JPH03291777A - Recognition candidate output control method for character recognizing device - Google Patents

Abstract

PURPOSE:To reduce the load of a user for the work of confirming and correcting a candidate character by learning the appearance order and appearance frequency of the exact character among the recognition candidate characters, selecting the recognition candidate character with much higher accuracy and reducing the number of recognition candidate characters to be displayed. CONSTITUTION:While referring to a recognition dictionary 14, a character recognition part 1 takes out the recognition candidate characters corresponding to candidate character selection and stores these taken-out characters into a recognition candidate character storage part 2. In respect to these stored recognition candidate characters, a candidate character selection part 3 refers to a second candidate character storage part 6 and selects characters to be a second candidate character group, and these characters are sent to a confirmation/correction part 4 and displayed on a display part 7. Through a keyword 8, the user designates the exact character out of the displayed second candidate character group, and the correction part 4 sends the correct character to an order learning part 5. When the definite correct character is received, the learning part 5 calculates the real order of the correct character while referring to the storage part 2. Further, the learning part 5 updates contents stored in the storage part 6 by changing the total number of times for generating the correct character in the area of the corresponding storage part 6, and the number of times for generating the relevant order.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a character recognition device, that is, an optical character reading device (hereinafter referred to as OCR: 0pticalChar).
The present invention relates to a method for controlling the output of recognition candidate characters in an OCR reader, and particularly to a method for controlling the output of recognition candidate characters in an OCR suitable for efficiently selecting a correct character from among recognition candidate characters.

[Conventional technology]

OCR is a device that reads characters using optical means, and it automatically recognizes characters written or printed on a paper medium and performs the function of inputting them into a computer.

Actual uses of OCR include reading postal codes,
It is used to read the contents of tax payment slips, electricity and gas bill payment slips, etc.

Among OCR, those that target alphanumeric kana are Habo1
A recognition rate of 0.00% was obtained.

However, OCR for kanji has a wide variety of character types, fonts, and character sizes.
Furthermore, the complexity of the patterns requires more complex character recognition methods.

Regarding this type of OCR, please refer to the IEICE publication “Electronic Information and Communication Handbook/1988J (published by Ohmsha)”
PP, 1677-1678, and PP, 2687
-2688.

The OCR described below is all OCR for kanji.
It is about.

In conventional OCR, when reading and recognizing characters, first, a plurality of recognition candidate characters are output.

Among the recognition candidate characters, either only - number of correct answers are included, or all correct answer characters are not included. In other words, most of the recognition candidate characters are not correct characters.

For example, in the case of OCR that outputs six recognition candidate characters for one character, at least 15 recognition candidate characters are not correct characters.

Furthermore, since OCR outputs candidate characters that are similar in shape, the recognition candidate characters naturally include many characters that do not form words.

In conventional OCR technology, multiple recognition candidate characters are displayed on a display device, etc. for one character, and the user selects the correct character from among them. Ta.

[Problem to be solved by the invention]

In conventional OCR, all the recognition candidate characters obtained through recognition are displayed on a display device or the like, and the user selects the correct character from among them.

As a result, the user has to perform a lot of work to check and correct the recognition results, which is a major cause of reducing the throughput of data entry operations using character recognition.

When displaying recognition candidate characters in this way, conventional OCR
is the order of the probability of each character being read as a correct character, that is, the order of accuracy (
Recognition candidate characters were output at positions 1, 2, ..., m).

However, the accuracy is determined in advance, and the method for determining it differs depending on each OCR. Also, O
Due to factors such as the characteristics of the CR and the way the characters are written, there have been problems such as the correct characters not necessarily appearing in a highly accurate ranking.

The purpose of the present invention is to solve the problems of these conventional techniques, learn the order of appearance of correct characters among recognition candidate characters and their frequency of appearance, select recognition candidate characters with higher accuracy, and create recognition candidates. OCR that reduces the number of characters and reduces the burden of checking and correcting candidate characters.
An object of the present invention is to provide a recognition candidate character output control method.

[Means to solve the problem]

In order to achieve the above object, the OCR recognition candidate character output control method of the present invention has a probability that the character code of each character of the output first candidate character group is extracted as a correct character set arbitrarily in advance. A recognition candidate character storage section that stores the characters in a high order, and the total number of times that characters are extracted as correct characters from the first candidate character group stored in this recognition candidate character storage section, and the probability. A second candidate character storage unit that stores the number of appearances corresponding to the ranking of Based on the candidate characters stored in the recognition candidate character storage unit, further selecting a second candidate character group with a high probability of being extracted as a correct character, and from the thus selected second candidate character group, a step of manually extracting the specified correct character; and recognizing the order of appearance of the extracted correct character in the recognition candidate character storage;
The present invention is characterized by including the step of correcting the number of appearances of the correct character corresponding to the appearance order in the second candidate character storage unit and the total number of times the correct character has been extracted as the correct character.

[Effect]

In the present invention, O and CR learn the rank of the determined correct character in the first candidate character group, that is, the rank in the recognition candidate character storage unit, and use this learned result for each character as the second candidate character group. Store in character storage.

Then, when a similar first candidate character group is output to the recognition candidate character storage section, it is further extracted as a correct character based on the learning results for each character stored in the second candidate character storage section. Select candidate characters with high probability.

In this way, when displaying recognition candidate characters on a display device, the appearance similarity data for each recognition candidate character from the past stored in the second candidate character storage section is referred to. The system determines whether the rankings are appropriate and selects recognition candidate characters to be displayed. This makes it possible to limit the displayed recognition candidate characters to only characters with a higher probability of being extracted as correct characters.

Since the candidate characters displayed on the display device are narrowed down, the user can quickly determine the correct character.

Further, after the user has determined the correct character, the appearance order of the correct character is learned, and the learning result is reflected in the second candidate character storage section. This further limits the recognition candidate characters displayed on the display device to only characters with a high probability.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the configuration of an OCR according to the present invention.

An optical scanning/photoelectric conversion unit 11 that travels on a form 10 and converts the shading of characters on the form lo into an electrical signal, removes noise from the character pattern obtained by the optical scanning/photoelectric conversion unit 11, and cuts out a pattern for - characters. a preprocessing unit 12 that performs
The recognition processing unit 9 compares the features obtained in step 3 with the recognition dictionary 14, performs the recognition candidate character output control of the present invention, and determines recognition candidate characters; and the user determines the correct character. It consists of a display section 7 and a keyboard 8 used for

Furthermore, the recognition candidate character processing unit 9 that performs the processing of the present invention compares the features obtained by the feature extraction unit 13 with the recognition dictionary 14, recognizes the read characters, and outputs a corresponding recognition candidate character group. a character recognition unit 1 for processing, a recognition candidate character storage unit 2 for storing a group of recognition candidate characters output from the character recognition unit 1;
From the recognition candidate character group stored in the recognition candidate character storage unit 2,
Further, a candidate character selection section 3 selects a second candidate character group with a high probability of being correct, and a correct character is extracted from the second candidate character group selected by the candidate character selection section 3 based on the user's specifications. It consists of a confirmation/correction section 4, a ranking learning section 5 that learns the processing results of the confirmation/correction section 4, and a second candidate character storage section 6 that stores the results of the ranking learning section 5 for each character.

The second candidate character group selected by the candidate character selection section 3 is displayed on the display section 7 via the confirmation/correction section 4, and the user can specify the correct character from this second candidate character group. , the confirmation/correction section 4 is notified via the keyboard 8.

In addition, the second candidate character storage unit 6 has a storage area corresponding to each character, as will be described later in FIG. The configuration is such that the number of appearances, the number of appearances in the second rank, the number of appearances in the m-th rank, etc. are recorded.

The following describes the OCR with this configuration, particularly the processing operation of the recognition processing section 9 according to the present invention.

FIG. 2 is a flowchart showing one embodiment of the processing operation of the recognition processing section in FIG. 1 according to the present invention.

The operation of each processing section in the configuration shown in FIG. 1 will be explained.

First, the character recognition unit 1 refers to the recognition dictionary 14, extracts m recognition candidate characters corresponding to the characters obtained by the feature extraction unit 13, and outputs them to the recognition candidate character storage unit 2 ( Step 201). At this time, each of the m recognition candidate characters is output in a preset order.

The operations up to this point are similar to the conventional OCR recognition processing operations.

Once all m recognition candidate characters are stored in the recognition candidate character storage unit 2 (step 202), the candidate character selection unit 3 next stores the m recognition candidate characters stored in the recognition candidate character storage unit 2. For each candidate character, refer to the contents of the second candidate character storage unit 6 (step 203) and select characters that can be part of the second candidate character group, that is, characters that have a high probability of being extracted as correct characters (step 203). 204). Then, the selected second candidate character group is sent to the confirmation/correction unit 4 (step 205).

The confirmation/correction unit 4 displays the second candidate character group sent from the candidate character selection unit 3 on the display unit 7 (step 206). The user specifies the correct character from the second candidate character group displayed on the display section 7, and inputs the correct character manually via the keyboard 8 (step 2.7.208). And confirmation correction section 4
sends the correct characters input by the user to the ranking learning section 5 (
Step 209).

When the ranking learning unit 5 receives the determined correct character from the confirmation/correction unit 4, it refers to the recognition candidate character storage unit 2 to determine the actual ranking of the correct character (step 2] 0).

Furthermore, the rank learning unit 5 changes the total number of appearances in the area of the second candidate character storage unit 6 corresponding to the correct character and the number of appearances of the corresponding rank, and stores them in the second candidate character storage unit 6. The contents are updated (step 211).

In this way, according to the present embodiment, each time a character is read and a correct character is determined, the contents are stored in association with the order of appearance in the group of recognition candidate characters and the frequency of appearance of the correct character. Update.

Thereby, the candidate character selection unit 3 can select and display candidate characters with higher accuracy from the first recognition candidate character group.

FIG. 3 is an explanatory diagram showing an example of the display configuration of the second candidate character group on the display unit by the confirmation/correction unit in FIG. 1.

On the screen of the display unit 7 in FIG. 1, there is a first-rank correct character candidate field 31, a second-rank correct character candidate field 32,
Furthermore, a second character candidate group 30 consisting of the correct character candidate column 33 of the first rank as the last candidate is displayed.

The user selects the second candidate character group 3o displayed on the screen.
The correct character is recognized from , and the correct character is input via the keyboard 8 in FIG.

In this embodiment, the candidate characters are narrowed down by the candidate character selection unit 3, and the number of candidate characters displayed is small, allowing the user to easily recognize and specify the correct character.

FIG. 4 is an explanatory diagram showing the storage configuration of the second candidate character storage section in FIG. 1.

The second candidate character storage section 6 is composed of an IC memory, a magnetic disk device, etc., and stores a certain character, for example, a candidate character "
In the storage area 40 corresponding to "Ma" 41, the total number of times this character appears (N, ) 401, and the number of times it appears in the first rank (N, )
402, the number of appearances in the second rank (N,) 403, and the number of appearances in the m-th rank (Nm) 404 are recorded.

The candidate character selection section 3 in FIG. 1 refers to this stored content and further selects the second candidate character group shown in FIG. 2 from the first candidate character group in the recognition candidate character storage section 2.

The operations of the candidate character selection section 3 and the ranking learning section 5 in FIG. 1 will be explained in more detail below using the storage configurations of recognition candidate characters shown in FIGS. 3 and 4. do.

FIG. 5 is a flowchart showing one embodiment of the processing operation of the candidate character selection section in FIG. 1 according to the present invention.

First, a standard "T" for the proportion of occurrences in which a recognition candidate character group can become a correct character candidate is set in advance (step 5
0]).

Then, the m stored in the recognition candidate character storage section 2 in FIG.
The following operations are performed for each of the recognition candidate characters by referring to the contents of the second candidate character storage section 6.

Now, any one of the recognition candidate characters is extracted (step 502). For example, let this be "X". Then, based on the contents of the second candidate character storage section 6 corresponding to this character rX4, the minimum rank in which the appearance rate of the character rXJ is equal to or higher than ITJ is determined (step 503).

That is, the second candidate character storage unit 6 corresponding to the character "x"
The total number of appearances is rNo(X)J, and the number of appearances at the i-th rank is rN i (X)J (i = 1).

2, ..., m).

At this time, Find the minimum rkJ.

That is, the minimum rank 1fJ in which the appearance rate of the character "X" is equal to or greater than "T" is determined. In addition, ITJ is usually 0
．． It is sufficient to set it to 8 to 0.9.

Next, this minimum rank rkJ is compared with the rank (k, (X)) of the character rXJ in the recognition candidate character storage unit 2 (step 504).

Then, if rko(X)≦k” (step 505)
, it is determined that the character rXJ is adopted as a candidate character (
Step 506).

Further, in step 505, if "k.(X)>k", it is determined that the character "x" is not adopted as a candidate character (step 508).

In step 505, if it is determined that rko(X)≦k and the character "X" is to be adopted as a candidate character, the character "X" is sent to the confirmation/correction unit 4 (step 507
).

The candidate character selection section 3 in FIG. 1 performs the above operations for all recognition candidate character groups, and sends the recognition candidate character selection results to the confirmation/correction section 4.

FIG. 6 is a flowchart 16 showing an embodiment of the processing operation of the confirmation/correction section in FIG. 1 according to the present invention.

The confirmation/correction unit displays the candidate character group sent from the candidate character selection unit 3 in FIG. 1 as the second candidate character group 30 in FIG. 3 (step 601), and allows the user to select the correct character (step 601).
Steps 602 and 603). That is, in the second candidate character group 30 in FIG. 3, the first-rank candidate character 31 and the second-rank candidate characters 32 to 33 are displayed separately, and the position indicator such as a cursor key or mouse is displayed. is used to make the user select the correct character (step 603), and the correct character,
For example, send rYJ to the rank learning section 5 (step 60
4).

FIG. 7 is a flowchart showing an embodiment of the processing operation of the ranking learning section in FIG. 1 according to the present invention.

When the ranking learning unit receives the confirmed correct character “Y” from the confirmation/correction unit 4 shown in FIG. 1, the ranking learning unit refers to the recognition candidate character storage unit 2 shown in FIG. The rank, for example, "j" is determined (step 701).

Next, the area of the second candidate character storage unit 6 in FIG. 1 corresponding to the correct character "Y" is referred to, and the total number of appearances FN is determined. (Y)'' and the number of appearances rNj(Y) of the jth rank are updated (step 702).

That is, the contents are updated as No(Y)←N, (Y)+1 Nj(Y)←Nj(Y)+1. However, "←" represents assignment.

In this way, the correct character is stored in the second candidate character storage unit 6 in correspondence with its appearance order among recognition candidate characters and appearance frequency, and each time the correct character is determined, the second candidate character is The contents stored in the storage unit 6 are updated. This makes it possible for the candidate character selection unit 3 of FIG. 1 to select and display candidate characters with a higher degree of accuracy from among the recognition candidate characters.

Although the content recorded in the second candidate character storage section 6 is the number of appearances, it is obvious that the same effect can be obtained even if the appearance ratio is determined and recorded.

As described above with reference to FIGS. 1 to 7, in this embodiment, the candidate character selection unit 3 refers to the stored contents of the second candidate character storage unit 6 and determines the probability that the character can be a correct character. Only high-value characters are selected from the first candidate character group in the recognition candidate character storage section 2 as the second candidate character group. This reduces the number of candidate characters displayed on the display section 7, making it easier for the user to select the correct character, and reducing the time required to confirm and correct character recognition results.

Furthermore, the memory contents of the second candidate character storage unit 6 are changed based on the processing results of the ranking learning unit 5 each time a correct character is determined.
Updated. As a result, the candidate character selection section 3 can obtain more accurate information, and therefore can more accurately perform selection of the second candidate character group.

Then, the load on the user when specifying the correct character can be reduced.

Furthermore, if this embodiment is applied to the preprocessing of the matching process that compares the recognition candidate characters with the word dictionary and estimates the correct characters,
It becomes possible to shorten the verification time.

[Effects of the Invention] According to the present invention, the order of appearance of correct characters among recognition candidate characters and their frequency of appearance are learned, the recognition candidate characters with higher accuracy are selected, and the number of recognition candidate characters to be displayed is determined. To reduce the burden on a user in checking and correcting candidate characters.

[Brief explanation of the drawing]

The drawings show embodiments of the present invention, FIG. 1 is a block diagram showing an embodiment of the configuration of an OCR according to the present invention, and FIG. 2 shows the processing operation of the recognition processing section in FIG. 1 according to the present invention. FIG. 3 is an explanatory diagram showing the operation of displaying the second candidate character group on the display unit by the confirmation/correction unit in FIG. 1, and FIG. 4 shows the storage configuration of the second candidate character storage unit in FIG. FIG. 5 is a flowchart showing an example of the processing operation of the candidate character selection section in FIG. 1 according to the present invention, and FIG. FIG. 7 is a flowchart showing an example of the processing operation of the ranking learning section in FIG. 1 according to the present invention. 1 Character recognition section, 2: Recognition candidate character storage section, 3 Candidate character selection section, 4: Confirmation correction section, 5: Rank learning section, 6: Second candidate character storage section, 7: Display section, 8: Keyboard, 9 Recognition processing unit, 10: Form, ] 1: Optical scanning/photoelectric conversion unit, 1
2: Preprocessing unit, 13° feature extraction unit, 14: Recognition dictionary, 3
0: Second candidate character group, 31: Correct character candidate field of first rank, 32 Correct character candidate field of second rank, 33, Correct character candidate field of mth rank, 40: Storage area, 41: Candidate character [J], 401: Total number of occurrences (N.), 402: Number of occurrences in first rank (N,), 403: Number of occurrences in second rank (N,), 404: Number of occurrences in mth rank Number of occurrences of (
Nm). Figure 2 Figure 0 1 Figure 5 Figure 735- Door 02

Claims (1)

[Claims] (1) Possibility that each character written on a paper medium and read using optical means will be recognized as a set of pixels for each character, and that the correct character will match the character represented by the set of pixels. A recognition candidate character output control method for a character recognition device that outputs a first candidate character group consisting of a certain plurality of characters in a predetermined order of a probability of being extracted as a correct character set arbitrarily. Regarding the recognition candidate character storage means for storing the character codes of each character of the first candidate character group in the above-mentioned order, and the characters extracted as correct characters from the first candidate character group stored in the recognition candidate character storage means. , the total number of times extracted as correct characters, and
and a second candidate character storage means for storing the number of appearances corresponding to the ranking, and the recognition candidate characters are stored in the recognition candidate character storage means based on the total number of times and the number of appearances stored in the second candidate character storage means. A step of further selecting a second candidate character group from the candidate characters that has a high probability of being extracted as a correct character, and a step of manually extracting a specified correct character from the selected second candidate character group. , recognize the appearance order of the correct character in the recognition candidate character storage means, and determine the number of appearances of the correct character in the second candidate character storage means corresponding to the appearance order, and the total number of times the correct character has been extracted as a correct character. A method for controlling the output of recognition candidate characters in a character recognition device, the method comprising the step of: