JPH06301817A - Character recognition device - Google Patents

Abstract

PURPOSE:To reduce the time and the labor required for confirmation, correction, or the like of the recognition result by changing the display color and the printing color of a character when the reliability of the recognition result is low. CONSTITUTION:In a line width validity discriminating part 72, an average line width calculating part 74 stores a line width W of the input pattern of each character in one field (one item) or one row of a slip, which is calculated by a line width calculating part 20, in a line width register 76. The average line width calculating part 74 averages the line width of characters in one field to obtain an average line width WM and obtains the difference between this average line width and the stored line width W of the input pattern of each character in the pertinent field and compares this difference with line width thresholds WH and WL as reference line width and sends the character number and the result, which indicates that the reliability of the recognition result is low, to a control part 32 in the case of W-WM>WH or W-WM<WL. The control part 32 presents the display color or the printing color of the pertinent character in a display part 80 or a printing part 82 based on the result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device which has good operability and high reliability.

[0002]

2. Description of the Related Art Conventionally, various character recognition devices have been developed and put into practical use. It is expected that the work of inputting other Japanese printed documents, such as forms, for creating various databases using these character recognition devices will increase more and more, in order to save labor and speed up input. Printed Kanji OCR (O to recognize printed Kanji optically
The need for a CR (optical character recognition device) is increasing.
Prior to the description of the character recognition device of the present invention, Japanese Patent Publication No. 60-
An example of a conventional typical character recognition device (hereinafter sometimes simply referred to as OCR) disclosed in No. 38756 will be described with reference to FIG.

FIG. 2 is a block diagram showing an OCR for explaining the conventional technique. In the figure, reference numeral 10 is a reading recognition unit of the character recognition device. In this reading recognition unit 10,
Reference numeral 12 is an optical read input unit for optically reading Chinese characters printed on a form or the like and converting them into an optical signal input, 14 is an optical signal input, 16 is a photoelectric conversion unit, 18 is a pattern register, 20
Is a line width calculation unit, 22 is a character frame detection unit, 24 is a sub pattern extraction unit, 26 is a feature matrix extraction unit, 28 is an identification unit,
30 is a character name output terminal, 32 is each component 12, 16, 18, 20, 22, 24, 26, 28 in the read recognition unit 10.
Is a control unit for controlling. Therefore, the reading unit 50 is configured by the optical reading input unit 12 and the photoelectric conversion unit 16,
The pattern register 18, the line width calculation unit 20, the character frame detection unit 22, the sub pattern extraction unit 24, the feature matrix extraction unit 26, and the identification unit 28 constitute a recognition unit 60.

A character recognition operation in the conventional apparatus will be described with reference to FIG. First, character information such as printed Chinese characters is read by the optical read input unit 12 and converted into an optical signal input 14 of an original pattern, and this optical signal input 14 is converted into binary quantized electrical signals by a photoelectric conversion unit 16. Signal 34. The pattern register 18 stores this electric signal 34. At the time of this storage, the character is divided into, for example, 100 × 100 cells, and each binary data is stored in the pattern register 18. The line width calculator 20 calculates the line width W of the input pattern (original pattern) and outputs the line width signal 36.
The line width calculation unit 20 has a shift register configuration similar to a known filter circuit, and calculates the line width W using, for example, a known approximation formula (1) shown below. W = 1 / {1- (Q / A)} (1) In the equation (1), Q is the pattern register 18 (2 ×
2) is the number of points where all the points in the window become black points when scanned by the window, and A is the number of all black points in the pattern register 18.

The sub-pattern extraction unit 24 extracts a vertical sub-pattern (VSP) and a horizontal sub-pattern (HS) from the original pattern.
P), the right diagonal sub-pattern (RSP) and the left diagonal sub-pattern (LSP) are extracted and a sub-pattern extraction signal 38 is output. FIG. 3 shows an example of extracting the original pattern and the sub-pattern. The original pattern is shown in FIG.
(B) is VSP, (C) is HSP, (D) is RSP,
(E) shows the LSP sub-patterns, respectively.
The sub-pattern extraction unit 24 performs a vertical scan on the pattern register 18 over the entire surface and extracts a vertical sub-pattern (VSP) based on the relationship between the continuous length of black bits and the line width calculated by the line width calculation unit 20. Similarly, the horizontal scan scans the horizontal sub-pattern (HSP), and the diagonal scan (45 °) scans the right diagonal sub-pattern (RS).
Left diagonal sub-pattern (LSP) is extracted by scanning left diagonal (45 °) of P).

The character box detector 22 is provided with a pattern register 18
A character frame circumscribing the character pattern in the pattern register 18 is detected from the electric signal 34 from the above, and the resulting character frame signal 40 is sent to the feature matrix extraction unit 26.
The feature matrix extraction unit 26 sets (N × M) (for example, an area corresponding to the character frame of the original pattern in an unillustrated sub pattern register storing sub patterns for each direction) (for example,
(N = M = 5). For example, the character is 100 ×
It is divided into 100 cells, and when N = M = 5, each area has 20 × 20 cells.

Here, a method of extracting a feature matrix will be described by taking the vertical sub-pattern VSP as an example. The feature matrix extraction unit 26 sets (N × M) counters (not shown) provided for each divided area to “0”, horizontally scans the character frame in the main scanning direction from left to right, and for each sub-scanning. And the black bit coordinate position (X _WB , Y _n ) when the white bit (character back shadow part) changes to the black bit (character line part) and the black bit coordinate position when the black bit changes to the white bit (X _BW ,
Y _n ) is detected and the midpoint coordinate position (X _B , Y _n ) is calculated by the equation (2). X _B = (X _WB + X _BW ) / 2 (2) Next, it is determined where the midpoint coordinate position exists in the divided area, and the content of the counter of the determined divided area is incremented by one.

This processing is performed on the entire surface of the character frame, and the (N × M) counter value provided in each divided area is the (N × M) -dimensional line length matrix of the vertical sub-pattern VSP. Similar processing is performed for each HSP, RSP, and LSP sub-pattern. However, the horizontal sub-pattern HSP scans vertically in the main scanning direction from top to bottom and horizontally in the sub-scanning direction from left to right. Left and right diagonal sub-patterns RSP and LS
P scans the main scanning direction vertically from top to bottom, the sub-scanning direction horizontally from left to right, or scans the main scanning direction horizontally from left to right and the sub-scanning direction vertically from top to bottom.

Next, the feature matrix extraction unit 26 normalizes the extracted line length matrix by the size of the character to create a feature matrix. The method is as follows, where one element of the line length matrix before normalization is k _ij , the element after normalization is L _ij , the horizontal length of the character frame is ΔX, and the vertical length is ΔY. It is processing. (1) Vertical sub-pattern (VSP) matrix L _ij = k _ij / ΔY (3) (2) Horizontal sub-pattern (HSP) matrix L _ij = k _ij / ΔX (4) (3) Diagonal sub-pattern In the case of (RSP, LSP) matrix L _ij = k _ij / {(ΔX) ² + (ΔY) ² } ^1/2 (5) By the above processing, the feature matrix extraction unit 26 finally represents the original pattern. A (N × M) × 4} -dimensional normalized feature matrix is created, and the feature matrix signal 42 is generated.
Is output.

The identification unit 28 is a standard character mask (f _m ) holding a feature matrix for standard character patterns.
And the feature matrix (f _i ) extracted by the feature matrix extraction unit 26, the category name of the standard character mask that gives the minimum value of the distance (D) of the expression (6) that has been used conventionally is a character name. Character name output terminal 3 as output 44
Output from 0. _{D = Σ (f m -f i} ) 2 (6)

By the way, in the above-described conventional OCR, if the resolution of the reading unit 50 including the optical reading input unit 12 and the photoelectric conversion unit 16 is insufficient, the pattern is crushed and the characteristics to be extracted become unstable. It caused a false recognition. FIG. 4 shows an example of a pattern which is crushed due to lack of resolution of the reading unit 50, and a pattern of "roar" of 10 points of medium Gothic is read at a resolution of 16 lines / mm. Therefore, in order to prevent it from being crushed, the conventional print Kanji OCR
When inputting a Japanese print document with, I confirmed that the size and font of the type of the Japanese print document were within the proper range before inputting.

[0012]

However, in the printed Kanji OCR, there are many types of typefaces, such as the Mincho typeface, such as the Hoshomincho type, the Middle Mincho type, and the Taiming type. Since there are many typefaces, such as different typefaces for each printing maker, general users who are not specialists in typefaces recognize Japanese printed documents with the printed kanji OCR and check whether the displayed or printed recognition results are correct. It is necessary to carefully observe and confirm and correct the recognition result. Therefore, the conventional device has a problem that it takes time and labor for confirmation and correction, resulting in poor operability. The above problem is not limited to the printed Kanji OCR, and is the same for any OCR including a handwritten OCR.

An object of the present invention is to eliminate the above-mentioned conventional problems, and to provide a character recognition device which has good operability and high reliability.

[0014]

In order to solve the above-mentioned problems, the present invention comprises a reading section for photoelectrically converting a character pattern described on an input medium and outputting character pattern data.
In a character recognition device including a recognition unit that recognizes the character pattern data and a display unit that displays the recognition result, or a printing unit, the reliability of the character recognition result is determined based on the line width of the character pattern data. A character line width validity determining unit is provided, and the method of displaying or printing the recognition result is visually changed based on the reliability of the character recognition result.

[0015]

According to the present invention, the reliability of the character recognition result is judged based on the line width of the character, and the display method or the printing method of the recognition result is visually changed based on the judgment result. Therefore, the time and labor required for post-processing of the recognition result can be reduced, and the above problem can be solved.

[0016]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, the same components as those shown in FIG. 2 are designated by the same reference numerals, and detailed description thereof will be omitted.

The character recognition apparatus of the present invention comprises a character line width validity determining section 72 including an average line width calculating section 74 and a line width register 76. This character line width validity determination unit 72
Determines the reliability of the recognition result of the character to be recognized during the character recognition process. In this case, the control unit 32 is operated to change the display color or the print color on the display unit 80 or the print unit 82, thereby making it possible to notify the operator that the recognition result of the recognized character is low in reliability.

The average line width calculation unit 74 includes one or more characters based on the line width (W) information of the line width signal 36 for each input character pattern output from the line width calculation unit 20 of the read recognition unit 10. The average line width W _M is calculated.

The line width register 76 receives the line width signal 36 of each character from the line width calculation unit 20 in calculating the average line width W _M in the average line width calculation unit 74 and receives the line width W of each character. Is stored once, and each line width W corresponding to the number of characters to be averaged can be read by the average line width calculation unit 74 as required. The control unit 32 controls the constituent components 74 and 76 of the character line width validity determining unit 72 and the display unit 80.
And for controlling the printing unit 82 in various ways.

Next, the operation of the line width validity determining unit 72 for notifying the user that the reliability of the recognition result is low using the line width will be described. Now, as an example, the character carrier of the characters to be recognized is used as a form, and the resolution in the reading unit is 16 lines / mm.

The average line width calculation unit 74 determines the line width W for each input pattern of each character in the field or line calculated by the line width calculation unit 20 for one field (one item) or one line of the form. Stored in register 76, line width register 7
An address counter (not shown) 6 is incremented. It should be noted that the range and the number of characters in one field are defined in advance depending on the form of the form. Then, the average line width calculation unit 74 averages the line widths of the characters for one field stored in the line width register 76 to obtain the average line width W _M , and the field stored in the line width register 76 is averaged. obtains the difference between the line width W of each input pattern of each character of the inner line width as a reference line width threshold value W _H, W _L (W _H, W _L is a constant, W _H = 1.0 in this embodiment , W _L = −1.0), and if Expression (7) or Expression (8) is satisfied, the validity determination result that the reliability of the character number and the recognition result is low is sent to the control unit 32. W-W _M > W _H (7) W-W _M <W _L (8)

The line width thresholds W _H and W _L as the reference values
Is predetermined for each field according to the recognition purpose (for example, reading character type or writing instrument) or the recognition condition (for example, resolution), and the value is stored in the memory (not shown) provided in the character line width validity determining unit 72. It is stored so that it can be read freely.

The control unit 32 instructs the display color or the printing color of the character in the display unit 80 or the printing unit 82 based on the validity judgment result corresponding to the character number of each character. In this case, the user should check the recognition result while paying attention only to the specific color of the display section or the printing section (in this case, there are two kinds of validity judgment results of formula (7) and formula (8), so there are two colors). Good.

It will be clear that the invention is not limited to the embodiments described above, but that many variants or modifications can be made. For example, the configuration of the recognition unit is not limited to the conventional configuration shown in FIG. 1, and the feature is extracted using the line width of the character pattern and the recognition result is output based on the extracted feature. If so, the configuration does not matter.

In addition to the above embodiment, the average line width W _M may be the average of all the characters recognized until the character is recognized. In that case, it is preferable to store the number of characters up to the character in the line width register.

Further, in the above-described embodiment, the difference between the average line width W _M of one field or one line and the line width W of each character in the field is compared with the reference line width thresholds W _H and W _L. However, as shown in Expression (9) or Expression (10), the reference line width thresholds W _THH , W _THL (W _THH , W _THL are constants, directly for each character.
In this embodiment, W _THH = 7.0 and W _THL = 2.0) can be compared. In this case, equation (9) or equation (10)
If the condition is satisfied, it is determined that the reliability of the recognition result is low. W> W _THH (9) W <W _THL (10)

Further, although the form has been described as the character carrier in the above-mentioned embodiments, this character recognition device can be applied to any suitable character carrier other than the form.

Further, in the above-mentioned embodiment, the line width signal is inputted from the line width calculation section to the average line width calculation section to the line width register of the character line width validity judgment section, but instead of this, Alternatively, the line width calculator may directly input the line width register.

Further, although the display color or the print color has been described in the above-mentioned embodiment, the present invention is not limited to this, and any color such as luminance, reversal, blinking, underline, etc. can be visually distinguished.

The above-mentioned carrier means handwriting, printing,
By any other suitable method, paper, plastic,
Text, information may be recorded on cloth, wood, metal, or any other suitable material.

[0031]

As is apparent from the above description, according to the present invention, the average line width of characters in a character string including an input character is compared with a predetermined threshold value, and as a result of the comparison, the character is recognized. When it is found that the reliability of the result is low, the display color or print color of the recognition result of the character is changed.
It is possible to reduce the time and labor required for checking and correcting the recognition result, as compared with the related art.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a character recognition device of the present invention.

FIG. 2 is a block diagram showing a conventional character recognition device.

FIG. 3 is a diagram showing an example of extracting an original pattern and a sub-pattern.

FIG. 4 is a diagram showing an example of a collapsed pattern.

[Explanation of symbols]

 10 Read recognition unit 12 Optical read input unit 16 Photoelectric conversion unit 18 Pattern register 20 Line width calculation unit 22 Character frame detection unit 24 Sub pattern extraction unit 26 Feature matrix extraction unit 28 Identification unit 30 Character name output terminal 32 Control unit 50 Read Part 60 Recognition part 72 Character line width validity judgment part 74 Average line width calculation part 76 Line width register 80 Display part 82 Printing part

Claims (1)

[Claims] 1. A reading unit for photoelectrically converting a character pattern described on an input medium and outputting character pattern data,
A recognition unit for recognizing the character pattern data, and a display unit for displaying the recognition result or a character recognition device including a printing unit, wherein the reliability of the character recognition result is determined based on the line width of the character pattern data. A character recognition device, comprising: a character line width validity determining unit, and visually changing a method of displaying or printing a recognition result based on the reliability of the character recognition result.