JPH0581477A - Character segmenting method - Google Patents

Abstract

PURPOSE:To perform correct segmentation at all times even in a document which has a separate character or contact character in a document image of indeterminate character side. CONSTITUTION:A temporary character whose line-directional size is larger than estimated character side is detected as the contact character, it is decided whether the character before the contact character is a separate character or nonseparate character, and the segmentation head position of the contact character is determined. Further, the widths of the characters before and behind the character are referred to so as to decide whether part of the separate character is segmented independently or not as to a obtained temporary character string, and temporary characters are mutually coupled according to the decision 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 (OCR) for optically reading a document having an irregular pitch and converting image data of characters contained in the document into a character code.
The present invention relates to a character cutout method that can correctly cut out each character even when a separated character or a character that contacts an adjacent character exists in a document.

[0002]

2. Description of the Related Art In OCR, in order to convert image data into character codes, it is necessary to cut out the image data on the paper for each character and perform character recognition. As a character cutting technique for that purpose, first the character string is separated by a cutting candidate line, and the average pitch estimated based on the shape of the character string (pitch means the interval between one character and the next character) There is known a method of searching for a pair of candidate lines in which the intervals between the candidate lines are substantially equal to the average pitch and cutting out a non-separated figure included in the pair of candidate lines as one character (Japanese Patent Publication No. 2-23905). See the bulletin).

Furthermore, for a figure included in an uncut portion adjacent to an already cut area, an error correction method is also known in which the distances to the front and rear cut areas are measured and integrated into one of the areas. (See the same publication). This error correction method is shown in FIG. 8 (the same as FIG. 6 of the publication).
Will be explained. For example, when cutting out the vertical letters "Tokyo", "east" was cut out correctly, but "Kyo"

[0004]

[Outer 1]

It is assumed that only one is cut out as one character. That is,

[0006]

[Outside 2]

[0007]

[0008]

[Outside 3]

It is assumed that they are separated into and. In such a case, first, the distances p2 and p3 from the figure Q sandwiched between the cutout areas to the upper and lower cutout areas are measured, and when the p2 is smaller than a constant value ρ7, the figure Q is set in the upper area, p2
If is smaller than a certain value ρ 7, the figure Q is integrated into the lower area. If p2 and p3 do not satisfy certain conditions, the reject processing is performed by clearing the cut-out flag and the cut-out position flag in the upper and lower areas.

By performing the processing as described above, a character string in which the character pitch is not given, a character string in which the characters themselves are equal in size, but the characters are not arranged at a constant pitch, and the preceding and succeeding characters are completely eliminated. It is said that the character can be effectively and accurately cut out even for a character string including symbols having different pitches.

[0011]

However, a separated character (a character separated into left and right or top and bottom peculiar to kanji and kana in Japanese documents is a character such as "kawa" and "i" in horizontal writing. , In the case of vertical writing, such as "Kyo") and non-separated characters exist, and when the separated characters and the non-separated characters come into contact with each other, only the character cutting method as described above is effective. There is a problem that characters cannot be cut out.

For example, when the character string "Well, ready" is cut out, when the above character cutting method is applied, the portion "," remains as a figure sandwiched between the cutout areas. As shown in FIG. 9, if the distance between "te" and "," is p2, and the distance between "," and "quasi" is p3,
If p2 <ρ7 <p3, "," will be combined with "te" and the wrong cut out will be made.

Even if "preparation" is cut out correctly, the left small portion of "ga-sei" that follows

[0014]

[Outside 4]

If the front-to-back intervals p5 and p6 are in the relation of p5 <ρ7 <p6 (see FIG. 10), they are combined with "bi" and

[0016]

[Outside 5]

Therefore, the cutout is erroneous. Alternatively,

[0018]

[Outside 6]

It is also possible to cut out as one character. Therefore, an object of the present invention is to solve the above technical problem, and in a document in which separated characters and non-separated characters are mixed and the characters are in contact with each other, even when the character size is unknown or the character size varies greatly, It is an object of the present invention to provide a character cutout method capable of accurately cutting out each character required for character recognition.

[0020]

In order to achieve the above object, a character cutting method according to claim 1 estimates a character size W from a shape of a character string, obtains a projected image of the character string image data, and The cluster of pixels forming the projected image is cut out as a temporary character, and for the temporary character Ci whose width is larger than the character size W, the temporary character Ci-1 immediately before the temporary character Ci is a non-separated character. Whether or not it is a part of a separator character is at least the width of the dummy character Ci-1 and the dummy character C.
Judgment is made based on the interval before and after i-1, the cut-out position of the temporary character Ci is corrected according to the judgment result, and the cut-out position is cut out at a position which is behind the corrected cut-out position by the character size W or in the vicinity thereof. Whether or not a certain character Cj of the temporary character string obtained as a result of the cutout is a part of the separated character is determined by the temporary character Cj and the temporary character Cj adjacent before or after the temporary character Cj. -1 or temporary character Cj +
1 is determined based on the width of the temporary character Cj-1 and the distance between the temporary characters Cj-1 and Cj or the width of the temporary character Cj + 1 and the distance between the temporary characters Cj and Cj + 1, This is a method of combining temporary characters in accordance with the result of this determination.

In the character cutting method according to the second aspect of the present invention, the temporary character Ci that is regarded as a contact character is regarded as the temporary character Ci.
It is determined whether the temporary character Ci + 1 that is adjacent to the next character after is a non-separated character or a part of the separated character. Only the point of outputting is different from the method of claim 1.

[0022]

FIG. 1 is a block diagram for explaining the method of the invention described in claim 1. First, the character size W is estimated. For this purpose, for example, the character string image data is projected in the row direction, the row height is obtained from the projected image, and the average value or the maximum value of the row height for each row is obtained (steps N1 and N2). The line height is made equal to the character size because Japanese characters are almost square.

Next, the character string image data is projected in a direction perpendicular to the rows to determine temporary cut-out positions from the projected images, and the blocks of pixels forming those projected images are extracted as temporary characters (steps). N3). Further, the width of the provisional character is determined (step N4), and the provisional character Ci having a width of the provisional character larger than the character size W is cut out together with the contact character (because it is in contact with another figure, it is cut out together with the other figure. Character which is adjacent to the contact character and is a non-separated character or a part of the separated character.
Judgment is made based on the width of i-1 and the space before and after the temporary character Ci-1 (step N5). If it is a part of the separated character, the cut-out position of the temporary character Ci is corrected according to this determination result, and for example, character string image data in the vicinity of the position after the corrected cut-out position by the character size W is corrected. The contact character can be correctly separated by cutting out the projected image at the minimum position (step N6).

Next, it is judged whether or not each temporary character Cj is a part of the separated character in the temporary character string obtained as a result of being cut out by the above procedure. For example, if the temporary character Cj and the temporary character Cj-1 immediately before the temporary character Cj are separated by mistake, the width of the temporary character Cj-1 and the temporary character Cj-1 and the temporary character Cj-1. This is determined based on the distance from Cj, and the temporary characters are combined according to the determination result (step N7).

A second aspect of the present invention is a method applied when a character is cut out from the end of a line, and for the temporary character Ci regarded as a contact character, one of the temporary characters Ci is used.
Only the difference from the method of claim 1 is that it is determined whether the next adjacent dummy character Ci + 1 is a non-separated character or a part of the separated character.

[0026]

Embodiments will be described in detail below with reference to the accompanying drawings showing embodiments. FIG. 2 is a diagram illustrating an operation for implementing the character cutting method according to the present invention.
It is a block diagram which shows the basic composition of CR. This OC
R is a light source 2 for illuminating a reading surface 1 on which characters to be read are written, and an optical image formed on the light receiving surface 6a by the reflected light from the reading and the back surface as an electric signal in a handy housing. The scanner 4 is provided with a two-dimensional image sensor 6 for conversion.

When the scanner 4 is applied to the reading surface 1, the reflected light from the reading surface 1 is reflected by the reflecting mirror 3 and further guided to the light receiving surface 6a of the image sensor 6 through the lens system 5, and this light receiving surface 6a. An optical image of the character is formed on 6a. The image sensor 6 converts this optical image into an electric signal and gives it to the binarization circuit 7. The binarization circuit 7 discriminates the output signal of the image sensor 6 by an appropriate threshold value to determine "1" or "0".
To a binary signal.

The output signal of the binarization circuit 7 is applied to the main body processing section 8 by decoding the code 9. The main body processing unit 8 temporarily stores the signal from the code 9 in the image memory 81.
With respect to the image stored in the image memory 81, the character is cut out by the one-character cutout circuit 82 by a predetermined process by the CPU, and the character recognition processing is performed on each image cut out by the character recognition circuit 83. Is done. This recognition processing is performed by calculating the difference between the character pattern and the characteristic amount of the character pattern stored in advance in the recognition dictionary, and the character is recognized. Then, the recognition result is given to the editing unit 84, and if the recognition result is correct, the editing unit 84 outputs the result, and if it is incorrect, the recognition result is ignored and the next reading is started.

Of the series of processes described above, the character cutout process performed by the one-character cutout circuit 82 will be described in more detail with reference to the flowchart. The character cutout process is roughly classified into a contact character detection process, a preceding tentative character determination process, a contact character separation process, and a separated character integration process. The contact character detection, the immediately preceding tentative character determination process, and the contact character separation process are performed by the procedure of FIG. First, the image data stored in the image memory 81 is projected in the row direction, and each portion divided by white pixels is set as one row. Find the maximum height of each line and use this as the character size Wma
Let x.

Next, each row is projected in a direction perpendicular to the row direction to cut out a set of black pixel blocks as a temporary character string. From this temporary character string, a character having a width W in the row direction larger than the character size Wmax is regarded as a contact character (step S1). Taking Fig. 4 as an example, "Well,
"Preparation" of "where you are ready"

[0031]

[Outside 7]

And are contact characters. A dummy character regarded as a contact character is represented by Ci, and a previous dummy character in the same line is represented by Ci-1. Then, it is determined whether the temporary character Ci-1 is a full-width non-separated character (step S2), a half-width character (step S3), or a second half of the separated character (step S4). This judgment is determined based on the following numerical values. The width Wi-1 of the temporary character Ci-1, the temporary character Ci-1 and the temporary character Ci that is one behind the temporary character Ci-1.
And the space Si-1, the width Wi-2 of the temporary character Ci-2 immediately before the temporary character Ci-1, the temporary character Ci-2 and the temporary character Ci- immediately after the temporary character Ci-
The interval Si-2 from 1, the width Wi-3 of the temporary character Ci-3 that is two characters before the temporary character Ci-1, the temporary character Ci-3 and the temporary character Ci- that is one character after the temporary character Ci-
Distance from 2 Si-3 First, the full-width non-separated character is determined by the temporary character Ci-
It is determined whether or not the width Wi-1 of 1 in the row direction is Wi-1> Wmax xP1 (1). Here, the parameter P1
Is a parameter for determining full-width non-separated characters, and 1
It is set to a smaller number close to 1, for example, 0.72.

If Wi-1 ≤ Wmax × P1 (2), it is determined that the character is not a full-width non-separated character, so it is determined whether it is a half-width character. This determination is made by Wi-1 + Si-1> Wmax × P2 (3) or Si-1> Wi-1 × P3 and Si-1> Wmax × P4 (4) or Wmax × P5 <Si-2 + Wi- It is judged whether any one of the judgment formulas of 1 + Si-1 <Wmax × P6 (5) is satisfied. The parameter P2 determines whether or not the total of the temporary character Ci-1 and the width Si-1 of the blank portion immediately after it is approximately one character size, and is set to a number smaller than 1 and close to 1, for example, 0.83. .. The parameter P3 determines whether or not the width Si-1 of the blank portion immediately after the provisional character Ci-1 is almost equal to the width Wi-1 of the provisional character in order to call it a half-width character.
It is set to a number smaller than 1 and close to 1, for example, 0.85.
Since the parameter P4 is a half-width character, the dummy character C
It determines whether the width Si-1 of the blank portion immediately after i-1 is extremely smaller than the character size Wmax, and is set to 0.18, for example. The parameters P5 and P6 determine whether or not the sum of the width Wi-1 of the temporary character Ci-1 and the widths Si-2 and Si-1 of the blank portions immediately before and after that is approximately one character size. P5 is smaller than 1 and 1
Is set to, for example, 0.85, and the parameter P6 is set to a number larger than 1 and close to 1, for example, 1.2.

FIG. 5 shows an example determined by the equation (4). In FIG. 5, the temporary character C2 immediately before the contact character C3
The width W2 of "," and the width S2 of the blank part immediately after
Is substituted into Eq. (4). Since S2> 0.85W2 and S2> 0.18Wmax are satisfied, it is understood that the dummy character C2 is a half-width character.

If none of the above judgment formulas (3), (4) and (5) is satisfied, it cannot be judged as a half-width character.
Determine if it is the second half of the separator. This is Wmax × P7 <Wi-2 + Si-2 + Wi-1 ≦ Wmax × P8 (6) Wi-1> Wmax × P9 (7) Wi-3 + Si-3 + Wi-2> Wmax (8) It is determined by whether the expressions are satisfied at the same time.

Here, the parameters P7 and P8 are temporary characters C.
It determines whether or not the total width of the temporary character Ci-2 immediately preceding i-1 and the space immediately after it and the temporary character Ci-1 is approximately one character size, and the parameter P7 is smaller than 1. 1
Is set to, for example, 0.68, and the parameter P6 is set to a number close to 1, for example, 1.0. Parameter P9
Determines whether or not the temporary character Ci-1 has a width that can be said to be the latter half of the separated character, and is set to 0.18, for example. The formula (8) determines whether the total width of the temporary character Ci-3, which is two digits before the relevant temporary character Ci-1, and the subsequent blank portion, and the subsequent dummy character Ci-2, is approximately one character size. It is an expression.

On the basis of the above-mentioned criteria, the immediately preceding character C
It can be seen that i-1 is a full-width non-separated character, a half-width character, or the second half of the separated character. If NO in any of steps S2, S3 and S4, the preceding character Ci-1 is the first half of the separated character and should be combined with the character Ci. I understand. Figure 5
In the example, it is the temporary character C6 '.

[0038]

[Outside 8]

The temporary character C5 'one before

[0040]

[Outside 9]

It is understood that none of the conditions of steps S2, S3 and S4 correspond to the provisional character C5 'being the first half of the separated character. Next, the contact character is separated. Therefore, the cutting start position is determined (step S in FIG. 3).
5, 6). If the temporary character Ci-1 is a full-width non-separated character, a half-width character, or the latter half of the separated character, the cut-out start position is performed from the start of the temporary character Ci (step S5). In the example of FIG. 5, since the temporary character C2 is a half-width character, the cutout start is from the start A of the temporary character C3.

If the temporary character Ci-1 is the first half of the separated character, the cut-out head position is from the head of the immediately preceding virtual character Ci-1 (step S6). In the example of FIG. 5, since the temporary character C5 'is the first half of the separated character, the cutout head is corrected to the head B of the temporary character C5'. The range of the cutout position is a range from the cutout start position to Wmax × P10 to Wmax. Here, the parameter P10 is, for example, 0.
Set to 9.

When the range of the cutout position is determined as described above, the actual cutout position is uniquely determined. For this purpose, for example, the position where the projected value of the black pixel is the smallest may be determined as the contact character separation position (step S7). The separation process in step S7 will be described in detail with reference to the flowchart (FIG. 6). The position of Wmax x P10 from the cutting _start position is x _start , and the position of Wmax from the cutting _start position is x
Define as _end . The projection value of the black pixel at the position x _start is h _start , the projection value of the black pixel at the position x _end is h _end , and the minimum projection value of the black pixels is h _min . In step S71, the parameter x is set to x _start and the parameter h is set to h _start . In step S72, the parameter x is increased by Δx, and in step S74 h _min is compared with the projection value h _x of the black pixel at this x. If h _{mi n}
If> h _x , this h _x is rewritten to h in step S75.
_min , and x is set to x _ext in step S76. Then, the process returns to step S72, the parameter x is increased by Δx again, and the same processing as described above is repeated. Step S7
If x exceeds x _{end in} 3, the black pixel at the position of x _ext at that time is converted into a white pixel, and the process ends (step S
77).

By this processing, the position where the actual cutting is performed can be determined. For example, referring to FIG. 5, from the beginning A of the temporary character C3 to Wmax × P10 or Wma.
The minimum position x _{ext in} the range up to x becomes the new cut-out position, and the minimum position x _ext ′ in the range from the beginning B of the temporary character C5 'to Wmax × P10 to Wmax becomes the new cut-out position.

In the character string obtained as described above, some of the separated characters are still mixed. Therefore, the character separation process is performed by using the character size Wmax. That is, paying attention to the temporary character Cj whose width Wj in the row direction is Wj <Wmax × P11, and the distance Sj from the next temporary character Cj + 1 is Sj <Wmax × P12 and Wj + Sj + Wj + 1 ≤Wma.
If x, the temporary character Cj and the next temporary character Cj + 1 are combined to determine a new character.

The distance Sj-1 from the preceding provisional character Cj-1 is Sj-1 <Wmax xP12 and Wj-1 + Sj-1 + Wj ≤
If it is Wmax, the temporary character Cj and the previous temporary character Cj-1 are combined and determined as a new character. Here, the parameter P11
Is a number smaller than 1 for determining whether the temporary character Cj is a part of the separated character and is set to, for example, 0.65.
The parameter P12 is a temporary character Cj-1, which is adjacent to the temporary character Cj.
It is a number for determining whether the interval with Cj + 1 is sufficiently small, and is set to, for example, 0.26.

A concrete example is shown in FIG. Figure 7
In the above, paying attention to the temporary character C3, ",", it is assumed that the distance S3 from the next temporary character C4 satisfies S3 <0.26 Wmax. However, since W3 + S3 + W4 is larger than Wmax, the temporary character is C3, ",".
It can be seen that the following temporary character C4, "quasi", should not be combined.

Next, in FIG. 7, it is a temporary character C6.

[0049]

[Outside 10]

Paying attention to the following, the space S6 between the next temporary character C7 and
Satisfies S6 <0.26 Wmax and W6 + S6 + W7 <Wmax. In this case, it is the temporary character C6

[0051]

[Outside 11]

And the next temporary character C7

[0053]

[Outside 12]

"And" can be made by combining and. Further, in FIG. 7, it is a temporary character C7.

[0055]

[Outside 13]

Paying attention to the above, since the distance S6 from the previous provisional character C6 satisfies S6 <0.26Wmax and W6 + S6 + W7 <Wmax,

[0057]

[Outside 14]

The same result can be obtained by combining with and forming "ga". In the above embodiment, the example of cutting out from the front end of the line was dealt with, but when cutting out from the rear end of the line, it becomes the same as viewing the horizontally written characters from the back side, and the completely same procedure. Can be processed by.

[0059]

As described above, according to the character cutting method of the first aspect, a temporary character having a width in the line direction larger than the estimated character size is detected as a contact character, and the character before the contact character is a separated character. Since it is determined whether the character is a non-separated character and the cut-out leading position of the contact character is determined, the contact character can be accurately separated.
In addition, with respect to the temporary character string thus obtained, it is determined whether or not a part of the separated characters is cut out independently, by referring to the widths of the preceding and following characters, and according to the determination result, It is also possible to combine characters. Therefore,
Even in a document in which a separated character or a contact character is included in a document image having an indefinite character size, correct clipping can always be performed.

According to the character cutout method of the second aspect, the correct cutout can always be performed even when the cutout is performed from the rear end of the line.

[Brief description of drawings]

FIG. 1 is a block diagram illustrating a procedure of the present invention.

FIG. 2 is a block diagram showing a basic configuration of an OCR that implements the character cutout method of the present invention.

FIG. 3 is a flowchart showing an outline of contact character detection, immediately preceding provisional character determination processing, and contact character separation processing.

FIG. 4 is a diagram showing a specific example of a contact character detection process.

FIG. 5 is a diagram showing a specific example of correction of a cutout position of a contact character and separation processing of the contact character.

FIG. 6 is a flowchart showing a procedure for determining a separation position of a contact character.

FIG. 7 is a diagram illustrating a specific example explaining an integration process of a part of separated characters.

FIG. 8 is an uncut portion Q in the conventional character cutting method.
It is a figure which shows the specific example explaining the integration method of.

FIG. 9 is a diagram showing a specific example for explaining the problems of the conventional character cutting method.

FIG. 10 is a diagram showing a specific example for explaining the problems of the conventional character cutting method.

[Explanation of symbols]

 1 Reading Surface 4 Scanner 8 Main Processing Unit 81 Image Memory 82 1 Character Cutout Circuit

Claims (2)

[Claims] 1. An image signal representing a read object including characters is acquired, and image data for each character to be recognized is cut out from character string image data corresponding to a character string included in the acquired image signal. In the character cutout method,
A character cutting method characterized by adopting steps (a) to (d). (a) The character size W is estimated from the shape of the character string. (b) Obtain the projected image of the character string image data, and cut out the blocks of pixels forming the projected image as temporary characters. (c) Regarding a temporary character Ci whose width is larger than the character size W, determines whether the adjacent temporary character Ci-1 before the temporary character Ci is a non-separated character or a part of the separated character. At least the width of the temporary character Ci-1 and the space between the temporary character Ci-1 and the virtual characters before and after the temporary character Ci-1 are used for determination, and the cutout position of the temporary character Ci is corrected according to the result of the determination, and the corrected cutout The character size W is cut out from the position at or after the position. (d) Whether the temporary character Cj of the temporary character string obtained as a result of being cut out in the step (c) is a part of the separation character is determined by the temporary character Cj and the previous or one character of the temporary character Cj. Regarding the temporary character Cj-1 or the temporary character Cj + 1 that is adjacent to the next one, the width of the temporary character Cj-1 and the temporary characters Cj-1 and Cj
And the width of the temporary character Cj + 1 and the distance between the temporary characters Cj and Cj + 1, and the temporary characters are combined if necessary according to the result of this determination. 2. An image signal representing a read object including characters is acquired, and image data for each character to be recognized is cut out from character string image data corresponding to a character string included in the acquired image signal. In the character cutout method,
A character cutting method characterized by adopting steps (a) to (d). (a) The character size W is estimated from the shape of the character string. (b) Obtain the projected image of the character string image data, and cut out the blocks of pixels forming the projected image as temporary characters. (c) For a temporary character Ci whose width is larger than the character size W, the temporary character Ci + 1 immediately after the temporary character Ci and adjacent to the temporary character Ci + 1
Is a non-separated character or a part of the separated character, based on at least the width of the dummy character Ci + 1 and the space between the dummy character Ci + 1 and the preceding and following dummy characters. Accordingly, the cut-out position of the temporary character Ci is corrected, and the cut-out position is cut out at a position preceding the corrected cut-out position by the character size W or in the vicinity thereof. (d) Whether the temporary character Cj of the temporary character string obtained as a result of being cut out in the step (c) is a part of the separation character is determined by the temporary character Cj and the previous or one character of the temporary character Cj. Regarding the temporary character Cj-1 or the temporary character Cj + 1 that is adjacent to the next one, the width of the temporary character Cj-1 and the temporary characters Cj-1 and Cj
And the width of the temporary character Cj + 1 and the distance between the temporary characters Cj and Cj + 1, and the temporary characters are combined if necessary according to the result of this determination.