JPH0934987A - Character recognized result correcting method and character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a high-operability character recognizing device. SOLUTION: This device is provided with a feature extracting part 16 for extracting the feature of an input pattern, dictionary part 18 in which the standard feature of a character is stored, character recognizing part 20 for performing the character recognition of the input pattern by comparing the feature with the standard feature, display part 24 for displaying this recognized result, and keyboard 26 for designating one of erroneously read characters for which the erroneously recognized results are displayed on the display part. Then, a control part 28 is provided to retrieve the input pattern having a feature similar to the feature of the designated erroneously read character by comparing the feature of the designated erroneously read character with the other feature and to display the recognized result of the retrieved input pattern on the display part while adding a marker to it.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition result correction method for correcting a recognition result obtained by reading a character pattern on a medium and a character recognition device using the same.

[0002]

2. Description of the Related Art Conventionally, when correcting a recognition result of a character pattern, an operator searches every character that is erroneously recognized (misread character) and replaces it with a correct character to correct it. In this correction method, it is necessary to carefully observe the similar characters, which requires a long time for correction and is a burden on the operator. Therefore, in order to reduce the burden on the operator and the time required for the correction, for example, a document: Media Drive Research Laboratory Catalog "W
inReader Plus v. 2.0 ”proposes a method for collectively searching and replacing misread characters based on the character name of the recognition result.

[0003]

However, in the technique disclosed in the above-mentioned document, since characters are collectively searched and replaced based on the recognition result, characters that do not need to be replaced may be searched and replaced. Therefore, the operator has to confirm the search result every time the search is performed in order to remove the characters that do not need to be replaced.

For this reason, it has been desired to realize a character recognition result correction method and a character recognition device which can retrieve misread characters more accurately and have good operability.

[0005]

[Means for Solving the Problems]

(First Invention) According to the character recognition result correction method of the first invention of this application, the character recognition of the input pattern is performed by comparing the characteristics of the input pattern with the standard characteristics stored in the dictionary section. Then, after displaying the recognition result of this character recognition on the display unit, when correcting this erroneous recognition result, specify one of the misread characters that displayed the incorrect recognition result, and The characteristics of the input pattern of the misread character and the characteristics of the input pattern other than the specified misread character are compared, and the input pattern having the characteristics similar to the specified characteristics of the misread character is searched and searched. It is characterized in that the recognition result corresponding to the input pattern is displayed with a marker.

(2nd invention) According to the character recognition device of the 2nd invention which relates to this application, it has the feature extraction section which extracts the feature of the input pattern, it possesses the dictionary section where standard feature is housed, features And a standard feature for comparing the input pattern with a character recognition unit, a memory for storing the recognition result and the feature of the character recognition, and a display unit for displaying the recognition result. And an input unit for designating one of the misread characters whose incorrect recognition result is displayed, and the characteristics of the input pattern of the specified misread character,
By comparing the characteristics of the input pattern other than the specified misread character, the input pattern having the characteristics similar to the characteristics of the input pattern of the specified misread character is searched, and the input pattern corresponding to the searched input pattern is searched. It is characterized by comprising a control unit for displaying a recognition result with a marker. The characters recognized in these inventions include general symbols.

[0007]

According to the character recognition result correction method of the first invention and the character recognition device of the second invention related to this application, when correcting a misread character, one of the misread characters in the displayed recognition result is corrected. And the characteristics of the specified misread character,
The characteristics other than the designated misread character are compared to search for an input pattern having a feature similar to the designated misread character.

As a result, when the operator looks at the display of the recognition result and designates one character of the misread characters, the character of the input pattern having similar characteristics to the designated misread character can be retrieved. it can. At the time of this search, unlike the first character recognition pattern, the characteristics of the input pattern of the designated misread character and the characteristics of the other input patterns are compared. Therefore, at the time of character recognition, non-identical input patterns can be excluded from the input patterns misread as the same pattern, and only the input patterns similar to the characteristics of the designated misread character input pattern can be searched.

In these inventions, the recognition result corresponding to the searched input pattern is displayed with a marker. As a result, the operator need only pay attention to the marked characters. For this reason, it is possible to suppress occurrence of omission of correction due to oversight of misread characters, and further it is possible to reduce the time required for correction. Further, since the markers are attached only to the characters having similar characteristics, it is possible to reduce erroneous corrections and correct them accurately.

Therefore, according to these inventions, it is possible to realize a character recognition result correction method and a character recognition device which reduce the burden on the operator at the time of correction and have good character and operability.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The first embodiment of the present application with reference to the drawings
An embodiment of the character recognition result correcting method of the invention of 1) and the character recognition device of the 2nd invention will be described together. It should be noted that the drawings to be referred to merely schematically show the shapes, sizes, and positional relationships of the respective constituent components to the extent that these inventions can be understood. Therefore, these inventions are not limited to the illustrated examples.

(Regarding Configuration of Character Recognition Device) First, FIG. 1 shows a block diagram of a character recognition device of this embodiment.
The character recognition device of this embodiment includes a scanning unit 10, an image storage unit 12, and a character cutting unit 14 as a configuration for obtaining an input pattern. In the scanning unit 10,
An image is captured from a medium to be read on which characters or the like to be recognized for characters are described. The image storage unit 12 also stores the image captured by the scanning unit 10. Further, in the character cutout unit 14, the image storage unit 1
A character pattern is cut out for each character from the image stored in 2 to obtain an input pattern.

The character recognition apparatus of this embodiment further comprises a feature extraction unit 16 for extracting the features of the input pattern.
It also comprises a dictionary unit 18 in which standard features of characters are stored. In addition, a character recognition unit 20 that performs character recognition of the input pattern by comparing the characteristics with the standard characteristics is provided, and a memory 22 that stores the recognition result and characteristics of the character recognition is provided. Further, it is provided with a display unit 24 for displaying the recognition result. Further, the keyboard 26 is provided as an input unit for designating one of the misread characters for which an incorrect recognition result is displayed on the display unit.

Then, the feature of the designated misread character is compared with the feature other than the designated misread character, and an input pattern having a feature similar to the feature of the designated misread character is searched and retrieved. The controller 28 is provided with a marker attached to the recognition result corresponding to the inputted input pattern and displaying it on the display.

(Regarding Character Recognition Method) First, character recognition of characters, symbols and the like described on the medium 30 to be read shown in FIG. 2 is performed using a conventionally known character recognition method. The medium to be read shown in FIG. 2 includes a kanji character “K” shown by 100 in the drawing and a katakana “E” shown by 102 in the drawing.

In this embodiment, when character recognition is performed, a feature matrix obtained by creating sub patterns is used as a feature. For this purpose, first, the scanning unit 1 is moved over the medium 30 to be read.
The image is optically scanned at 0, and the obtained image is stored in the image storage unit 12. Next, the character cutting unit 14 cuts out an input pattern for each character from the stored image. This input pattern is input to the feature extraction unit 16 that extracts the features of the input pattern.

Next, FIG. 3 is a block diagram showing the configuration of the feature extraction unit 16 of the character recognition apparatus of this embodiment. In this embodiment, character recognition is performed using a feature matrix obtained by creating sub patterns. Therefore, the input pattern input to the feature extraction unit 16 is first input to the pattern register 32 and the line width calculation unit 34.

The line width calculator 34 calculates the line width (W) of the input pattern. In calculating the line width (W), first, when each part of the input pattern is scanned by a 2 × 2 window of 4 pixels, the number M of windows in which all 4 pixels are black pixels and the input pattern are calculated. The total number of black pixels A is counted. Then, from the number M of windows and the number A of all black pixels, the line width (W) is counted by the following equation (1).

W = A / (A−M) (1) Next, in the sub pattern extraction unit 36, the vertical scanning is performed on the entire pattern register 32, and the black bit continuous length and line width calculation unit 34 obtains it. The vertical sub-pattern (VSP) is extracted from the relationship with the obtained line width (W). Furthermore, VS
Similarly to P, a horizontal scan is performed to extract a horizontal sub-pattern (HSP), a right diagonal 45 ° scan is performed to extract a right diagonal sub-pattern (RSP), and a left diagonal 45 ° scan is performed to perform a left diagonal sub-pattern ( LSP).

Here, the upper part of FIG. 4 shows a sub-pattern of the kanji "Kou". The original patterns of the input patterns, VSP, HSP, RSP and LSP are shown in order from the left end of the upper part of FIG. However, there is no LSP sub-pattern in the kanji "Kou", so the LSP is blank. Also, in the upper part of FIG.
The sub-pattern of katakana "e" is shown. The original pattern of the input pattern, VSP, HSP, in order from the left end of the upper part of FIG.
RSP and LSP are shown. However, since there is no RSP or LSP sub-pattern in the katakana "e", RS
P and LSP are blank. Then, as will be described later, a four-dimensional feature matrix (shown in the lower part of FIGS. 4 and 5) for one input pattern is extracted from the four sub patterns.

Next, the input pattern on the pattern register 32 is input to the character frame detector 38. Character frame detector 38
Then, the character frame circumscribing the input pattern is detected.

Next, the character frame detected by the character frame detection unit 38 is input to the character frame division determination unit 40. The character frame division determination unit 40 determines the division point coordinates on the X axis and the Y axis used for dividing the inside of the character frame into N × M regions.
However, N and M represent natural numbers, and in this embodiment, N =
Let M = 5. In this embodiment, the horizontal direction of the character frame is the X axis and the vertical direction is the Y axis.

Next, in the feature matrix extraction unit 42, the division point coordinates determined by the character frame division determination unit 40,
A feature matrix for each input pattern is extracted using the four sub patterns extracted by the sub pattern extraction unit 36 and the line width (W) calculated by the line width calculation unit 34.

In extracting the characteristic matrix, first, the character area frames of the four sub patterns are divided into N × M areas by the division point coordinates. Next, the number of black pixels (B _ij ) in each divided area is counted. Next, the character line length (L _ij ) of each area is calculated from the number of black pixels (B _ij ) and the line width (W) using the following (2).

L _ij = B _ij / W, where 1 ≦ i ≦ N and 1 ≦ j ≦ M.

Next, the character line length (L _ij ) is normalized by the size of the input pattern. For example, in the feature matrix of VSP, the length ΔY of the character frame in the Y direction, in the feature matrix of HSP, the length ΔX of the character frame in the X direction, and
In the feature matrix of RSP and LSP, (ΔX
Normalize by ² + ΔY ² ) ^1/2 . Then, the value obtained by normalizing the character line length is used as the value of each area to create a (N × M) × 4 dimensional feature matrix.

Here, the lower part of FIG. 4 shows the characteristic matrix of the Chinese character "Kou". The feature matrices corresponding to the input patterns VSP, HSP, RSP, and LSP are shown in order from the left end of the lower part of FIG. However, the kanji "Kou"
, There are no LSP sub-patterns, so all values in the feature matrix corresponding to LSP are 0. In addition, the lower part of FIG. 5 shows a feature matrix of katakana “D”. Input patterns VSP and HS in order from the left end of the lower part of FIG.
The characteristic matrix corresponding to each of P, RSP, and LSP is shown. However, the katakana "E" does not have the sub-patterns of RSP and LSP, so RSP and LSP
All the values in the feature matrix corresponding to are 0.

Next, the feature matrix extracted by the feature extraction unit is input to the recognition unit 20 and the control unit 28.

In the identification unit 20, the distance D between the extracted feature matrix (f _i ) and each feature matrix (f _m ) of standard character patterns as standard features stored in the dictionary unit is known. It is calculated using the following equation (3).

D = {Σ (f _i −f _m ) ² } ^1/2 (3) Incidentally, a plurality of standard character patterns are usually prepared for the same character.

Then, ordering is performed from the smallest distance between the feature matrix corresponding to each of the four sub patterns of the input pattern and the feature matrix of the standard character pattern. Then, the character name corresponding to the standard character pattern having the smallest distance D is input to the control unit 28 as the recognition result of the input pattern.

The feature matrix and the recognition result of the input pattern input to the control unit 28 are stored in the memory 22.

Further, the processing from the character segmentation to the storage of the recognition result in the memory 22 is repeated for each character on the medium 30 to be read. After obtaining the recognition result for each character on the medium 30 to be read, the memory 22
The recognition result corresponding to each input pattern stored in is displayed on the display unit 24.

By the way, as a feature matrix of a dictionary used in character recognition, a plurality of types of feature matrices are usually prepared as standard features for one character. For this reason, for example, the standard feature of the kanji "Kaku" may be closer to the distance D than the standard feature of the katakana "E", although it is the katakana "E". obtain. A display example 50 of the recognition result in which such misreading is made is shown in FIG. This display example 5
At 0, the input pattern originally having the katakana "e" is recognized as all the kanji "work" as shown by 104 in the figure.

(Method of Correcting Character Recognition Result) Next, a method of correcting the character recognition result will be described with reference to the flowchart of FIG. After the recognition result is displayed (a) on the display unit as described above, first, the operator looks at the recognition result displayed on the display unit 24 and, if necessary, designates one misread character using the keyboard 26. (B).

Next, it is judged whether or not there is a correction (c).
If there is no correction, the recognition result stored in the memory 22 is output. On the other hand, if there is a correction, the presence or absence of the marker in the recognition result is determined (d). If there is a marker, the process returns to the display of (a) and the character with the marker is corrected. On the other hand, when there is no marker, a similar input pattern is searched (e).

Here, the operator, looking at the display, corrects and designates one character, which is originally the katakana “E” but recognized as the Chinese character “K”, into the Katakana “E”. At this time, the correct recognition result (that is, the katakana “E”) is input, and the position information in the display example of the corrected character is also specified. For the position information, the feature matrix of the input pattern of the designated misread character is stored in the memory 22.
Used to read from inside.

When the misread character is designated, the control unit 28
The characteristic matrix of the designated misread character is read from the memory 22. Then, the feature of the designated misread character is compared with the feature other than the designated misread character stored in the memory 22, and an input pattern having a feature similar to the feature of the designated misread character is compared. Search for.

Here, having similar features means that the distance between the feature matrices as the features of the input pattern is calculated using, for example, the above-mentioned equation (3), and this distance is a feature matrix below a certain threshold. It is good to judge that they are similar features.

Since the feature matrix extracted from the character pattern of the misread character is used as a reference in the search for this similar pattern, as in the case of the first character recognition, for example, “K” of kanji and “E” of katakana are used. Can be suppressed.

Next, the control unit 28 causes the display unit 24 to display the recognition result corresponding to the retrieved input pattern with a marker. Here, an underline is added as a marker. Here, FIG. 6B shows a display example 60 of the display result with the marker. In this display example, 106 in the figure
The underline is added only to the kanji “K”, which is the misread character that should be originally “Ka” in “”.

Therefore, the operator only needs to correct the underlined recognition result by looking at the recognition result displayed with the marker. Every time the underlined recognition result is corrected using the keyboard, the recognition result display is updated and the underline of the corrected part disappears.
And you can correct the misread characters one by one until all the underlines disappear.

Further, for example, when there is a misread character of a type other than the above-mentioned katakana "E", the correction process is repeated from the above-mentioned process (a) for designating the misread character according to the flowchart shown in FIG. For example, the katakana "E"
Other examples of misreading include the kanji “8” and katakana “ha”, the kanji “10” and katakana “na”, the kanji 1,000 and katakana “chi”, and the symbol “/” and katakana "No" etc. are mentioned.

In the above-described embodiments, the examples in which these inventions are configured under specific conditions have been described, but many modifications and variations can be made to these inventions. For example, in the above-described embodiment, the sub-pattern is created and the feature matrix is used as the feature of the input character, but in these inventions, for example, the similarity method may be used as the feature. In the similarity method, it is advisable to compare the feature vectors instead of the feature matrix to search for input patterns having similar features.

In the above-described embodiment, the marker is underlined. However, in these inventions, the marker is surrounded by ruled lines, the character is highlighted, or the character is highlighted. You may blink it.

Although the keyboard is used as the input unit in the above-described embodiments, the input unit is not limited to this in these inventions, and for example, a mouse may be used.

[0047]

According to the character recognition result correction method of the first invention and the character recognition device of the second invention related to this application,
In correcting the misread character, one of the misread characters displayed with the incorrect recognition result is specified, the specified misread character is compared with the specified feature other than the misread character, and the specified An input pattern having characteristics similar to the characteristics of the misread character is searched.

As a result, if the operator looks at the display of the recognition result and designates one character of the misread characters, the character of the input pattern having the similar characteristics to the designated misread character can be retrieved. it can. At the time of this search, unlike the first character recognition, the characteristics of the input pattern of the designated misread character are compared with the characteristics of the other input patterns. Therefore, at the time of character recognition, non-identical input patterns can be excluded from the input patterns misread as the same pattern, and only the input patterns similar to the characteristics of the designated misread character input pattern can be searched.

In these inventions, the recognition result corresponding to the retrieved input pattern is displayed with a marker. As a result, the operator need only pay attention to the marked characters. For this reason, it is possible to suppress occurrence of omission of correction due to oversight of misread characters, and further it is possible to reduce the time required for correction. Further, since the markers are attached only to characters having similar characteristics, it is possible to reduce erroneous corrections and correct them accurately.

Therefore, according to these aspects of the invention, it is possible to realize a character recognition result correction method and a character recognition device which are easy to operate and have a reduced burden on the operator.

[Brief description of drawings]

FIG. 1 is a block diagram for explaining a configuration of a character recognition device according to an embodiment.

FIG. 2 is a read medium according to an embodiment.

FIG. 3 is a block diagram for explaining the configuration of a feature extraction unit that constitutes the character recognition device of the embodiment.

FIG. 4 is a diagram for explaining a sub-pattern and a feature matrix of a Chinese character “ko”.

FIG. 5 is a diagram for explaining a sub-pattern of katakana “D” and a feature matrix.

FIG. 6A is a display example of a recognition result at the time of initial character recognition, and FIG. 6B is a display example of a recognition result with a mark.

FIG. 7 is a flowchart for explaining a recognition result correction process.

[Explanation of symbols]

 10: scanning unit 12: image storage unit 14: character cutting unit 16: feature extraction unit 18: dictionary unit 20: recognition unit 22: memory 24: display unit 26: keyboard (input unit) 28: control unit 30: medium to be read 32: pattern register 34: line width calculation unit 36: sub pattern extraction unit 38: character frame detection unit 40: character frame division determination unit 42: feature matrix extraction unit 50, 60: display example of recognition result

Claims (2)

[Claims] 1. The character of the input pattern is recognized by comparing the characteristic of the input pattern with the standard characteristic stored in the dictionary unit, and the recognition result of the character recognition is displayed on the display unit. In correcting the recognition result, one character of the misread characters for which an incorrect recognition result is displayed is specified, and the characteristics of the input pattern of the specified misread character and the input pattern other than the specified misread character are specified. It is characterized in that an input pattern having a feature similar to the feature of the specified misread character is searched by comparing with the feature, and a recognition result corresponding to the searched input pattern is displayed with a marker attached. Character recognition result correction method. 2. A character extracting section for extracting a characteristic of an input pattern, a dictionary section storing standard characteristics, comparing the characteristic with the standard characteristic, and performing character recognition of the input pattern. A recognition unit, a memory for storing the recognition result of the character recognition and the feature, a display unit for displaying the recognition result, and one of the misread characters for which an incorrect recognition result is displayed on the display unit. An input unit for designating a character is provided, and the characteristics of the input pattern of the specified misread character and the characteristics of the input pattern other than the specified misread character are compared,
A control unit for searching for an input pattern having characteristics similar to the input pattern of the specified misread character and displaying a recognition result corresponding to the searched input pattern with a marker attached. Character recognition device.