed on the display device 18 to notify the user of a frequently corrected character.

Specifically, the characters having a high correction frequency include:
In the first edit processing unit 31 and the second edit processing unit 32, for example, a character that has been corrected 20 times or more can be used, and the presence of such a character increases the frequency of correction. The work efficiency has been reduced.

Therefore, by notifying the user of a character having a high frequency of correction as described above, the user activates the dictionary management unit 10 to check the standard pattern and the standard feature amount of the character in the dictionary 5. The character can be corrected so that the recognition processing section 4 can accurately recognize the character.
That is, by notifying the corrected character information (information of frequently corrected characters), the dictionary management unit 10 allows the dictionary 5 for recognition to be notified.
Are updated (learned), and the recognition accuracy of this recognition processing system can be sequentially improved.

In the above-described example, the user is allowed to correct the standard pattern and the standard feature amount of the character in the dictionary 5, but the dictionary management unit 10 automatically updates the dictionary 5. (Learning) is also possible. That is, the modified character information from the first edit processing unit 31 and the second edit processing unit 32 is provided to the dictionary management unit 10, and the first edit processing unit 31 and the second edit processing unit 32, based on the corrected character information from 32, automatically updates a standard pattern or a standard feature amount of a frequently corrected character in a direction in which the correction frequency decreases.
The dictionary 5 can be automatically updated (learned).

Note that such a process of updating the dictionary 5 may be performed, for example, for each page, or at the time when the editing process of the entire n-page document (for example, n-page meeting minutes manuscript) is completed. May be done.

It is considered that the updating (learning) of the dictionary 5 improves the recognition accuracy of the recognition processing unit 4 and decreases the frequency of correction. In addition, the extraction rate w can be reduced.

For example, the sampling rate w in the second editing unit 32 is set to, for example, 100
%, The processing is performed, and when the dictionary 5 is updated, the sampling rate w is changed from 100% to, for example, 80%. When the dictionary 5 is updated, the sampling rate w is set to 80. The sampling rate w can be reduced each time the dictionary 5 is updated, such as changing the setting from% to, for example, 60%. This makes it possible to gradually increase the work efficiency each time it is used.

Further, in the present invention, the second edit processing section 32 outputs the recognition result of the extracted page.
When editing (confirmation of editing process results), modification, etc.
It also has a function of outputting information (for example, notifying the user) of information on how much the correction has been made, that is, information on the correction rate. In the case where the user also has such a function, the user can know the degree of the check correction omission in the first editing process by looking at the information of the correction rate.
That is, in the editing processing in the second editing processing unit 32, if the correction rate is high, it is understood that there are many confirmation correction omissions in the first editing processing. This allows the user to
If skip correction is used in the first editing process,
The threshold of the certainty factor is changed based on the correction rate information,
It is possible to reduce the degree of check and correction leakage in the first editing process. For example, when the correction rate is high, the certainty threshold can be reset higher, while when the correction rate is low, the certainty threshold can be reset lower.

It is also possible to automatically change the certainty threshold value without depending on the user. That is, the threshold of the certainty factor can be automatically updated according to the output of the correction rate information.

In the above description, the sampling processing is not performed in the first editing processing section 31 and the sampling processing is performed in the second editing processing section 32. On the contrary, the first editing processing section 31 does not. It is also possible that the sampling process is performed at 31 and the sampling process is not performed at the second editing unit 32. In this case, for example, the first edit processing unit 31
Does not perform the skip correction processing, the second edit processing unit 3
In step 2, the skip correction process may be performed. Here, the fact that the sampling process is not performed is as follows.
This means that the extraction rate w is set to 100%.

In the above description, for convenience of explanation, the editing processing unit 8 is composed of the first editing processing unit 31 and the second editing processing unit 32, but the first editing processing unit 31,
Each function of the second edit processing unit 32 can be provided to one edit processing unit 8.

In the above description, the extraction rate w is
Although set in page units (set as the number of extracted pages for all pages), it is also possible to set not only in page units but also in arbitrary units (for example, line units or block units).

In other words, according to the present invention, at least two editing processes are performed on the recognition result, and at that time, the extraction rates of the recognition results are made different from each other in each editing process. When a recognition result is extracted at a sampling rate, an editing process is performed only on the extracted recognition result.

Next, a specific example of the processing operation of the character recognition system having such a configuration will be described. The character recognition system of the present invention is a software package (information recording medium) 2
When provided in the form of "0", the user sets the information recording medium 20 in the medium driving device 21 and loads the character recognition system software into the RAM 13, for example. At this stage,
The CPU 11 can execute processing according to, for example, character recognition system software loaded on the RAM 13.

When the scanner 14 is, for example, an ADF,
When a plurality of documents are set on the scanner 14, for example, the scanner 14 sequentially reads the plurality of documents. For simplicity, if each document is a single-sided single-sheet document,
One document corresponds to one page, and therefore, for example, n (n
When ≧ 1) originals are read, n pages of document images are obtained for each page, and these are sequentially stored in the document image file 15.

The CPU 11 performs a character recognition process for each page of the document images of n pages stored in the document image file 15 by using the dictionary in the dictionary file 16, and recognizes the recognition result for each page as text. These are sequentially stored in the text file 17.

At this stage, for example, at this stage, the user edits the recognition results (text) for n pages stored in the text file 17 in page units, such as confirmation and correction in page units (first editing). Processing).
That is, the user can correct an erroneously recognized character or the like for each page. This editing process is performed on all pages n by the above-described first editing processing unit 31. In this case, this editing process can be performed by moving the cursor with a mouse or the like to a place where the user needs to correct, but if the skip correction function as described above is provided, the editing process is performed at the position of the low confidence character. , The cursor may be automatically skipped, and the recognition result of the low certainty character may be confirmed and corrected by the user. That is, for example, a low confidence character is searched from the beginning of the recognition result (text) for one page, and when a low confidence character is detected, a cursor is automatically set to this character,
When the user confirms or corrects the character and performs a key operation indicating that the character has been confirmed or corrected, the cursor is automatically skipped to the next low confidence character and the user is prompted to confirm the character. It is also possible to perform a display control for confirming and correcting. This eliminates the need for the user to look through the entire document at the time of confirming and correcting the recognition result, and can increase the work efficiency.

When editing such as confirmation and correction of the recognition results for all pages n is completed, the editing results (recognition results) for all pages n are stored in the first result file 41a.

At this stage, the user can perform the second editing process on the recognition result. That is,
Editing processing can be further performed on the recognition result (editing result) stored in the first result file 41a. This editing processing is performed by the second editing processing unit 32 at a predetermined extraction rate w from the recognition results (editing processing results) for all pages n stored in the first result file 41a (set in advance). An arbitrary page is randomly extracted at the extracted extraction rate w). That is, the user can correct an erroneously recognized character or the like in the extracted page. In this case, this editing process is performed without, for example, activating the skip correction function. Recognition result of the page extracted in this way (editing processing result)
When editing such as confirmation and correction is completed, the result of the editing process is stored in the second result file 41b. In this manner, the first and second editing processes can be performed. Further, if necessary, the third editing process can be performed.

When the first and second editing processes are performed as described above, the dictionary management unit 10 uses the correction history before and after the first editing and the correction history before and after the second editing. Then, for the frequently corrected characters, the dictionary 5
(Eg, a user pattern dictionary or a user language dictionary) is maintained (updated, etc.). The dictionary maintenance can be performed by, for example, an expert of this processing. Information on how many characters with a high correction frequency are present can be easily confirmed by the "correction character information" in the correction history.For example, after updating a standard pattern for a certain character, a certain amount of information can be obtained. If it is determined that it is safe to register the updated standard pattern for this character in the dictionary 5 while observing the period, the updated standard pattern for this character can be registered in the dictionary 5 and used.

When the user performs the second editing process, the user can obtain information on the correction rate at that time. In other words, the user can confirm at an arbitrary timing how much omission of confirmation correction has occurred in the first editing process, based on the correction rate information obtained in the second editing process. In addition, the setting of the certainty threshold can be changed to an appropriate one.

In the above example, the recognition result (editing result) for n × w pages is extracted at a predetermined extraction rate w from the recognition result (editing result) for n pages. In the range up to n, n × w random number values are generated.
Although the pages are randomly extracted (extracted), the recognition result for n × w pages at a predetermined extraction rate w
As a method of extracting (editing result), any extraction method other than the method of generating random numbers and extracting them at random can be used. For example, if the first page (the cover part) must be sampled and inspected, the random sampling method as described above uses the first page.
(Cover part) may not be extracted. Therefore, in such a case, an n × w page can be extracted from the n page and extracted by an arbitrary operation method that always extracts the first page (the cover portion).

[0101]

As described above, according to the first to twelfth aspects of the present invention, the editing work efficiency can be significantly improved in editing such as confirmation and correction of the recognition result.

In particular, according to the first, fifth, sixth, and eleventh aspects of the present invention, at least two editing processes are performed on the recognition result. The sampling rate of the result is made different from each other, and in each editing process, when the recognition result is extracted at a predetermined sampling rate, the editing process is performed only on the extracted recognition result, so that the recognition result is confirmed and corrected. Editing can be performed efficiently.

According to the second, seventh and twelfth aspects of the present invention, when the certainty factor of the recognition result is calculated together with the recognition result in the pattern recognition process, a predetermined editing process is performed in the first editing process. Since the recognition result with the certainty higher than the threshold value is skipped and only the recognition result with the lower certainty is corrected, the user does not have to look through the entire document when confirming or correcting the recognition result, thereby improving work efficiency. Can be increased.

According to the third and ninth aspects of the present invention, when information on a character having a high frequency of correction is output as corrected character information by the editing processing means, the dictionary data is output based on the corrected character information. By performing the update, the recognition rate can be improved.

According to the fourth and eighth aspects of the present invention, the rate of correction made in the second editing process is output as a correction rate, and the certainty threshold is updated in accordance with the correction rate. By doing so, the recognition rate can be improved.

According to the tenth aspect of the present invention, the sampling rate is reduced each time the dictionary is updated, so that the working efficiency can be gradually increased each time the dictionary is used.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration example of a character recognition system according to the present invention.

FIG. 2 is a diagram illustrating an example of a hardware configuration of the character recognition system in FIG. 1;

FIG. 3 is a diagram illustrating an example of a document for a certain page.

FIG. 4 is a view showing a display example for confirming and correcting the recognition result or the editing processing result of the document shown in FIG. 3;

FIG. 5 is a diagram showing an example in which a recognition result or an edit processing result displayed on one side of a reference line and a document image (character image) displayed on the other side of the reference line are always linked on a display screen. FIG. 4 is a diagram for explaining a function of scrolling or moving.

FIG. 6 is a diagram showing an example in which a recognition result or an editing result displayed on one side of a reference line and a document image (character image) displayed on the other side of the reference line are always linked on a display screen; FIG. 4 is a diagram for explaining a function of scrolling or moving.

FIG. 7 is a diagram showing a display example for confirming and correcting the recognition result of the document of FIG. 3;

FIG. 8 is a diagram illustrating an example of an initial screen of an editing process.

FIG. 9 is a diagram illustrating an example of a sampling setting screen.

FIG. 10 is a flowchart illustrating a processing example of a first editing processing unit.

FIG. 11 is a flowchart illustrating a processing example of a second editing processing unit.

FIG. 12 is a flowchart illustrating a processing example of a second editing processing unit.

FIG. 13 is a diagram showing a display example of a modified character information screen.

[Explanation of symbols]

 Reference Signs List 1 image input unit 2 document image storage unit 4 recognition processing unit 5 dictionary 6 recognition result storage unit 8 editing processing unit 9 editing processing result storage unit 9a first result storage unit 9b second result storage unit 10 dictionary management unit 11 CPU Reference Signs List 12 ROM 13 RAM 14 Scanner 15 Document image file 16 Dictionary file 17 Text file 18 Display device 19 Input device 20 Information recording medium 21 Medium drive device 31 First result storage unit 32 Second result storage unit 40 Condition setting unit 41 Editing Processing result file 41a First result file 41b Second result file 42 Temporary file

Claims (12)

[Claims] 1. A recognition result editing method for performing an editing process on a recognition result obtained by performing a pattern recognition process on a predetermined document image, wherein the recognition result is edited at least twice. In each editing process, the recognition result sampling rates are made different from each other, and in each editing process, when the recognition results are extracted at a predetermined sampling rate, the editing process is performed only on the extracted recognition results. Recognition result editing method to feature. 2. The recognition result editing method according to claim 1, wherein when the certainty factor of the recognition result is calculated together with the recognition result in the pattern recognition process, the certainty factor that is equal to or more than a predetermined threshold value is used in the first editing process. A recognition result editing method characterized in that the recognition result is skipped and only the recognition result with low confidence is corrected. 3. The recognition result editing method according to claim 1, wherein in each editing process, a correction history before and after editing is managed, and information on a character having a high correction frequency is output based on the correction history. Recognition result editing method. 4. The recognition result editing method according to claim 1, wherein a rate of correction in the second editing process is output as a correction rate. 5. A recognition dictionary, and recognition processing means for performing pattern recognition processing on a document image using the dictionary,
Editing processing means for performing editing processing on the recognition result by the recognition processing means, wherein the editing processing means performs at least two times of editing processing on the recognition result,
In each editing process, the recognition result sampling rates are made different from each other, and in each editing process, when the recognition results are extracted at a predetermined sampling rate, the editing process is performed only on the extracted recognition results. Characteristic pattern recognition system. 6. The pattern recognition system according to claim 5, wherein the sampling rate is set to be updatable. 7. The pattern recognition system according to claim 5, wherein when the recognition processing means calculates the confidence level of the recognition result together with the recognition result, the editing processing means performs a predetermined threshold processing in the first editing processing. A pattern recognition system characterized in that the above recognition result of certainty is skipped and only the recognition result of low certainty is corrected. 8. The pattern recognition system according to claim 7, wherein said editing processing means further has a function of outputting, as a correction rate, a ratio of corrections made in a second editing process. A pattern recognition system, wherein the degree threshold value can be updated in accordance with a correction rate output from the editing processing means. 9. The pattern recognition system according to claim 5, further comprising dictionary management means for managing said dictionary, wherein said dictionary management means stores information on characters frequently corrected by said editing processing means. A pattern recognition system having a function of updating a dictionary based on corrected character information when is output as corrected character information. 10. The pattern recognition system according to claim 9, wherein a sampling rate is reduced each time the dictionary is updated. 11. A recognition result obtained by performing a pattern recognition process on a predetermined document image is subjected to at least two editing processes, and in each editing process, the recognition result sampling rates are made different from each other. In each editing process,
An information recording medium characterized by recording a program for performing an editing process only on the extracted recognition result when the recognition result is extracted at a predetermined sampling rate. 12. A program for calculating a certainty factor of a recognition result together with a recognition result in a pattern recognition process, and for editing a recognition result, a program for skipping and correcting a character having a certainty factor equal to or higher than a predetermined threshold value is recorded. An information recording medium characterized in that: