JPH07182459A - Table structure extracting device - Google Patents

Abstract

PURPOSE:To provide a table structure extracting device which can properly and accurately extract the ruled lines, the intersecting points, etc., out of the printed documents and also can omit the pre-processing for correction of the tilt. CONSTITUTION:A black pixel string detecting part 12 scans the compressed binary date and detects the horizontal and vertical black pixel strings. A segment extracting part 13 constructs the paths by the continuous horizontal vertical black pixel strings which are adjacent to each other and extracts the horizontal and vertical segments on both paths by a straight line deciding rule. A ruled line candidate constructing pert 21 successively detects the continuous horizontal and vertical segments adjacent to each other by a method that detects the next adjacent segments to the present segments based on the overlapping between both segments end the continuous black pixel strings adjacent to each other. Then the horizontal and vertical ruled line candidates are obtained from detected horizontal end vertical segments respectively. A table structure extracting part 31 checks the crossing between the rectangular intersecting point candidates to the vertical ruled line candidate on the basis of the horizontal ruled line candidate and vice versa. Thus the genuine intersecting points and the genuine horizontal end vertical ruled lines are acquired.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a table structure extracting device for extracting ruled lines and intersections in a printed document.

[0002]

2. Description of the Related Art A conventional table structure extraction technique is, for example, "A Study on Tapping on Tables with Complex Structures".
ble Recognition with complex structure) "(Proceedings of the 37th National Information Processing Society of Japan (Late 1988) National Convention)
There is one shown in. FIG. 2 shows the system configuration of this conventional example. In the figure, 10 is a tilt correction unit that detects and normalizes the tilt of an image input by a tilt correction algorithm in order to solve the tilt problem of the input image. 20 is the marginal distribution method (Prejection Profile)
e, The histogram of the image is used as the reference for ruled line extraction. The histogram is mainly divided into a horizontal histogram and a vertical histogram. ) To eliminate the skew and skew of the paper, project the image after the skew correction in the horizontal direction and the vertical direction, and make the height above a certain threshold from the histogram (column chart) as a ruled line candidate. It is a ruled line candidate extraction unit. 30 is a ruled line candidate, in the case of a CLOSE type table, those whose both ends are in contact with the outer frame ruled line are extracted as ruled lines, and in the case of an OPEN type table with no outer frame ruled line. Is a ruled line extraction unit that adds an outer frame ruled line for convenience. Reference numeral 40 divides the image into a plurality of blocks by the extracted ruled lines, returns each of the divided block regions to the ruled line candidate extraction unit 20 to perform ruled line extraction, and performs recursive processing until there are no ruled line candidates in each block region. This is a block division unit.

With the above configuration, the table structure in the print document is extracted.

[0004]

However, the above-mentioned conventional apparatus has the following problems. First problem: As shown in FIG. 3, when extracting ruled line candidates by the marginal distribution method, inclination correction (preprocessing) must be performed. In this case, the sine (si
n) It takes time because it needs to be calculated. Here, the sine calculation for each pixel means the following calculation.

The detected tilt angle is A. (X,
For each Y), X ′ = XcosA−YsinA, Y ′ = Xsi
Finding nA + YcosA, the pixel value of (X, Y) is put into (X ′, Y ′). Second problem: In a method in which an image is repeatedly divided into blocks by ruled lines and a height higher than a certain threshold is used as a ruled line candidate from a histogram, for example, an irregular table as shown in FIG. It is difficult to extract correctly. That is, the horizontal histogram indicated by arrow 1 is smaller than the horizontal histogram indicated by arrow 2. However, there is a ruled line in the part indicated by arrow 1, and there is no ruled line in the part indicated by arrow 2. FIGS. 4 (2) to 4 (6) show erroneous results caused by the extraction process and its limit by the conventional technique.

Third problem: marginal distribution method (Prejection Pr
It is not possible to determine the presence or absence of a segment in the ruled line and the actual crossing status by extracting ruled line candidates by (ofile) and extracting those with both ends contacting the sides of the block as ruled lines. For example, in the case of a ruled line in which a vertical direction line is missing in the center square of the table as shown in FIG. 5, a rough extraction process is performed, resulting in erroneous recognition.

The present invention has been made for the purpose of solving the above-mentioned problems and providing a table structure extracting apparatus which can obtain a correct extraction result.

[0008]

In order to solve the above-mentioned problems, in the invention of claim 1, a table structure extracting device for extracting a table structure in a printed document by using the overlapping and adjacency relations of black pixel rows is provided. An image compression unit that sets the result of the logical OR of the n * n pixel values inside the n * n square to the input primitive binary digital data to 1 pixel, and the compressed binary digital data And a black pixel series detection unit that detects horizontal black pixel series and vertical black pixel series by scanning in the vertical and vertical directions, and is continuously adjacent to the horizontal and vertical black pixel series detected by the black pixel series detection unit. A line segment is constructed by constructing a path (path) that is a non-intersection point from the vertical black pixel series and the horizontal black pixel series that are determined to be each, and then extracting the horizontal line segment and the vertical line segment in this configured path by the straight line discrimination rule. Department, The horizontal line segment and the vertical line segment extracted by the line segment extraction unit are successively adjacent to each other by using the method of detecting the next adjacent line segment by overlapping and continuously adjoining black pixel lines in the line segment. A horizontal line segment and a vertical line segment are detected, and a ruled line candidate configuration unit that configures horizontal ruled line candidates and vertical ruled line candidates from the detected horizontal line segment and vertical line segment, respectively, and a horizontal line configured by the ruled line candidate configuration unit. For the ruled line candidate and the vertical ruled line candidate, it is determined whether or not the ruled line candidate overlaps before and after the ruled line and whether or not there is a portion that extends from the ruled line to the front and back in the consecutive black pixel series. On the other hand, it has a table structure extraction unit that detects a true horizontal ruled line, a vertical ruled line, and an intersection by performing an intersection inspection on a horizontal ruled line candidate with reference to the vertical ruled line candidate. It is characterized in.

According to the second aspect of the invention, the line segment extracting section forms a path from each of a series of vertically adjoining vertical black pixels and a series of horizontal black pixels, and the black pixel series on the path are sequentially arranged on this path. The feature is that the horizontal line segment and the vertical line segment on the route are extracted by a straight line discriminating rule for inspecting the start point coordinate value and the end point coordinate value. Claim 3
In the invention, the ruled line candidate extraction unit detects the next adjacent line segment by overlapping the horizontal line segment and the vertical line segment and the black pixel string that is continuously adjacent to the horizontal line segment and the vertical line segment, and the adjacent line segment. A horizontal line segment and a vertical line segment that are consecutively adjacent to each other in order are detected by using the method of processing the situation where characters and ruled lines are in contact with each other by comparing the widths of It is characterized in that it constitutes ruled line candidates and vertical ruled line candidates.

According to another aspect of the present invention, the table structure extraction unit extends frontward and rearward from the ruled line by overlapping the front and rear of the ruled line candidate with respect to the horizontal ruled line candidate and the vertical ruled line candidate and continuously adjoining black pixel lines. Rectangle intersection candidates and rectangle intersection candidates for the line segments that make up the ruled line candidates are obtained, and then, based on the horizontal ruled line candidates, vertical ruled line candidates or vertical ruled line candidates, horizontal ruled line candidates, rectangular crossing points It is characterized in that it detects true horizontal ruled lines, vertical ruled lines and intersections by detecting intersections between candidates.

[0011]

With the above structure, in the invention of claim 1,
The following operation is performed in a table structure extraction device that extracts a table structure in a printed document by using the overlapping and adjacency relationships of a series of black pixels each including a single or a plurality of black pixels. The image compression unit uses n * n (here, * is replaced with X (multiply) to prevent misreading with X (X) for the input binary digital data,
Represents multiplication. Further, n is an integer value of 2, 3, 4, ..., Etc.), and the result of the logical OR operation of the n * n pixel values inside the square is defined as one pixel. (For this reason, if only one pixel is 1, the n * n data is 1.)
The black pixel series detection unit scans the compressed binary digital data in the horizontal and vertical directions to detect horizontal black pixel series and vertical black pixel series. The line segment extraction unit constructs a path (path) from each of a series of vertically adjoining vertical black pixels and a series of horizontal black pixels, so that the portion that is an intersection is first removed, and the horizontal line in this constructed path is determined by the straight line discrimination rule. Extract minutes and vertical line segments. The ruled line candidate constituent parts are successively adjacent to the extracted horizontal line segment and vertical line segment in order by using a method of detecting the next adjacent part by overlapping and continuous adjacent black pixel series in the line segment. Detects horizontal and vertical line segments. Then, a horizontal ruled line candidate (finally, a line that does not form a table may be determined in some cases) and a vertical ruled line candidate are constructed from each of the detected horizontal and vertical line segments. The table structure extraction unit determines whether or not the configured horizontal ruled line candidate and the vertical ruled line candidate have overlaps before and after the ruled line candidate and a portion extending continuously from the ruled line in the consecutive black pixel series. Since the table structure is formed by the crossing of horizontal and vertical lines, the intersection check is performed on the vertical ruled line candidate with the horizontal ruled line candidate or with the horizontal ruled line candidate with the vertical ruled line candidate, and the true horizontal Detects ruled lines, vertical ruled lines and intersections.

According to the second aspect of the present invention, the line segment extracting unit forms a path from each of a series of vertically adjoining vertical black pixels and a series of horizontally adjoining black pixels. Then, a horizontal line segment and a vertical line segment on the route are extracted by a straight line discriminating rule that sequentially inspects the start point coordinate and the end point coordinate value for each black pixel string on the route. According to a third aspect of the present invention, the ruled line candidate extraction unit detects the next adjacent line segment by the overlapping of the line segments with respect to the horizontal line segment and the vertical line segment and the black pixel string that is continuously adjacent, and the adjacent line segment. The horizontal line segment and the vertical line segment that are consecutively adjacent to each other in order are detected using the method of processing the situation where the character and the ruled line are in contact by comparing the width with the line segment
A horizontal ruled line candidate and a vertical ruled line candidate are configured from the detected horizontal line segment and vertical line segment, respectively.

According to another aspect of the invention, the table structure extracting unit extracts the horizontal ruled line candidate and the vertical ruled line candidate from the ruled line by overlapping the consecutive ruled line candidates before and after the ruled line candidate and continuously adjoining the black pixel string. A rectangular intersection candidate that extends back and forth and a rectangular intersection candidate of the line segments that form the ruled line candidate are obtained, and then a horizontal ruled line candidate is used as a reference for a vertical ruled line candidate or a vertical ruled line candidate is used as a reference. Then, the intersections between the rectangular intersection candidates are inspected to detect true horizontal ruled lines, vertical ruled lines, and intersections.

[0014]

1 is a system configuration diagram of an embodiment of a table structure extracting apparatus according to the present invention. A document, that is, original document data, is converted into binary digital data by a scanner and processed before the table structure is extracted by this apparatus. Further, the binary digital data which is the chart portion is cut out by the area identification processing and stored in the buffer memory (not shown) of the present apparatus. Hereinafter, the configuration and operation of each part will be described based on this drawing. In the figure, reference numeral 11 denotes an image compression unit which divides the inputted original binary digital data into a square region having some 4 * 4 pixels and compresses the divided region. A logical OR operation is performed on 4 * 4 = 16 pixel values in each of the divided square areas, and the result is set to one pixel. Reference numeral 12 denotes a black pixel series detection unit that scans the compressed image in the horizontal and vertical directions to detect the horizontal black pixel series and the vertical black pixel series. Reference numeral 13 denotes a line segment extraction unit that forms a path from each of a series of vertically adjoining vertical black pixels and a series of horizontal black pixels, and then extracts a horizontal line segment and a vertical line segment on the path according to a straight line determination rule. The extraction principle will be described later with reference to FIGS. 7 and 8. 21 uses a method of detecting the next adjacent line segment by tracing the horizontal line segment and the vertical line segment extracted by the line segment extraction unit and the consecutive black pixel series that overlap each other in the line segment. The horizontal line segment and the vertical line segment which are successively adjacent to each other are sequentially detected, and finally, a ruled line candidate configuration unit that forms a horizontal ruled line candidate and a vertical ruled line candidate from the detected horizontal line segment and vertical line segment. In addition,
The processing flow and processing example will be described later with reference to FIGS. 9 and 10. Reference numeral 31 denotes a black ruled line that continuously overlaps and is adjacent to the configured horizontal ruled line candidate and vertical ruled line candidate before and after the ruled line candidate, and defines a portion extending from the ruled line to the front and back. Is a table structure extraction unit that performs true cross rule inspection on vertical ruled line candidates or horizontal ruled line candidates on vertical ruled line candidates to detect true horizontal ruled lines, vertical ruled lines, and intersections. The processing flow and processing example will be described later with reference to FIG.

In the following, the operation of each of the above parts will be specifically and more specifically described in order. First, with respect to the operation of the black pixel consecutive detection unit 12, as an example, a case of scanning the compressed image shown in FIG. 6A will be described as an example. In this case, scanning is performed in the horizontal direction and the vertical direction to detect the horizontal black pixel string shown in FIG. 6B and the vertical black pixel string shown in FIG. 6C. Among them, (y; x, x ′) indicates a horizontal black pixel string constituted by a black pixel string from the start point x to the end point x ′ on the horizontal scanning line y,
(X, x ′; y) indicates a vertical black pixel string having continuous black pixels X to X ′ at the vertical scanning line Y. Hereinafter, in order to simplify the description, the horizontal black pixel series will be mainly described. The processing for the vertical black pixel series is basically the same as the processing for the horizontal black pixel series.

The line segment extraction unit 13 forms a path from each of a series of vertically adjoining vertical black pixels and a series of horizontal black pixels, and extracts a horizontal line segment and a vertical line segment on the path according to a straight line discrimination rule. . First, according to the following rules or procedures (1) to (3), a path is formed from a series of vertically adjoining vertical black pixels and a series of horizontally adjoining black pixels. Rule (1) R ₁ = (y ₁ ; x ₁ , x ₁ ') and R ₂ = (y ₂ ; x ₂ ,
x ₂ '), if x ₁ <x ₂ ' and x ₁ '> x
If _2, it is defined that R ₁ and R ₂ overlap. FIG. 7 shows an example thereof.

Rule (2) R ₁ = (y ₁ ; x ₁ , x ₁ ') and R ₂ = (y ₂ ; x ₂ ,
against x ₂ '), if, y ₂ = y ₁ + ₁ and R ₁ and R
_{If 2} overlaps, it is determined that R ₁ and R ₂ are adjacent to each other. Rule (3) For R ₁ , R ₂ , ... R _n that are continuously adjacent (this is called because it consists of a large number of adjacent R _i and R _{i + 1} ), if before and after each R _i If at least one scanning line is adjacent to one pixel series, it is defined that one path is composed of R ₁ , R ₂ , ... R _n . FIG. 8 shows, as an example of this, two paths constituted by a series of black pixels that are continuously adjacent to each other.

Then, a straight line segment on the route is extracted according to the straight line discrimination rule. The straight line discriminating rule used here is the following rule (4), and the start point coordinate value and the end point coordinate value are inspected for each black pixel string in each path in order. Rule (4) For R ₁ , R ₂ , ... R _n , which is a part of the path, if the following conditions are satisfied, R ₁ ,
It is determined that R ₂ , ... R _n form a straight line segment. (1) | x _i −x _{i + 1} | <between adjacent black pixel strings
1 and | x _{i ′} −x _{i + 1 ′} | <1 holds. (2) In sequence, K black pixel rows are treated as one segment,
For each segment, | x ^{a (a} represents the horizontal line on the x and later also the same.) -X _j 1│ <1 and | x ^a '-x _j' | is <. Among them, R ^a is a first-th black pixel run that dwell in a certain segment ((y ^a;. Which x ^a, is represented by x ^a ')), R _j Other R ^a, in the segment It is a black pixel string. (3) The length is larger than a certain threshold value. (4) When the width is divided by the length, the obtained value is smaller than a certain threshold value.

The above condition (1) and condition (2) are given here.
Indicates that the line segment has end points corresponding to positions and is composed of a series of black pixels having a constant length. Further, in consideration of the actual condition of the frame in printing, when the line segment is not curved, it is determined that the trivial inclination constitutes a ruled line. For example, when K = 14, tan ^-1 (1/14)
If the inclination is (that is, the inclination is within 4 degrees), the ruled line may be present.

The conditions (3) and (4) represent that the line segment has slenderness within a certain range. That is, the length of R ₁ , R ₂ , ... R _n is n, and the width is the length of the frontmost black pixel string (that is, x ₁ ′ −x ₁ +1). Taking FIG. 8 as an example, one vertical line segment is extracted from the right route, and two vertical line segments are extracted from the left route.

The ruled line candidate construction unit 21 successively and sequentially adjoins the horizontal line segment and the vertical line segment by using the method of detecting the next adjacent line segment by the overlap in the line segment and the consecutive black pixel sequence. The horizontal line segment and the vertical line segment that are present are detected, and the horizontal ruled line candidate and the vertical ruled line candidate are configured from the horizontal line segment and the vertical line segment detected above. The operation flow in that case is shown in FIG. Hereinafter, a detailed description will be given with reference to FIG. (S1) If there is no "not inspected" line segment, the operation ends. If there is, go to (S2). (S2) The y value of the first black pixel string of the frontmost line segment from the "non-inspected" line segments is set as the starting line segment. (S3) The next adjacent line segment is detected by the overlap between the line segments and the continuous black pixel series. The rules and procedures in that case are shown below.

Rule (5) Line segment L₁And L₂At L₁The last black pixel that makes up
R is R_lastAnd L ₂The first black pixel array that makes up
R_firstAnd A line segment if the following conditions are met:
L₂Is L₁It is determined that the line segment is adjacent to. (1) y_last<Y_firstAnd R_lastAnd R_firstOverlap (2) L₂Is y while satisfying condition (1)_firstIs the best
It is small (3) R_lastFrom R₁, R₂, ... R_nThrough y_first
And are adjacent to each other. And R₁, R₂, ... R_nIs all
R_lastOverlap with (4) L₂Is not an adjacent line segment to another line segment. FIG. 10 shows an example of applying the above rule to a vertical line segment.
Is shown. In this figure, (1) is R₂And R
_lastL do not overlap₂And L₁Is not an adjacent line segment
That is the case. Similarly (2) is L₂Is L₁Adjacent line
This is the case when it is considered to be minutes. (S4) When not detected, each L₁, L₂, ... L_nAgainst
Then, the ruled line candidates are sequentially determined. (S5) When adjacent line segments are detected, the
Since the line segment that the ruled line touches is processed,
Paying attention to the fact that the line segment is wider than the
Compare the width with the width of the adjacent line segment. The width of the line segment here
Is the length of the frontmost black pixel string forming the line segment. (S6) If the difference between the widths of these two line segments is small, the line
Record the minutes as adjacent line segments. (S7) Whether or not the width of the adjacent line segment is the same as the width of the start line segment
After inspecting, if the widths are not the same, mark that
Attach it and return to (S3) to detect the next adjacent line segment.
It (S8) If the line segment is smaller in width than the adjacent line segment,
The adjacent line segment is recognized as the line segment where the character and the ruled line are in contact.
Check the line segment and return to (S3)
And detect again. (S9) If the line segment is wider than the adjacent line segment,
The line segment is recognized as a line segment in which characters and ruled lines are in contact.
Therefore, the fact that it is a "non-ruled line" is recorded. And (S
Returning to 3), the adjacent "non-ruled line" line segment is recorded.
Detect line segment. (S10) If the adjacent line segments can be detected in (S3)
If not, it is recorded according to the following rules (6)
Whether ruled line candidates can be constructed from line segments that are consecutively adjacent
To judge.

Rule (6) Line segments L that are continuously adjacent₁, L₂, ... L_nAgainst
However, if the following conditions are met, these L₁, L
₂, ... L_nIt is determined that the ruled line candidate is constructed from the. (1) Marks that "widths are not the same" or "non-ruled lines"
(2) The length is larger than a certain threshold (3) If the width is divided by the length, the obtained value is a certain threshold.
In Rule (6), condition (2) and condition (3) are
Ruled line candidates indicate that they have slenderness. here,
L₁To L_nThe length at is L_nThe last black pixel that makes up
L from the value of y₁The y of the front black pixel string that composes
One shall be added to the difference value obtained by subtracting the value. Well
L₁To L_nThe width of the ruled line candidate in is L ₁Make up
It is the length of the frontmost black pixel string.

In FIG. 1, the table structure extraction unit 31 moves forward and backward from the ruled line on the basis of the overlap of the ruled line candidate with respect to the horizontal ruled line candidate and the vertical ruled line candidate and the consecutive black pixel series. Detects (defines) the rectangular intersection candidates that extend and the rectangular intersection candidates that form the ruled line candidate, and then detects the horizontal ruled line candidates as the vertical ruled line candidates or the vertical ruled line candidates as the horizontal ruled lines. The rule is confirmed by performing an intersection inspection between the rectangular intersection candidates on the candidates and detecting a true horizontal ruled line, a vertical ruled line, and an intersection.

With respect to the procedure, a rectangular intersection candidate is detected according to the following rule (7) for each horizontal ruled line candidate and vertical ruled line candidate. It should be noted that this also considers shortening the time for determining the presence or absence of an intersection by avoiding geometric calculation. Rule (7) For a ruled line candidate composed of line segment candidates L ₁ , L ₂ , ... _Li , L _{i + 1,} ..., L _n , a rectangle composed of those satisfying the following conditions is set as an intersection candidate. To do. Hereinafter, description will be given with reference to FIG. (1) L _i and L _{i + 1} between the last (after the larger in the y-coordinate of the scan line) black pixel run constituting the L _i is R _i, one of and constituting the L _{i + 1} The last black pixel string (the smaller one in the y coordinate of the scanning line is first) is R _{i + 1} , that is, a rectangle formed by x _i to x _i ′ and y _i to y _{i + 1.} It is determined that (2) If the previous black pixel string forming L ₁ is R ₁ ,
Previously, a black pixel string which overlaps R ₁ and is continuously adjacent is detected, and at a distance y _u . That is, it is determined that the rectangle is composed of x ₁ to x ₁ ′ and _yu to y ₁ . (3) _If the last black pixel string R _n forming L _n exists, a black pixel string that overlaps with R _n and is continuously adjacent is detected, and even up to y _b . That is, it is determined that the rectangle is composed of x _n to x _n ′ and y _n to y _b .

FIG. 11B shows an example of vertical ruled line candidates defined as above. Then, the intersection inspection between the rectangular intersection candidates is performed on the vertical ruled line candidates by the horizontal ruled line candidates. In the figure, 1, 2 and 3 correspond to (1), (2) and (3) of Rule 7 above, respectively. The inspection gives a true intersection and a true horizontal ruled line,
Since vertical ruled lines can be obtained, a table structure can be acquired.

The operation of the table structure extracting device of the present embodiment having the above-described structure will be described below with reference to the processing example of FIG. First, the image compression unit 11 performs a compression process on the input binary digital data. The state after this compression processing is shown in FIG. Black pixel consecutive detection unit 12
Scans the compressed binary digital data in the horizontal and vertical directions to detect horizontal black pixel series and vertical black pixel series. This state is shown in FIG.

The line segment extraction unit 13 forms a path from each of a series of vertically adjoining vertical black pixels and a series of horizontally adjoining black pixels. Then, for each black pixel string in this configured path,
The true horizontal line segment and the vertical line segment on the route are extracted by applying the straight line discriminating rule for inspecting the start point coordinate value and the end point coordinate value in order. This state is shown in FIG. Here, nine horizontal line segments and eight vertical line segments are extracted.

The ruled line candidate construction unit 21 detects the next adjacent line segment by the overlap of the horizontal line segment and the vertical line segment in the line segment and the continuous black pixel series and the adjacent line segment. A horizontal line segment and a vertical line segment as ruled line candidates that are successively adjacent to each other are detected by using a method of processing a situation in which a character and a ruled line are in contact with each other by comparing the widths.
This state is shown in (4) of FIG. In this case, three horizontal ruled line candidates and four vertical ruled line candidates are configured from the detected horizontal line segment and vertical line segment, respectively.

The table structure extraction unit 31 determines whether or not the horizontal ruled line candidate and the vertical ruled line candidate are overlapped before and after the ruled line candidate and whether or not there is a portion extending continuously from the ruled line in the consecutive black pixel series. The determination is made, and the rectangular intersection candidate and the rectangular intersection candidate of the line segment that constitutes the ruled line candidate are found on the judgment. This state is shown in FIG. In this case, there are 24 hollow rectangles. Then, the intersection inspection between the rectangular intersection candidates is performed on the vertical ruled line candidate by the horizontal ruled line candidate or the horizontal ruled line candidate by the vertical ruled line candidate. The inspection determines horizontal ruled lines and vertical ruled lines and the intersections of the two. This state is shown in FIG. In this case, finally, three horizontal ruled lines, four vertical ruled lines and twelve intersections are correctly obtained. 13 and 14 from the tables shown in FIGS. 4 and 5, respectively.
Extract the exact table as shown in.

Although the present invention has been described above based on the embodiments, it goes without saying that the present invention is not limited to the above embodiments. That is, for example, the following is also included in the present invention. (1) The input data is general binary data, and the table is extracted as it is. (2) The logical OR operation of image compression is not a square area in units of 4 * 4 pixels, but a square area in units of 3 * 3 pixels. (3) The term “black pixel” as used in the present specification means a pixel forming a ruled line, and is not limited to a chromatically “black” color. For example, if ruled lines are printed in “green” color on “white” paper, the pixels that make up this “green” color are also “black pixels”.
Corresponds to. (4) For the convenience of manufacturing or the like, one indispensable constituent element (requirement) of the present invention is divided into a plurality of parts, conversely, a plurality of constituent elements are integrated, or these are combined appropriately.

[0032]

As described above, according to the present invention, the following effects can be obtained. (1) The line segment candidate extraction unit and the ruled line candidate detection unit configure horizontal ruled line candidates and vertical ruled line candidates. Therefore, the problem that the tilt correction (pre-processing), which has been indispensable in the past, requires a long time is solved. (2) A true horizontal ruled line,
Vertical creases and intersections are obtained. Therefore, the problems that an irregular table cannot be extracted and the presence / absence of a segment in a ruled line and the actual intersection situation cannot be determined are solved. (3) The image compressing unit calculates the logical OR of the n * n pixel values inside the n * n (n = 2, 3, 4, ...) Square with respect to the input primitive binary digital data. The result to be calculated is one pixel. For this reason, the amount of data to be processed can be reduced, and breaks in ruled lines can be corrected.

Therefore, the present invention can be applied to the field of office automation equipment such as a form recognition system, and its practical effect is very large.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of a table structure extraction device according to the present invention.

FIG. 2 is a configuration diagram of a conventional table structure extraction device.

FIG. 3 is a diagram showing an example when applied to a tilt image by a conventional peripheral distribution method.

FIG. 4 is a diagram showing a table structure extraction process and its result according to a conventional technique.

FIG. 5 is a diagram showing a table structure extraction process and its result according to a conventional technique.

FIG. 6 is a diagram showing a black pixel string.

FIG. 7 is a diagram for explaining the definition of overlap of a series of black pixels.

FIG. 8 is a diagram showing an example of configuring a route.

FIG. 9 is an operation flow chart of a ruled line candidate construction unit in the embodiment of the present invention.

FIG. 10 is a diagram for explaining adjacent line segments.

FIG. 11 is a diagram for explaining a rectangular intersection candidate.

FIG. 12 is a diagram showing a table structure extraction process and its result in the embodiment of the present invention.

FIG. 13 is a diagram showing a table structure extraction process and its result in the embodiment of the present invention.

FIG. 14 is a diagram showing a table structure extraction process and its result in the embodiment of the present invention.

[Explanation of symbols]

 11 Image Compressor 12 Black Pixel Consecution Detector 13 Line Segment Extractor 21 Ruled Line Candidate Constructor 31 Table Structure Extractor

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] December 8, 1994

[Procedure Amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0017

[Correction method] Change

[Correction content]

Rule (2) R ₁ = (y ₁ ; x ₁ , x ₁ ') and R ₂ = (y ₂ ; x ₂ ,
against x ₂ '), if, y ₂ = y ₁ + ₁ and R ₁ and R
If ₂ overlap, it determined that R ₁ and R ₂ are adjacent. Rule (3) For R ₁ , R ₂ , ... R _n that are continuously adjacent (this is called because it consists of a large number of adjacent R _i and R _{i + 1} ), if before and after each R _i If at least one scanning line is adjacent to one pixel series, it is defined that one path is composed of R ₁ , R ₂ , ... R _n . FIG. 8 shows, as an example of this, two paths constituted by a series of black pixels that are continuously adjacent to each other.

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0018

[Correction method] Change

[Correction content]

Then, a straight line segment on the route is extracted according to the straight line discrimination rule. The straight line discriminating rule used here is the following rule (4), and the start point coordinate value and the end point coordinate value are inspected for each black pixel string in each path in order. Rule (4) For R ₁ , R ₂ , ... R _n , which is a part of the path, if the following conditions are satisfied, R ₁ ,
It is determined that R ₂ , ... R _n form a straight line segment. (1) | x _i −x _{i + 1} | <between adjacent black pixel strings
1 and | x _i ′ −x _{i + 1} ′ | <1 holds. (2) In sequence, K black pixel rows are treated as one segment,
For each segment, | x ^{a (a} represents the horizontal line on the x and later also the same.) -X _j | a <1 | <1 and | x ^a '-x _j'. Among them, R ^a is a Oh Ru FIRST black pixel run in the segment ((y ^a;., Represented by ^{x a, x a '))} , R j Other R ^a, the segment It is a series of black pixels. (3) The length is larger than a certain threshold value. (4) When the width is divided by the length, the obtained value is smaller than a certain threshold value.

[Procedure 3]

[Document name to be amended] Statement

[Name of item to be corrected] 0025

[Correction method] Change

[Correction content]

With respect to the procedure, a rectangular intersection candidate is detected according to the following rule (7) for each horizontal ruled line candidate and vertical ruled line candidate. It should be noted that this also considers shortening the time for determining the presence or absence of an intersection by avoiding geometric calculation. Rule (7) For a ruled line candidate composed of line segment candidates L ₁ , L ₂ , ... _Li , L _{i + 1,} ..., L _n , a rectangle composed of those satisfying the following conditions is set as an intersection candidate. To do. The following, as shown in FIG. 11
This will be described with reference to (1) . (1) L _i and L _{i + 1} between the last (after the larger in the y-coordinate of the scan line) black pixel run constituting the L _i is R _i, one of and constituting the L _{i + 1} The last black pixel string (the smaller one in the y coordinate of the scanning line is first) is R _{i + 1} , that is, a rectangle formed by x _i to x _i ′ and y _i to y _{i + 1.} It is determined that (2) If the previous black pixel string forming L ₁ is R ₁ ,
Previously, a black pixel string which overlaps R ₁ and is continuously adjacent is detected, and at a distance y _u . That is, it is determined that the rectangle is composed of x ₁ to x ₁ ′ and _yu to y ₁ . (3) _If the last black pixel string R _n forming L _n exists, a black pixel string that overlaps with R _n and is continuously adjacent is detected, and even up to y _b . That is, it is determined that the rectangle is composed of x _n to x _n ′ and y _n to y _b .

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0026

[Correction method] Change

[Correction content]

FIG. 11 (2) shows an example of a vertical ruled line candidate as defined above. Then, the intersection inspection between the rectangular intersection candidates is performed on the vertical ruled line candidates by the horizontal ruled line candidates. In this figure, (1), (2), and (3) correspond to (1), (2), and (3) of the above rule (7) , respectively. The above inspection gives a true intersection,
Moreover, since a true horizontal ruled line and a vertical ruled line can be obtained, a table structure can be acquired.

[Procedure Amendment 5]

[Document name to be corrected] Drawing

[Name of item to be corrected] Fig. 11

[Correction method] Change

[Correction content]

FIG. 11

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display location G06K 9/36 9/46 A 9289-5L

Claims (4)

[Claims] 1. A table structure extraction device for extracting a table structure in a printed document by using the overlapping and adjacency relations of a series of black pixels, in an n * n square inside the inputted original binary digital data. An image compression unit that uses the result of the logical OR operation of n * n pixel values as one pixel, and detects the horizontal black pixel series and the vertical black pixel series by scanning the compressed binary digital data in the horizontal and vertical directions. And a vertical black pixel series and a horizontal black pixel series which are determined to be continuously adjacent to each other among the horizontal and vertical black pixel series detected by the black pixel series detection section. , And a line segment extraction unit for extracting the horizontal line segment and the vertical line segment in the route constructed by the straight line determination rule, and the line segment overlapped with the horizontal line segment and the vertical line segment extracted by the line segment extraction unit. A horizontal line segment and a vertical line segment that are successively adjacent to each other by using the method of detecting the next adjacent line segment by a series of consecutive black pixels that are adjacent to each other, and then the detected horizontal line segment and vertical line segment are detected. A ruled line candidate forming section that forms a horizontal ruled line candidate and a vertical ruled line candidate, respectively, and the horizontal ruled line candidate and the vertical ruled line candidate formed by the ruled line candidate forming section are overlapped and consecutively adjacent to each other before and after the ruled line candidate. The presence or absence of a portion extending from the ruled line to the front and rear is determined by a series of black pixels, and a horizontal ruled line candidate is used as a reference for a vertical ruled line candidate or a vertical ruled line candidate is used as a reference for a horizontal ruled line candidate to perform a true horizontal check. A table structure extracting device having a ruled line, a vertical ruled line, and a table structure extracting unit for detecting intersections. 2. The line segment extraction unit forms a path from a series of vertically adjoining vertical black pixels and a series of horizontally adjoining black pixels, respectively.
Further, a horizontal line segment and a vertical line segment in the route are extracted by a straight line discriminating rule for sequentially inspecting the start point coordinate value and the end point coordinate value for each black pixel string in the route. Item 1. The table structure extraction device according to item 1. 3. A method for detecting the next adjacent line segment by the ruled line candidate extraction unit by the overlap of the line segments with respect to the horizontal line segment and the vertical line segment and the black pixel string that is continuously adjacent and the adjacent line segment. A horizontal line segment and a vertical line segment that are consecutively adjacent to each other in order are detected by using the method that handles the situation where characters and ruled lines are in contact by comparing the widths with the minutes. 2. The table structure extracting device according to claim 1, wherein each of them constitutes a horizontal ruled line candidate and a vertical ruled line candidate. 4. The table structure extraction unit includes a rectangular intersection candidate, which extends forward and backward from the ruled line, in an overlap before and after the ruled line candidate with respect to the horizontal ruled line candidate and the vertical ruled line candidate and in a continuous black pixel series. Find the rectangular intersection candidates for the line segments that make up the ruled line candidate, and then use the horizontal ruled line candidate as the reference for the vertical ruled line candidate or the vertical ruled line candidate as the reference for the horizontal ruled line candidate. 2. The table structure extraction device according to claim 1, wherein the detection is performed to detect true horizontal ruled lines, vertical ruled lines, and intersections.