JPH05334490A - Table recognizing device - Google Patents

Abstract

PURPOSE:To provide a table recognizing device capable of accurately segmenting respective frames constituting a table even in the case of a table omitting a part or the whole of vertical and horizontal ruled lines. CONSTITUTION:The device is provided with a character block extracting means 11 for extracting a character block from a table image and a positional relation identifying means 12 for identifying positional relation between character blocks extracted by the means 11 and outputting data expressing the structure of the table. Since the structure of the table is recognized by using the arrangement of character blocks, the structure of the table can be accurately recognized even when part or the whole of vertical and horizontal ruled lines is omitted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a table recognition device for recognizing a table structure from a table image in the field of document image processing.

[0002]

2. Description of the Related Art Conventional table recognition methods include a method using a peripheral distribution of a table area and a method of converting a ruled line forming a table into a vector line segment and extracting a rectangular frame surrounded by the ruled line. Are known. A method using the marginal distribution is disclosed in, for example, Japanese Patent Laid-Open No. 2-61775, and a method using a vector line segment is, for example, Japanese Patent Laid-Open No. 1-21935.
8 publications are available.

The method using the peripheral distribution described in Japanese Patent Laid-Open No. 2-61775 takes the peripheral distribution of an image in a table area and estimates the position of a ruled line from a mountain having a height higher than a certain threshold from the histogram of the peripheral distribution. Then, the ruled line of the outer frame having the ruled line position on the outermost side of the table is taken out. Next, a ruled line contacting both ends of this outer frame is obtained, and the outer frame is divided into a plurality of rectangular frames by the ruled line. Further, the same processing is recursively performed on each of the divided rectangular frames to extract a rectangular frame surrounded by ruled lines. The latter method disclosed in Japanese Patent Laid-Open No. 1-129358 recognizes a table by tracing the vector line segments and checking the positional relationship between the extracted rectangular frames.

These systems are based on the assumption that the ruled lines that make up the table are not omitted, but there are quite a few tables that are actually used in a document in which some of the ruled lines are omitted. In the method described in Japanese Patent Laid-Open No. 2-264386, even if the ruled lines on both sides of the table are omitted, the rectangular frame can be taken out correctly. That is, it is a method of determining whether there are ruled lines on both sides of the table from the vertical ruled lines and the horizontal ruled lines extracted from the table image and virtually generating vertical ruled lines on both sides of the table when there is no ruled line.

[0005]

Conventionally, there are various types of tables used in documents. FIG. 2 shows an example thereof. The table in FIG. 2 (a) is a table in which all ruled lines are aligned, (b) is a table in which ruled lines on both sides are omitted, and (c) and (d) are both sides. In addition to the above ruled lines, a table having vertical ruled lines and horizontal ruled lines which are omitted, and (e) is a table in which all ruled lines are omitted. Of these, the tables of (a) and (b) can be dealt with by the conventional technique, but vertical ruled lines omitted in addition to the ruled lines on both sides as in the tables of (c) and (d), When there are horizontal ruled lines and when all ruled lines are omitted like the table in (e), it is not possible to correctly recognize the structure of the table and extract the character strings in meaningful units as a table. It was An object of the present invention is to provide a table recognition device capable of accurately cutting out each frame forming a table even in a table in which some or all of vertical and horizontal ruled lines are omitted. is there.

[0006]

The table recognition device of the present invention comprises a character block extracting means (11 in FIG. 1 and 81 in FIG. 8) for extracting a character block from a table image, and the character block extracting means. It is provided with a positional relationship identifying means (12 in FIG. 1, 82 in FIG. 8) for identifying the positional relationship between the extracted character blocks and outputting data representing the structure of the table as a basic configuration. According to the present invention, the positional relationship between the character blocks extracted by the character block extracting means is identified by the positional relationship identifying means. Since the character blocks in the table generally have a positional relationship in which they are regularly arranged as constituent elements of the table, the structure of the table can be recognized by looking at the positional relationship between the character blocks. In the past, since the frame forming the table was sought by paying attention only to the ruled lines of the table, it is impossible to accurately recognize the structure of the table in which some or all of the vertical ruled lines and the horizontal ruled lines are omitted. Although there is a problem, according to the present invention, since the structure of the table is recognized by using the arrangement of the character blocks, the problem can be solved.

According to one aspect of the present invention, in the basic configuration described above, the character block extracting means is a character rectangle extracting means (see FIG. 1) for obtaining a rectangular area surrounding a block of pixels in which characters are written. 111) and the distance between the respective character rectangles obtained by the character rectangle extracting means, and all the character rectangles whose distances are smaller than a certain threshold are integrated as one character block. Character block rectangle extracting means (112 in FIG. 1). ) Is provided. The threshold value may be determined by examining statistics of the distance between the entire character rectangles, or by determining what percentage of the width of the character is the standard.

According to another aspect of the present invention, in the above basic structure, the character block extracting means separates the characters in the table from the ruled lines and vectorizes the ruled lines (see FIG. 811), a character area extraction means (812 in FIG. 8) for extracting a rectangular area in which a character should be written as a character area based on the ruled line vector data obtained by the ruled line vectorization means, and the character. For each character area obtained by the area extracting means, a character rectangle extracting means (8 in FIG. 8) for obtaining a rectangular area surrounding a block of pixels in which characters are written.
13) and the distance between the character rectangles obtained by the character rectangle extraction means, and all character rectangles smaller than a certain threshold are set to 1
And a character block rectangle extraction unit (814 in FIG. 8) that is integrated as one character block. This has a configuration in which ruled line vectorization means and character area extraction means are added to the character block extraction means described in the previous paragraph (0007). According to this aspect, the ruled line vectorization unit obtains the ruled line, the character region extraction unit checks the region sandwiched by the ruled lines to grasp each character region in which the character is to be written, and the character rectangle is extracted in each of the character regions. As a result, the character block can be accurately extracted.

According to another aspect of the present invention, in the above basic structure, the positional relationship identifying means uses a character block rectangle extracted by the character block extraction process as a constituent frame, and a line direction of the constituent frame. Row extracting means (121 in FIG. 1, 821 in FIG. 8) for identifying the arrangement of columns, and column extracting means for identifying the arrangement in the column direction of the constituent frames that make up the table extracted by the constituent frame identifying means (122 in FIG. 1). , 822 in FIG. 8). Further, in a more specific aspect thereof, the row extracting means is configured to extract a group of constituent frames having the same y-coordinate of the center of each constituent frame within a predetermined error range as the same row, and extract the column. The means are arranged to extract groups of constituent frames in which the x-coordinate of the center of each constituent frame is the same within a predetermined error range as the same column. Since the character blocks in a table are generally arranged along the rows and columns of the table, the table structure is determined by grouping rows and columns by checking the arrangement in the row and column directions using the character block rectangle as a frame. Can be extracted.

Further, according to another aspect of the present invention, in the above basic structure, a rectangular frame extracting means for extracting a rectangular frame surrounded by ruled lines is further provided, and the rectangular frame surrounded by the ruled lines in the positional relationship identifying means. The character blocks in the table are treated equally and each positional relationship is identified. That is, this table recognition device is a character block extracting means (11 in FIG. 11) for extracting a character block from a table image.
3) and a rectangular frame extracting means for extracting a rectangular frame surrounded by ruled lines forming a table from the table image (112 in FIG. 11).
And a positional relationship identifying means for identifying the positional relationship between the rectangular frame extracted by the rectangular frame extracting means and the character blocks extracted by the character block extracting means, and creating data representing the structure of the table (see FIG. 11). 114) and. By treating the rectangular frame surrounded by the ruled line of the table and the character block in the table equally, even a frame in the table not surrounded by the ruled line is one component in the table as a character block. (A) in FIG. 2,
Not only the table of (b) but also the tables of (c), (d), and (e) can be recognized accurately.

In the above invention, the rectangular frame extraction means is based on the connection relation between the ruled line vectorization means (1121 in FIG. 8) for converting the ruled line image into vector data and the ruled line vector output by the ruled line vectorization means. First rectangular frame extraction means (1121 in FIG. 11) for obtaining a rectangular frame, and a rectangular frame in which some ruled lines are omitted from a ruled line vector whose one end is not connected to any other ruled line vector 2 rectangular frame extracting means (1123 in FIG. 11).

In the above invention, the positional relationship identifying means is
According to one aspect thereof, a component frame identification for identifying a component frame forming a table from a rectangular frame formed by the ruled lines of the table extracted by the rectangle extraction means and a character block rectangular frame extracted by the character block extraction processing. Means (114 in FIG. 11)
1) and the row extracting means (1 in FIG. 11) for identifying the arrangement in the row direction of the constituent frames forming the table extracted by the constituent frame identifying means.
142) and column extraction means (1 in FIG. 11) for identifying the arrangement in the column direction of the constituent frames that make up the table extracted by the constituent frame identification means.
143). Further, the constituent frame identification means is
In a specific aspect, with respect to the rectangular frame extracted by the rectangular extraction means, a character block within the rectangular frame is extracted, and when there are a plurality of character blocks, the plurality of character blocks are respectively determined as constituent frames. However, when there is a single character block, the rectangular frame is determined as the constituent frame. By determining (recognizing) the constituent frame in this way, even if there is a rectangular frame of ruled lines in which the ruled lines are partially omitted as shown in FIG. Can be accurately determined.

[0013]

【Example】

(First Embodiment) FIG. 1 is a diagram showing the configuration of the first embodiment of the present invention. The table recognition device of this embodiment is for recognizing the structure of a table by checking the arrangement state of a character block consisting of a series of characters, and extracting the character block from the character image in the table image as shown in FIG. Character block extraction unit 1
1 and a positional relationship identifying unit 12 for identifying the positional relationship between the character blocks extracted by the character block extracting unit 11 and obtaining data representing the structure of the table.

The character block extraction unit 11 calculates a character rectangle extraction processing unit 111 that obtains a rectangular area that surrounds a block of pixels in which characters are written, and a distance between each character rectangle obtained by the character rectangle extraction processing unit 111. A character block rectangle extraction processing unit 112 that obtains and integrates a character rectangle whose distance is smaller than a certain threshold as a character block. Further, the positional relationship identifying unit 12 receives a character block rectangle extracted by the character block extracting process as a constituent frame, and a line extraction processing unit 121 for identifying the arrangement in the row direction of the constituent frame,
Column extraction processing unit 122 for identifying the arrangement of the constituent frames in the column direction
And a table structure storage unit 123 that stores the identification result of the positional relationship.
It consists of

The processing of each section of the present embodiment configured as above will be described in detail. The image to be processed in this embodiment is a table image obtained by separating a table area from a document image including a table input by an image input device such as an image scanner. The table area separating means is a table area separating device that is specified by an operator on a screen with a pointing device such as a mouse, or a table area separating device that automatically separates images based on image attributes (see, for example, Japanese Patent Laid-Open No. 2-210586). ) And the like, all of which are known techniques. The character rectangle extraction processing unit 111, as shown in (a) and (b) of FIG. 3, for the character image portion in the front image, a block of pixels 31, 32, 33, 34 in which characters are written. Rectangular area 3 surrounding
Find 5, 36, 37, 38. That is, when the table image is written with a background pixel value of 0 and a character / line pixel value of 1, a block having a pixel value of 1 is taken out to obtain its rectangular area. At this time, when the two rectangular areas overlap each other, they are represented by a rectangular area 39 which can include the two rectangular areas 37 and 38 as shown in FIG. 3B. The method of extracting the rectangular area of the character is a well-known technique (see, for example, Japanese Patent Laid-Open No. 2-267678), and detailed description thereof will be omitted.

Further, the character block rectangle extraction processing unit 112.
Then, the distance between the character rectangles calculated by the character rectangle extraction processing unit 112 is calculated, and all character rectangles smaller than a certain threshold are set to 1
Performs the process of integrating as one character block. The threshold used in this processing may be determined by examining statistics of the distances between the entire character rectangles, or may be determined as a few% of the size of the character, and the method of determining the threshold is not specified here. When this process is applied to the table without ruled lines shown in FIG. 4A, the result is as shown in FIG. The rectangular frame of the character block obtained as a result of these processes is given an identifier, and the position (x coordinate, y coordinate), width, height, etc. of the rectangular frame is stored as data in an appropriate memory.

FIGS. 5a and 5b are diagrams showing the flow of processing of the character block rectangle extraction processing unit 112. Figure 5
Reference character a indicates a processing procedure for obtaining the threshold value for grouping the character rectangles into blocks. As a storage unit that stores constants and intermediate results necessary for processing, a constant N, a variable sum _w that stores the totaling result of the width of the character rectangle, a variable sum _h that stores the totaling result of the height of the character rectangle, width, and height Thresholds T _w and T _h and a variable i are prepared. First, as an initial setting, the total number of character rectangles extracted by the character rectangle extraction processing unit 111 is set to N, and sum _w , sum _h, and i are set.
Are set to 0 (step 501). And
It is determined whether i does not exceed N (step 50).
2) When i <N, sum _w and sum _h are character rectangles C
The width and height of i are added (step 503), and the added value is divided by 2N (step 504). Then, while increasing i by 1 (step 505), the processes of steps 502 to 505 are repeated until i becomes larger than N. i
Is greater than N, the width and height thresholds T _w ,
T _h is obtained as a value of ½ of the average width of the character rectangle.

When the threshold value is obtained by the process of FIG.
The process of grouping the character rectangles into blocks is performed. Set variables j and B to 0 (step 50).
6). It is determined whether the character rectangle C _j has been registered in any of the character blocks CB (step 507). If registered, the variable j is incremented by 1 to process the next character rectangle (step 517). Is determined in step 507, if the character rectangle C _j was still registered, registers the character rectangle C _j into character blocks CB _B (Step 50
8). The registered character rectangle C _j becomes the leading character rectangle of one character block CB _B. Next, the character rectangle whose distance from the registered character is within the threshold T _w or T _h is searched for and registered in the character block CB _B. Therefore, first, the variable k is set to j (step 5).
09). Then, it is checked whether or not the character rectangle C _k is already registered in any of the character blocks (step 510). If it has not been registered, the distance D between CB _B and C _k is calculated (step 511). It is checked whether or not the obtained distance D is within the threshold T _w or T _h (step 512). Distance D
If There were within the range of the threshold value T _w or T _h, and add the character rectangle C _k into character blocks CB _B, to change the size of the CB _B (step 513). When it is determined in step 510 that the character rectangle C _k has been registered, when it is determined in step 512 that the distance D is not within the range of the threshold T _w or T _h , and additional processing in step 513. When you have finished, to find the next character rectangle, k = k
After setting it to +1 (step 514), it is determined whether or not the processing for all the character rectangles has been completed (step 51).
5) If the processing is not finished yet, the processing of steps 510 to 514 is repeated for the set next character rectangle. When it is determined in step 515 that k <N is not satisfied, B = B + in order to obtain the next character block.
The value is set to 1 (step 516) and j = j + 1 (step 517). While j <N, steps 507-
The process of step 518 is continued, and when j <N is not satisfied, the process ends (step 519).

Next, the function of the positional relationship identifying section 12 will be described. The positional relationship identifying unit 12 is composed of the three processing units, the row extraction processing unit 121 and the column extraction processing unit 122 as described above, and will be described below in order. In this embodiment, the character block extracted by the character block extraction unit 11 is registered as it is as a constituent frame. FIG. 4C shows the configuration frame.

Row extraction processing section 121 and column extraction processing section 122
Then, the character block rectangles extracted by the character block rectangle extraction processing unit 121 are regarded as the constituent frames forming the table, and their arrangement is identified. 6a and 6b are flowcharts of the row extraction process, and FIGS. 7a and 7b are flowcharts of the column extraction process. As shown in the figure, the coordinates of the center points of all the constituent frames are calculated, and in the line extraction processing, the constituent frames in which the Y coordinates of the central points of the constituent frames are arranged within a certain error range are identified as rows of the table, In the column extraction processing, the constituent frames in which the X-coordinates of the center points of the constituent frames are arranged within a certain error range are identified as columns in the table.

That is, in the line extraction process, the process shown in FIGS.
As shown in b, first, the total number of constituent frames is set to a variable N (step 601). The Y coordinates of the center points of all the constituent frames are obtained and stored in the array CB (step 602). Search for the one with the maximum height in all the rectangular frames, and set the height to 1
/ 2 is the value of the error range threshold T _h (step 60)
3). Next, the array CB of Y coordinates of the character blocks is sorted in ascending order (step 604). Then, i = G = 0, y
= CB _i and clear the row array (step 60)
5). Next, the Y coordinate CB _i of one constituent frame from the array CB
Is taken out and it is judged whether or not it has been registered in the row array (step 607). For the unregistered component frame CB _i , it is judged whether or not the distance from y is within the threshold value T _h.
CB _i −y | <T _h is determined (step 60
8) If the distance from y is within the threshold value T _h , the configuration frame corresponding to CB _i is stored in the row array (step 609). When the character block has been registered, and when the distance from y is not within the range of the threshold value T _h ,
Set i = i + 1 to retrieve the next character block (step 610). Judgment and registration processing in the same row array for the new character block taken out (step 60)
7 to 609). When the process progresses and i <N is not satisfied (determination in step 606), the extraction process for one row ends, and the process proceeds to the flow of FIG. 6b to perform the extraction process for the next row. The contents of the row array are output as the Gth row information (step 611). Next, the row array is cleared, and i = 0 and G = G + 1 are set (step 61).
2). Then, the constituent frame that should be the head of the (G + 1) th row is searched. That is, the composition frames are taken out one by one from the beginning, and it is judged whether or not the composition frames are already registered in any of the rows (step 614), and the first unregistered composition frame found should be the head of the G + 1th row. The frame y is designated, and i = 0 is set (step 616), and the process proceeds to the extraction processing of one row in steps 606 to 610 of FIG. 6a. When it is determined in step 613 that i <N is not satisfied,
That is, when there is no unregistered component frame, the line extraction process is terminated.

The column extraction process is shown in FIGS. 7a and 7b.
It ’s the same as the row extraction process
Except for the fact that it is present, almost the same processing is performed. That is, all
The X coordinate of the center point of the component frame of is stored in the array CB (step
702) and sort in ascending order (step 704). all
½ of the maximum width in the rectangular frame is the threshold T of the error range_wof
Value (step 703), i = G = 0, x = CB_i
To clear the row array (step 705). Next
And the configuration frame CB stored in the array CB._iTake one by one
Outgoing and unregistered component frame CB_iFor x
The distance T is the threshold T of the error range_wWhether it is within the range
Is determined (steps 707 to 708) and is within the error range.
CB if _iStore the configuration frame corresponding to to the column array
(Step 709). Extraction process for one column ends
Then, to perform the extraction process of the next column, go to the flow of FIG. 7b.
move on. Next, the first unregistered structure that should be the beginning of the next column.
I searched for a frame (steps 714-715) and found it
Et al., The extraction process of one column in steps 706 to 710 of FIG. 7a.
Move to. Column extraction process when there are no unregistered components
End the reason.

Row extraction processing section 121 and column extraction processing section 1
By the processing of 22, the constituent frames are grouped into rows and columns as shown in (d) and (e) of FIG. The output data is
For example, it is stored in the table structure storage unit 123 in a format in which the row number and the column number are given to the identification number representing the configuration frame, and the system can be used in an arbitrary system such as a word processor.

As described above, according to the first embodiment, the character blocks extracted by the character block extraction unit 11 are used as a table configuration frame, and the table structure including the rows and columns is extracted. The table structure can be recognized even in a table having no ruled lines as shown in FIG. Even in the case of a table having ruled lines, the table structure can be recognized based on only the character blocks by the first embodiment.

(Second Embodiment) FIG. 8 is a diagram showing the configuration of the second embodiment of the present invention. The table recognition device of this embodiment is for recognizing the structure of a table by extracting a character area in a table based on a ruled line, extracting a character block in the character area, and checking the arrangement state of the extracted character block. There
Similar to the configuration of the first embodiment shown in FIG. 1, a character block extraction unit 81 for extracting character blocks from a table image and a positional relationship between the character blocks extracted by the character block extraction unit 81 are identified to form a table structure. And a positional relationship identifying unit 82 that generates data representing The second embodiment is different from the first embodiment in the configuration of the character block extraction unit 81, and the character rectangle extraction processing unit 8 is different.
A unit for determining a character area including a ruled line vectorization processing unit 811 and a character area extraction processing unit 812 is added to the preceding stage of 13. The ruled line vectorization processing unit 81
1 is to separate the characters and ruled lines in the table and vectorize the ruled lines. Further, the character area extraction processing unit 812
Based on the ruled line vector data obtained by the ruled line vectorization processing unit 811, a rectangular area in which a character should be written is extracted as a character area.

In order to vectorize only the ruled lines of the table, it is necessary to divide the table into lines and characters. For this separation processing, an existing method similar to the processing for separating a character and a line segment in a figure can be used. It should be noted that, when the technique of the Japanese Patent Application No. 3-290299 “Character / figure separation device” (inventor Noboru Shimizu), which was previously filed by the applicant of the present application, is used, accurate separation processing with few errors can be performed faster. It can be carried out. The character / graphic separation device will be briefly described. As shown in FIG. 9, this is a feature extraction unit 91 that extracts two or more features of each black pixel block in the input image.
And an initial cluster center determination unit 92 that obtains an initial cluster center using the feature extraction result of the feature extraction means 91.
And a region determination unit 93 that determines a region by performing clustering using the feature extraction result of the feature extraction unit 91 and the determination result of the initial cluster center determination unit 92. As the feature amount of each black pixel block, for example, the area of the black pixel block, the flatness ratio, the complexity of the contour line, or the like can be used. When such a feature amount is extracted by the feature extraction unit 91, the initial cluster center determination unit 92 then obtains the center of the initial cluster using the distribution of the extracted feature amounts of the black pixel blocks. The area determination unit 93 uses the initial cluster centers determined by the initial cluster center determination unit 92 for two or more feature amounts of the extracted black pixel clusters, and should belong to each black pixel cluster. Determine the area.

The ruled line area of the separated table converts the binary image into vector data having start points and end points of characteristic points such as end points, polygonal lines, intersections, and branch points. The method for converting this vector data is based on existing technology (for example, Technical Report PRL83-8, PRL85-24, PRL86-).
89, JP-A-2-210586, JP-A-2-10
5265, etc.) may be used, and the description thereof is omitted here.

FIGS. 10a and 10b are diagrams showing the flow of the extraction processing of the character area extraction processing unit 812. The ruled line vector obtained by the ruled line vectorization processing unit 811 is divided into vertical ruled lines VR and horizontal ruled lines HR (step 1001), each is counted, the number of vertical ruled lines is stored in V, and the horizontal ruled lines are stored in H. (Step 1002). The presence / absence of a horizontal ruled line is determined (step 1003). If there is no horizontal ruled line, the number of character areas R is set to 1 and the size of the area is set as the size of the input image (step 1009). If there is a horizontal ruled line, the number of character areas R is set to H-1, and i is set to 0 (step 10).
04). Next, the horizontal ruled lines are sorted in ascending order of the Y coordinate (step 1005). Then, numbers are assigned to the character areas in ascending order of the Y coordinate. That is, the area delimited by the i-th horizontal ruled line and the i + 1-th horizontal ruled line is set as the i-th character area (step 1007). When i <R is not satisfied (step 1006), the number assignment is completed,
The process moves to the processing of the character area by the vertical ruled lines shown in FIG. 10b.

The presence or absence of vertical ruled lines is determined (step 101
0), if there is no vertical ruled line, the number R of character areas is set to 1 and the size of the area is set as the size of the input image (step 101).
8). If there is a vertical ruled line, the content of the number R of character areas by the horizontal ruled line obtained in the processing of FIG. 10A is transferred to R1, and R + V
-1 is set, and i, j, and k are set to 0 (step 1011). Next, the vertical ruled lines are sorted in ascending order of the X coordinate (step 1012). Then, for each area delimited by the horizontal ruled line, the area delimited by the vertical ruled line is obtained (steps 1014 to 1017). That is, regarding the j-th area delimited by the horizontal ruled lines, the area delimited by the i-th and i + 1-th vertical ruled lines is set as the k-th character area (step 1013). This numbering is sequentially i and k
Is incremented by 1 and repeated until it is determined that there is no unprocessed area delimited by the vertical ruled line in the j-th area (steps 1014 and 1015). Then, in order to process the area separated by the next horizontal ruled line, j is set to 1
And i is cleared to 0. Then, steps 1013 to 1017 are repeated until the processing is completed for the last one of the areas delimited by the horizontal ruled lines, that is, until it is determined that j <R1 is not satisfied. As described above, the character area delimited by the ruled lines is extracted, and the result is passed to the character rectangle extraction processing unit 813.

The operation of the processing units after the character rectangle extraction processing unit 813 is basically the same as that of the first embodiment. However,
The character rectangle extraction processing and the character block extraction processing are performed using the information of the character area extracted by the character area extraction processing unit 812. Therefore, the extraction of the character rectangle becomes easier and more reliable, and with respect to the character block, there is no error in detecting characters that are close to each other with the ruled line as one block, and the character block can be reliably extracted. You can

(Third Embodiment) FIG. 11 is a block diagram showing a third embodiment of the present invention. The table recognition device according to this embodiment includes a character / ruled line separation processing unit 1110 for separating a character portion and a ruled line portion included in a table image, and a ruled line forming a table from the ruled line image separated by the character / ruled line separation processing unit 1110. Rectangular frame extraction unit 1 for extracting a rectangular frame surrounded by
120, a character block extraction unit 1130 for extracting a character block rectangular frame forming a table from the character images separated by the character / ruled line separation processing unit 1110, and a rectangular frame extraction unit 1
The rectangular frame extracted by 120 and the positional relationship identifying unit 1140 that identifies the positional relationship between the character blocks extracted by the character block extracting unit 1130 and creates data representing the structure of the table, and the positional relationship identifying unit 1140 Table structure storage unit 11 for storing data representing the structure of the opened table
44 and.

The rectangular frame extraction unit 1120, based on the ruled line vectorization processing unit 1121 for vectorizing the ruled line image and the ruled line vector output from the ruled line vectorization processing unit 1121,
A complete rectangular frame extraction processing unit 1122 that extracts a complete rectangular frame surrounded by a ruled line vector, and an incomplete rectangular frame in which some of the ruled lines are omitted and there is no part of the rectangular frame is extracted to compensate for the lack. Incomplete rectangular frame extraction processing unit 1 that creates a rectangular frame
And 123.

The character block extraction unit 1130 includes a character area extraction processing unit 1131, a character rectangle extraction processing unit 1131,
The character block rectangle extraction processing unit rectangle extraction unit 11 is provided. The character area extraction processing unit 1131 is a rectangular frame extraction unit 1.
Each of the areas surrounded by the rectangular frame obtained by 120 is determined as a character area, and the character / ruled line separation processing unit 1110 is executed.
The character image from is cut out for each character area and passed to the character rectangle extraction processing unit 1132. Character rectangle extraction processing unit 113
2 is for obtaining a rectangular area surrounding a block of pixels in which characters are written. Character block rectangle extraction processing unit 1133
Is to obtain the distance between the character rectangles obtained by the character rectangle extraction processing unit 1132 and integrate all the character rectangles whose distance is smaller than a certain threshold value as one character block.

The positional-relationship identifying unit 1140 identifies the positional relationship between the rectangular frame extracted by the rectangular-frame extracting unit 1120 and the character blocks extracted by the character block extracting 113. Processing unit 1141, row extraction processing unit 1142, column extraction processing unit 1
143 and a table structure storage unit 1 for storing the extraction results thereof
It consists of 141. The configuration frame identification processing unit 1141
The constituent frame forming the table is identified from the rectangular frame composed of the ruled lines of the table extracted by the rectangle extracting means and the character block rectangular frame extracted by the character block extracting means. The line extraction processing unit 1142 includes the component frame identification processing unit 1
The arrangement in the row direction of the constituent frames forming the table extracted in 141 is identified, and the column extraction processing unit 1143 determines that the constituent frame identification processing unit 1
It identifies the arrangement in the column direction of the constituent frames that make up the table extracted in 141.

The operation of this embodiment configured as described above will be described. The character / ruled line separation processing unit 1110 can use an existing method similar to the processing for separating a character and a line segment in a figure. By using the technique of Japanese Patent Application No. 3-290299 "Character / Figure Separation Device" mentioned in the second embodiment, an accurate separation process with few errors can be performed at a higher speed. The information of the ruled line image separated here is output to the rectangular frame extraction unit 1120, and the information of the character image is output to the character block extraction unit 1130.

The ruled line image is a ruled line vectorization processing unit 1121.
Is vectorized with. That is, the binary image is converted into vector data having characteristic points such as end points, polygonal lines, intersections, and bifurcation points as start points and end points. As a method of converting this vector data, the existing technology described above may be used. The converted ruled line vector data is a complete rectangular frame extraction processing unit 11 for extracting a complete rectangular frame surrounded by the ruled line vector.
22 and an incomplete rectangular frame extraction processing unit 1123 that extracts an incomplete rectangular frame in which a part of the ruled lines is omitted and a part of the rectangular frame is omitted to form a rectangular frame.

The complete rectangular frame extraction processing unit 1122 extracts a complete rectangular frame surrounded by the ruled line vector based on the ruled line vector data. 12a and 12b are a flow chart of the process. In the rectangular frame in the table, vertical vector data is connected to the left and right of one horizontal vector data, and horizontal vector data is connected below it. Therefore, each horizontal vector data is examined and vector data satisfying the condition is searched. Fill in the rectangular frame configuration table shown in FIG. First, the number of all vector data forming the table is counted (step 1201). The processes of steps 1202 to 1212 below are applied to all vector data. Horizontal vector data V that is the upper ruled line of the rectangular frame
Search for _i (step 1203). This is because the horizontal vector data can be found because the angle between the vector data and the horizontal line is less than or equal to a certain threshold. Since the horizontal vector data V _i found here may become the upper ruled line of the k-th rectangular frame, the vector data V _{i is written in} the column of the upper ruled line of the k-th rectangular frame of the rectangular frame configuration table 141.
Is registered (step 1204). Next, the vector data forming the right side of the rectangular frame W _k is searched (step 12).
05). That is, in contact with the right end of the end point of the vector data V _i, and the end point of more not in contact with the vector data V _i performs the processing of finding the vertical vector data such that below the vector data V _i. The vertical vector data can be easily obtained because the angle formed by the perpendicular is less than or equal to a certain threshold. Since the vector data found in this step may form the right ruled line of the rectangular frame W _k , it is registered in the right ruled line of the k-th rectangular frame of the rectangular frame configuration table 141 (step 1206). Similarly, a rectangular frame W
The left ruled line of _k is searched (step 1207) and registered in the column of the left ruled line of the kth rectangular frame of the rectangular frame configuration table 141 (step 1208). Further, find the vector data that touches the lower sides of the right and left ruled lines that have just been found (step 120).
9), it is registered in the column of the lower ruled line of the kth rectangular frame of the rectangular frame configuration table 141 (step 1210). If at least one ruled line is not found in the above processing, all the registrations of the kth rectangular frame in the rectangular frame configuration table 141 are discarded, and a rectangular frame composed of other vector data can be registered. Reset to. The rectangular frame configuration table 141 when the above processing is applied to the table of FIG. 13 is as shown in FIG. Further, as another example, in the processing on the table shown in FIG. 15, the rectangular frame configuration table 141 becomes as shown in FIG. Further, the rectangular frame configuration table 141 is rewritten to a rectangular frame table 171 represented by the X coordinate, Y coordinate of the upper left corner of each rectangular frame, and the width and height of the rectangle for convenience of the subsequent processing. The rectangular frame table of FIG. 14 is as shown in FIG.

The incomplete rectangular frame extraction processing unit 1123 extracts an incomplete rectangular frame in which a part of the ruled lines is omitted and a part of the rectangular frame is omitted, and the insufficient part is supplemented to extract it as a rectangular frame of the table. 18a and 18b are a flow chart of the process. First, unregistered vectors other than the vector data registered in the rectangular frame configuration table as the elements of the rectangular frame extracted by the complete rectangular frame extraction processing unit 1122 are extracted (steps 1801 to 1806). To do that, first
The total number of vector data is set in N, and i = k = 0 is cleared (step 1801). The vector data V _i is taken out, and it is checked whether or not it is registered in the rectangular frame configuration table (step 1803). If it is not registered, it is registered in the vector column VV (step 1804) and counted by the counter k (step 1805). ). Then, i = i + 1 is set to retrieve the next vector (step 1806). If the vector data V _i is registered in the rectangular frame configuration table, the process moves to the next vector as it is (step 1806). When i reaches N, that is, when the registration processing of the unregistered vector is completed for all the vectors (step 1802), the horizontal / vertical vector connecting the two nearest endpoints in the unregistered vector string VV is complemented ( Step 1807). The supplemented number is n
And Let k + n be the total number of vectors and i = m =
It is cleared to 0 (step 1808). Search for horizontal vector data VV _i which is the upper ruled line of the rectangular frame (step 18).
10). This is because the horizontal vector data can be found because the angle between the vector data and the horizontal line is less than or equal to a certain threshold. Here found horizontal vector data VV _i is, m-th rectangular frame W so on is likely to be a border of _m, the vector data in the column on borders of the m-th rectangular frame incomplete rectangular frame configuration table to register the VV _i (step 1811). Then search for vector data constituting the right side of the rectangular frame W _m (step 1812). That is, in contact with the right end of the end point of the vector data VV _i, and the end point of more not in contact with the vector data VV _i performs the processing of finding the vertical vector data such that below the vector data V _i. The vertical vector data can be easily obtained because the angle formed by the perpendicular is less than or equal to a certain threshold. Since the vector data found in this step may form the right ruled line of the rectangular frame W _m , it is registered in the right ruled line of the m-th rectangular frame in the incomplete rectangular frame configuration table (step 1813). Similarly looking left border of the rectangular frame W _m (step 1814), incomplete rectangular frame configuration table m
It is registered in the left ruled line of the second rectangular frame (step 181).
5). Further, the vector data which touches the lower side of the right ruled line and the left ruled line thus obtained is found (step 1816), and registered in the column of the lower ruled line of the m-th rectangular frame W _m of the incomplete rectangular frame configuration table (step 1817). .. If at least one ruled line is not found in the above processing, all the registrations of the m-th rectangular frame W _{m in} the incomplete rectangular frame configuration table are discarded, and a rectangular frame composed of other vector data is deleted. Reset to allow registration. FIG. 20B shows an example of an incomplete rectangular frame configuration table, which represents the incomplete rectangular frame portion of the table in FIG. Further, the incomplete rectangular frame configuration table is set to the X coordinate, Y coordinate of the upper left corner of each rectangular frame for convenience of the subsequent processing.
Rewrite to the rectangular frame table expressed by coordinates, width and height of rectangle.

Next, the processing of the character block extraction unit 1130 will be described. The character area extraction processing unit 1131 finds a rectangle in which a character should be written and which is delimited by ruled lines, and registers it in the character area table. In the present embodiment, since the rectangular frame is extracted by the complete rectangular frame extraction unit and the incomplete rectangular frame extraction unit, it may be registered in the character area table. In the example of FIG. 19, a character area surrounded by two complete rectangular frames and a character area within four incomplete rectangular frames are obtained. As another example, for a table lacking ruled lines as shown in FIG. 21A, this processing is performed by supplementing the ruled lines as shown in FIG. 21B and then as shown in FIG. A character area 211 including a plurality of character blocks is extracted. Subsequent processing proceeds for each of the character areas obtained here. By thus obtaining the character regions and extracting the character blocks for each of the character regions, it is possible to prevent the extraction of the character blocks that straddle the ruled line.

Next, the character rectangle extraction processing 1132 will be described. Here, for each character area obtained by the character area extraction processing unit 1131, a rectangular area enclosing a block of pixels in which characters are written is obtained. That is, when the table image is written with a background pixel value of 0 and a character / line pixel value of 1, a block having a pixel value of 1 is taken out to obtain its rectangular area.
When the two rectangular areas overlap, they are represented by a rectangular area 39 that can include the two rectangular areas 37 and 38 as shown in FIG. Since the method of extracting the rectangular area of the character is an existing technique, detailed description thereof will be omitted.

Further, the character block rectangle extraction processing unit 113
In step 3, the distance between the character rectangles calculated by the character rectangle extraction processing unit 1132 is calculated, and all the character rectangles smaller than a certain threshold are integrated into one character block. The details of the processing are basically the same as the processing of the character block rectangle extraction processing unit 112 in the first embodiment, and are shown in the flowcharts of FIGS. 5a and 5b. Since this flow chart has already been described in the first embodiment, description thereof will be omitted here. However, in the case of the first embodiment, whether or not to integrate into a block is determined only by the threshold value. However, this embodiment refers to the information of the character area and only if the same character area exists. Integrate into a character block. As a result, it is possible to prevent the extraction of character blocks that cross ruled lines.

Finally, the operation of the positional relationship identifying section 1140 will be described. The positional relationship identification 114 further includes a component frame identification processing unit 1141, a row extraction processing 1142, a column extraction processing 1
It is composed of three processing units 143. In the component frame identification process,
The frame that actually forms the structure of the table is a process of identifying and selecting whether the frame is a rectangular frame composed of ruled lines of the table or a frame of a character block. Since there should be at least one character block inside the rectangular frame, if the number of character blocks inside the rectangular frame is counted and two or more character blocks are confirmed, the A block is registered as a constituent frame, and when only one character block exists, a rectangular frame is registered as a constituent frame.

FIG. 22 is a flow chart showing details of the above-mentioned component frame identification processing. A variable i that specifies the elements of the array of the complete rectangular frame w, where N is the total number of complete rectangular frames and M is the total number of character blocks, variable C that specifies the elements of the identified constituent frame array, and each complete rectangular frame Each variable s for counting the character blocks included in the is set to 0 (step 220).
1). The variables j and s that specify the elements of the array of the character block CB _j are set to 0 (step 2202). The character block CB _j is extracted, and it is determined whether or not it is included in the complete rectangular frame W _i (step 2203). If it is included, s is incremented (step 2204), and if it is not included, nothing is done. Then, j is incremented to fetch the next character block (step 220).
5). The above steps 2204 to 2205 are sequentially repeated until it is determined that there are no unprocessed character blocks (step 2206). In this way
When all the character blocks have been checked for one complete rectangular frame, it is determined whether or not there are a plurality of character blocks s included in the complete rectangular frame (step 220).
7) If there are a plurality of them, s character blocks included in the complete rectangular frame W _i are registered as constituent frames (step 2208). Since s pieces have been registered, C is set to C + s (step 2209). On the other hand, when the number s of character blocks included in the complete rectangular frame W _i is 1, the complete rectangular frame W _i
i is registered as the Cth component frame (step 221).
0) and C are incremented (step 2211).
Through the above processing, when the related constituent frame is identified for one complete rectangular frame, i = i + 1 is set to perform similar processing for the complete rectangular frame, and the process returns to step 2202. When it is determined that i <N is not satisfied (step 2213), all the processes are completed. 23A and 23B show an example of the result of extracting the constituent frames. FIG. 10A shows an example of the table, and FIG. 9B shows the constituent frames extracted from the table of FIG.

Row extraction processing section 1142 and column extraction processing section 11
At 43, the arrangement of the frames forming the table extracted by the configuration frame identification processing unit 1141 is identified. That is, the coordinates of the center points of all the constituent frames are obtained, and the row extraction processing unit 1142 identifies the constituent frames in which the Y coordinates of the central points of the constituent frames are arranged within a certain error range as a row of the table, and performs the column extraction processing. In 1143, the constituent frames in which the X coordinates of the center points of the constituent frames are arranged within a certain error range are identified as columns in the table. The details of this processing are shown in FIG. 6A and FIG. 6B for the row extraction processing flow and in FIG.
a and FIG. 7b. For a detailed description of these processes,
This is the same as that described in the first embodiment. The results of the row extraction processing 152 and the column extraction processing 153 are shown in FIGS.
5 shows.

After performing such processing, the extracted rows and columns are numbered in order, and the data of the word processor table is described according to this numbering, so that the table input as an image can be written in the word processor. It is possible to convert it into a table that can be edited. Further, when the characters are cut out by using the configuration frame, when the ruled lines are insufficient on both sides of the table as shown in the table of FIG. 24A, the table of FIG. The second and third columns are rectangular frames with ruled lines,
The rest can be extracted as constituent blocks that are character blocks.
It is also possible to easily input a table lacking lines into the character recognition device. For example, when there are insufficient ruled lines on both sides of the table as shown in the table of FIG. 24 (a), the second and third columns of the table have ruled lines as shown in FIG. 24 (b). It is possible to extract a constituent frame which is a rectangular frame consisting of, and the rest is a character block. Further, when all the vertical ruled lines and part of the horizontal ruled lines are omitted as in the table shown in FIG. 25A, the constituent frames are all character blocks as shown in FIG. 25B. Become. Further, when a part of the vertical ruled lines and the horizontal ruled lines of the table is omitted as in the table shown in FIG. 26A, a rectangular frame composed of ruled lines can be extracted in the table. Due to the fact that it contains multiple character blocks within it, FIG.
As shown in (b), all the constituent frames are character blocks. For any type of table as shown in FIGS. 24, 25, and 26, the present embodiment can accurately extract the row and column structure of the table. In the present embodiment, at least the table in which the ruled lines are written has been described as an object, but by performing the same processing, the present invention can be applied to a table including no ruled lines at all. Furthermore, it is possible to add a structure as a table to a sentence that is not explicitly written as a table, for example, a document of a clause.

[0046]

According to the present invention, the positional relationship between the character blocks extracted by the character block extracting means is identified by the positional relationship identifying means. Since the character blocks in the table generally have a positional relationship in which they are regularly arranged as constituent elements of the table, the structure of the table can be recognized by looking at the positional relationship between the character blocks. In the past, since the frame forming the table was sought by paying attention only to the ruled lines of the table, it is impossible to accurately recognize the structure of the table in which some or all of the vertical ruled lines and the horizontal ruled lines are omitted. Although there is a problem, according to the present invention, since the structure of the table is recognized by using the arrangement of the character blocks, the problem can be solved.

According to the aspect in which the character area extracting means of the present invention is provided, the character area extracting means extracts the character area by using the information of the ruled line, and the character block is extracted for each character area. Therefore, there is no possibility that character rectangles that are close to each other with a ruled line in between will be extracted as one block, the character blocks can be extracted with high accuracy, and the structure of the table can be accurately recognized.

In the present invention, a rectangular frame extracting means for extracting a rectangular frame surrounded by ruled lines is provided, and the positional relationship identifying means treats the rectangular frame surrounded by the ruled lines and the character blocks in the table equally and identifies each positional relationship. In such a mode, by treating the rectangular frame surrounded by the ruled lines of the table and the character block in the table equally, even the frame in the table not surrounded by the ruled lines is treated as the character block. 1 in the table
Since it is identified as one component, a table in which some or all of the ruled lines are omitted can be recognized as accurately as a table in which all the ruled lines are complete.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a first exemplary embodiment of the present invention.

2A to 2E are diagrams showing examples of tables used in a document.

3A and 3B are diagrams showing an example of a character rectangle.

4A to 4E are views for explaining recognition of a table in which all ruled lines are omitted.

FIG. 5a is a diagram showing a flow of character block extraction processing.

FIG. 5b is a diagram showing a flow of character block extraction processing (continuation from FIG. 5a).

FIG. 6a is a diagram showing a flow of a line extraction process.

FIG. 6b is a diagram showing a flow of line extraction processing (continuation of FIG. 6a).

FIG. 7a is a diagram showing a flow of column extraction processing.

FIG. 7b is a diagram showing a flow of column extraction processing (continuation from FIG. 7a).

FIG. 8 is a diagram showing a configuration of a second exemplary embodiment of the present invention.

[Figure 9]

FIG. 10a is a diagram showing a flow of character region extraction processing in the second embodiment.

FIG. 10b is a diagram showing a flow of character area extraction processing in the second embodiment (sequel to FIG. 10a).

FIG. 11 is a diagram showing a configuration of a third exemplary embodiment of the present invention.

FIG. 12a is a diagram showing a processing flow of a complete rectangular frame extraction processing unit.

FIG. 12b is a diagram showing a processing flow of the complete rectangular frame extraction processing unit (continuation of FIG. 12a).

FIG. 13 is a diagram showing an example of vector data forming a table.

FIG. 14 is a diagram showing a diagram showing an example of a rectangular frame configuration table.

FIG. 15 is a diagram showing another example of vector data forming a table.

FIG. 16 is a diagram showing another example of the rectangular frame configuration table.

FIG. 17 is a diagram showing an example of (a) a rectangular frame table and (b) a character area table.

FIG. 18a is a diagram showing a flow of incomplete rectangular frame extraction processing.

FIG. 18b is a diagram showing a flow of incomplete rectangular frame extraction processing (continuation of FIG. 18a).

FIG. 19 is a diagram showing an example of vector data forming a table in which some ruled lines are omitted.

20 shows an example of a rectangular frame configuration table corresponding to the table of FIG. 19, (a) showing a complete rectangular frame configuration table, and (b) showing an incomplete rectangular frame configuration table.

FIG. 21 is a diagram for explaining extraction of a character area.

FIG. 22 is a diagram showing a processing flow of a component frame identification processing unit.

FIG. 23 is a diagram showing an example of extraction of a configuration frame, rows, and columns, (a) is an example of a table in which the vertical ruled line at the right end is omitted, (b)
Indicates an extracted constituent frame, (c) indicates an extracted row, and (d) indicates an extracted column, respectively.

FIG. 24 is a diagram showing another example of extraction of a configuration frame, rows and columns, where (a) is an example of a table in which vertical ruled lines at the left and right ends are omitted, and (b) is an extracted configuration frame ( c) shows the extracted row, and (d) shows the extracted column, respectively.

FIG. 25 is a diagram showing another example of extraction of a configuration frame, rows and columns, (a) is an example of a table in which all vertical ruled lines are omitted,
(B) is the extracted configuration frame, (c) is the extracted line,
(D) shows each extracted example.

FIG. 26 is a diagram showing another example of extraction of a configuration frame, rows, and columns, (a) is an example of a table in which some vertical and horizontal ruled lines are omitted, and (b) is an extracted configuration. A frame, (c) shows an extracted row, and (d) shows an extracted column, respectively.

[Explanation of symbols]

11, 81 ... Character block extraction unit, 111, 813 ... Character rectangle extraction processing unit, 112, 814 ... Character block rectangle extraction processing unit, 12, 82 ... Positional relationship identification unit, 121, 8
21 ... Row extraction processing unit, 122, 822 ... Column extraction processing unit,
123, 823 ... Table structure storage unit, 811 ... Ruled line vectorization processing unit, 812 ... Character area extraction processing unit, 1110 ... Character / ruled line separation processing unit, 1120 ... Rectangular frame extraction unit, 112
1 ... Ruled line vectorization processing unit, 1122 ... Complete rectangular frame extraction processing unit, 1123 ... Incomplete rectangle extraction processing unit, 1130 ...
Character block extraction unit 1132 ... Character rectangle extraction processing unit,
1133 ... Character block rectangle extraction processing unit, 1140 ... Positional relationship identification unit, 1141 ... Configuration frame identification processing unit, 1142
... Row extraction processing unit, 1143 ... Column extraction processing unit, 1144 ...
Table structure storage unit, 31, 32, 33, 34 ... Black pixel block, 3
5, 36, 39 ... Character rectangle, 161, ... Composition frame, 171 ...
Rectangular frame table, 172 ... Character area table 211 ... Character area.

Claims (15)

[Claims] 1. A character block extracting unit for extracting a character block from a table image, and a positional relationship identifying unit for identifying a positional relationship between the character blocks extracted by the character block extracting unit and generating data representing a table structure. A table recognition device having means. 2. A character / ruled line separating means for separating a front image into a character image and a ruled line image is provided, and the character image separated by the character / ruled line separating means is input to the character block extracting means. The table recognition device according to claim 1. 3. The character block extracting means calculates a value of 1 based on a character rectangle extracting means for obtaining a rectangular area surrounding a block of pixels in which characters are written and a distance between the character rectangles obtained by the character rectangle extracting means. The table recognition device according to claim 1, further comprising a character block rectangle extraction unit that integrates the above character rectangles as a character block. 4. The character block rectangle extraction means obtains the distance between the character rectangles obtained by the character rectangle extraction means, and integrates consecutive character rectangle groups with a distance smaller than a certain threshold value into one character block. The table recognition device according to claim 3, wherein the table recognition device is performed. 5. The character block extraction means is a character / ruled line separation means for separating characters in the table from the ruled lines, a ruled line vectorization means for vectorizing the ruled lines separated by the character / ruled line separation means, and a ruled line vectorization. 1 based on the distance between each character rectangle obtained by the character rectangle extraction means and the character area extraction means for extracting a rectangular area in which a character should be written as a character area based on the ruled line vector data obtained by the means.
A character block rectangle extracting means for integrating the above-mentioned character rectangles as a character block is provided.
The described table recognition device. 6. The positional relationship identifying means regards the character block rectangle extracted by the character block extracting means as a constituent frame of a table, and the row extracting means for identifying the arrangement of the constituent frame in the row direction; and the constituent frame. 2. The table recognition device according to claim 1, further comprising a column extracting means for identifying the arrangement in the column direction. 7. The row extracting means extracts a group of constituent frames having the same y-coordinate of the center of each constituent frame within a predetermined error range as the same row, and the column extracting means extracts each constituent frame. 7. The table recognition device according to claim 6, wherein groups of constituent frames having the same x-coordinate of the center within a predetermined error range are extracted as the same column. 8. A rectangular frame extracting means for extracting a rectangular frame surrounded by ruled lines forming a table from a target table area, a character block extracting means for extracting a character block from the target table area, and the rectangular frame. A table recognition device comprising: a rectangular frame extracted by the extraction means; and a positional relationship identification means for identifying the positional relationship between the character blocks extracted by the character block extraction means. 9. A character / ruled line separating means for separating a front image into a character image and a ruled line image is provided, and the character image separated by the character / ruled line separating means is input to the character block extracting means to extract the rectangle. 9. The table recognition device according to claim 8, wherein the ruled line image separated by the character / ruled line separating means is input to the means. 10. The rectangular frame extraction means is based on a ruled line vectorization means for converting the ruled line image separated by the character / ruled line separation means into vector data, and a connection relationship between the ruled line vectors output by the ruled line vectorization means. And a second rectangular frame extracting means for estimating a rectangular frame in which some of the ruled lines are omitted from a ruled line vector whose one end is not connected to any other ruled line vector. 9. The table recognition device according to claim 8, further comprising means. 11. The character block extracting means obtains a character rectangle extracting means for obtaining a rectangular area surrounding a block of pixels in which characters are written, and a distance between each character rectangle obtained by the character rectangle extracting means, 9. The table recognition device according to claim 8, further comprising character block rectangle extraction means for integrating one or more character rectangles as a character block based on the distance. 12. The character block extracting means obtains the character area extracting means for extracting a rectangular area in which a character is to be written as a character area based on the output of the rectangular frame extracting means, and the character area extracting means. For each character area, one or more character rectangles based on the distance between the character rectangle extracting means for obtaining a rectangular area enclosing a block of pixels in which characters are written and each character rectangle obtained by the character rectangle extracting means. 9. The table recognition device according to claim 8, further comprising a character block rectangle extraction unit that integrates as a character block. 13. The positional relationship identifying means identifies a constituent frame forming a table from a rectangular frame composed of ruled lines of the table extracted by the rectangle extracting means and a character block rectangle extracted by the character block extracting means. And a row extracting means for identifying the arrangement in the row direction of the constituent frames constituting the table extracted by the constituent frame identifying means, and a column direction of the constituent frames constituting the table extracted by the constituent frame identifying means. 9. The table recognition device according to claim 8, further comprising a column extraction unit for identifying a row. 14. The component frame identifying means extracts a character block in the rectangular frame extracted by the rectangular extracting means, and when there are a plurality of character blocks, the plurality of character blocks are extracted. 14. The table recognizing device according to claim 13, wherein each of the above is determined as a constituent frame, and when there is a single character block, a process of determining a rectangular frame as a constituent frame is performed. 15. The line extracting means is arranged at the center y of each constituent frame.
A group of constituent frames whose coordinates are the same within a predetermined error range is extracted as the same row, and the column extracting means of the constituent frames whose x-coordinates of the centers of the respective constituent frames are the same within a predetermined error range. 14. The table recognition device according to claim 13, wherein the groups are extracted as the same column.