JPH04130979A - Character picture segmenting method - Google Patents

Abstract

PURPOSE:To obtain the character frame of a specified character composed of only line segments vertical in a row direction by deciding a condition concerning the length and integrated width of a character element circumscribed rectangle having width narrower than a prescribed value and a following character element circumscribed rectangle. CONSTITUTION:At a character picture segmentation processing part 2, the linkage of black picture elements on a document picture is searched, the group of linked black picture elements is extracted as a character element, and the coordinates of the circumscribed rectangle of the character element (such as the coordinates of a diagonal apex, for example) are extracted as a character element data. On the other hand, the horizontal and vertical distances between the circumscribed rectangles of the character elements are calculated, the group of the character element circumscribed rectangles having the both distances smaller than a certain threshold value is segmented as one row, and the data is stored in a character element data memory 7 as well. A parameter setting part 8 monitors the entire character element data in the character element data memory 7, sets a relative parameter used for character segmentation (character element integration and character pattern synthesization) and stores the set parameter in a parameter memory 9. According to the condition decision using the parameter in the parameter memory 9, a character element integration part 10 starts the integration of the character element circumscribed rectangles, which are regarded as the same character, successively from the character element at the head of the row so as to correct the character element data in the character element data memory 7.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for cutting out a character image from a document image input as a binary image in a character recognition system.

[Conventional technology]

In character recognition systems, two
Lines are cut out from the document image input as a value image, character images are cut out for each line, and character recognition is performed.

Methods for cutting out character images are roughly divided into a method using projection and a method using a circumscribed rectangle of connected black pixels.

However, no matter which method you use, a block of projection or a single circumscribed rectangle does not always directly turn into data for one character, so there is no technology to combine or separate the data to obtain single character data. is necessary.

Conventionally, the circumscribed rectangle of connected black pixels (graphemes) is extracted, and the width of a single circumscribed rectangle or a combination of multiple circumscribed rectangles is compared with a preset standard character width, and the width that can be considered as one character is calculated. A method for cutting out a character image using a circumscribed rectangle or a combination thereof as a character frame (Japanese Patent Laid-Open No. 117670/1983
No.) is known.

There is also a known method in which a single circumscribed rectangle or a combination of regions is regarded as a character image, character recognition is performed, and the character image is determined by evaluating the certainty of the result (patented in 1988 by the same applicant). Application No. 133424).

[Problem to be solved by the invention] Characters consisting only of vertical line segments, such as the vertically written Chinese numerals "-1, 2, 3" and the horizontally written kanji "kawa", are
They often do not have the width of one character, and if they appear consecutively, a certain recognition result can be obtained no matter where they are separated.

Therefore, when such characters appear consecutively, cutting errors are likely to occur in the conventional method, even if the preceding and following blank widths are considered.

SUMMARY OF THE INVENTION An object of the present invention is to provide a character image cutting method that can accurately cut out even when character images such as Chinese numerals consisting only of line segments perpendicular to the row direction appear consecutively.

1. Means for Solving the Problems] The present invention provides a character image cutting method in which a character frame is obtained by integrating grapheme circumscribing rectangles extracted from a line image, and a character image is cut out from a line image. Focusing on the smaller grapheme circumscribing rectangle, a conditional judgment is made regarding the length and width of the drawn grapheme circumscribing rectangle and the subsequent grapheme circumscribing rectangle, and the drawn grapheme is determined according to the result of the condition judgment. By controlling the integration of a circumscribing rectangle and a subsequent grapheme circumscribing rectangle, a character frame for a specific character consisting only of line segments perpendicular to the line direction is obtained.

[Function] In the case of a specific character consisting only of line segments perpendicular to the line direction, such as the vertically written Chinese numeral ``-23'', each line segment that makes up each character is extracted as a grapheme, but The width of the circumscribing rectangle is sufficiently small, and the length of each line segment has a proportional relationship specific to the character. The overall width of a character is also character specific.

Therefore, according to the present invention described above, when the vertically written character string "-23" is taken as an example, it is possible to accurately cut out the character string as follows.

First, the width of the glyph-eme circumscribing rectangle for "-" is sufficiently small, so we will pay attention to it, but the width when integrated with the next glyph-eme circumscribing rectangle (corresponding to the upper line segment of "2") is Due to the presence of spaces, the standard spacing between characters is too large, so
The noticed grapheme circumscribing rectangle is determined to constitute a separate character from the next grapheme circumscribing rectangle and is not integrated, and the image of "-" can be cut out using the noticed grapheme circumscribing rectangle as a single character frame. can.

Next, we will focus on the grapheme circumscribing rectangle corresponding to the upper line segment of "2", but the character width is the standard character width since there is no inter-character space between the characters. It's not too much. Furthermore, the length (length of a line segment) of the character element circumscribing rectangle of interest is shorter than the length of the next character element circumscribing rectangle. Based on these conditions, it is possible to determine that the two character element circumscribing rectangles constitute "2" and integrate them, and cut out the image of "2" using the integrated rectangle as a character frame.

Next, pay attention to the grapheme circumscribing rectangle corresponding to the top line segment of "three". The width integrated with the next grapheme circumscribing rectangle is not excessive as a standard character width, but the relationship in length is the opposite of "=". Furthermore, the integrated width up to the next grapheme circumscribing rectangle is not excessive as a standard character width, and the length of the last grapheme circumscribing rectangle is larger than the length of the central grapheme circumscribing rectangle. Based on these conditions, it is determined that the circumscribed rectangle of the character element 311 constitutes r3J, and they are integrated, and the image of 3J can be cut out using the integrated rectangle as a character frame.

〔Example〕

FIG. 1 is a block diagram of a character recognition system according to an embodiment of the present invention.

A document image to be processed is input as binary image data from a scanner or the like, and a 0 character image cutting unit 2 stored in the original image memory 1 cuts out character images of the document image in the original image memory 1. The 0 character recognition processing unit 4 performs processing and stores the extracted character image in the character image memory 3.
The characters are recognized by normalizing the character image, extracting features, and comparing it with a dictionary, and the recognition results are stored in the recognition result memory 5.

In the character image extraction processing section 2, a line extraction section 6 extracts one line of character elements from the document image and stores the character element data in the character element data memory 7. For example, by searching for connections of black pixels on a document image, extracting a block of connected black pixels as a grapheme (a group of black pixels that make up a character or a part of a character), The coordinates of the diagonal vertices) are extracted as character element data. Additionally, the horizontal and vertical distances between the grapheme circumscribing rectangles are determined, and a collection of grapheme circumscribing rectangles whose distance in both directions is smaller than a certain threshold is cut out as one line, and that data is also stored in the grapheme data memory 7. . Such line cutting is more likely to cause document skew than line cutting by line direction projection.

Parameter setting 8B looks over the entire grapheme data in the grapheme data memory 7, sets relative parameters for use in character extraction, grapheme integration, and character pattern synthesis, and stores them in the parameter memory 9. Although the grapheme integration is performed multiple times, the parameters are reset based on the revised grapheme data in the grapheme data memory 7 each time the process is completed.

The parameters to be set are as follows.

(a) Line height (distance from the bottom edge to the top edge of a horizontally written line, distance from the vertically written left edge to right edge of a vertically written line) This is the maximum height of the circumscribed rectangle of the character element forming the line. However, the height of a character element circumscribing rectangle is the distance from the bottom edge of a horizontally written line, and from the left side of a vertically written line to the farthest side of the character element circumscribing rectangle.

(b) Standard character width (width is the size in the line direction) This is a value slightly larger than the line height, for example, [line height] + 1 (dot).

(c) Minimum character spacing This is the minimum interval between adjacent grapheme circumscribing rectangles.

However, only character element circumscribing rectangles with a certain width, for example, half or more of the standard character width, are determined.

Note that before the first character element integration process, the following initial value is set.

[Initial value ko=-[line height ko/30 (a negative value means that it overlaps the grapheme circumscribing rectangle) (d) Maximum character width The grapheme circumscribing rectangle that makes up the line has the maximum width. do.

(e) Knth This is the maximum width of a grapheme circumscribing rectangle that can be considered as a grapheme consisting of only one line segment perpendicular to the line, for example, [line height]
/8.

The grapheme integration unit 10 integrates grapheme circumscribing rectangles that are considered to be the same character, starting from the first grapheme in a line, based on conditional judgment using parameters in the parameter memory 9, and Modify grapheme data. This character element integration processing consists of a normal processing mode and an exception processing mode.

First, the contents of the exception handling mode will be explained.

FIG. 2 is a flowchart for explaining the exception handling mode. However, Wl is the character width of the grapheme of interest,
W2 is the character width of the grapheme next to the grapheme of interest, W3
is the character width of the grapheme next to W2, Ll is the length of the grapheme of interest, L2 is the length of the grapheme next to the grapheme of interest, L3 is the length of the grapheme after W2 , Wl, 2 is the character width when the grapheme of interest and the next grapheme are integrated, W
l.

2.3 is the character width when the grapheme of interest and the two following graphemes are integrated.

Certain characters that are broken down into graphemes consisting of one line segment perpendicular to the line direction, such as ``kawa'' in a horizontally written line or ``-23'' in a vertically written line, and whose graphemes have wide intervals, are If the prime spacing is sufficiently wider than the grapheme spacing of other characters, it is possible to extract in the normal processing mode, but there are few documents that satisfy such conditions. This exception processing mode is a mode in which character element integration processing is performed to accurately extract such specific characters.

The character width of the grapheme 1 of interest (size in the row direction of the grapheme circumscribing rectangle) wl is compared with the parameter Knth (step ■), and if w 1 < Knth,
The character element l of interest is assumed to consist of only one line segment perpendicular to the line direction, and the exception handling mode is entered.

When entering the exception handling mode, first, a comparison is made between the character width 1.2 when the noticed grapheme 1 and the next grapheme 2 are integrated and the standard character width (step ■), and wl, 2 ≥ product standard character. If it is the width, the character element 1 of interest is determined as a single character element. For example, in the character string "-23" in a vertical line as shown in Figure 3, if we pay attention to the first character "-", we can see that wl,
Since it has a standard character width of 2, it is determined as a single character element.

In step ■, wl, 2 (when the standard character width is reached, the length of the grapheme 1 of interest (size in the direction perpendicular to the row direction of the circumscribing rectangle) Ll and the length of the next grapheme 2 L2 (step ■), and if L2>Ll, character element 1
and grapheme 2 are integrated to form one grapheme. For example, the grapheme of the line segment above “two” shown in Figure 3 is the grapheme of interest 1.
If so, the condition of step (■) is satisfied, so the two graphemes constituting "2" are definitively integrated into one grapheme.

When L2≦L1 in step ■, the noted grapheme 1
, compare and judge the character width w1.2.3 that integrates the next grapheme 2 and the next grapheme 3 with the standard character width (step ■), and if wl,2.3<standard character width, the character A comparison is made between the length L2 of element 2 and the length L3 of character element 3 (step ■), and L3) if L2, character elements 1.2.3 are definitively integrated. For example, if the grapheme of the top line segment of [3J shown in Figure 3 is the noted grapheme l, the condition of step ■ is satisfied, so the three graphemes that make up ``3'' are Deterministically integrated.

As is clear from the above explanation, the grapheme integration result shown in Figure 4 is obtained for character strings that are difficult to extract in the normal processing mode shown in Figure 3, so each character can be correctly extracted. .

Next, the contents of the normal processing mode will be explained using FIGS. 5 to 9. However, for convenience, it is assumed here that the lines are written horizontally.

■) Two adjacent glyph circumscribing rectangles a and b are conditional expressions (1
) and (2) are satisfied at the same time, they are integrated into one glyph-eme circumscribing rectangle (see Figure 5).

Wab≦[Standard character width] Conditional expression (
1) However, Wab is the character width of the integrated rectangle of a and b.

X5b-Xea<[Minimum character spacing Conditional expression (2) where Xsb is the X coordinate of the starting point of b, and Xea is the X coordinate of the ending point of a.

Conditional expression (1) indicates that the width of the grapheme circumscribing rectangle after integration does not exceed Aab [standard character width], and conditional expression (2) indicates that a,
This indicates that the interval Gab of b is smaller than the [minimum character interval].

However, if conditional expression (3) is met, the characters are integrated even if the standard character spacing is exceeded.

Xeb (Xeb Conditional expression (3) where Xeb is the end point xll mark of b.a,
Assume that the positional relationship of b is that a is 61 on the left (near the beginning of the line) and b is on the right.

That is, as shown in FIG. 6, two completely overlapping grapheme circumscribing rectangles a and b where Dab<O are combined even if Wab≧[standard character width].

2) In 1) above, if conditional expression (1) is satisfied but conditional expression (2) is not satisfied, the right grapheme circumscribing rectangle (referred to as b) satisfies all of the following conditional expressions (4) to (7). If it satisfies the above criteria, it is considered to be a circumscribed rectangle of a voiced or handakuten character and is exceptionally integrated (see Figure 7).

wb≦fth Conditional expression (4) Hb≦fth
Conditional expression (5) (Hb is the height of b) Yeb -TsaS f th
Conditional expression (6) (Yeb is the end point y coordinate of b, Yea is a
starting point y coordinate)Xsb-Xea((minimum character spacing + α
Conditional expression (7) where fth (threshold) = [row height]
/β. α and β are empirical values, for example α=2, β
=3 is selected.

3) Although conditional expressions (1) and (2) are satisfied in 1) above, if the right side grapheme circumscribing rectangle (referred to as b) falls under the following conditional expressions (8) to (10), punctuation marks should be used. It is assumed that there is such a thing and it is not integrated as an exception (see Figure 8).

wb≦fth Conditional expression (8) Hb≦fth
Conditional expression (9) base-Yeb≦f th/7
Conditional expression (10) where base is the baseline yJ! ! The target, 7, is an experience value and is selected as 4, for example.

Each time the above-described integration process in the normal processing mode is completed for one line of grapheme data (excluding the grapheme determined in the exception processing mode), the parameter setting unit 8 restarts the process based on the revised grapheme data. is set, and the parameter memory 9 is rewritten. Using the rewritten parameters, the grapheme integration unit 10 re-executes the process in the normal process mode.

In this way, the grapheme integration process is repeated while dynamically modifying the parameters. This repetition ends, for example, when the parameter no longer changes, or when the amount of change in the parameter becomes less than or equal to a certain threshold.

FIG. 9 shows an example of character element integration in the normal processing mode for horizontally written lines. The line cutting unit 6 extracts character elements as shown in (b) from the line image shown in (a). These graphemes are integrated as shown in (c) in the first integration process, and as shown in (d) in the second integration process.

At the stage when the integration process is completed as shown in (d),
Most of the grapheme circumscribing rectangles correspond to the circumscribing rectangle (character frame) of one character. That is, circumscribed rectangles of most characters are generated. However, in this example, graphemes that should be integrated may remain separated, such as the small J.

Since there is a possibility that such unintegrated graphemes may remain,
In the character pattern synthesis section 11, the character image extraction processing unit 2 does not simply cut out a character image from the original image using each character element circumscribing rectangle after the integration process as a character frame, but instead extracts the character image from the original image by using continuous character element circumscribing rectangles as [maximum The characters are combined to the extent that they do not exceed the character width (stored in the parameter memory 9), and each of the combined rectangles (including the circumscribed rectangle of one grapheme) is used as a character frame to cut out character image candidates from the original image. Output to image memory 3. However, for a character element determined in the exception processing mode of character element integration processing, only character image extraction is performed for that character element alone.

The character recognition processing section 4 performs character recognition on the character image candidates stored in the character image memory 3. however,
For character image candidates with overlapping graphemes, the recognition results of each are evaluated, and the character image candidate that is most likely to be a character is selected (in other words, character extraction is determined based on the character recognition results), and only that recognition result is selected. The recognition result is output to the memory 5.

The method of recognizing a combination of such graphemes and determining character segmentation by evaluating the recognition results is detailed in the specification and drawings of the patent application No. 133424 of 1988.

〔Effect of the invention〕

As explained above in detail, according to the present invention, when specific characters consisting only of vertical line segments (for example, Chinese numerals r-23J in vertical lines) appear consecutively, too,
Accurate character image extraction becomes possible.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a character recognition system according to an embodiment of the present invention, FIG. 2 is a flowchart for explaining an exception handling mode for character element integration, and FIG. 3 is a character string subject to the exception handling mode. FIG. 4 is a diagram showing the result of the character string integration process shown in FIG. 3. FIGS. The figure shows an example of grapheme integration. l...Original image memory, 2...Character image cutout processing unit, 3...Character image memo 1c 4...Character recognition processing unit,
5... Recognition result memory, 6... Line cutting section, 7.
...Characteristic data memo 1c8...Parameter setting section, 9.Parameter memory, 10.Characteristic integration section, 11.Character pattern synthesis section. 3rd ward, 4th figure, 3rd ward, 4th figure, 9th figure, W Kyu bar, middle axe, little horse girl, abbess, 9th Qo-h <[Short text W gift] tz W ≦Chito h 1-1t, j 5h Lctb'=fsoh <

Claims (1)

[Claims] (1) In a character image extraction method in which a character frame is obtained by integrating character element circumscribing rectangles extracted from line images and character images are extracted from line images, attention is paid to character element circumscribing rectangles whose width is smaller than a predetermined value, A conditional determination is made regarding the length and width of the noted grapheme circumscribing rectangle and the following grapheme circumscribing rectangle, and the width of the noted grapheme circumscribing rectangle and subsequent grapheme circumscribing rectangle is determined according to the result of the conditional judgment. A character image cutting method characterized by obtaining a character frame of a specific character consisting only of line segments perpendicular to the line direction by controlling integration.