JPH04291479A - Ruled line recognition method and table processing method - Google Patents

Abstract

PURPOSE:To recognize discontinuous ruled lines like dotted lines, broken lines, or chain lines and to process even a table including discontinuous ruled lines. CONSTITUTION:Black runs and white runs of each line in the main scanning or subscanning direction are extracted, and a histogram of length of respective runs is obtained. When the value of a peak part of the histogram exceeds a certain value, the continuity range of black and white runs which has length corresponding to this peak part is extracted as a ruled line candidate, and ruled line candidates connected between lines are integrated to recognize a discontinuous ruled line.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to recognition of ruled lines and table processing in binary images of documents and forms.

0002

[Prior Art] Generally, when a character recognition device processes a document or a form, the image is divided into a character area, an image area (photos, diagrams, etc.), and a table area, and different processing is performed on each area. many.

Conventionally, in processing a table area, a rectangle in which black pixels are connected is determined from a binary image of the table area, and a rectangle having a length longer than a certain value is recognized as a ruled line. Then, the area surrounded by the recognized ruled lines is recognized as one frame of the table, an image within the frame is cut out, and characters are cut out and recognized from this image. Examples of such processing methods are Japanese Patent Application No. 1-314519 and Japanese Patent Application No. 1-312868 filed by the applicant of the present application.
No., Japanese Patent Application No. 2-134876 and Japanese Patent Application No. 2-1348.
This is described in the specifications and drawings attached to patent applications such as No. 78.

0004

[Problem to be solved by the invention] According to the conventional method,
If a table contains even one non-solid ruled line, such as a dotted line, the table frame cannot be recognized correctly, and as a result , there was a problem that an error occurred in character recognition within the table. However, in recent years, tables containing non-solid ruled lines have been frequently used, and there has been an increasing demand for their proper processing.

[0005] Accordingly, an object of the present invention is to provide a ruled line recognition method and a table processing method that enable accurate processing even for tables including non-solid ruled lines such as dotted lines.

[0006]

[Means for Solving the Problem] According to the invention of claim 1, in a binary image, black runs and white runs of each line in the main scanning direction or sub-scanning direction are extracted, and the length of each run is determined. Find the histogram of . When the value of the peak part of this histogram exceeds a certain value, a continuous range of black runs and white runs with the length corresponding to the peak part is extracted as a ruled line candidate, and connected to the ruled line candidate between lines. By performing processing to integrate objects, ruled lines in the main scanning direction or the sub-scanning direction are recognized.

According to the second aspect of the invention, ruled line candidates are extracted by the method based on the histogram, and
Black runs with a length longer than a certain length are also extracted as ruled line candidates through threshold processing of the length of black runs, and ruled lines are recognized.

According to the invention of claim 3, in a binary image of a table area of a document or a form, the ruled lines in the main scanning direction and the sub-scanning direction are recognized by the method of claim 2, and the information of the table is determined from the information of the ruled lines. A frame area is recognized, and characters within the frame area are extracted and recognized.

[0009]

[Operation] In the case of a dotted line, a broken line, or a chain line, there are continuous black runs and white runs of approximately constant length, so the histogram of the runs shows a peak at each length. Furthermore, if the ruled line is valid and has a certain length or longer, the peak value of the histogram will exceed the certain value. Therefore, claim 1
According to the invention, non-solid ruled lines such as dotted lines, broken lines, or chain lines can be correctly recognized. Also,
According to the invention of claim 2, which includes ruled line recognition based on black run threshold processing, both non-solid and solid ruled lines can be recognized.

[0010] According to the third aspect of the invention, any table including only solid ruled lines, only non-solid ruled lines, and a mixture of solid and non-solid ruled lines may be surrounded by ruled lines. It is possible to correctly recognize a frame area, cut out characters in a frame area, and recognize characters.

[0011]

[Embodiment] FIG. 1 is a block diagram of a processing device according to an embodiment of the present invention, FIG. 2 is a flowchart of processing, and FIGS. 3 to 6
is a diagram for explaining ruled line recognition.

In the binary image input unit 1, binary data of a document image is inputted using a scanner or the like and stored in the binary image memory 2 (processing step 100). Table area recognition unit 3
, the table area is automatically recognized from the image data in the binary image memory 2, or the table area is recognized by manually specifying the area using a mouse etc., and the image data of the table area is recognized. It is cut out and stored in the table area image memory 4 (processing step 102).

Next, processing for extracting ruled line candidates (processing steps 104 to 111) is performed on the table area image data using a predetermined fixed number of lines in each of the main scanning direction and the sub-scanning direction. Execute in units of (several lines). The details of this process are as follows.

The run extraction unit 5 extracts a black run and a white run from one line of interest in the table area image data,
The extracted run information (for example, the coordinates of the start point and end point, the coordinates and length of the start point, etc.) is stored in the black run memory 6b and the white run memory 6.
w (processing step 104).

In the histogram calculating section 7, the run information of the line of interest is read from the black run memory 6b and the white run memory 6w, and a histogram is calculated for the lengths (run lengths) of the black run and white run, respectively. The result is stored in the histogram memory 8 (processing step 106).

The ruled line candidate extraction unit 9 refers to the histogram data of black runs and white runs in the histogram memory 8 and the information in the run memories 6b and 6w, and selects possible ruled lines from runs on the line of interest. A run is selected, a continuous range of the selected run is extracted as a ruled line candidate, and its information (for example, the coordinates of the starting point and ending point, the starting point and length, etc.) is stored in the ruled line candidate memory 10 (processing step 108). In other words, when a histogram of black runs as shown in Figure 3 and a histogram of white runs as shown in Figure 4 are obtained, the lengths (hatched areas) with histogram values exceeding a certain value at their respective peaks are Select a run and use its continuous range as a ruled line candidate.

The basis of such a ruled line candidate extraction algorithm is that for dotted, broken, or chain line ruled lines,
If there are consecutive black runs and white runs of a specific length, and the run histogram shows a peak at each length, and if the ruled line is longer than a certain valid length, the peak value will exceed a certain value. . Therefore, in the case of the ruled line A shown in FIG. 6, a continuous range B of black runs and white runs on each of several consecutive lines is extracted as a ruled line candidate.

Next, the run threshold processing unit 11 refers to the black run memory 6b, extracts black runs on the line of interest whose length is longer than a certain threshold value as ruled line candidates, and stores this information in the ruled line candidate memory 10. (processing step 110). With this process, it is possible to extract the black runs that make up the solid ruled lines. In the example of FIG. 5, a black run D whose length exceeds the threshold value is extracted as a solid ruled line candidate, but a short black run D is not a candidate.

When the above processing loop for extracting ruled line candidates (processing steps 104 to 110) is completed for the fixed number of lines, this processing loop is exited at processing step 111. Then, the ruled line extraction unit 12 refers to the information in the ruled line candidate memory 10, checks the connectivity between lines for the ruled line candidates within the unit of a fixed number of lines, and performs processing to integrate the connected ruled line candidates. The integrated ruled line candidate after processing or the unintegrated ruled line candidate is recognized as a ruled line, and its information (such as the coordinates of the diagonal vertices of the ruled line rectangle) is stored in the ruled line memory 13 (processing step 11
2). In the example of FIG. 6, the ruled line candidates extracted from three consecutive lines are connected, so a rectangle that integrates them is recognized as a single ruled line B.

The ruled line recognition processing loop (processing steps 104 to 114) for each fixed number of lines explained above is repeated in the main scanning direction and the sub-scanning direction, and when this is completed for the entire table area, this loop is exited and Proceed to processing for character recognition.

First, in the frame recognition unit 14, the ruled line memory 1
3, the frames surrounded by the recognized ruled lines are recognized, and the coordinates of each frame (coordinates of two diagonal vertices, etc.) are stored in the frame coordinate memory 15 (processing step 116). Next, in the frame area extraction unit 16, by referring to the information in the frame coordinate memory 15, image data of each frame area is extracted from the table area image data in the table area image memory 4 and stored in the frame image memory 17. (Processing step 118). The line cutting unit 18 cuts out image data of a line (character string) from the data in the frame area image memory 17 by, for example, a method using projection in the main scanning direction, and stores it in the line image memory 19 (processing step 120). Next, the character cutting unit 20 cuts out character image data from this line image data by, for example, a method using projection in the sub-scanning direction, and passes it to the character recognition unit 21 (processing step 122
). The character recognition unit 21 performs character recognition on this character image data by, for example, feature extraction and matching with a recognition dictionary, and outputs the result (processing steps 124, 126).

[0022]

[Effects of the Invention] As explained above, according to the present invention, black runs and white runs are extracted, a histogram of their lengths is calculated, and a selected ruled line is likely to be constructed based on this histogram. It is possible to recognize non-solid ruled lines such as dotted lines, dashed lines, or chain lines by extracting a continuous range of runs as ruled line candidates and performing processing to integrate the connected ones to recognize ruled lines. In addition, by combining the method of extracting black lines of a certain length or more as ruled line candidates by thresholding the length of black runs, and integrating the connected lines to recognize them as ruled lines, it is possible to distinguish between solid ruled lines and non-solid ruled lines. It is possible to recognize both
Furthermore, with this method, even for tables with non-solid ruled lines, it is possible to successfully extract the frame area and perform character recognition, greatly expanding the range of tables that can be processed. .

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a processing device according to an embodiment of the present invention.

FIG. 2 is a schematic flowchart of processing.

FIG. 3 is a graph showing an example of a histogram of black run lengths.

FIG. 4 is a graph showing an example of a histogram of the length of white runs.

FIG. 5 is an explanatory diagram of ruled line candidate extraction by threshold processing.

FIG. 6 is an explanatory diagram of the extraction of ruled line candidates using a histogram and the recognition of ruled lines by integrating the candidates.

[Explanation of symbols]

1 Binary image input section
2 Binary image memory 3 Table area recognition unit
4 Table area image memory 5 Run extraction section
6b Black run memory 6w White run memory
7 Histogram calculation section 8 Histogram memory
9 Ruled line candidate extraction unit 10 Ruled line candidate memory
11 Run threshold processing section 12 Ruled line extraction section
13 Ruled line memory 14 Frame recognition section
15 Frame coordinate memory 16 Frame area extraction section
17 Frame image memory 18 Line cutting section
19 row image memory 20 character cutting section
21 Character recognition section

Claims (3)

[Claims] Claim 1: In a binary image, extract the black run and white run of each line in the main scanning direction or the sub-scanning direction, obtain a histogram for the length of each run, and calculate the value of the peak part of this histogram. When a certain value is exceeded, a continuous range of black runs and white runs with a length corresponding to the peak part is extracted as a ruled line candidate, and a process is performed to integrate the lines connected to this ruled line candidate. Characteristic ruled line recognition method. 2. The ruled line recognition method according to claim 1, wherein threshold processing is performed on the lengths of the extracted black runs to extract black runs with a length longer than a certain length as ruled line candidates. [Claim 3] In a binary image of a table area of a document or a form, the ruled lines in the main scanning direction and the sub-scanning direction are recognized by the ruled line recognition method according to claim 2, and the frame area of the table is recognized from the information of the ruled lines. A table processing method characterized by cutting out characters within a frame area and recognizing the characters.