JPH03230289A - Method for determining ruled line and character recognizing device - Google Patents

Abstract

PURPOSE:To accurately decide ruled lines from a table type document whose format is not fixed by applying normal histogram to a line decided as a temporary ruled line candidate to decide the ruled line and converting a line decided as a real ruled line from the temporary ruled line candidate into a ruled line candidate. CONSTITUTION:A normal histogram is formed from a table type document image. Then, an OR image based upon the OR data of black picture elements of two picture elements having the same positional coordinates in the ruled line direction coordinates of two continuous lines is formed and the number of black picture elements on each line is counted up by means of the OR image to form an OR histogram. The normal histogram is furthermore applied to a line decided as a temporary ruled line candidate to decide the probability of the ruled line and the line decided as a real ruled line is set up as the final ruled line candidate. Even in a table type document whose format is not fixed, the ruled lines can be accurately recognized and accurate character recognition can be attained.

Description

[Detailed Description of the Invention] (Summary) Regarding a character recognition device using image input, the present invention aims to accurately determine ruled lines from a tabular document with an inconsistent format, and a normal histogram of the number of black pixels of a reading line.
For the black pixels of two consecutive lines, (7) Create an OR histogram in which the black pixels are counted for each line using the OR image, and calculate the peak of the histogram data when the histogram data in each line of the OR histogram changes from increasing to decreasing. By comparing the height or width of the peak of any line of interest with the histogram data of adjacent peaks or the lines before and after, the line of interest is identified as a true ruled line and as a likely ruled line. For lines that are determined to be temporary ruled line candidates, the ruled line is further determined using a normal histogram, and the lines that are determined to be true ruled lines by the normal histogram are selected from the temporary ruled line candidates. It has a configuration that converts it to .

[Industrial application field]

The present invention relates to a method for recognizing ruled lines in a tabular document in a character recognition device, and a character recognition device that includes a ruled line deletion means and performs character recognition after deleting the ruled lines.

(Prior Art) A character recognition method in a conventional character recognition device is shown in FIG.

In a conventional character recognition device, first, a document reading means 91 such as an image scanner performs a reading process using characters, ruled lines, etc. of a document 90 as images.

Next, the character pattern separating means 92 separates and extracts the pattern of each character from the read image,
Character recognition means 93 performs character recognition on the pattern extracted one character at a time.

The separation process in the character pattern separation means 92 is as follows:
First, from the read image, recognize a block that is a collection of lines (94), then recognize a line that is a collection of characters from the collection block of lines, and separate it into -line units (95)
. Furthermore, after the lines are separated, the characters within the line are separated into one character space, and character recognition is performed for each separated character.

[Problems to be Solved by the Invention] When a character recognition device reads a tabular document with an inconsistent format as an image, the ruled lines are read as images along with the characters.

Conventional character reading devices perform line separation and character separation processing based on the read image that includes the ruled lines, so the ruled lines may be confused with part of the character pattern.
It became difficult to perform correct character separation, reducing character recognition accuracy.

SUMMARY OF THE INVENTION An object of the present invention is to accurately recognize ruled lines in a tabular document whose format is not constant, and to perform highly accurate character recognition as image data that does not include ruled lines after deletion of ruled lines.

[Means for Solving the Problems] The present invention first creates a normal histogram from an image of a tabular document by counting the number of black pixels for each line in the ruled line direction during reading and scanning of document data.

Next, OR for the black pixel of two pixels with the same position coordinates in the coordinates in the ruled line direction of two consecutive lines
Create an OR image using data.

Then, the number of black pixels on each line is counted using the OR image to create an OR histogram.

Then, the peak of the histogram data when the histogram data in each line of the OR histogram changes from increasing to decreasing is determined, and regarding the peak, ruled line candidates determined to be true ruled lines and temporary ruled line candidates determined to be similar to ruled lines are determined. identify

Therefore, lines that are determined to be ruled line candidates are finally treated as ruled line candidates, and for lines that are determined to be temporary ruled line candidates, the certainty of the ruled line is further determined using a normal histogram, and the lines that are determined to be true ruled lines are is now used as the final ruled line candidate.

After that, the ruled line candidates were deleted and character recognition was performed using image data of only the character parts.

FIG. 1 shows the basic configuration of the ruled line determination method of the present invention.

In the figure, 1 is a tabular document, 2 is the original image of the document read by the reading means, and 3 is the original image 2. Count the number of black pixels for each line in the ruled line direction during scanning of the document data. 4 is a normal histogram (NR histogram), which is a continuous 2
OR image created by taking OR data for black pixels with the same position coordinates in the ruled line direction of two lines, 5 is O
This is an OR histogram created by counting the number of black pixels in each line in the R image.

The flow in the figure shows the principle of the ruled line determination method of the present invention.

[Operation] The principle of the ruled line determination method of the present invention will be explained according to the flow numbers in the figures.

■ Read the characters and ruled lines of the tabular document 1 as an image, and create an original image 2 consisting of black pixels and white pixels.

(2) Using the read image data, an NR histogram 3 is created by counting the number of black pixels for each line in the ruled line direction during reading and scanning of the document data.

(2) Create an OR image using OR data for black pixels from pixels in two consecutive lines from the original image 2, count the number of black pixels in each line, and create an OR histogram 5 for the black pixels.

■、■ For OR histograms, detect the peak where the histogram data changes from increasing to decreasing, and determine the width of the histogram by comparing the height of any peak and the adjacent peak or comparing the histogram values before and after the peak. do. Thereby, a ruled line determination process is performed to determine whether a line with a peak obtained by the OR histogram is a ruled line candidate determined to be a true ruled line, a temporary ruled line candidate determined to be a ruled line, or not a ruled line.

■、■ For a line determined to be a temporary ruled line candidate by the OR histogram, the width of the peak is checked using the histogram values of the lines before and after the peak using NR histogram 3 for the corresponding line, and whether or not the line is a true ruled line is determined. judge.

(2), (2) As a result, peaks determined to be true ruled line FA candidates are converted from temporary ruled line candidates into ruled line candidates.

(2) [Phase] As a result of peak determination in the normal histogram, peaks that are determined not to be ruled lines are treated as temporary ruled line candidates.

(2), (2) As a result of the peak determination based on the OR histogram, a peak that is determined to be a true ruled line is finally determined as a ruled line candidate at that point.

(2), (2) In the peak determination result based on the OR histogram, a peak that is determined to be not a ruled line is finally determined to be not a ruled line at that point.

According to the recognition method of the present invention, even if there is a missing or misaligned line in the original image, it is corrected by the OR image, and the line is determined based on the corrected image. For temporary ruled line candidates that are difficult to determine, ruled line determination is further performed using the normal histogram of the original image, so that accurate ruled line determination can be performed.

〔Example〕

FIG. 2 shows ruled lines determined by the ruled line determining method of the present invention;
An example configuration of a character recognition device that performs character recognition after deleting ruled lines is shown.

In the figure, 21 is a tabular document, 22 is a tabular document 21
document reading means such as an image scanner that reads characters and ruled lines as images, 23 a ruled line determining means for determining ruled line portions from the read image, and 24 a ruled line deletion processing means for deleting ruled lines from the image based on the determination result of the ruled line determining means. , 25 is a character recognition processing means for recognizing characters from an image from which ruled lines have been removed, 26 is a microprocessor that executes programs such as a ruled line determination program and a character recognition program, and 27 is an image data storage for storing data such as the read image. A shared memory 28 is a work memory 30 that performs operations such as ruling line determination processing and character recognition processing.
31 is the data of the judgment result of the ruled line judgment process 30;
32 is a process of separating patterns of each character from the image from which ruled lines have been removed, and 33 is a process of character recognition for each separated pattern of one character.

Next, a ruled line recognition method in a character recognition device having an apparatus configuration according to an embodiment of the present invention will be explained with reference to FIGS. 3 to 17.

An example of a method for recognizing horizontal ruled lines will be described below, but recognition of vertical ruled lines can also be processed using a similar method.

The ruled line recognition method of the present invention will be explained below in the order of the figure numbers from FIG. 3 to FIG. 17.

(1) As shown in FIG. 3(a), first, prior to the ruled line recognition process, a compressed image is created for the input original image.

In Figure (a), a compressed image 41 is an example of an image obtained by data-compressing the original image 40.

Compression is performed by deleting a portion of white pixels, etc., so that the compression ratio in the direction perpendicular to the extracted ruled line (vertical direction in the figure) is the distance (g) between the ruled line and the character on the ruled line on the compressed image.
Be sure to do this to the extent that there is a gap of one pixel or more.

By compressing the original image, the character image portion becomes a lumped image, making it easier to distinguish it from the ruled line portion.

(2) Next, as shown in FIG. 3(b), the compressed image 4
2 into 166 pixels, and create a histogram 43 for the number of black pixels in a direction perpendicular to the direction of the ruled line to be extracted (vertical direction in the figure).

The reason for dividing into 16 pixel values is that when the ruled line recognition condition is that the slope of the ruled line is within -3° to +3°,
This is to ensure that the ruled lines can be read accurately.

Alternatively, as shown in FIG. 4, a histogram 46 is created by reading the inclination 44 set in advance on the document, tracing the image of the compressed image 45 along the inclination of the ruled lines, and counting the black pixels. do.

In the present invention, for a compressed image (original image) of an original image, a normal (or
mal) histogram (hereinafter referred to as NR histogram), an OR image created by OR data of black pixels of a given pixel line and the next line, and an OR histogram that counts the number of black pixels in each line. Create two types of histograms.

FIG. 5(a) shows an example of an NR histogram.

In the figure, 47 is an original image of a compressed original image.

48 is an NR histogram based on the number of black pixels for the original image 47.

511(b) shows an example of an OR histogram.

In the figure, 49 is the original image, 50 is an OR image created using OR data for pixels of an arbitrary line and pixels of the next line, and 51 is an OR image created by counting the black pixels of each line of the OR image. This is an OR histogram.

The reason why an OR histogram is created in the present invention will be explained with reference to FIG.

In the figure, line l of the original image 52.53 is a ruled line, but if the ruled line is missing as shown in the figure, the NR histogram 54 of that line will be as shown in the NR histogram ■, ■. , the histogram value becomes small, and it becomes difficult to distinguish the line as a ruled line.

In the OR histogram 54' created by the OR image of line 1 and black pixels of line 2, the histograms are ■'', ■.

This makes it clear that it is a ruled line.

In addition, in cases where the ruled lines are misaligned, such as line 4 of the original image 52 and line 5 of the original image, the NR histogram 54 will be isolated by - lines like ■ and ■, but OR In the histogram, ■゛, ■°
Since the ruled lines are characterized as shown in FIG.

(3) In the present invention, ruled lines are first determined using the OR histogram created as described above.

A method for determining ruled lines from an OR histogram will be explained with reference to FIGS. 7(a) and 7(b).

First, as shown in Figure (a), a position with a peak where the histogram value changes from increasing to decreasing is extracted.

In the case of the illustrated OR histogram, four peaks from peak 0 to peak 3 are extracted.

Next, the peaks are arranged in descending order (downward in the figure), two peaks are set as a set, and it is determined whether the line with the peak is a ruled line or not.

First, if the values of the two peaks are both the maximum value (
For peak 0 and peak 1), it is determined that there is a possibility that a line with an upper peak becomes a line with a ruled line, and peak O is extracted as a ruled line determination candidate.

If this condition is not met, the smaller of the two peaks in the set is selected and set as PK (in the case of peak 1 and peak 2, peak 2 is set as PK).

Then, find the line (vlly) that takes the minimum value of the histogram values between the two peaks, and set that value to ■0 (
In the case of peak 1 and peak 2, vlly2 is set to ■0). In addition, in the case of peak 3 at the edge, the value of peak 3 is P
K, and vlly4 is ■0.

(2) Regarding 0, compare it with a preset valley darkness value (approximately 4) and set the smaller one as VL.

If PK and ■0 satisfy the following relational expression A, a set of 2
The higher peak among the two peaks is determined to have a possibility of becoming a ruled line peak, and is extracted as a ruled line determination candidate.

Conditional expression A: (PK-VO)>VL In the case of peak 1 and peak 2, the selected peak 2 satisfies this condition, so peak 1 is selected as a ruled line determination candidate.

(4) Next, using the above method, the histograms of the ruled line determination candidates are arranged in ascending order (
(in the upper direction in the figure) in descending order, examine the histogram value and width, and extract true ruled line candidates and temporary ruled line candidates.

The determination conditions will be explained below.

Define the following values as processing parameters.

Peak threshold l Peak dark value 2 Ratio to peak dark value 1 Line width threshold 2 = pti = Pt2 = pp = Ltl = Lt2 (Appropriate value = 5) (Appropriate value = 2) (Regular value = 3) (Normal value = 2 mm) (Normal value = 3 mm) Next, regarding the above-mentioned darkness value, the following conditions are determined using the OR histogram value of the ruled line determination candidate as the peak value.

1. Let thl be the difference between the peak value and Ptl.

Peak value - Ptl = th1 2. Let th2 be the quotient obtained by dividing the peak value by Pt2.

Peak value/Pt2=th2 3. Set the smaller of thl and th2 as th3.

4. Let th4 be the quotient of the peak value divided by pp.

Peak value/P p = t h 4 5, the smaller one of th3 and th4 is set as TH2, and the larger one is set as THI.

6. Next, scan from the peak position in descending and ascending order, find the number of lines whose histogram values are larger than THI, and calculate this number by w.
Let it be l.

7. Similarly, scan and check the number of lines whose histogram values are in a range larger than TH2, and set this number of lines as w2.

8. If the above Wl and W2 satisfy any one of the following conditional expressions, the peak is determined to be a true ruled line candidate.

Conditional expression B Ltl≧w1 Conditional expression CLt2≧w2 9. If the peak does not satisfy either conditional expression B or conditional expression 〇, check its histogram value and calculate the maximum value (=16
), the peak is determined to be a temporary ruled line candidate that is determined to be a ruled line.

In other cases, the peak is neither a true ruled line candidate nor a temporary ruled line candidate.

Hereinafter, true ruled line candidates will be simply referred to as ruled line candidates.

Figure 7(b) shows the histogram and THI, T in the determination method based on the peak value and width of the OR histogram.
H2, wl, w2 are shown. The figure shows the case where w1=2 and w2-4.

This completes the process of extracting ruled line candidates and temporary ruled line candidates for peaks based on the OR His 1-gram.

As a result of the above OR histogram processing, peaks determined to be ruled line candidates are finally determined to be ruled line candidates.

(5) On the other hand, regarding temporary ruled line candidates, next, ruled line determination is performed using the NR histogram.

As a result of the judgment using the OR histogram, the line determined as a temporary ruled line candidate is determined by the NR histogram as described above (
The relationship between the peak value and the dark value is investigated in the same manner as in 4), and if the line is determined to be a ruled line candidate in the NR histogram, the temporary ruled line candidate is converted to a ruled line candidate.

In other cases, it is left as is as a temporary ruled line candidate.

The above relationship is shown in FIG.

In the figure, (a) is the determination result obtained using the OR histogram, and (b) is the determination result obtained using the NR histogram.

Based on the determination of the OR histogram, the peaks of lines 1 and 4, which are temporary ruled line candidates, are determined using the NR histogram.

As a result, the result was that it was a true ruled line, so
This shows that lines 1 and 4 are converted into true ruled line candidates.

(6) Next, the continuity of the ruled lines is determined for the ruled line candidates and temporary ruled line candidates located in different pixel areas obtained through the above processing.

FIG. 9(a) is an explanatory diagram of a process of determining the continuity of a temporary ruled line candidate with a ruled line candidate and converting a temporary ruled line candidate that is continuous with the ruled line candidate into a true ruled line candidate.

In the figure, the dotted line portions are temporary ruled lines, and the blacked-out portions represent ruled line candidates determined to be ruled line candidates.

Continuity of ruled lines is defined as whether a temporary ruled line is on the same line as a ruled line candidate and is in direct contact with either the left or right side of the ruled line candidate, or is in contact with a temporary ruled line candidate that is also in contact with a ruled line candidate on the same line. Assume that the ruled line candidates have continuity.

For example, in Figure (a-1), temporary ruled line candidates S2 and Sl have continuity with respect to ruled line candidate EO.

Further, it is assumed that a temporary ruled line in a pixel area one diagonally above or below the ruled line candidate has continuity with the ruled line candidate9.For example, the temporary ruled line candidate S3 in the figure has continuity with the ruled line candidate EO.

Further, it is assumed that a temporary ruled line candidate diagonally above or below a temporary ruled line candidate that is continuous with the ruled line candidate also has continuity with the true ruled line candidate.

For example, it is assumed that the temporary ruled line candidate So, 34 in the figure has continuity with the true ruled line candidate.

As a result of the determination of continuity with the ruled line candidates, temporary ruled line candidates determined to have continuity are converted into ruled line candidates.

(a-2) in FIG. 9(a) shows the relationship between the converted ruled line candidates and temporary ruled line candidates.

Temporary ruled line candidates 5O3S1, S2 in Figure (a-1)
, S3, and S4 are continuous with the true ruled line EO, so (a
-2) Convert to ruled line candidates as shown.

Temporary ruled line candidates S7, S8, and S9 in FIG. (a-1) are left as temporary ruled line candidates because they have no continuity with the ruled line candidates.

(7) After the temporary ruled line candidate conversion process of converting the temporary ruled line candidate into a ruled line candidate in (6) above, a double line to single line conversion process is performed for the ruled line candidate.

The double line conversion process examines the interval between line candidates that line up in the y direction (in descending order perpendicular to the direction of the line to be extracted),
The distance between the double lines is determined in advance by the darkness value (1 mm).
The process below (the degree is appropriate) is the process of deleting one ruled line.

FIG. 9(b) is an explanatory diagram of the double line conversion process.

The ruled line candidates L1 and L2 in Figure (b-1) have an interval of 1 in the y direction.
(1 mm), which is less than the distance value, delete the negative ruled line and make it a single line as shown in Figure (b-2). Note that, among the double ruled line candidates, the one with the smaller NR histogram value is deleted. In Figure (b-1), other ruled line candidates are spaced apart from each other by a distance greater than the darkness value, so deletion processing is not performed.

(8) For the ruled line candidates extracted in the NR histogram and OR histogram, trace the image of the line of the ruled line candidate in the horizontal direction on the compressed image, and extract the left and right end points of the ruled line candidate. Then, the coordinates of the end points of the ruled lines in the original image are calculated backward from the coordinates of the end points on the compressed image.

When tracing borders on a compressed image, missing borders,
Since the ruled lines may be interrupted due to misalignment, blurring, etc., please check the line you are trying to trace, the one above it, and the one above it.
An OR image is created for the black pixels of the next lower line, and the continuity of the black pixel data is determined on the OR image.

FIG. 1O shows a tracking processing method for recognizing the end points of ruled lines.

In the figure, 55 is the original image, 56 is the original image 55
This is an OR image of an arbitrary line and black pixels of three lines one above and one below.

When tracing line 1, the continuity of black pixels on line 1 on line 1 on OR image 56 of black pixels of line 1 in original image 55 and lines 0 and 2, which are the lines above and below it, is determined. Continuity is determined by detecting a point where continuity breaks. Then, the coordinates of that point are read and tracking for that ruled line is stopped.

A method for recognizing left and right end points of a ruled line will be explained with reference to FIG.

A in the figure indicates histogram information, and E in the figure indicates a ruled line candidate.

B in the figure is a compressed image.

The histogram information A is sequentially scanned from the left in the vertical direction indicated by the arrow.

As a result, when a ruled line candidate is detected in Ll on the image of the figure, the corresponding ruled line image Kl in the image of figure B is traced.

The tracking process for the ruled line image is performed using a white pixel darkness value representing the gap indicated by a constant value set in advance (1 mm if dotted lines are also extracted at the same time, Qm if only solid lines are extracted).
Track to the position where there is a gap (G) larger than l and s (where m is an appropriate value).

For example, in FIG. B, for the ruled line image Kl, tracking processing is performed between the left end LP L and the right end RPI.

In this case, the gap G was detected at the position (1, 3), but
This histogram information is a candidate for a ruled line, so
In this case, further track to the right from the position of LP3,
Track until the next void is detected.

When tracing 3 to the ruled line image from LP3, a gap was detected at (3°3), and 3 was traced to the left edge of LP3 in the ruled line image.
The right end is RP 3. And with that void (3.3
) is not a ruled line candidate, so
Tracing of the ruled line image of this line ends.

As mentioned above, when tracing a ruled line image, not only the positions that are ruled line candidates in the histogram information (Fig. 11A) are traced, but also the positions that are not ruled line candidates are traced. If the image is continuous with the image, it is tracked to the position where the gap exists.

Similarly, since L2 is also a ruled line candidate on histogram information A, when it is detected, tracking processing is performed on the corresponding ruled line image 2 in image B. and,
As the end points, the coordinates of the left end LP2 and right end RP2 are read.

Through the above processing, LPI-RPI, LP2-RP2
The coordinates of the left and right end points of the three ruled lines LP3 to RP3 are detected.

The flow of the ruled line recognition method in the embodiment of the present invention is shown in the 12th section.
This is shown in FIGS.

FIG. 12 shows the flow of processing from reading a document to extracting a histogram peak.

FIG. 13 shows the flow from the process of determining ruled line candidates to the process of determining ruled line candidates using the NR histogram regarding the peaks extracted from the OR histogram.

FIG. 14 shows the flow from the middle of the ruled line candidate determination process using the NR histogram to the continuity determination process.

FIG. 15 shows the flow of double line determination processing.

FIG. 16 shows the flow of processing up to creating a tracking pattern for detecting end points of ruled line candidates.

FIG. 17 shows a flow from the middle of processing for tracing candidate ruled lines using an OR image, following deletion of ruled lines, to character recognition, and the end of processing.

The flow from FIG. 12 to FIG. 17 will be briefly explained in accordance with the order of numbers shown in the figures.

■ Read the original document using an image scanner, etc.

■ Create a compressed image from the scanned document image.

■ Count the black pixel data of each line from the compressed image and calculate N
Create an R histogram.

(2) Create an OR image from the compressed image using OR data for black pixels of an arbitrary line and the next line, count the black pixels, and create an OR histogram.

(2) In the OR histogram, scan the OR histogram value vertically in the ruled line direction and extract the peak where the histogram value changes from increasing to decreasing.

■ Examine the peak valley value, peak height, and peak width from the extracted peaks to determine whether the peak is a peak that exhibits the characteristics of a ruled line.

(2) A line with a peak that is a temporary ruled line candidate based on the OR histogram is determined to be a ruled line candidate based on the NR histogram.

■ Determine the continuity of the extracted ruled line candidates.

■ For continuous ruled lines, extract double ruled lines and delete one side of the double ruled lines.

[Phase] Detects a ruled line that performs ruled line tracing processing to read the end points of the ruled line.

Create data.

■ Calculate the coordinates of the left end of the next ruled line on the same line that is adjacent to it across the gap, and start tracing the ruled line.

■ Create an OR image of the black pixels of the newly traced line with the lines above and below it.

[Phase] Delete the extracted true ruled lines from the read image of the document.

■ Separate character patterns for each character using the image with ruled lines removed and perform character recognition.

〔Effect of the invention〕

According to the present invention, ruled line recognition is performed using the OR image of pixels of an arbitrary line and the next line.
No ruled lines are overlooked due to missing or misaligned lines, and if a ruled line is determined to be a temporary ruled line candidate based on OR data, the ruled line is further determined based on the original image, so the reading accuracy of ruled lines is excellent. ing.

Further, since the ruled lines are deleted after the ruled lines are recognized and character recognition is performed only by character patterns, highly accurate character recognition is possible even in tabular documents.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the basic configuration of the ruled line determination method of the present invention. FIG. 2 is a diagram showing the configuration of an apparatus according to an embodiment of the present invention. FIG. 3(a) is a diagram showing an example of a compressed image. FIG. 3(b) is a diagram showing an example of a histogram. FIG. 4 is an explanatory diagram when the original document is set tilted. FIG. 5(a) is a diagram showing an example of an NR histogram. FIG. 5(b) is a diagram showing an example of an OR histogram. FIG. 6 is an explanatory diagram of an OR histogram. FIG. 7(a) is a diagram showing an example of a method for explaining peaks in an OR histogram. FIG. 8(a) is a diagram showing the determination result based on the OR histogram. FIG. 8(b) is a diagram showing the determination result based on the NR histogram. FIG. 9(a) is a diagram illustrating temporary ruled line candidate conversion processing. FIG. 9(b) is an explanatory diagram of the double line conversion process. FIG. 10 is a diagram showing a ruled line tracing method for recognizing end points. FIG. 11 is a diagram showing a method for recognizing the end points of ruled lines. FIG. 12 is a diagram showing the flow of the recognition method of the present invention. FIG. 13 is a diagram showing the flow of the part following FIG. 12 of the recognition method of the present invention. FIG. 14 is a diagram showing the flow of the part following FIG. 13 of the recognition method of the present invention. FIG. 15 is a diagram showing the flow of the part following FIG. 14 of the recognition method of the present invention. FIG. 16 is a diagram showing the flow of the part following FIG. 15 of the recognition method of the present invention. FIG. 17 is a diagram showing the flow of the part following FIG. 16 of the recognition method of the present invention. FIG. 18 is a diagram showing a conventional character recognition method. In the figure, there is a tabular document, a binary image, :NR histogram, ;OR image, :OR histogram.

Claims (2)

[Claims] (1) In a method for recognizing ruled line data in a character recognition device that inputs document data including ruled lines as image data, a normal histogram is used to count the number of black pixels for each line in the ruled line direction during scanning of document data. , O for the black pixel of the pixel on each line with the same coordinates in the ruled line direction in two consecutive lines
Create an OR histogram by counting the black pixels of an arbitrary line line by line using the OR image created by the R data, and determine the peak of the histogram data when the histogram data in each line of the OR histogram changes from increasing to decreasing. Then, by comparing the peak height or width of any line of interest with the histogram data of adjacent peaks or lines before and after, the line of interest is divided into candidate ruled lines that are true ruled lines and hypothetical lines that are likely to be ruled lines. Line candidates and non-ruled lines are determined, and lines that are determined to be true ruled lines by the OR histogram are finally considered as ruled line candidates. Lines that are determined to be temporary ruled line candidates are further confirmed as ruled lines using a normal histogram. 1. A ruled line determination method characterized by determining the similarity of a line and converting a line determined to be a true ruled line from a temporary ruled line candidate to a ruled line candidate based on a normal histogram. (2) Character recognition characterized by comprising a ruled line deletion means for deleting a ruled line candidate determined to be a true ruled line by the ruled line determination method according to claim (1), and character recognition is performed after deleting ruled line data. Device.