JPH04352295A - System and device for identifing character string direction - Google Patents

Abstract

PURPOSE:To correctly identify the direction of a character string on various documents including a document whose character spacing is larger than its line spacing since the direction of the character string is identified by detecting the state of blank lines (or blank column) obtained from image data of an input document. CONSTITUTION:The device consists of a means 6 which extracts the extent of horizontal character arrangement from the inputted image data, a means which extracts the extent of vertical character arrangement, and a means 5 which compares the extents of horizontal character arrangement and vertical character arrangement with each other to identify the character string direction in the image.

Description

[Detailed description of the invention]

[Purpose of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character string direction determining method for determining the direction of a character string written in an input document, and a character string direction determining device for determining the direction of a character string.

0002

2. Description of the Related Art When automatically reading documents written vertically or horizontally, it is necessary to determine the direction of character strings in the input document and determine the correct reading direction prior to character recognition processing. Conventional character string direction determination methods and devices are realized by estimating character spacing and line spacing from an input document and comparing them based on the empirical fact for printed documents that character spacing is smaller than line spacing.

By the way, in recent years word processors and DTP
As systems have become more widespread, it has become easier to create printed documents. Even if the documents created in this way use the same size paper, the relationship between the sizes of the character spacing and line spacing varies depending on the creator, and it cannot necessarily be said that the relationship ``character spacing < line spacing'' holds true. In such cases, a character string direction determination method that simply assumes the property that ``the character spacing is smaller than the line spacing'' has a problem in that a determination error occurs.

0004

[Problem to be Solved by the Invention] Conventional character string direction determination methods and devices have determined the character string direction based on the empirical fact that the line spacing is wider than the character spacing. There has been a problem in that it is not possible to correctly determine the direction of character strings in documents that are larger than the line spacing.

The present invention has been made in view of the above problems, and provides a character string direction determining method and a character string direction determining device that can accurately determine the direction of character strings even when reading a document in which the character spacing is larger than the line spacing. The purpose is to provide. [Structure of the invention]

[0006]

[Means for Solving the Problems] The present invention extracts character candidate rectangles from input image data, calculates the distance between character candidate rectangles that are adjacent to each other in the horizontal and vertical directions, and calculates the distance between character candidate rectangles that are not occupied by the character candidate rectangles. A character string direction determination method characterized by having means for detecting a state of a data area, and means for determining a character string direction from the detected state of the data area, and such a character string direction determination method A character string direction determining device is provided, which is characterized by performing character string direction determination.

[0007]

[Operation] According to the present invention, the degree of character arrangement in the horizontal and vertical directions is extracted, and based on the extraction result of the degree of character arrangement in the horizontal and vertical directions, parts other than character strings of the input document,
Since the direction of the character string is determined by detecting the state of so-called blank lines (or blank columns), it is possible to correctly determine the direction of the character string in various documents, including documents where the character spacing is larger than the line spacing. can.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram for explaining the character string direction determination method of the present invention. First, the procedure of the character string direction determination method of the present invention will be outlined.

[0009] For an input image inputted by an image input device such as an image scanner, a "black connected rectangle generation unit" 1 extracts a circumscribing rectangle of a portion of the image information in which black points are connected. Based on the circumscribed rectangle of the extracted black connected part,
The character size estimating unit 2 estimates the rectangular size of frequently occurring black connected rectangles or the average rectangular size of black connected rectangles as the size of character rectangles in the input document. When the size of the character rectangle is estimated, the "character candidate rectangle extraction unit" 3 compares the rectangle size of the previously obtained circumscribed rectangle of the black connected part and the estimated character rectangle,
Black connected rectangles close to the rectangular size of the estimated character rectangle are extracted as a character candidate rectangle group.

The "intercharacter space estimating unit" 4 calculates the distance between character candidate rectangles in the vertical and horizontal directions for pairs of character candidate rectangles that are adjacent in the vertical and horizontal directions from the extracted character candidate rectangle group. By performing statistical processing on this distance, for example, the mode or average value is estimated as the size of the space between characters in the vertical or horizontal direction in the input document. By comparing the vertical character spacing with the horizontal character spacing, it is determined that the characters are consecutively arranged in the direction with the narrowest character spacing. ”5, the direction of the character string is determined. In other words, the first character string direction determination section compares the vertical character spacing with the horizontal character spacing, and if the vertical character spacing is larger than the horizontal character spacing by a certain setting value, Determines that the character string direction of the input document is horizontal, and if the horizontal space between characters is larger than the vertical character space by a certain setting value, determines that the character string direction of the input document is vertical. However, if the difference in size between the vertical character space and the horizontal character space is less than a certain set value, the process moves to the next step without determining the character direction.

For example, as shown in FIG. 2, when character candidate rectangles are arranged horizontally and vertically with approximately the same character spacing, the first character string direction determination section described above does not determine the character string direction. Then, the data is sent to the "character sequence degree extraction section" 6 for detailed determination. Here, we can determine the direction of character strings in an input document by quantifying the state of the character arrangement in areas other than the blank area following the "end of the paragraph" (the shaded area in the figure). Estimate and determine. This is done by counting the number of character candidate rectangle pairs in which the interval between adjacent character candidate rectangles is not larger than the intercharacter space estimated by the intercharacter space estimating section 4 by more than a certain degree.

FIG. 3 is an enlarged view of the shaded area shown in FIG. Here, we will focus on character candidate rectangle A. Character candidate rectangle A and character candidate rectangle a or c horizontally adjacent to each other are considered to be adjacent to each other at a distance approximately equal to the estimated horizontal intercharacter space. These character candidate rectangle pairs are determined by the "degree of horizontal character alignment"
is counted. Further, the distance between character candidate rectangle A and character candidate rectangle b adjacent to it in the downward direction in the drawing is also considered to be approximately equal to the estimated vertical space between characters. This pair of character candidate rectangles is counted as the "degree of vertical character arrangement."

On the other hand, the character candidate rectangle d adjacent to the character candidate rectangle A in the upward direction in the drawing is spaced apart from the character candidate rectangle by a distance larger than the size of one character candidate rectangle. Character candidate rectangular pairs exhibiting an interval larger than the inter-character space to a certain extent are not counted in the degree of character arrangement in the vertical direction. This counting operation is performed for all character candidate rectangle pairs. Considering the character candidate rectangles adjacent to the shaded area in Figure 2, since the distance between character candidate rectangles that are vertically adjacent to each other is at least one character candidate rectangle, these are each counted as the degree of vertical character arrangement. Not done. Since the horizontally long blank lines shown in the drawings are counted only in the vertical direction and not in the horizontal direction, the connection direction of the character strings can be determined from the magnitude relationship of these counted values.

According to this method, even when it is difficult to determine the character direction due to the difference in the size of the vertical character space and the horizontal character space, character direction can be determined with high accuracy. becomes possible. The operation of each component will be explained in detail below.

The black connected rectangle generation unit 1 focuses on the black background data of the input image, generates a rectangle that circumscribes the shape of the continuous black background data, and calculates the coordinates of this rectangle as data. The input image is captured as binary white and black data corresponding to the scanning pitch of a reading device such as an image scanner. For example, an XY coordinate system is provided in the captured input image, and black background data is detected by scanning in the X direction. When black background data is first detected, it is detected whether black background data exists in a portion adjacent to this data in the vertical and horizontal directions. If black background data exists in an adjacent portion, detection is continued to see if black background data exists in a further continuous portion, and if black background data does not exist in an adjacent portion, detection of black background data is stopped there. Then, when a continuous black background part is no longer detected for all adjacent parts, a rectangle that circumscribes the shape of the continuous black background part is generated. FIG. 4 shows an example of generating black connected rectangles. Then, the coordinates of a black connected rectangle circumscribing this black background portion are obtained as data. In this way, black background data is detected and scanned two-dimensionally in the horizontal and vertical directions for all image data obtained from the input image. The coordinate data of all the obtained black connected rectangles and the horizontal and vertical side length data of the rectangles are imported as width w and height h (see Fig. 5), and the data is sent to the character size estimation unit 2. give. In this case, the coordinate data of the black connected rectangle can be the coordinates of one corner of the rectangle and the lengths of two sides in the horizontal and vertical directions, or the coordinates of two corners located diagonally of the rectangle, Alternatively, the coordinates of the four corners of the rectangle may be used as data.

The character size estimating unit 2 estimates the size of a character rectangle constituting the main text of the input document from the width w and height h of the black connected rectangle obtained by the black connected rectangle generation unit. This uses the width w and height h of the obtained black connected rectangle as data, and calculates the width W and height of the circumscribed rectangle of the characters that make up the main text of the input document, respectively, and calculates the value with the highest frequency of appearance (or average value). Used as an estimated value of H. In this case, when estimating the size of a character candidate rectangle, the width w and height h of the character rectangle can be estimated by using each independently. At this stage, it is also possible to remove w and h that are larger than a certain set value or smaller than another set value.

By performing this operation, it is possible to eliminate dirt, spots, etc. on the paper surface that are unrelated to the input document, and it is also possible to eliminate image data that is relatively larger than character data such as inserted drawings and headings.

Next, the character candidate rectangle extraction unit 3 uses the width W and height H of the circumscribing rectangle assumed by the character size estimation unit 2 to extract character candidate rectangles from among the black connected rectangles taken in as data. Select those to be used as this is,
W×n1 < w < W×n2
(However, 0<n1 <n2)...Formula 1 and H×n3 < h < H×n4
(However, 0<n3<n4)...This is performed by selecting those that both satisfy the conditional expression 2.

Here, n1, n2, n3, and n4 represent certain constants. For example, n1 = n3 = 0.2,
It is assumed that n2 = n4 = 4. By performing this process, you can create "
You can remove "parts of text" or black connecting rectangles such as inserted drawings.

Next, a character space estimating section 4 estimates a horizontal character space and a vertical character space from the character candidate rectangle selected by the character candidate rectangle extraction section 3. Refer to FIG. 6 to explain the operation of this estimator. This represents a certain character candidate rectangle i and a character candidate rectangle j horizontally adjacent thereto among the character candidate rectangles extracted by the steps up to the character candidate rectangle extraction unit 3 described in this embodiment. be. The coordinates of the upper left corner of character candidate rectangle i are (xi1, yi1), the coordinates of the lower right corner are (xi2, yi2), and the coordinates of the upper left corner of character candidate rectangle j are (xi1, yi1), and the coordinates of the lower right corner are (xi1, yi1). xj2
, yj2). At this time, in order to estimate the character space in the horizontal direction, the character space estimation unit 4 performs the following steps.
min(yi2, yj2)-max(yi1, yj1
)>H×α...Formula 3 (however, min(a, b) is a
, b, whichever has the smaller value. ) (also max(a
, b) represents the larger value of a and b. ) Detect character rectangle candidate pairs that satisfy the conditions. Here, H is the height of the character estimated by the character size estimation unit 2, and α
is a certain constant, for example α=1/3.

This condition is for selecting only those pairs of character candidate rectangles that are adjacent in the horizontal direction and have a small deviation in the vertical direction. What is out of place,
Remove them as inappropriate character candidate rectangle pairs. and formula 3
For a character candidate rectangle pair that satisfies the condition, the horizontal distance between character candidate rectangles of the character candidate rectangle pair (xj1-xi
Find 2). This horizontal distance between character candidate rectangles is obtained from a pair of character candidate rectangles in the input document, and serves as data for estimating the distance between characters in the horizontal direction. This process is performed for all extracted character candidate rectangles, and among the horizontal distances between the character candidate rectangles, the value (or average value) with the highest frequency of appearance is determined, and this is calculated as the horizontal distance between characters. It is estimated that

Similarly, reference will be made to FIG. 7 to explain the operation of this estimator. This represents a character candidate rectangle i and a character candidate rectangle j vertically adjacent thereto among the character candidate rectangles extracted by the steps up to the character candidate rectangle extraction unit 3 described in this embodiment. be. The coordinates of the upper left corner of character candidate rectangle i are (xi1, yi1), the coordinates of the lower right corner are (xi2, yi2), and the coordinates of the upper left corner of character candidate rectangle j are (xi1, yi1), and the coordinates of the lower right corner are (xi1, yi1). x
j2, yj2). At this time, in order to estimate the vertical character space, min(xi2, xj2)-max(xi1, xj
1)>W×α...Formula 4 (however, min(a, b) is
Represents the smaller value of a and b. ) (also max(
a, b) represents the larger value of a and b. ) Detect character rectangle candidate pairs that satisfy the conditions. Here, W is the width of the character estimated by the character size estimation unit 2, and α
is a certain constant, for example α=1/3. This condition is to select characters that have a small horizontal deviation among vertically adjacent character candidate rectangle pairs, and this condition selects characters that have a large horizontal deviation among adjacent character candidate rectangle pairs. are removed as inappropriate character candidate rectangle pairs. Then, for a pair of character candidate rectangles that satisfy the condition of Equation 4, the vertical distance between character candidate rectangles (yj1-yi2) of the pair of character candidate rectangles is determined. This vertical distance between character candidate rectangles becomes data for estimating the vertical distance between characters obtained from the character candidate rectangles of the input document. This process is performed for all extracted character candidate rectangles, and among the vertical distances between the character candidate rectangles, the value (or average value) with the highest frequency of appearance is calculated, and this value is calculated as the vertical intercharacter space VS We estimate that.

In order to efficiently perform the work in the character space estimating section 4, the following operation may be added. Before starting the work for estimating the horizontal character space HS in the character space estimating unit, for all character candidate rectangles extracted by the character candidate rectangle extraction unit 3,
Sorting is performed in ascending order based on the x-coordinate value of the upper left corner of the character candidate rectangle. As a result, character candidate rectangles are ordered in descending order of x-coordinate, so that an adjacent character candidate rectangle j horizontally adjacent to a certain character candidate rectangle i always appears after character candidate rectangle i. Therefore, when searching for character candidate rectangles adjacent to character candidate rectangle i in the horizontal direction, the search can be started for character candidate rectangles following character candidate rectangle i.
When a character candidate rectangle that satisfies the condition of Equation 3 is first found after starting the search, this can be used as the distance between character candidate rectangles in the horizontal direction (xj1-xj2).

Similarly, before starting the work for estimating the vertical inter-character space VS in the inter-character space estimating section, for all the character candidate rectangles extracted by the character candidate rectangle extraction section 3, the upper left corner of the character candidate rectangle is Sort in ascending order based on the y-coordinate value of the corner. As a result, character candidate rectangles are ordered in ascending order of y coordinate, so that an adjacent character candidate rectangle j that is vertically adjacent to a certain character candidate rectangle i always appears after character candidate rectangle i. Therefore, when searching for a character candidate rectangle vertically adjacent to character candidate rectangle i, character candidate rectangle i
Since the search can be started for the character candidate rectangles that follow the The distance can be (yj1-yj2).

The first character string direction determining unit 5 determines the horizontal character space HS extracted by the character space estimating unit 4.
and the vertical character space VS are compared. This is based on the empirical rule that in normal sentences, the spacing between letters is narrower in the direction where there is a semantic connection, and the spacing between the letters is wider in the direction where there is no semantic connection. That is, as shown in FIG. 8, in the area where VS>HS+th1...Equation 5 holds (ie, area 1), the character string direction in the input document is determined to be horizontal, that is, horizontal writing,
VS<HS+th2...In a region (ie, region 3) where Equation 6 holds true, the character string direction in the input document is determined to be vertical, that is, vertical writing. Then, th2<(VS-HS)<th1...In the case of a region where Equation 7 holds true (ie, region 2), it is assumed that discrimination is impossible. Here, th1 and th2 are set to certain threshold values, and in this method, t
The character spacing is allowed to be larger than the line spacing up to the size of h1 and th2. By determining the character string direction of the input document in this manner, it is determined whether the character strings of the input document are connected horizontally, vertically, or cannot be determined. However, VS=HS+th1...
It is predetermined that the case where the condition of Equation 8 is satisfied is included in either region 1 or region 2. Also, VS=HS
+th2...The case where the condition of formula 9 is satisfied is predetermined to be included in either region 2 or region 3.

If the first character string direction determining unit 5 determines that the character string cannot be determined, the character sequence degree extraction unit 6 extracts the horizontal character sequence degree and the vertical character sequence degree. The degree of character arrangement is a numerical representation of the character string-likeness in that direction. For example, for character candidate rectangles that are horizontally adjacent to each other as shown in FIG.
in(yi2, yj2)-max(yi1, yj1)>
Count character candidate rectangle pairs that satisfy H×β...Equation 8 and xj1-xi2<HS+th2...Equation 9. This count value is defined as the degree of character arrangement in the horizontal direction HC. Further, as shown in FIG. 7, character candidate rectangles that are vertically adjacent to each other are min(xi2, xj2)-max(xi1, xj
1) Count character candidate rectangle pairs that satisfy Equation 10 and yj1-yi2<VS+th3 Equation 11. This count value is defined as the degree of character arrangement in the vertical direction VC. Here, for example, β may be set to 1/2. th2 and th3 are threshold values. This value indicates that the horizontal and vertical size distribution of the extracted character candidate rectangles has a certain width, and that the intervals between the extracted character candidate rectangles are not necessarily uniform. Therefore, it was decided to add redundancy coefficients to reduce the possibility of making incorrect judgments regarding the width. The meanings of these discriminants are as follows. The meaning of Equation 10 is to select a pair of character candidate rectangles that are vertically adjacent and have a small deviation in the horizontal direction.
Among character candidate rectangle pairs adjacent in the vertical direction, those that are largely shifted in the horizontal direction are removed as inappropriate character candidate rectangle pairs. In addition, what Equation 11 means is that the distance between characters (yj1-yi2) of a pair of vertically adjacent character candidate rectangles is a certain amount or more compared to the vertical character space VS estimated by the character space estimation unit 4. This means that they have a relationship as a pair of character candidate rectangles that are not large. That is, according to the conditions of equations 10 and 11,
For pairs of character candidate rectangles that are adjacent in the vertical direction (corresponding to the condition of Equation 8), pair of character candidate rectangles in which the distance between the character candidate rectangles is larger than the vertical intercharacter space VS is defined as the character arrangement in the vertical direction. It will not be counted in the degree VC.

[0027] As a result, for a pair of character candidate rectangles that are adjacent in the vertical direction, if the pair of character candidate rectangles has an interval of one character or more in the vertical direction, the condition can be set such that the pair is not counted in the degree of character arrangement in the vertical direction VC. can. According to the conditions of Equations 10 and 11 above, by appropriately setting β and th3, it is possible to distinguish between a blank line (or blank column) and a continuous string. becomes. FIG. 9 shows an example of input data from which only character candidate rectangles are extracted. Among pairs of character candidate rectangles adjacent in the vertical direction, those whose distance between character candidate rectangles is approximately the same as the line spacing estimated by the character space estimation unit 4 (for example, character candidate rectangle a1 and character candidate rectangle b1) are sent to VC. The distance between character candidate rectangles is considerably larger than the line spacing estimated by the character space estimation unit 4, especially those whose distance is larger than one character candidate rectangle (for example, character candidate rectangle a4). and character candidate rectangle c4) are not counted as VC. Similar operations can be expressed as Equation 8 and Equation 9.
The degree of character arrangement in the horizontal direction HC for the blank part of the input image that exists in the horizontal direction
Count.

This operation is performed for all extracted character candidate rectangles, and the state of blank areas in the input document is measured. When there is a blank area that is continuous in the horizontal direction as shown in FIG. 9, a pair of character candidate rectangles (for example, character candidate rectangles a4 and c4
, or a5 and c5), the degree of character arrangement in the vertical direction VC of that part is not counted, and similarly if there is a blank part that is continuous in the vertical direction, the degree of character arrangement in the horizontal direction of that part is counted. HC is not counted. As a result, for example, when the character direction of the input document is horizontal, if there is a blank line following the "end of paragraph", HC will be counted more than VC in this part. Conversely, if the text direction of the input document is vertical,
If there is a blank string following "End of Paragraph," VC is counted more than HC in that part. Therefore, if HC is larger than VC, it can be determined that the input document has characters arranged horizontally, and if VC is larger than HC, it can be determined that the input document has characters arranged vertically. It becomes possible.

Then, in the second character string direction determining section 7,
The HC and VC extracted by the character arrangement degree extraction unit 6 are compared to determine the final direction of the character string. Here, for example, when HC-VC>γ1...Equation 12 (however, γ1 is a real number greater than or equal to 0), the character string direction of the input document is the horizontal direction, and VC-HC>γ2...Equation 13 (however, γ2 is (a real number greater than or equal to 0), the character string direction of the input document is set to be vertical. And equation 12, equation 13
If it cannot be determined based on these conditions, it is ultimately determined that it is impossible to determine. This completes the procedure of the character string direction determination method of the present invention.

FIG. 10 is a block diagram when the present invention is applied to the document reading device 11. That is, the document image input section 8
The character string direction determining section 9 determines the character string direction for the input document image, and determines whether the input document image is continuous in the horizontal direction or continuous in the vertical direction. , and supplies the result to the subsequent document reading section 10. As a result of the above processing, the character string direction of the input document is determined and the information is supplied to the document reading section 10, but the content of the information differs depending on the control method of the document reading section 10. For example, if the document reading unit 10 adopts a control method in which the reading direction differs depending on the character string direction of the input document, the character string direction determination unit 9 will cause the document reading unit 10 to read the characters along with the input document image. Provides column-wise information. Furthermore, if the document reading section 10 is controlled so that the document reading direction is constant, the document reading section 10 converts the input document image so that the reading direction and the character string direction match.
Supply to 0. The present embodiment can be summarized as follows.

First, the size distribution of the circumscribed rectangle of the shape in which the image data is connected is determined from the image data of the input document, and the size of the image data connection rectangle with the highest frequency of appearance is calculated based on the size of the characters in the document. presume. Image data concatenation rectangles with a size approximately equal to the estimated character size are selected as character candidate rectangles, and from the horizontal and vertical distribution of these character candidate rectangles, the horizontal and vertical character spacing of the document is determined. Determine the size of the space. Compare the horizontal character spacing with the vertical character spacing, and if the horizontal character spacing is greater than the vertical character spacing by more than the set value, the characters in the input document It is determined that the columns are arranged vertically. Conversely, if the vertical inter-character space is larger than the horizontal inter-character space by a set value or more, it is determined that the character strings of the input document are aligned in the horizontal direction.

Next, if the above conditions cannot be used, the spacing between horizontally and vertically adjacent character candidate rectangles for the document and the previously determined horizontal and vertical directions are determined. The character spacing is compared and the number of pairs of character candidate rectangles with approximately equal spacing is counted in both the horizontal and vertical directions. The count value counted by this operation represents the number of pairs of character candidate rectangles that are adjacent to each other with an interval similar to the average character spacing in the horizontal and vertical directions.For example, the size of one character candidate rectangle Pairs of character candidate rectangles that are adjacent to each other at a distance greater than or equal to this distance are not included. In normal sentences, blank lines (or blank columns) occur in areas where there is a line break at the end of a paragraph, so this area consists of a pair of adjacent character candidate rectangles with an interval similar to the average character spacing. not counted as. Therefore, by comparing the sizes of the count values counted in the horizontal and vertical directions through the above operation, the apparent arrangement of character candidate rectangles in the horizontal and vertical directions can be identified, and the character strings of the input document can be identified. Determine direction. That is, if the count value counted in the horizontal direction is larger than the count value counted in the vertical direction by a set value or more, it is determined that the character strings are arranged in the horizontal direction. Further, if the count value counted in the vertical direction is greater than the count value counted in the horizontal direction by a set value or more, it is determined that the character strings are arranged in the vertical direction. The above operation makes it possible to appropriately determine the character string direction. (Example 2)

In the first embodiment, the first character string direction determining unit 5 compares the values of the horizontal character space HS and the vertical character space VS extracted by the character space estimating unit 4. We decided to determine the character string direction of the input document. Therefore, the conditional expressions used for discrimination are as follows: VS>HS+th1...Equation 5 VS<HS+th2...Equation 6 th2<(VS-HS)<th1...Equation 7VS=HS
+th1...Equation 8 VS=HS+th2...Equation 9 However, it is also possible to use the following conditions instead of these conditional expressions. In the area where VS>HS...Equation 5-2 and VS/HS>th1'...Equation 5-3 hold (hereinafter referred to as area 4), the character string direction in the input document is determined to be horizontal, and HS>V
In an area (hereinafter referred to as area 6) where S...Equation 6-2 and HS/VS>th2'...Equation 6-3 hold, the character string direction in the input document is determined to be vertical. And 1/th2'<(VS/HS)<th1'...Formula 7-
2 (where th2'≠0) holds (hereinafter referred to as region 5), it is assumed that the determination is impossible. Here th1',
It is assumed that th2' is a certain threshold value. By determining the character string direction of the input document in this manner, it is determined whether the character string direction of the input document is horizontal, vertical, or cannot be determined. However, it is predetermined that the case where the condition of VS/HS=th1'...Equation 8-2 is satisfied is included in either region 4 or region 5. Also, HS/VS
=th2'...The case where the condition of Equation 9-2 is satisfied is predetermined to be included in either region 5 or region 6. Even if such a discriminant is used, the same effects as in the embodiment described above can be obtained. Similarly, the conditions for the following discriminant expressions HC-VC>γ1...Equation 12 VC-HC>γ2...Equation 13 used in the second character string direction determination unit 7 explained in the previous example are changed as follows. It is possible to do so. In the area where HC>VC...Equation 12-2 and HC/VC>th1'...Equation 12-3 hold, the character string direction in the input document is determined to be horizontal, and VC>HC...Equation 13-2 and VC /HC>th2'...In a region where equation 13-3 holds true, the character string direction in the input document is determined to be vertical. Even if such a conditional expression is used, the same effect as in the above embodiment can be obtained. (Example 3)

In the first embodiment, a method has been described in which the first character string direction determining section 5 performs the determination and then the second character string direction determining section 7 performs the determination, but the determination conditions are changed as described below. Therefore, it is possible to perform character string direction determination even by reversing the order of the determination methods.

This will be explained using FIG. 11. The steps described above include generating black concatenated rectangles for the input image, estimating the size of characters in the input document, extracting character candidate rectangles, and estimating the space between characters in the horizontal and vertical directions. Since this embodiment is the same as that of the embodiment, the explanation will be omitted.

Next, the degree of character arrangement extraction unit 6 extracts the degree of character arrangement in the horizontal direction and the degree of character arrangement in the vertical direction, and extracts the degree of character arrangement in the horizontal direction and the vertical direction of the input image. The number of character candidate rectangle pairs having a distance between character candidate rectangles that is approximately equal to the intercharacter space is counted. Based on this counting result, the second character string determining section 7 performs "character string direction determination." In other words, using the degree of character arrangement in the vertical direction VC and the degree of character arrangement in the horizontal direction HC, HC-VC>γ1...Equation 12 (where γ1 is a real number greater than or equal to 0), the character string direction of the input document is set horizontally. When VC-HC>γ2 (Equation 13) (where γ2 is a real number greater than or equal to 0), the character string direction of the input document is the vertical direction. And equation 12, equation 13
If it cannot be determined based on the conditions, the process moves to the next step without making any determination.

For input documents that have not been determined by the second character string direction determining section, the first character string direction determining section 5 determines the character string direction. In other words, the character space estimation unit 4
Compare the values of the horizontal character space HS extracted with the vertical character space VS, and find that VS>HS+
In an area where th1...Equation 5 holds, the character string direction in the input document is determined to be horizontal, and in an area where VS<HS+th2...Equation 6 holds, the character string direction in the input document is determined to be vertical. Then, in the case of a region where th2<(VS-HS)<th1...Equation 7 holds true, the character string direction determination is ended without performing the final determination. In this way, even if the order of the first character string direction determining section 5 and the second character string direction determining section 7 described in the previous embodiment is reversed, character string direction determination is possible.

In this case as well, the conditional expressions used for judgment in the first character string discriminator and the second character string discriminator are Equations 5-2 to 13-3.
By changing as follows, the direction of the character string can also be determined and the same effect can be obtained.

In all of the above embodiments, the first character string direction determining section and the second character string direction determining section are provided as character string direction determining means, but if the input document has a smaller character spacing than the line spacing. Of course, high accuracy can be obtained even if the character string direction is determined based on the determination result of the first character string direction determining section 5, provided that the document has certain characteristics.

[0040]Also, the entire input document is input to the character string direction determining section to determine the character string direction, but for example, text regions (headings, (text, explanatory text for diagrams, photographs, etc.)
may be input to the character string direction determining section to determine the character string direction in each text area. In this way, it is possible to determine the character string direction in each text area, and it is possible to accurately determine the character string direction even for a document in which vertically written and horizontally written texts are mixed.

Furthermore, in the English document reading device,
If the input device is vertically long (or horizontally long) and a long horizontally (or vertically) long document must be rotated 90 degrees to be input, the character string direction of the input document image is determined and the image is By performing the process of restoring the
The input document can be read in the correct string direction.

[0042]

[Effects of the Invention] According to the present invention, the direction of a character string is determined by detecting the state of blank lines (or blank columns) obtained from image data of an input document, which cannot be determined correctly using conventional methods. The character string direction can be correctly determined in various documents, including those in which the character spacing is larger than the line spacing.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a character string direction determination method.

FIG. 2 is a diagram showing an example of character candidate rectangles generated from an input image.

FIG. 3 is a diagram showing an example of character candidate rectangles generated from an input image.

FIG. 4 is a diagram showing an example of generating black connected rectangles.

FIG. 5 is a diagram showing an example of coordinate data of black connected rectangles.

FIG. 6 is a diagram showing pairs of character candidate rectangles adjacent in the horizontal direction.

FIG. 7 is a diagram showing pairs of character candidate rectangles that are vertically adjacent to each other.

FIG. 8 is a diagram showing an example of a determination area for character string direction determination.

FIG. 9 is a diagram showing an example of input data in which only character candidate rectangles are extracted.

FIG. 10 is a diagram when this method is applied to a document reading device.

FIG. 11 is a block diagram showing another configuration of a character string direction determination method.

[Explanation of symbols]

1 Black connected rectangle generation part 2 Character size estimation part 3 Character candidate rectangle extraction part 4 Character space estimation part 5 First character string direction determination unit 6 Character sequence degree extraction part 7 Second character string direction determination unit 8 Document image input section 9 Character string direction determination unit 10 Document reading section 11 Document reading device 14 Judgment section

Claims (7)

[Claims] Claim 1: means for extracting the degree of character arrangement in the horizontal direction; means for extracting the degree of character arrangement in the vertical direction;
A character characterized by comprising: means for comparing the degree of character arrangement in the horizontal direction and the degree of character arrangement in the vertical direction; and determining means for determining the direction of the character arrangement based on the comparison result from the comparing means. Column direction discrimination method. [Claim 2] Comparing the sizes of the horizontal inter-character space and the vertical inter-character space to determine the character string direction,
A character string direction determining method further comprising determining the character string direction using the character string direction determining method according to claim 1. [Claim 3] For the horizontal inter-character space, use the mode value of the distance between horizontally adjacent character candidate rectangles, and for the vertical inter-character space, use the mode value of the distance between vertically adjacent character candidate rectangles. 3. The character string direction determination method according to claim 2. 4. The horizontal character space uses the average distance between horizontally adjacent character candidate rectangles, and the vertical character space uses the average distance between vertically adjacent character candidate rectangles. 3. The character string direction determination method according to claim 2. 5. Means for extracting a plurality of character candidate rectangles from input data, means for determining a character string direction by determining the distance between different character candidate rectangles, and a means for extracting a data area not occupied by the character candidate rectangles. 1. A character string direction determination method characterized by comprising means for determining character string direction by converting a state into a numerical value. 6. An image input unit; a character candidate rectangle extraction unit that extracts a plurality of character candidate rectangles from input data;
A first character string direction determining unit determines the character string direction by determining the distance between different character candidate rectangles, and a second character string direction determining unit determines the character string direction by quantifying the state of the area not occupied by the character candidate rectangles. A character string direction determining device comprising a character string direction determining unit. [Claim 7] An image input section for inputting image data;
A black connected rectangle generation unit that generates a rectangle circumscribing an area where black image data is continuous, and calculates the size of the generated black connected rectangle and calculates the size of the black connected rectangle that appears most frequently among the plurality of black connected rectangles. a character size estimation unit that selects a size; a character candidate rectangle extraction unit that extracts a character candidate rectangle by comparing the estimated character size with the size of the generated black connected rectangle; A character space estimating unit that estimates the horizontal and vertical character space by calculating the distance between two adjacent character candidate rectangles, and a horizontal character space and a vertical character space. a first character string direction determining unit that determines the character string direction by comparing the sizes of the characters; , a first counting unit that counts the number of character candidate rectangle pairs having the relationship HL<HC+th1, a distance VL between two vertically adjacent character candidate rectangles, and a space VC between characters in the vertical direction. Toga, VL
A second counting section that counts the number of character candidate rectangle pairs having a relationship such as <VC+th2, and a comparison between the counted value counted by the first counting section and the counted value counted by the second counting section. a second character string direction determination unit that determines the character string direction by performing the following steps; and a determination unit that outputs a character string direction determination result from the determination result of the first character string direction determination unit and the determination result of the second character string direction determination unit. A character string direction determining device comprising: