JP3582734B2 - Table vectorizer - Google Patents

Description

[0001]
[Industrial applications]
The present invention relates to a table vectorizing apparatus for vectorizing a table such as a form or a drawing.
[0002]
Vectorizing tables such as forms and drawings is an important technology in the fields of OCR (optical character reading device), map processing, document processing, and the like. For example, in OCR, it is required to recognize characters from not only a form dedicated to OCR but also a form available on the market.
[0003]
By automatically vectorizing the form table, the table can be identified and the table can be extracted, thereby improving the OCR recognition rate and adding value. In the drawings, vectorization of a table such as a parts table is indispensable.
[0004]
[Prior art]
FIG. 10 is an explanatory diagram of a conventional technique. In FIG. 10, 1 indicates a front image, 2 indicates a front portion, and 3 indicates a projected image (the number of dots).
[0005]
Conventionally, in a table vectorizing apparatus, when vectorizing a table portion 2 from a table image 1 such as a form or drawing, vectorization is performed by extracting ruled lines of the table.
In this case, the ruled lines forming the table are projected on the vertical axis (Y axis) and the horizontal axis (X axis) (the number of dots) on the assumption that the ruled lines constituting the table are horizontal or vertical with respect to the table image 1. I was seeking from 3. The outline of this vectorization processing is as follows.
[0006]
First, a form or the like is read from a scanner or the like to obtain a front image 1 (an image in the memory). Thereafter, a projection image 3 in the vertical axis direction and the horizontal axis direction is obtained for the front image 1.
This projected image 3 has a value obtained by accumulating the number of dots in each axis (X, Y) direction of the XY coordinates (for example, if a dot in a black portion is “1” and a dot in a white portion is “0”). , Dot “1”).
[0007]
The projected image 3 also obtains not only the ruled lines of the table but also the dots such as characters and numerals of the table portion 2 (dots other than the ruled lines are included as noise).
Therefore, in order to extract only the ruled line of the table portion 2, it is determined whether or not the projected image (the number of dots) 3 is equal to or larger than a predetermined threshold value. In this way, only the ruled lines can be extracted.
[0008]
[Problems to be solved by the invention]
The above-described conventional device has the following problems.
Forms and drawings are slightly inclined when read by an image input device such as a scanner. In this case, a large inclination can be corrected by coordinate conversion.
[0009]
However, even if the inclination is, for example, 1 ° or less, the straight line has an inclination of several pixels due to the quantization error in the image, but this inclination cannot be corrected. If the inclination cannot be corrected, the conventional method as described above cannot uniquely determine a threshold used for extracting a ruled line.
[0010]
An object of the present invention is to solve such a conventional problem, and to accurately vectorize a table portion even when an input table image is slightly inclined.
[0011]
[Means for Solving the Problems]
FIG. 1 is a diagram illustrating the principle of the present invention. In FIG. 1, the same components as those in FIG. 10 are denoted by the same reference numerals. 5 is a table vectorizing device, 6 is an image input unit, 7 is a projection unit, 8 is a mask image generation unit, 9 is a ruled line search unit, 10 is a first frame memory, and 11 is a second frame memory. Show.
[0012]
The present invention is configured as follows in order to solve the above problems.
(a) : A table vectorization device for obtaining a projection image on the vertical axis (Y axis) and the horizontal axis (X axis) from the input table image 1 and vectorizing the table portion 2, For pixels corresponding to the table portion of FIG. 1 , on the horizontal axis (X-axis), a horizontal line segment is ignored, and only the vertical line segment is extracted to extract a projected image. On the axis), a projection unit that obtains a projection image of a ruled line constituting a table by ignoring a vertical line segment and extracting a projection image limited to only a horizontal line segment, and a projection of the ruled line A mask image generation unit 8 that draws a straight line having the same width as the image on a memory (second frame memory 11) in the horizontal and vertical directions and generates the same as a mask image, and a mask generated by the mask image generation unit According to the image, the ruled line is searched according to the ratio of the length between intersections and the number of pixels between intersections, and the table part is vectorized. A table vectorization apparatus provided with a ruled line search unit 9 for converting into a rule .
[0014]
(b) : In the configuration (a) , when the ruled line search unit 9 searches for a ruled line in accordance with the mask image, an intersection of straight lines is extracted from the mask image, and a ratio of the number of pixels to the distance between the extracted intersections is calculated as follows. A table vectorization device for determining the existence of a ruled line between the intersections.
[0015]
(c) : In the configuration (a) , when the ruled line search unit 9 searches for a ruled line in accordance with a mask image, a table vectorizing apparatus that searches for a ruled line using a tree structure of a straight line.
(d): in the construction (a) to (c), the vertical performed vectorization unit, and the horizontal component processing unit that performs vectorization processing of the horizontal component of the table image 1, the vectorization processing of the vertical component ingredients A table vectorizing apparatus, comprising a processing unit, dividing an input table image 1 into a horizontal component and a vertical component, and performing parallel processing by the horizontal component processing unit and the vertical component processing unit.
[0016]
[Action]
The operation of the present invention based on the above configuration will be described with reference to FIG.
The following processing is performed on the front image 1 stored in the first frame memory 10 by inputting a form, a drawing, and the like from the image input unit 6.
[0017]
First, the projection unit 7 extracts a projection image of a vertical / horizontal ruled line for a pixel corresponding to the table portion 2 in the table image 1. In this case, the projection on the horizontal axis (X axis) is limited only to the vertical line segment (ignoring the horizontal line segment), and conversely, the vertical axis (Y axis) is projected in the horizontal direction. Limit to line segments only (ignore vertical line segments).
[0018]
This allows the projection images to be clearly distinguished from each other, whereas conventionally, the projection images contact each other and cannot be separated.
Next, labeling is performed on each projected image on the vertical axis (Y axis) and the horizontal axis (X axis). In this process, the number of pixels (the number of dots) is counted for each label, and only an image reaching a certain threshold is left (noise component is removed).
[0019]
Subsequently, in accordance with the projection image, a straight line having the width of the image is drawn from the vertical / horizontal direction to fill the image to generate a mask image. After that, the intersection of the ruled lines in the table is calculated by extracting the coordinates of the intersection of the straight lines from the mask image. Finally, a ruled line is searched to vectorize the table portion. In this case, the coordinates are extracted also for the intersection of the end of the table image and the straight line.
[0020]
Then, it is checked whether or not a ruled line exists between the intersections. As a result, if a ruled line exists, the coordinates of both ends are recorded.
As described above, it is possible to accurately vectorize the table portion even for a subtle inclination of the input table image. In addition, when the vectorization processing of the table portion is performed, the processing of the horizontal component and the vertical component is performed in parallel, so that the processing speed can be increased.
[0021]
【Example】
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
2 to 9 are views showing an embodiment of the present invention. In FIGS. 2 to 9, the same components as those in FIGS. 1 and 10 are denoted by the same reference numerals. Reference numeral 12 denotes a first projection memory, and reference numeral 13 denotes a second projection memory.
[0022]
§1: Description of the configuration of the table vectorization device ... See FIG. 2 FIG. 2 is a device configuration diagram of the embodiment. Hereinafter, the configuration of the table vectorization device will be described with reference to FIG.
[0023]
The table vectorizing device 5 is a device provided in various processing devices such as a personal computer, a workstation, and OCR.
The table vectorizing device 5 includes an image input unit 6, a projection unit 7, a mask image generation unit 8, a ruled line search unit 9, a first frame memory 10, a second frame memory 11, and a first projection memory. 12, a second projection memory 13 and the like.
[0024]
The image input unit 6 reads a form, a drawing, and the like and inputs a table image, and is configured by an optical device such as a scanner or a camera.
The projection unit 7 projects the input table image in the vertical axis (Y axis) and horizontal axis (X axis) directions. The mask image generation unit 8 generates a mask image, and the ruled line search unit 9 performs a ruled line search process according to the mask image (details will be described later).
[0025]
§2: Outline of processing of embodiment ... refer to FIG. 3 FIG. 3 is a processing flowchart of the embodiment. Hereinafter, the processing outline of the embodiment will be described with reference to FIG. Note that S1 to S5 in FIG. 3 indicate each processing number.
[0026]
The following processing is performed on a table image (see a conventional example) obtained by inputting a form or drawing from a scanner.
In this case, the input table image is a binary digital image. In the table image, the pixel value of the table portion (ruled line portion) is “1”, and the other values are “0”, and f (x, y) It expresses. In addition, it is assumed that the table portion has been subjected to inclination correction (correctable large inclination correction) by performing processing such as coordinate transformation on the table image.
[0027]
In the input table image, a process (S1) of extracting a projected image (the number of dots) of a vertical / horizontal ruled line is performed on pixels corresponding to the table portion.
Specifically, the projection on the horizontal axis (X axis) is limited only to the vertical line segment (ignoring the horizontal line segment), and conversely, the horizontal axis (Y axis) is projected horizontally. Limited to line segments in the direction only (ignoring line segments in the vertical direction). This allows the projection images to be clearly distinguished from each other, whereas conventionally, the projection images contact each other and cannot be separated.
[0028]
Next, labeling is performed on each of the projected images on the vertical axis and the horizontal axis (S2). In this process, the number of pixels (the number of dots) is counted for each label, and only an image reaching a certain threshold is left (noise component is removed). When setting the threshold value, the ruled line of the table is sufficiently longer than the characters and the like, and thus can be easily determined.
[0029]
Subsequently, a straight line having an image width is drawn to fill the image according to the projection image, and a mask image is generated (S3). Note that an image obtained by drawing a straight line as described above is referred to as a “mask image”. Thereafter, the intersection of the ruled lines in the table is calculated by extracting the coordinates of the intersection of the straight lines from the mask image (S4).
[0030]
Finally, a ruled line is searched (S5), and the table portion is vectorized. In this case, the coordinates are extracted also for the intersection of the end of the table image and the straight line. Then, it is checked whether or not a ruled line exists between the intersections. As a result, if a ruled line exists, the coordinates of both ends are recorded.
[0031]
§3: Process description of the embodiment FIG. 4 is a process description of the embodiment FIG. 1 (projection image extraction process), FIG. 5 is a process description of the embodiment FIG. 2 (mask image generation process), FIG. 6 is a process description of the embodiment FIG. 3 (ruled line intersection calculation processing), FIG. 7 is a processing explanatory diagram of the embodiment (ruled line search processing), and FIG. 8 is a processing explanatory diagram of the embodiment (vector hierarchical processing). Hereinafter, each process of the embodiment will be specifically described.
[0032]
(1): Description of Table Image Input Processing The table image input processing is performed by the image input unit 6. That is, the image input unit 6 optically reads a form or drawing and converts a table into a digital image (table image).
[0033]
In the input table image, the table portion has a value of “1”, and the other portions are binary images having a value of “0”.
In this example, the size of the table image is N × M pixels. The input table image is temporarily stored in the first frame memory 10. In this case, it is assumed that the size of the first frame memory is the same as the size of the front image (N × M pixels).
[0034]
(2): Description of Projected Image Extraction Process—See FIG. 4 The projection unit 7 performs a projected image extraction process on the input table image. In this processing, the projection unit 7 sets a mask of, for example, 9 × 1 pixels and 1 × 9 pixels, and scans the first frame memory 10 separately in the horizontal and vertical directions.
[0035]
That is, when scanning in the horizontal direction, a mask of 9 × 1 pixels (9 pixels in the horizontal direction and 1 pixel in the vertical direction) is used, and when scanning in the vertical direction, 1 × 9 pixels (1 pixel in the horizontal direction) is used. Pixel, 9 pixels vertically).
[0036]
As a result of scanning the first frame memory 10 using the mask, for example, if the number of pixels “1” is 80% or more of the total number of pixels in the mask, If there is a pixel in which the number of pixels including “1” is “1” continuously for at least half of the number of pixels in the mask, the central pixel of the mask is left.
[0037]
However, pixels to be left in the horizontal axis (X-axis) direction are left in the second frame memory 11, and pixels to be left in the vertical axis (Y-axis) direction are left in the first frame memory 10. In this case, the original table image (image at the time of input) stored in the first frame memory 10 is deleted.
[0038]
With the above processing, pixels that become noise are removed, and the image left in the horizontal direction and the pixel left in the vertical direction are separately stored in different frame memories.
After that, projection processing is performed on the vertical axis and the horizontal axis from the table images stored in the first frame memory 10 and the second frame memory 11, and a projection image is obtained (the number of pixels is accumulated).
[0039]
Then, as shown in FIG. 4, the first frame memory 10 is scanned in the horizontal direction, and the number of pixels having the value “1” in each column is stored in the first projection memory 12. At this time, the number of pixels (the number of dots) is stored in the address corresponding to each column.
[0040]
Further, the second frame memory 11 is scanned in the vertical direction, and the number of pixels having the value “1” in each column is stored in the second projection memory 13. At this time, the number of pixels (the number of dots) is stored in the address corresponding to each column.
[0042]
(3): Description of Image Labeling Process—See FIG. 4 The mask image generation unit 8 compares the projection images stored in the first projection memory 12 and the second projection memory 13 A labeling process is performed by assigning a number to each (a lump portion having the number of dots).
[0043]
For example, in FIG. 4, one label (for example, a number) is attached to a portion where one projection (for three pixels), such as “3, 9, 1”, is projected.
(4): Description of mask image generation processing ... see FIG. 5 Next, as shown in FIG. 5, a line having the same width as the width of each block (pixel value = 1) is stored in the second frame memory 11. Draw horizontally and vertically (draw a straight line to fill the memory).
[0044]
In this case, for example, if the block has a width of 3 pixels, a straight line having a width of 3 pixels is drawn. The straight line portion (cross-shaped image) drawn in the horizontal / vertical direction in this manner is called a “mask image”.
The mask image is generated as described above.
[0045]
(5): Explanation of Rule Line Intersection Calculation Process--See FIG. 6 From the mask image generated as described above, the coordinates (x, y) of the line intersection are extracted as shown in FIG. In this case, the intersection of the straight lines is a set of positions having the highest value (the number of dots is large) in the projected image obtained by drawing each straight line in the vertical / horizontal direction.
[0046]
That is, in the example of FIG. 6, the coordinates of the intersection of the straight line drawn at the position "9" of the projected image and the straight line drawn at the position "8" of the projected image are extracted. In this way, the intersections of the ruled lines are extracted.
[0047]
(6): Description of ruled line search processing--see FIG. 7 After the ruled line intersection is detected as described above, the ruled line search unit 9 performs ruled line search processing as follows.
[0048]
For example, as shown in FIG. 7, if there is a straight line having a width A, the first frame memory 10 is referred to, and if a pixel of “1” exists within the width A, it is assumed that a ruled line exists. I do.
[0049]
If the number of pixels assumed to have a ruled line between the intersections occupies a predetermined ratio (for example, 80%) of the length (shortest pixel number) between the intersections, it is assumed that a ruled line exists between the intersections. The coordinates of the end points of the ruled line are stored in the memory 14. This process is repeated at all intersections.
[0050]
In order to obtain a ruled line that is as long as possible, the ruled line search unit 9 sequentially selects intersections connected by straight lines, starting from the distant ones, and inspects whether a ruled line exists between them.
[0051]
(7): Description of vector hierarchization processing ... See FIG. 8 For example, as shown in FIG. 8, the straight lines between intersections are hierarchized as a scale.
[0052]
For example, if there is a straight line AE, the straight line AE is composed of a slightly shorter straight line AD and a straight line BE (they may partially overlap). The straight line AD includes a straight line AC and a straight line BD. When such a relationship is represented by a tree structure, a hierarchical structure is obtained as shown in FIG.
[0053]
In this case, if there is no ruled line on the straight line AE, the presence of the straight line AD and the straight line BE is checked, and similarly, the presence of the ruled line on each straight line is checked. If a ruled line exists on the straight line AE, the existence of a ruled line on the straight line constituting the ruled line is not confirmed.
[0054]
§4: Description of the parallel processing device ... refer to FIG. 9 FIG. 9 is a configuration diagram of the parallel processing device of the embodiment. By using, for example, a parallel processing device as shown in FIG. 9 as an apparatus for performing the vectorization processing, the processing speed can be increased.
[0055]
This apparatus divides an input table image into a horizontal component and a vertical component, and performs horizontal ruled line extraction processing and vertical ruled line extraction processing as parallel processing.
As shown in the figure, the table vectorizing device 5 is provided with a horizontal component processing unit 15 and a vertical component processing unit 16. These processing units are respectively provided with a projection unit 7 and a mask image generation unit 8. , A ruled line search unit 9 is provided.
[0056]
In this apparatus, the table image input from the image input unit 6 is input to the horizontal component processing unit 15 and the vertical component processing unit 16, and the above-described processing is performed as parallel processing for each of the horizontal component and the vertical component. The function of each unit is the same as that of the apparatus shown in FIG.
[0057]
【The invention's effect】
As described above, the present invention has the following effects.
{Circle around (1)} Even if the input table image is slightly inclined, the table can be accurately vectorized.
[0058]
{Circle around (2)} By using a straight tree structure, ruled lines constituting a table can be made as long as possible.
{Circle around (3)} The processing speed can be increased by using a parallel processing device as a device for performing the table vectorization process.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating the principle of the present invention.
FIG. 2 is an apparatus configuration diagram of an embodiment.
FIG. 3 is a processing flowchart of the embodiment.
FIG. 4 is a process explanatory diagram (projected image extraction process) of the embodiment.
FIG. 5 is a process explanatory diagram (mask image generation process) of the embodiment.
FIG. 6 is an explanatory diagram of a process in the embodiment (process of calculating an intersection of ruled lines);
FIG. 7 is an explanatory diagram of a process in the embodiment (a ruled line search process).
FIG. 8 is an explanatory diagram of a process in the embodiment (a vector hierarchical process).
FIG. 9 is a configuration diagram of a parallel processing device according to an embodiment.
FIG. 10 is an explanatory diagram of a conventional technique.
[Explanation of symbols]
1 Table image 2 Table part 5 Table vectorization device 6 Image input unit 7 Projection unit 8 Mask image generation unit 9 Ruled line search unit 10 First frame memory 11 Second frame memory

Claims (4)

From the input table image, in the vertical axis, and in the table vectorization device for vectorizing the table portion to obtain a projected image on the horizontal axis,
For pixels corresponding to the table portion of the table image , the horizontal axis (X-axis) ignores horizontal line segments and extracts only the vertical line segments to extract projected images. (Y-axis), a projection unit that obtains a projection image of a ruled line forming a table by ignoring a vertical line segment and extracting a projection image limited to only a horizontal line segment ;
A mask image generation unit that draws a straight line having the same width as the projection image of the ruled line on a memory in the horizontal and vertical directions, and generates it as a mask image.
A table vector comprising a ruled line search unit for searching for a ruled line at a ratio of the length between intersections and the number of pixels between intersections according to the mask image generated by the mask image generation unit and vectorizing the table part. Device. When the ruled line search unit searches for a ruled line according to a mask image, the intersection of straight lines is extracted from the mask image,
2. The table vectorizing apparatus according to claim 1 , wherein the existence of a ruled line between the intersections is determined from the ratio of the number of pixels to the distance between the extracted intersections . 2. The table vectorizing apparatus according to claim 1 , wherein the ruled line data when the ruled line search unit searches for a ruled line in accordance with a mask image is represented by a straight tree structure . The vectorization device includes a horizontal component processing unit that performs a vectorization process of a horizontal component of a table image, and a vertical component processing unit that performs a vectorization process of a vertical component,
The input table image is divided into a horizontal component and a vertical component, and parallel processing is performed by the horizontal component processing unit and the vertical component processing unit . Table vectorizer.

Images