JP2803735B2 - Character recognition device including ruled lines - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] The present invention relates to a character recognition device, and more particularly, to a character recognition device for an image in which ruled lines and characters are mixed, in which a ruled line region and a character region can be simultaneously extracted by the same means. The purpose of the present invention is to achieve an improvement in recognition speed by realizing recognition rules by expressing ruled lines as character codes, converting all recognition results into character codes, and then shaping the ruled lines. A strip forming means for creating a plurality of strip-shaped processing sections in the vertical direction for the target image, and detecting an upper end and a lower end of a ruled line area or a character area for each processing section to detect an area for one line. Area detecting means, horizontal ruled line detecting means for determining whether the area for one row detected for each processing section by the area detecting means is a horizontal ruled line or a character string, and a vertical ruled line for each processing section. Detect presence The vertical ruled line detecting means and the horizontal ruled line detecting means are provided with a character recognizing means for recognizing a character string in an area of one line determined to be a character string.

[Industrial applications]

The present invention relates to a character recognition device, and more particularly, to a character recognition device for an image in which ruled lines and characters are mixed.

[Conventional technology]

As an apparatus for recognizing ruled lines and characters and displaying the recognition result for an image in which ruled lines and characters are mixed, a device that first extracts only ruled lines and then extracts characters while avoiding the ruled line existing area is used. Proposed.

At this time, as a method of extracting ruled lines, a method of extracting elements (primitives) considered as being a part of the ruled lines as much as possible from the entire target image and connecting the primitives to form the entire ruled line And, the image is divided into vertical and horizontal strips, and the vertical and horizontal ruled lines are extracted by taking histograms for the divided vertical and horizontal strips, and the candidate lines are extracted. A method has been proposed in which the entire ruled line is formed by determining the connection relation of the lines.

In addition, as a method of recognizing a character area, a method of extracting and recognizing a character in each rectangular area surrounded by a ruled line based on the extracted ruled line area and embedding the recognized character in a target rectangle. And so on.

[Problems to be solved by the invention]

However, according to such a conventional example, in order to extract a ruled line, a ruled line extracting unit different from the character area extracting unit is required, and information on the extracted ruled line is stored as coordinate values on the original image. Therefore, there is a problem that the recognition device is complicated.

In addition, since the character recognition is performed after the ruled line is extracted by the ruled line extracting means, there is a problem that the overall recognition speed is reduced.

The present invention realizes an improvement in recognition speed by enabling a ruled line area and a character area to be simultaneously extracted by the same means, expresses the ruled line as a character code, converts all the recognition results into a character code, and then executes the ruled line. It is an object of the present invention to realize a simplification of a recognition device by shaping an image.

[Means for solving the problem]

 FIG. 1 is a block diagram of the present invention.

First, there is provided a strip forming means 102 for creating a plurality of strip-shaped processing sections in the vertical direction with respect to the recognition target image 101 input and accumulated from, for example, an image scanner or the like.

Next, for each of the processing sections, there is provided an area detecting means 103 for detecting an upper end and a lower end of a ruled line area or a character area to detect an area for one line.

Subsequently, there is provided a horizontal ruled line detecting means 104 for determining whether the area for one row detected for each processing section by the area detecting means is a horizontal ruled line or a character string.

Further, for each processing section, there is provided a vertical ruled line detecting means 105 for detecting the presence of a vertical ruled line. For example, in each processing section, a region where black pixels are continuously present in the vertical direction is detected, a vertical histogram is calculated for each region, and each region is adaptively enlarged in the vertical direction. By calculating a new vertical histogram and determining the amount of change between these vertical histograms, it is identified whether or not each area is a vertical ruled line area. Further, the vertical ruled line detection means 105 is configured to remove the influence of blurred ruled lines by, for example, searching each region in the vertical direction from the end point of each region. In this case, the vertical ruled line detection means 105 adaptively changes the range to be searched further in the vertical direction from the end point of each region depending on, for example, whether each region is a region of a vertical ruled line.
In addition, even when the vertical ruled line detection means 105 determines that each region is not a vertical ruled line region, if the peak of the vertical histogram calculated for each region exists near the position of the vertical ruled line in the previous row, , Each area is identified as a vertical ruled line area.

Then, in the horizontal ruled line detecting means, a character recognizing means for recognizing a character string in an area of one line determined as a character string
Has 106.

In addition to the above configuration, in the present invention, the character recognition unit 106
Is converted into character codes, the vertical ruled lines and horizontal ruled lines detected by the vertical ruled line detecting means 105 and the horizontal ruled line detecting means 104 are converted into ruled line codes, and the character codes and ruled line codes are mixed. And a coding means 107 for outputting the result as a recognition result 108.

[Action]

According to the present invention, a ruled line region and a character region can be simultaneously extracted from an image (document) in which ruled lines and characters are mixed by horizontal ruled line detection means. As a result, the recognition speed can be improved.

In addition, by expressing the recognized characters and ruled lines as character codes and converting all the recognition results into character codes, the recognition device can be simplified.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 2 is an overall configuration diagram of an embodiment of the present invention.
FIG. 4 is an operation flowchart showing the overall operation of the present invention, and FIG.
The figure is an operation flowchart showing details of the line-by-line recognition processing in FIG.

The present embodiment includes a human interface unit 200,
It comprises a recognition processing unit 300 capable of transferring data with the human interface unit 200, and an image input unit 400 capable of transferring digital image data to the recognition processing unit.

First, the human interface unit 200 and the image input unit 400 will be described in detail. The human interface unit 200 includes a command input unit 201 required for the operator to operate the present embodiment, a display unit 202 for presenting characters or image information to the operator, and image information presented to the display unit 202 by the operator. And an image information input unit 203 for designating an image range. The human interface unit can be constituted by, for example, a personal computer which is the most common information acquisition means.
Reference numeral 1 denotes a keyboard, a display unit 202 is a CRT, and an image information input unit 203 is configured by a pointing device such as a mouse.

First, the operator inputs a command to start the operation of the present embodiment through the command input unit 201 (FIG. 3S
1).

Next, a command to start capturing an image is transferred to the image input unit 400, and the image input unit 400 starts reading a document in which ruled lines and characters set in advance by the operator are mixed (FIG. 3, S2). ). Image input unit 400
Is composed of, for example, a scanner or the like. Image input section
The digital image data read by 400 is recognized by the recognition processing unit 3.
Transferred to 00.

In the recognition processing section 300, recognition processing is performed as follows.

First, the digital image data transferred from the image input unit 400 is stored in the image storage unit 301. The coordinate system of the image stored in the storage unit has the origin at the upper left of the image, the right lateral direction is the positive direction of the x-axis, and the downward direction is the positive direction of the y-axis. Next, using the image data stored in the image storage unit 301,
The image reduction unit 302 creates layout image data in which the image data is reduced to 1/8 and stores the layout image data in the layout storage unit 303 (FIG. 3, S3).

The layout image data stored in the layout storage unit 303 is transferred to the display unit 202 and displayed. The operator specifies the recognition processing range using the image information input unit 203 while viewing the layout image displayed on the display unit 202 (S4 in FIG. 3). For example, the range is specified by a broken line shown in FIG. The specified range is converted into coordinate information and stored in the recognition information storage unit 304 as recognition area coordinates.

Next, the strip forming section 305 forms strips having a constant width (48 dot width in the present embodiment) in the vertical direction in order from the upper left coordinates of the recognition area coordinates stored in the recognition information storage section 304. (See FIG. 6), strip numbers i (i
= 0, 1, 2, 3, 4,...) (FIG. 3, S5).

Thereafter, the line-by-line recognition processing of FIG. 3 S6 is executed. This process is shown in the operation flowchart of FIG.

First, for each strip, the black pixel area detection unit 306 searches for black pixels (characters or ruled lines) using a black pixel sensor (not shown) (FIG. 4, S7). Black pixel sensor (yp
nt [i], i is a strip number), as shown in FIG. 6, has a width of two strips (a strip number i and a strip number i + 1),
The search for the black pixel is performed from the upper end of the strip, and it is detected whether or not the pixel on the black pixel sensor includes the black pixel. If the detection result changes from white to black, the y coordinate (ypnt [i]) when the detection result turns black is stored in the recognition information storage unit 304 as a black pixel start point (ypos1 [i]).
After determining ypos1 [i], the black pixel sensor performs further detection until the detection result changes from black to white. When the detection result changes from black to white, the y coordinate including the last black pixel is stored in the recognition information storage unit 304 as the temporary black pixel end point (ypos2 '[i]).

This temporary black pixel end point cannot correspond to blurring of characters and ruled lines, and is detected even if blurring, that is, the division of a black pixel has a width of one pixel. Therefore, even when there is a division of black pixels due to blurring, the lower search width determination unit 30 is set so that the black pixel area detection unit 306 can continue to detect black pixels.
7 determines the allowable width of black pixel division due to blurring (FIG. 4, S10). The determination unit 307 determines the black pixel start point (ypos
1 [i]) and the above-mentioned provisional black pixel end point (ypns2 ′ [i])
, A vertical histogram is obtained (hist1 [x]), and the maximum value of the vertical histogram (hist1 [x]) and the x coordinate (peak [i]) indicating the maximum value are obtained (see FIG. 7). Next, the values of the vertical histogram around the peak [i] are inspected, a portion where the value of the histogram becomes 0 is searched on both sides of the peak [i], and the starting coordinates (x_star) of the peak of the histogram are obtained.
t [i]) and the ending coordinate (x_end [i]) are determined (7th
See figure). The following processing is performed to estimate whether the area of the histogram determined by the start coordinates and the end coordinates is a vertical ruled line. In the vertical histogram, the range for taking the histogram is between ypos2 '[i] and ypos1 [i], but the range d = ypos2' [i] -ypos1 [i] +1 is set to 2
The magnification is doubled, and ypos1 [i] −0.5d to ypos2 ′ [i] +0.
In the range defined by 5d (see FIG. 8), a vertical histogram (hist2 [i]) between the x_start [i] and x_end [i] is obtained in the range, and the value is set to the aforementioned hist1 [i].
, And exceeds a predetermined threshold, it is estimated that the strip currently being searched for includes a vertical ruled line. The lower search width is set to 4 pixels for a strip estimated to include a vertical ruled line, and the lower search width is set to 1 pixel for other strips.

Next, using the downward search width, the black pixel end point determination unit 30
8 detects the end point of the black pixel (S13 in FIG. 4). Using the black pixel sensor (ypnt [i]), the black pixel end point determination unit 308 performs a downward search from the provisional black pixel end point (ypos2 '[i]) by the downward search width. As a result of the search, when a black pixel is detected, it is determined that the black pixel is separated due to blurring, and the detection of the temporary black pixel end point of the black pixel area detection unit 306 and the processing of the downward search width determination unit 307 are repeated. (FIG. 4, S14). As a result of the search, when the black pixel is no longer detected, it is determined that the end point of the black pixel has been detected. Finally, the detected y-coordinate of the black pixel is set as the black pixel end point ypos2 [i]. Is attached to the recognition information storage unit 304.

On the other hand, although the black pixel start point (ypos1 [i]) is determined, if black pixels are continuously present in the vertical direction and white pixels are not detected, detection is performed to a predetermined range. At this point, the y coordinate is set to the black pixel end point (ypos2
[I]), it is determined that this strip is a candidate including a vertical ruled line, the detection status flag is set to 6, and the information is stored in the recognition information storage unit 304 (FIG. 4, S8 → S9 → S11).

If no black pixel is detected in a predetermined range by the search for the black pixel sensor, the detection is stopped, and the y-coordinates of the point at which the detection is started and the point at which the detection is completed are respectively defined as the black pixel start point. (Ypos1 [i]), black pixel end point (ypo
s2 [i]), the detection state flag = 1 is attached to the strip, and the information is stored in the recognition information storage unit 304 (FIG. 4S
8 → S12).

The above-described processing of the black pixel region detection unit 306, the lower search width determination unit 307, and the black pixel end point determination unit 308 is repeated for each strip (S14 → S7 in FIG. 4), and for all the strips. Then, ypos1 [i] and ypos2 [i] are obtained. With this process, a black pixel area for one row is obtained, and then vertical ruled line candidates are extracted for each strip.

The vertical ruled line candidate detection unit 309 calculates the black pixel start point (ypos1
[I]) and the end point of black pixel (ypos2 [i]), take a vertical histogram (hist3 [x]), where the histogram value exceeds 80% of the value of ypos2 [i] -ypos1 [i] To explore. When a histogram value exceeding 80% is included, it is determined that a vertical ruled line exists, and as shown in FIG.
The point at which the height reaches 50% is defined as the starting point of the vertical ruled line (vkei [mv] .s
t) and an end point (vkei [mv] .en) (where mv is the number of vertical ruled line candidates), and are stored in the recognition information storage unit 304 (FIG. 4, S1).
Five).

Further, the vertical ruled line coordinate determination unit 310 includes a vertical ruled line candidate extraction unit 3.
The vertical ruled line coordinates are determined using the start and end points of the vertical ruled line candidates extracted in 09 (S16 in FIG. 4). In the strip including the vertical ruled line candidate, if the above-mentioned detection state flag is 6, the
As shown in FIG. 10, ypos1 [i] and ypos2 [i] are vertically determined by the black pixel start point and the black pixel end point (ypos2 [i] −ypo).
s1 [i]) is enlarged by 1/2 and a vertical histogram is obtained (vertical histogram 4).
v] .st, vkei [mv] .en compared to the vertical histogram 3
If it has increased by more than the specified ratio, it is regarded as a vertical ruled line. If it is not regarded as a vertical ruled line, it is deleted from the vertical ruled line candidate vkei. When the detection state flag is 3, if there is a vertical ruled line candidate at substantially the same position as the vertical ruled line position of the previous row, it is left as a vertical ruled line, and otherwise removed.

Next, the horizontal ruled line detection unit 311 determines a row including the horizontal ruled line (FIG. 4, S17). First, as shown in FIG. 11, in each of the strips, the detection state flag is ypos1 [i], ypos2 only at three places.
A vertical histogram during [i] is obtained (vertical histogram 5). The width of the histogram is determined by the vertical histogram 5, and if it is larger than a predetermined threshold value, it is determined that a ruled line or a character is included. Further, when it is determined that the line includes a ruled line or a character, the value of ypos2 [i] −ypos1 [i] is obtained. The information is stored in the recognition information storage unit 304. Also, ypos2 [i] -ypos1 [i]
Is smaller than the threshold value, it is determined that the strip is a strip including a horizontal ruled line, and the information is stored in the recognition information storage unit 304.

When the horizontal ruled line detection unit 311 determines that a horizontal ruled line is included, the one-line shaping processing unit 312 performs one-line-by-line shaping as preprocessing for determining horizontal ruled line coordinates (FIG. 5). S18). The horizontal ruled line detection unit 311 performs a one-line shaping process only on strips determined to have horizontal ruled lines.
When strips including a character string are mixed, one-line shaping processing is performed on all strips whose detection state flag is 3. The results are shown in FIG.

Then, it is determined whether or not there is a horizontal ruled line (S1 in FIG. 4).
9) If there is, the horizontal ruled line coordinate determining unit 313 determines the coordinates of the horizontal ruled line using the processing result of the one-line shaping processing unit 312 (S20 in FIG. 4). Take a vertical histogram between ypos1 [i] and ypos2 [i] after shaping, find the coordinates of a section where the value of the histogram is smaller than a predetermined threshold, and set the length to be greater than the predetermined threshold. If it is long, start position (hkeir [mh] .st) and end position (hkei [mh] .en) as horizontal ruled lines
Is stored in the recognition information storage unit 304.

On the other hand, when the horizontal ruled line detecting unit 311 determines that there is no horizontal ruled line and the character is included (S19 in FIG. 4), the recognizing unit 314 determines the character area calculated by the horizontal ruled line detecting unit 311. Perform character recognition. First, the one-line shaping processing unit 312 performs shaping in units of one line, and removes a vertical ruled line portion using the vertical histogram (vertical histogram 6) for one line after the shaping, and then uses the vertical histogram 6 Then, a character cutout process is performed (S21 in FIG. 4), the start coordinates and the end coordinates of the character area are obtained, compared with the vertical ruled line coordinates (vkei), and output is started from the one with the smaller coordinate value.

When a ruled line is output by the determination of the horizontal ruled line coordinates (S20 in FIG. 4), the ruled line code conversion unit 315 first converts the ruled line into a JIS ruled line code (S22 in FIG. 4). On the other hand, when characters are output by character segmentation (S21 in FIG. 4), the character recognition unit 316 performs character recognition and converts the characters into character codes (S23 in FIG. 4).

After the ruled lines or characters are converted into codes, the character codes are stored in the recognition information storage unit 304 according to the corresponding ruled lines or character coordinates, as conceptually shown in FIG. 3 (S24 in FIG. 4).

By the above processing, one line of the ruled line or the character was recognized and represented by the code. The process of FIG. 3 S6 shown in the operation flowchart of FIG. 4 is repeated as in FIG. 3 S26, so that recognition is performed for each line. In this case, the start position of the next line is Ypos2 after the shaping process described above
Immediately after [i] (FIG. 4, S25).

Then, after all the recognition is completed for the processing range designated by the image information input unit 203, the ruled line shaping unit 31
7 inserts or removes a blank code and a horizontal ruled code before and after the vertical ruled line in order to remove the discontinuity of the ruled line (S27 in FIG. 3), and the result is conceptually shown in FIG. Is displayed on the display unit 202 (S28 in FIG. 3).

〔The invention's effect〕

According to the present invention, a ruled line region and a character region can be simultaneously extracted by a horizontal ruled line detection unit from an image (document) in which ruled lines and characters are mixed, thereby improving the recognition speed. Becomes

In addition, the recognized character and the ruled line are expressed as character codes, and all the recognition results are converted into character codes, so that the recognition device can be simplified.

Also, by converting the ruled lines into character codes, the output result can be easily handled by general software.

[Brief description of the drawings]

FIG. 1 is a block diagram of the present invention, FIG. 2 is a block diagram of an embodiment of the present invention, FIG. 3 is an operation flowchart showing an overall processing flow of the present embodiment, FIG. FIG. 5 is an explanatory diagram of layout image range designation, FIG. 6 is an explanatory diagram of strips, ypos1, ypos2 ′, FIG. 7 is an explanatory diagram of vertical histogram 1, FIG. 8 is an explanatory diagram of a downward search width determining unit, FIG. 9 is an explanatory diagram of a vertical ruled line candidate detecting unit, FIG. 10 is an explanatory diagram of a vertical ruled line coordinate determining unit, and FIG. 11 is a horizontal ruled line detecting unit , FIG. 12 is a diagram showing an example of the processing result of the one-line shaping processing unit, FIG. 13 is a diagram showing an output example of the recognition processing result, and FIG. It is a figure showing an example. 101: recognition target image, 102: strip forming means, 103: area detecting means, 104: horizontal ruled line detecting means, 105: vertical ruled line detecting means, 106: character recognition means, 107: coding means , 108 ... Recognition result.

Claims (5)

(57) [Claims] 1. A strip forming means for creating a plurality of strip-shaped processing sections in a vertical direction for an image to be recognized, and an upper end of a ruled line area or a character area for each of the processing sections created by the strip forming means. An area detecting means for detecting a lower end to detect an area for one line; a horizontal ruled line detecting means for determining whether the detected area for one line is a horizontal ruled line or a character string; For each of the processing sections created by the strip forming means, a vertical ruled line detecting means for detecting the presence of a vertical ruled line, and in the horizontal ruled line detecting means, the area of the one line area determined to be a character string Character recognition means for recognizing a character string; code for the horizontal ruled line detected by the horizontal ruled line detection means, the vertical ruled line detected by the vertical ruled line detection means, and the character string recognized by the character recognition means Coding means Character recognition apparatus including a ruled line, characterized in that it comprises 2. The vertical ruled line detecting means detects an area where black pixels are continuously present in the vertical direction in each of the processing sections, calculates a vertical histogram for each of the areas,
By calculating a new vertical histogram by expanding in the vertical direction and determining the amount of change between these vertical histograms,
2. The character recognition device including a ruled line according to claim 1, wherein whether or not each of the regions is a vertical ruled line is identified. 3. The vertical ruled line detecting means removes the influence of blurred ruled lines by searching in a vertical direction from an end point of each region in each of the processing sections. A character recognition device including the ruled line according to claim 2. 4. The vertical ruled line detecting means performs a downward search by changing a search range from an end point of each area according to a result of identifying whether each area is an area of a vertical ruled line, 3. The character recognition device including a ruled line according to claim 2, wherein if a black pixel is detected, the search is further continued assuming that the black pixel is continuing, and if no black pixel is detected, the black pixel end point is set. 5. The vertical ruled line detecting means according to claim 1, wherein said vertical ruled line detecting means detects a peak of a vertical histogram calculated for each region at a position close to a position of a vertical ruled line in a preceding row even when the region is identified as not a vertical ruled line region. 3. The character recognition apparatus including a ruled line according to claim 2, wherein, when present, each of the regions is identified as a region of a vertical ruled line.