JP2004334743A - Method for extracting ruled line, method for encoding ruled line, and ruled line extracting program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To extract and encode ruled lines from document image data. <P>SOLUTION: Binary document images are acquired and connecting pixels are recognized from the images. The periphery lengths of the connecting pixels are found, the connecting pixels with larger periphery length than a predetermined threshold THC are determined to be the ruled lines, and the ruled lines are solely extracted. When setting the threshold value THC, the connecting pixels in the document images are stored, the number of pixels with the periphery length for every periphery length for each stored connecting pixel is counted, the local maximum point (x<SB>m</SB>', y<SB>m</SB>') as the farthest point from the straight line connecting a point (x<SB>1</SB>, y<SB>1</SB>) in the rank of the smallest periphery length and a point (x<SB>N</SB>, y<SB>N</SB>) in the rank of the largest periphery length in the relation between the periphery lengths and the number of pixels is found, and the periphery length of the local maximum point is set as the threshold value THC. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a method for extracting ruled lines from image data of a document.
[0002]
[Prior art]
In the advanced information society, many documents are digitized and can be used through various media. On the other hand, a large number of paper documents are still used in various situations. In recent years, however, the demand for automatic electronicization of these paper documents has been rapidly increasing in order to disclose information and improve work efficiency. An OCR (Optical Character Reader) greatly assists in the conversion from “paper” to “electronic data”. With the improvement of document image processing technology, OCR has become more practical. With the spread of computers including image scanners and the low price of high recognition rate OCR software, the environment that can be used in society has been improved. Is getting ready.
[0003]
However, with respect to document images containing components other than characters such as charts and ruled lines, the current OCR is often unable to cope. Some OCR software manually specifies the attributes of each area, but in that case, complete automation of the process cannot be performed. In addition, there is OCR software that limits the processing target to tabular documents and specializes in recognizing the table structure. However, in the output, the table structure is expressed with considerable redundancy. Is not concisely represented. That is, it merely maps ruled lines on a fine grid, and does not mean that the system understands the table structure.
[0004]
Tables are a type of expression that is very often used in society as a whole, and are not structurally complex like sentences, but are used to sort the contents by attribute and organize meanings. Is a particularly important element in the text. Therefore, in order to represent the table, the table must be understood not only by visually reproducing the structure but also by using a simple one without redundancy. In recent years, in order to promote electronic commerce (EC) and electronic data exchange (EDI), a technology called XML (extensible Markup Language) has been attracting attention. This attribute is used to make text processing easy, and if there is redundancy in the designation of an area, it becomes difficult to give an attribute to the data.
[0005]
By understanding such a table structure, a technique for converting the table structure (ruled lines) into electronic data is required as a premise for encoding the table structure. As a technique for converting a table structure into electronic data, an input document image is binarized, and the number of black pixels is determined for each row and each column in the vertical and horizontal directions of the image. There is known a method of extracting a ruled line by using a conventional technique (for example, see Patent Document 1).
[0006]
[Patent Document 1]
JP 2001-109888 A (Paragraph 0064, FIG. 18, etc.)
[0007]
[Problems to be solved by the invention]
However, it has been difficult to accurately extract ruled lines when the table described above is greatly inclined or in a document image in which the same characters are continuously arranged.
Therefore, an object of the present invention is to provide a method for easily and accurately extracting a ruled line.
[0008]
[Means for Solving the Problems]
In order to solve the above-mentioned problem, according to claim 1 of the present invention, there is provided a method for extracting a ruled line from a document image in which characters and ruled lines are mixed, wherein a step of obtaining a binarized document image, The method includes a step of recognizing, a step of calculating a perimeter of the connected pixel, and a step of determining a connected pixel whose perimeter is larger than a threshold value as a ruled line.
[0009]
According to such a method, ruled lines and other elements can be distinguished by determining whether or not the length of the perimeter of a connected pixel, that is, a portion recognized as a lump, in a binarized image is greater than a threshold value. it can. Here, the perimeter is the length of one circumference that follows the contour of the connected pixel, and is determined, for example, by the number of pixels forming the contour.
[0010]
In the invention of claim 2, when determining the threshold value, the step of storing connected pixels in the document image, and for each of the stored connected pixels, accumulates the perimeter for each class to calculate a frequency. Obtaining, on orthogonal coordinates using the perimeter and the frequency as coordinate axes, a point of a connected pixel that is the farthest from a straight line connecting a point in the class with the smallest perimeter and a point in the class with the largest perimeter. The determination is made including a step of obtaining a certain local maximum point and a step of setting the circumference of the local maximum point as the threshold value.
[0011]
In this way, when the perimeter is accumulated for each class and the frequency is counted, and the relationship between the perimeter and the frequency is formed on the rectangular coordinates, characters and the like have a high frequency with a small perimeter, while the ruled line is large. There is only low frequency at the perimeter. For this reason, there is a high frequency in a portion where the peripheral length is small, but the frequency is extremely low from a certain peripheral length. When a graph of the above relationship is created, a curve is sharply bent at a certain peripheral length. Then, in order to find a point where this curve bends sharply, a point farthest from a straight line connecting a point in the class with the smallest perimeter and a point in the class with the largest perimeter (in this specification, this is referred to as Point "). If this maximum point is set as a threshold value, it is possible to select only connected pixels having a perimeter longer than this threshold value and extract a ruled line.
Here, the class is referred to because the points that can be plotted on the rectangular coordinates are not continuous with respect to the perimeter, but exist discretely. Using the peripheral length of the local maximum point as the threshold value means that the connected pixel actually selected as the ruled line has the peripheral length of the class next to the threshold value.
[0012]
According to a third aspect of the present invention, there is provided a method for encoding a ruled line image extracted by the ruled line extracting method according to the first or second aspect. Recognizing and, at the node position, determining whether the node can be the upper right, upper left, lower right, or lower left corner of the table structure, and the upper right, upper left, lower right , Or referring to a table in which node numbers are made to correspond one-to-one to combinations of corner points at the lower left, and assigning node numbers to the nodes.
As described above, by determining the node positions of the ruled lines from the image from which the ruled lines are extracted, the node positions can be determined more accurately than by directly determining the node positions from the state where the ruled lines and characters are mixed. Then, for each node, a determination is made as to whether or not this node can be formed by using a table in which node numbers are made to correspond one-to-one, based on a combination of whether each node is located at the upper right, upper left, lower right, or lower left corners. The node number can be assigned only by using the node numbers. In these inventions, each step is executed by a computer.
[0013]
According to a fourth aspect of the present invention, there is provided a ruled line extracting program for extracting a ruled line from a document image in which characters and ruled lines are mixed, the computer comprising: means for inputting a binarized document image; , A means for determining the perimeter of a connected pixel, and a means for determining a connected pixel whose perimeter is greater than a threshold value as a ruled line.
[0014]
BEST MODE FOR CARRYING OUT THE INVENTION
Next, an embodiment of the present invention will be described with reference to the drawings as appropriate.
In the drawings to be referred to, FIG. 1 is a configuration diagram of a system for executing a ruled line extraction method and an encoding method according to an embodiment (hereinafter, referred to as “ruled line extraction encoding device”), and FIG. It is a flowchart which shows the outline of a process of such a ruled line extraction method and encoding method, and FIG. 3 is a functional block diagram of a ruled line extraction encoding device according to the embodiment.
[0015]
As shown in FIG. 1, a ruled line extraction encoding apparatus 1 includes a central processing unit (CPU) 11, a keyboard 13 and a mouse 14 as input devices, a display 16 as an output device, and a storage device 15. The computer is connected to a scanner 12 that reads a document D including ruled lines as a digital image (hereinafter, simply referred to as an “image”). The storage device 15 stores a computer program for realizing each function described later, and is loaded into the CPU 11 and executed as appropriate.
[0016]
The outline of the ruled line extraction method and the encoding method according to the embodiment will be described with reference to FIG. First, the document D is acquired as a binarized image using the scanner 12 (S1). Then, connected pixels connected to one from the acquired image are recognized (S2). Next, the perimeter (perimeter) is calculated for each connected pixel (S3). Next, it is determined whether or not the connected pixel is a ruled line from the length of the peripheral length (S4). Since the perimeter of the ruled line is longer than that of the character, the ruled line may be determined to be a ruled line if it is larger than a certain threshold. Then, a node is found from the ruled line, and the node is encoded, thereby encoding the ruled line (S5). At this time, encoding is performed by referring to the node encoding table 29 as shown in FIG.
[0017]
Next, details of the ruled line extraction encoding apparatus will be described with reference to FIG.
As shown in FIG. 3, the ruled line extraction encoding apparatus 1 includes a document image input unit 21, a connected pixel recognition unit 22, a perimeter calculation unit 23, a threshold value calculation unit 24, a ruled line determination unit 25, It has a node recognition unit 26, a node encoding unit 27, and a node encoding table 29.
[0018]
The document image input unit 21 is configured as software for operating the scanner 12 described above. The document image input unit 21 takes in the document D as a binarized image from the scanner 12 and stores the document D in the storage device 15. When an image is input as a color image or a monochrome multi-tone image, the image may be input and then binarized.
The image input by the document image input unit 21 is, for example, an original image shown in FIG. 4A or an original image shown in FIG.
[0019]
The connected pixel recognition unit 22 is a unit that recognizes a series of black pixels (connected pixels) from the binarized image of the document D input by the document image input unit 21. When a document in which characters and ruled lines are marked in white on a black background is input, it may be configured to recognize white pixels.
Various methods can be used for recognizing connected pixels.
For example, there is a method of labeling each pixel. This method refers to pixels around the target pixel (upper line and left), and if the surrounding pixels are black, assigns the same label to the target pixel as the surrounding pixels. By performing the processing on the entire image in order from the pixel of, a unique label can be attached to each connected pixel.
[0020]
The perimeter calculating unit 23 calculates the perimeter of each connected pixel recognized by the connected pixel recognition unit 22. The perimeter is the length of the outline of the connected pixel, and is obtained, for example, by counting the total number of pixels constituting the outline (for example, a pixel in which at least one of the upper, lower, left, and right pixels is not its own connected pixel). . Of course, the present invention is not limited to this method, and the perimeter may be obtained by another method. In addition, when there is a white pixel surrounded by black pixels like the inside of a ruled line, the perimeter may or may not be counted as the perimeter of this inner black pixel. Counting is preferable because characters and ruled lines can be more accurately distinguished.
[0021]
The threshold value calculation unit 24 makes a relationship between the circumference length of each connected pixel obtained by the circumference length calculation unit 23 and the frequency of the connected pixel having the circumference length, and determines a ruled line and a character from this relationship. This is the part that calculates the threshold value. The perimeter may be graded with an appropriate width according to the resolution of the acquired document image. For example, when the perimeter is counted by the number of pixels forming the periphery of the connected pixels, a class is created every 10 pixels, every 50 pixels, every 100 pixels, and the like, and each connected pixel is assigned to the class. Then, for each class, the number of connected pixels corresponding to the class is accumulated to obtain a frequency.
The relationship between the perimeter and the frequency is, for example, as shown in FIGS. Here, FIG. 5 is a plot of the connected pixels of the document (original image) of FIG. 4 (a), taking the class (perimeter) on the horizontal axis and the frequency for each class on the vertical axis. FIG. 7 is a plot of the connected pixels of the document (original image) of FIG. 6A, with the horizontal axis representing the class (perimeter) and the vertical axis representing the frequency for each class. As can be seen, the frequency distribution shows a high frequency at low classes (short perimeters), a sharp decrease in frequency with higher classes (longer perimeters), and lower frequencies up to the highest class. Distributed in This is because, in general, in a document image in which characters and ruled lines are mixed, the number of characters has a short perimeter and a large number of frequencies, whereas a ruled line has a long perimeter and a very small number of frequencies.
[0022]
A method of obtaining a class threshold for determining whether or not a ruled line is determined from the relationship between the perimeter and the frequency as shown in FIG. 5 or FIG. 7 will be described with reference to FIG. FIG. 8 is a diagram for explaining a method of obtaining a threshold value from the relationship between the perimeter and the frequency corresponding to FIG.
As shown in FIG. 8, the axis (horizontal axis) of the class of the original perimeter is the x-axis, and the axis of frequency (vertical axis) is the y-axis. Then, the point of the class with the smallest perimeter (x ₁ , Y ₁ ) And the largest perimeter class point (x _N , Y _N ) Is created, and this is defined as the x ′ axis. Next, in the direction orthogonal to the x 'axis (in FIG. 8, (x ₁ , Y ₁ )), The y 'axis is taken in the lower left direction. The coordinate system of x'y 'is called an S' system.
Then, in the S ′ system, the point where each point is projected on the y ′ axis is the maximum point (x _m ', Y _m '), And the perimeter of this class is set as the threshold THC.
The maximum point (x _m ', Y _m ') Is the point furthest from the x' axis, as is evident from the distribution of points in FIG.
[0023]
The ruled line determination unit 25 is a unit that determines whether each connected pixel is a ruled line based on the threshold value THC obtained by the threshold value calculation unit 24. That is, for each connected pixel, the perimeter of the connected pixel is compared with the threshold THC, and if the perimeter is larger than the threshold THC, it is determined that the pixel is a ruled line. I do.
In addition, the ruled line determination unit 25 generates an image in which portions (connected pixels) other than the ruled lines are deleted from the document image. The images from which the ruled lines have been extracted in this way are, for example, as shown in FIGS. 4B and 6B.
[0024]
The node recognizing unit 26 is a part that specifies a position of a ruled line in a document image to obtain a node position. For example, the node recognition unit 26 counts the number of black pixels for a line of pixels arranged in the vertical (y-axis) direction of the document image, and performs this for all x coordinates. FIGS. 9A and 9C illustrate the counted number of black pixels (accumulated value). FIG. 9 is a graph showing the cumulative value of black pixels for each of the x and y coordinates, where (a) is the x direction before character erasure, (b) is the y direction before character erasure, and (c) Is the x direction after character erasure, and (d) is the y direction after character erasure. As shown in FIGS. 9A and 9C, the cumulative value shows a high value intermittently on the x coordinate. The x-coordinate with the higher accumulated value is the coordinate at which the vertical line of the ruled line is located.
Similarly, the number of black pixels is counted for a line of pixels arranged in the horizontal (x-axis) direction of the document image, and this is performed for all y coordinates. This is illustrated in FIGS. 9B and 9D. As for the y-coordinate, the y-coordinate indicating the high cumulative value is the coordinate at which the horizontal line of the ruled line is located.
[0025]
(Identification of ruled line position)
There are various ways of recognizing the x-coordinate of the vertical line and the y-coordinate of the horizontal line.
For example, as shown in FIG. 10, three consecutive points of interest (k-1, k, k + 1) are taken on the x-axis, and the distribution (contour) of the accumulated value is tracked while moving them. As can be seen from FIGS. 9 (c) and 9 (d), when the ruled line extraction process is performed, the portion other than the ruled line (connected pixels) is deleted, and the cumulative value of the black pixels is calculated. It is flat and the boundary between the flat part and the ruled line is easy to detect.
Assuming that h (k) is the value of the accumulated value of the black pixel at the coordinate k and the difference between the accumulated values of the ruled line portion and the other portion is the threshold value THR, the following expressions (1) to (3) are obtained. Is the position where the ruled line starts (see FIG. 10A).
h (k + 1) −h (k)> THR (1)
h (k + 1) −h (k−1)> THR (2)
| H (k) -h (k-1) | <THR... (3)
After specifying the position where the ruled line starts, a position (k-1) satisfying both of the following expressions (4) to (6) is defined as a position indicating the end of the ruled line (see FIG. 10B).
h (k−1) −h (k)> THR (4)
h (k−1) −h (k + 1)> THR (5)
| H (k + 1) -h (k) | <THR (6)
When it is necessary to determine the coordinates of the node as one, the center position (coordinates) of the start position and the end position of the ruled line determined by these determinations may be used as the coordinate of the ruled line.
[0026]
Further, as shown in FIGS. 9C and 9D, there is a remarkable difference between the accumulated value of the ruled line position and the other accumulated values. Therefore, the threshold value THR of the accumulated value is set and the threshold value THR is set. If the cumulative value is larger than THR, it may be determined that the line is a ruled line portion, and if the cumulative value is smaller than THR, it may be determined that the coordinate is not the position where the ruled line is located. In this case, as a method of setting the threshold value THR, for example, a constant multiple of the average value of the accumulated values is set as the threshold value THR. When it is necessary to determine the coordinates of the node, a portion determined to be an adjacent ruled line is regarded as one ruled line, and appropriate coordinates, for example, the smallest coordinate or the central The coordinates may be the coordinates of the ruled line.
[0027]
(Node position)
Then, all combinations of the x-coordinate where the vertical line is located and the y-coordinate where the horizontal line is located are created, and the combined x and y coordinates become the coordinates of the node. For example, when it is determined that the x coordinate has a vertical line at 10, 20, and 50 and the y coordinate has a horizontal line at 100 and 120, all combinations of the x coordinate and the y coordinate, that is, (10, 100) ), (10, 120), (20, 100), (20, 120), (50, 100), and (50, 120) are the coordinates of the node.
FIG. 11 shows an example in which the positions of the nodes are obtained in such a manner and are superimposed on the ruled lines. In FIG. 11A, the positions of the nodes are indicated by circles. In addition, E. R. (Element Region) indicates an element area where characters and the like are arranged. In the example shown in FIG. 11, five vertical lines and five horizontal lines are recognized as rule lines, and as a result, 25 nodes combining the x coordinate of each vertical line and the y coordinate of each horizontal line are recognized. I have. Note that a point having neither a vertical line nor a horizontal line, such as a node N44 shown in FIG. 11A, is also recognized as a node having no line.
[0028]
The node encoding unit 27 determines from the shape whether each node can be the upper right, upper left, lower right, or lower left corner of the element region (this is referred to as “corner position determination”). This part encodes each node from a combination of possible angular positions of each node.
[0029]
(Angle position judgment)
The determination of the angular position is performed, for example, as follows. FIG. 12 is a diagram illustrating a method for determining the angular position. In FIG. 12A, each square represents a pixel, and pixels shaded with oblique lines represent black pixels. In FIG. 12A, a horizontal ruled line R1 and a vertical ruled line R2 intersect. In the horizontal ruled line R1, the coordinate ys is the start of the ruled line, and the coordinate ye is the end of the ruled line. In the vertical ruled line R2, the coordinate xs is the start of the ruled line, and the coordinate xe is the end of the ruled line.
The coordinates ys, ye, xs, and xe are specified by specifying the above-described node positions, and the angular position of one node need only be determined around the coordinates.
The determination of the corner position includes the following first to third steps for each corner. Since the basic processing is common to all four types of corners (upper right, upper left, lower right, lower left), here, an example of processing for determining whether a node can be the upper left corner will be described. I do.
[0030]
[First step]
An area D indicated by a thick line in FIG. 12A is set. The area D is an area obtained by moving the area limited by the thickness of the vertical and horizontal ruled lines by one pixel to the lower right (one pixel in the x direction and one pixel in the y direction).
Then, a point that can be the upper left of the element region is searched from the region D. That is, a point that satisfies the following equations (7) to (9) is searched. Note that 0 indicates a white pixel and 1 indicates a black pixel.
b (x, y) = 0 (7)
b (x, y-1) = 1 (8)
b (x-1, y) = 1 (9)
If there is a point that satisfies these conditional expressions, that point is referred to as a notice point T (x _t , Y _t ). If there is no such point, it is determined that the node does not constitute (cannot be) the upper left corner position at that time.
[0031]
[Second step]
The attention point T is moved by a predetermined number of pixels, for example, 200 pixels (pixels) along the black pixels forming the ruled line. The attention point T is moved such that the black pixel is on the right side in the traveling direction. That is, as shown by the downward arrow 2 in FIG. 12A, the pixel moves downward in the figure with the black pixel located on the right side in the traveling direction. Then, when the node is an inverted T-shaped node as shown in FIG. 12B and does not form the upper left corner, the point of interest T advances in a direction in which the x coordinate decreases and reaches xs. I will. Conversely, if the node can be at the upper left corner in the figure, the point of interest T should have a larger x-coordinate (away from xs) but not a smaller one. Therefore, when the x coordinate of the point of interest T reaches xs, it is determined that the node constitutes the upper left corner position, and when it does not reach xs, the node constitutes the upper left corner position.
Only in the first step, as shown in FIG. 12B, even when the vertical ruled line partially protrudes from the horizontal ruled line at the inverted T-shaped node, the condition that the upper left corner position is formed is satisfied. However, by this step, it can be correctly determined that such a node does not constitute the upper left corner position.
[0032]
[Third step]
The point of interest T is set at the initial coordinates (x _t , Y _t ), And shifts the point of interest by a predetermined number of pixels, for example, 200 pixels, along the black pixels forming the ruled line. In this case, the operation is performed such that the black pixel is on the left side in the traveling direction. For example, in FIG. 12A, as shown by a rightward arrow 3 in the figure, the vehicle travels rightward with the black pixel located on the left side in the traveling direction. As in the second step described above, if the node can be at the upper left corner, the point of interest T should have a larger y-coordinate (away from ys) but not a smaller one. Therefore, during this movement, if the y coordinate of the point of interest reaches ys, it is determined that the node does not constitute the upper left corner position. When this movement is completed to the last, it is determined that the node is the upper left component.
[0033]
The above processing of the first to third steps is also performed for the upper right, lower right, and lower left corner positions, and the result of the determination as to whether or not the corner positions can be obtained is stored in the storage device 15. Then, the above processing is performed for all nodes.
[0034]
(Coding)
If the corner position can be determined, a node number is assigned to each node based on a value (○ or ×) indicating whether or not a corner position can be obtained with reference to the node encoding table of FIG.
In the node encoding table 29 of FIG. 13, one node number is associated with each combination of the corner positions (upper left, upper right, lower left, lower right) of the occupied element area. For example, as shown in the second row of FIG. 13, a node that forms only the upper left corner position is a node number n ₁ Is related.
By replacing the nodes with the node numbers according to the node coding table, a matrix as shown in FIG. 11B (this is called a “node matrix”) can be obtained. FIG. 11 (b) is obtained by encoding each node of the table structure of FIG. 11 (a) according to the node encoding table of FIG.
The node number in FIG. ₀ To n ₉ Since each of the ten types is represented by a single-digit number from 0 to 9, it can be encoded as shown in FIG. In other words, each node can be expressed in text with one character number, which is convenient when treating a table as a code in a computer program language.
[0035]
(Post-processing of encoding)
After encoding the nodes, the node encoding unit 27 deletes redundant data portions to obtain compressed data excluding portions other than tables. For example, when a node matrix is obtained from a table T1 as shown in FIG. 14A and an image from which the underline U1 is extracted, the result is as shown in FIG. 14B. In FIG. 14B, the underlined U1 is a non-node (n ₀ ) And the first line is all n ₀ (In the figure, capital N to emphasize ₀ It is expressed by). Thus, non-nodes (n ₀ ) Are continuous, the data amount can be reduced by deleting the row or column. The data after deleting the row is as shown in FIG.
Whether or not to perform this post-processing is optional in the present invention.
[0036]
The operation of the ruled line extraction encoding apparatus 1 as described above will be described with reference to the flowchart of FIG.
First, the document image input unit 21 takes in the document D as an image by the scanner 12 and binarizes it (S11). For example, an image as shown in FIGS. 4A and 6A is input. Then, the connected pixel recognition unit 22 recognizes the connected pixels in the binarized image (S12). Next, the perimeter calculator 23 calculates the perimeter for each connected pixel (S13). Based on the perimeter, the threshold calculator 24 associates the perimeter with the frequency of the connected pixel having the perimeter by classifying the perimeter (S14). When this is illustrated, for example, a graph as shown in FIG. 5 or FIG. 7 is obtained. In this relationship (graph), points of the minimum class and the maximum class are connected by a straight line (S15), a point (class) farthest from the straight line is searched for, and the perimeter thereof is set as a threshold THC (S16).
Then, the ruled line determination unit 25 compares the perimeter of each connected pixel with the threshold value THC to determine whether or not each connected pixel is a ruled line (S17). Further, the connected pixels determined to be not ruled lines are deleted from the original document image (S18), and a new image is generated (see FIGS. 4B and 6B).
Next, the node recognizing unit 26 obtains the cumulative value of black pixels for each x coordinate and y coordinate of the new image generated by the ruled line determining unit 25 (S19). For example, distributions of accumulated values as shown in FIGS. 9C and 9D are obtained. Then, the position of the node is obtained by tracking the distribution of the accumulated value. That is, the cumulative value of the coordinate of interest is compared with the cumulative value of the adjacent coordinate, and whether or not the ruled line is present is determined based on whether or not the cumulative value is greater than the threshold value THR (see Equations (1) to (6)). The position is determined (S20).
Then, at each node position, the node encoding unit 27 determines whether the node can form the upper right, upper left, lower right, or lower left corner position of the element region (table structure) (S21). In this determination, it is determined whether or not the corner position can be reached ([first step]) only by the relationship with the adjacent pixel, and when the pixel is moved along the ruled line (black pixel), It is determined whether or not the angular position of the element coordinates can be formed based on whether or not a certain x-coordinate or y-coordinate is straddled ([second step] [third step]).
Lastly, the node encoding unit 27 refers to the node encoding table 29 to code each node from the combination of the angular positions (n ₀ ~ N ₉ ) (S22).
[0037]
According to the above embodiment, the following effects can be obtained.
First, in order to recognize the position of a ruled line, a relationship between a connected pixel and a perimeter is created. In this relationship, a distance between a straight line and a point is determined. Since the ruled line can be determined by simple processing, the processing speed can be increased.
Then, the position of the ruled line (the position of the node) is determined from the image after the ruled line is extracted, so that mistakes in extracting the ruled line can be reduced.
Further, for a combination of which of the upper right, upper left, lower right, and lower left corner positions the node can form, refer to the node encoding table corresponding to one-to-one, and change the node number from the combination. Since assignment and encoding are performed, ruled lines can be encoded very easily. Furthermore, by deleting non-node rows or columns as post-processing, unnecessary data can be discarded, and the data amount can be reduced.
Further, in the node encoding table of the present invention, since ten types of nodes are arranged, by assigning a number from 0 to 9 to each node model as a node number, the handling of the table structure on a computer program is simplified. It becomes.
[0038]
Since the present invention is a ruled line extraction / encoding method and a ruled line extraction program, it will not be described in detail. However, coded ruled line data is obtained by using the ruled line coding table and a , And connecting the nodes so as to extend the straight line of the node model, the ruled line (table structure) can be reproduced.
[0039]
【The invention's effect】
As described in detail above, according to the present invention, ruled lines can be easily extracted and encoded with a small amount of processing and a small amount of data.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a ruled line extraction encoding apparatus according to an embodiment.
FIG. 2 is a flowchart illustrating an outline of processing of a ruled line extraction method and an encoding method according to the embodiment;
FIG. 3 is a functional block diagram of a ruled line extraction encoding apparatus according to the embodiment.
FIG. 4 is an example of an input image, in which (a) shows an original image and (b) shows a ruled line after extraction.
FIG. 5 is a graph showing a relationship between a perimeter and a frequency in the example of FIG. 4;
FIG. 6 is an example of an input image, in which (a) shows an original image and (b) shows a ruled line after extraction.
FIG. 7 is a graph showing a relationship between a perimeter and a frequency in the example of FIG. 6;
8 is a diagram for explaining a method of obtaining a threshold value from the relationship between the perimeter and the frequency corresponding to FIG. 7;
9A and 9B are graphs showing cumulative values of black pixels for each of the x coordinate and the y coordinate, where (a) is the x direction before the character is erased, (b) is the y direction before the character is erased, and (c) is the character. The x direction after erasure, and (d) the y direction after erasure.
10A is a diagram illustrating a search for a position where a ruled line starts, and FIG. 10B is a diagram illustrating a search for a position where a ruled line ends.
11A is a diagram showing an extracted ruled line superimposed on a node, FIG. 11B is a matrix obtained by encoding the ruled line of FIG. 11A, and FIG. Shows a matrix quantified using letters.
12A and 12B are diagrams for explaining a method of determining a corner position of a ruled line, where FIG. 12A illustrates a case of a cross node and FIG. 12B illustrates a case of an inverted T-shaped node.
FIG. 13 is a node encoding table.
FIGS. 14A and 14B are diagrams for explaining post-processing of encoding. FIG. 14A shows an image after ruled line extraction, FIG. 14B shows a matrix after encoding, and FIG. 14C shows a matrix after post-processing.
FIG. 15 is a flowchart illustrating an operation of the ruled line extraction encoding apparatus according to the embodiment.
[Explanation of symbols]
11 CPU
12 Scanner
15 Storage device
21 Document image input means
22 Connected pixel recognition unit
23 Perimeter calculation unit
24 threshold calculator
25 Ruled line judgment unit
26 Node recognition unit
27 Node coding unit
29 node coding table

Claims (4)

A method of extracting ruled lines from a document image in which characters and ruled lines are mixed,
Obtaining an image of the binarized document;
Recognizing connected pixels;
Determining the perimeter of the connected pixel;
Determining a connected pixel whose perimeter is greater than a threshold value as a ruled line. In determining the threshold,
Storing connected pixels in the document image;
For each of the stored connected pixels, accumulating the perimeter for each class to obtain a frequency,
On the orthogonal coordinates using the perimeter and the frequency as coordinate axes, a local maximum point, which is a point of a connected pixel that is the farthest from a straight line connecting a point in the class with the smallest perimeter and a point in the class with the largest perimeter, Seeking steps;
Setting the perimeter of the maximum point as the threshold value. A method of encoding a ruled line image extracted by the ruled line extracting method according to claim 1 or 2,
Recognizing a node position where the ruled line intersects from the ruled line image;
At the node position, a step of determining whether the node can be the upper right, upper left, lower right, or lower left corner of the table structure,
Referencing a table in which node numbers are associated one-to-one with respect to combinations of the upper right, upper left, lower right, and lower left corners, and assigning node numbers to the nodes. The encoding method of the ruled line. To extract ruled lines from a document image that contains mixed characters and ruled lines,
Means for inputting an image of a binarized document,
Means for recognizing connected pixels,
Means for determining the perimeter of a connected pixel,
A ruled line extraction program for causing a connected pixel having a perimeter longer than a threshold value to function as a ruled line.

Images