JP2890307B2 - Table space separation device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial Application Field) The present invention relates to a document image recognition for recognizing a general document in which a mark for human recognition is not performed. The present invention relates to an apparatus for separating a table area from a document in which photographs and the like are mixed.

(Conventional technology) Conventional character / graphic separation processing is described in, for example, Shugai Iwaki's three authors, "Investigation of Introduction of Production System in Character / Graph Separation Processing", IEICE Technical Report PRL83-6.
It only separates characters and graphics, as shown in 3. In other words, for a document image including the table shown in FIG. 2A, both the lines and characters of the table (characters outside the table area [text characters and index characters, etc.]) and characters within the table area are included. Is separated and extracted only into the character image shown in FIG. 2B and the line image shown in FIG. 2C, and the table area including the characters in the table is separated, and FIG. ) Cannot be separated and extracted into a character region image as shown in FIG.

In such a conventional technique, the following problem arises because a table area including characters in a table cannot be extracted.

In document image recognition, efficient character recognition using context cannot be performed because the text region and the characters in the table cannot be distinguished during character recognition.

In a table, the same character, the same character, or the opposite character is often used in the same row (column). Character recognition cannot be performed.

In document editing, it is only possible to edit a table line, but not a meaningful table area.

(Problems to be Solved by the Invention) It is an object of the present invention to enable extraction of a table area including characters in order to efficiently perform document image recognition and document editing.

(Means for Solving the Problems) The present invention provides image input means for digitally inputting an image including a table, an image memory for storing the input image, and a table such as a hook-shaped or T-shaped table existing in a table area. Feature pattern extraction means for extracting a feature pattern at the intersection of the line segments constituting
Finding the outermost feature pattern among the extracted feature patterns, and confirming that the extracted feature pattern exists in a lattice shape, the coordinates of the outermost feature pattern are calculated using a table area. The table area separating apparatus includes table area coordinate detecting means for detecting the coordinates of a rectangle to be displayed, and table area extracting means for extracting a table area from an image memory based on the coordinates of the detected rectangular area.

(Operation) According to the present invention, a general document including a table, that is, a document that has not been marked for processing by humans, is digitally input from an image input unit, and its original image is stored in an image memory. A feature pattern extracting unit extracts a feature pattern of an intersection of line segments constituting a table such as a hook shape or a T-shape from the original image, and detects respective coordinates thereof. In the table area coordinate detecting means, determine the outermost feature pattern among the feature patterns extracted by the feature pattern extracting means, and confirm that the extracted feature pattern exists in a lattice shape, and The coordinates of the feature pattern located are detected as coordinates of a rectangle represented in the table area. For example, if the feature pattern is a hook-shaped pattern, it is checked whether or not the coordinates of the hook-shaped pattern are present in a grid pattern. Select a pair of coordinates and a hook-shaped pattern coordinate having the lower right coordinate, or a pair of a hook-shaped pattern coordinate having the upper right coordinate and a hook-shaped pattern coordinate having the lower left coordinate to select a table area. The coordinates of the rectangle to represent. If a T-shaped pattern is used as the feature pattern, it is checked whether or not the coordinates of the T-shaped pattern exist in a grid pattern. T with maximum and minimum values and y-coordinates
The coordinates of the character pattern are selected to be the coordinates of a rectangle representing the table area. Next, the table area extracting means separates the table area image and the character area image (the image outside the table area) from the original image stored in the image memory using the coordinates representing the table area from the table area coordinate detecting means. .

The present invention extracts a feature pattern of intersections of line segments constituting a table such as a hook shape or a T-shape existing in a table region, detects coordinates of a rectangle representing the table region based on coordinates of the feature pattern, Since the table area is configured to be extracted from the image memory by the coordinates of the rectangle representing the detected table area,
Unlike extracting only line segments of a table as in the related art, it is possible to extract the entire table area including characters in the table. Therefore, when the present invention is used for preprocessing of document image recognition, it is possible to provide a separated table region and character region having conditions necessary for performing document image recognition with high efficiency, which is extremely useful. is there.

(Embodiment) FIG. 1 shows an embodiment of the present invention. This table area separating apparatus includes an image input section 1 such as an image memory, an image memory 2, a hook-shaped pattern matching section 3, a table area coordinate. It consists of an extraction unit 4 and a table area extraction unit 5.

The image input unit 1 inputs a document image including text characters and tables. For example, the original image shown in FIG. 2A is input as binary digital data.

The image memory 2 is a memory for storing the input binary digital data.

The hook-type pattern matching unit 3 extracts the pattern representing the special feature of the upper left corner in the table as shown in FIG. 3A by pattern matching, and the hook-type pattern matching unit 31 shown in FIG. 3B. And a lower right hook type pattern matching unit 32 for extracting a pattern representing the characteristic of the lower right corner in the table as shown in FIG.

The upper left hook-shaped pattern matching unit 31 detects the hook-shaped pattern by the following process, for example, and obtains the coordinates of the corner (intersection).

The image in the image memory is scanned in the x direction from a certain image (xi, xj), and n bits are read and stored in the first register.

Data is generated by shifting the n-bit data of the first register one bit to the left, and stores the data in the second register.

The logical product of the data of the first register and the data of the second register is calculated and stored in the first register.

Steps and operations are repeated n-1 times, and when a bit representing a black pixel remains at the left end of the first register, the coordinates (xi, yj) are stored as hook-shaped pattern coordinate candidates.

Steps (1) to (4) are performed for all the coordinates of the image in the image memory.

 Next, the same procedure is performed in the y direction.

That is, the image in the image memory is scanned in the y direction from a certain coordinate (xi, yj), and n bits are read out and stored in the first register.

Data is generated by shifting the n-bit data of the first register one bit to the left, and stores the data in the second register.

The logical product of the data of the first register and the data of the second register is calculated and stored in the first register.

Steps and operations are repeated n-1 times, and when a bit representing a black pixel remains at the left end of the first register, the coordinates (xi, yj) are stored as hook-shaped pattern coordinate candidates.

Steps (1) to (4) are performed for all the coordinates of the image in the image memory.

A hook-shaped pattern candidate obtained by scanning in the x direction of the step is compared with a hook-shaped pattern candidate obtained by scanning in the y direction of the step, and a match is determined as a hook-shaped pattern.

The result of extracting the original image shown in FIG. 4A is as shown in FIG.

In the lower right hook type pattern matching unit 32,
The hook-shaped pattern shown in FIG. 3 (b) can be extracted by the same procedure as that shown in FIG. 3, and the result of extraction of the original picture shown in FIG. 4 (a) is as shown in FIG. .

The table area coordinate extracting unit 4 includes an upper left pattern matching unit 31
Table area upper left coordinate extracting unit 41 for extracting the coordinates of the uppermost left hook-shaped pattern from the hook-shaped patterns extracted by
And a table area lower right coordinate extracting unit 42 for extracting the coordinates of the lower right hook pattern among the hook patterns extracted by the lower right pattern matching unit 32.

The table region upper left coordinate extracting unit 41 finds the coordinate of the hook-shaped pattern extracted by the upper left pattern matching unit 31 that has the smallest X coordinate and the largest y coordinate, and the hook-shaped pattern forms a table. Is confirmed, and the coordinates are passed to the table area extracting unit 5. If you do not form a table,
Find the coordinates of the next upper left hook-shaped pattern, try to see if they form a table, and repeat until you find one. If not found, the table area extracting unit 5 is notified that the table does not exist.

To check whether the hook-shaped pattern forms a table,
This is performed depending on whether or not the same hook-shaped pattern exists in a lattice shape.

The table area lower right coordinate extracting unit 42 includes a table area upper left coordinate extracting unit 41
, And passes the lower right coordinates of the table to the table area extracting unit 5.

The table area extracting unit 5 calculates an image corresponding to the coordinates from the upper left coordinates of the table area passed from the table area upper left coordinate area unit 41 and the lower right coordinates of the table area passed from the table area lower right coordinate extracting unit 42. A rectangular area of the memory 2 is cut out, and a table area including characters in the table as shown in FIG. 2 (e) is extracted. In addition, an image other than the table area as shown in FIG. 2D can be obtained by filling the rectangular coordinates with white.

In this embodiment, the hook-shaped patterns shown in FIGS. 3A and 3B are used as the characteristic patterns indicating the table area. Instead, the hook-shaped patterns shown in FIGS. 3C and 3D are used. The table area may be configured to be separated by the upper right coordinate and the lower left coordinate.

As the characteristic pattern of the table area, a T-shaped pattern as shown in FIG. 5 may be used to extract the left and right ends and the upper and lower ends of the table to separate the table area. The left and right edges and the upper and lower edges are determined by selecting the coordinates of the T-shaped pattern having the maximum and minimum values of the x coordinate and the maximum and minimum values of the y coordinate. I do.

Further, in the embodiment, the example in which the hook-type pattern matching unit and the coordinate detecting unit have two systems and the processing is performed in parallel has been described. However, it may be configured that the processing is performed sequentially by one system.

(Effect of the Invention) As described above, the present invention provides a hook-shaped or T-shaped
A feature pattern of an intersection of line segments constituting a table such as a character shape is extracted, coordinates of a rectangle representing a table region are detected based on the coordinates of the feature pattern, and an image memory is obtained based on coordinates of the rectangle representing the detected table region. Since the table area is extracted from the table, it is possible to extract the entire table area including the characters in the table, unlike the conventional technique in which only the line segments of the table are extracted. In addition, since it is determined whether or not the extracted feature pattern exists in a grid pattern, the table area is determined, so that only a rectangular area that forms a table can be reliably identified as a table area. Therefore, when the present invention is used for preprocessing of document image recognition, it is possible to provide a separated table region and character region having conditions necessary for performing document image recognition with high efficiency, which is extremely useful. is there.

Further, according to the present invention, the characteristic pattern of the table such as the hook type or the T-shaped type existing in the table area can be constituted by a relatively simple logic circuit and hardware such as a shift register. And speeding up of processing can be realized.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention. FIGS. 2A and 2B are diagrams for explaining the manner of separation between a character and a table. FIG. 2A shows an original input image, FIG. 2B shows a character image extracted by a conventional method, and FIG. (D) is a diagram showing an example of a character region image, and (e) is a diagram showing an example of a table region image. FIGS. 3 (a), (b), (c) and (d) are diagrams showing examples of hook-shaped patterns used in the present invention. 4 (a), (b), and (c) are diagrams for explaining pattern matching, where (a) is an original image, (b),
(C) shows an example of a hook-shaped pattern extracted from the original image. FIG. 5 is a diagram showing an example of a pattern for extracting a table other than the hook-shaped pattern. 1 image input section, 2 image memory, 3 hook pattern matching section, 31 upper left hook pattern matching section, 31 lower right hook pattern matching section, 4
Table region coordinate extraction unit, 41 ... Table region upper left coordinate extraction unit, 42 ...
... Table area lower right coordinate extracting unit, 6... Table area extracting unit.

Claims (1)

(57) [Claims] 1. An image input means for digitally inputting an image including a table, an image memory for storing an input image, and features of intersections of line segments constituting a table, such as hooks and T-shapes, existing in a table area. A feature pattern extracting means for extracting a pattern; obtaining an outermost feature pattern among the extracted feature patterns; and confirming that the extracted feature pattern exists in a grid pattern. Table area coordinate detecting means for detecting the coordinates of the characteristic pattern to be detected as the coordinates of a rectangle representing the table area, and table area extracting means for extracting the table area from the image memory based on the coordinates of the detected rectangle representing the table area. A table region separating apparatus.