JP3476640B2 - Noise processing method and apparatus - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise processing method and apparatus for removing noise from an input paper image used for character recognition and determining image quality.

[0002]

2. Description of the Related Art In a character recognition process, an operator recognizes an input paper image input by a facsimile or the like and confirms and corrects it. For this reason, if noise is included in the input paper image input by a facsimile or the like, erroneous recognition will occur even in correctly written characters, and correction by the operator is required. Further, the time required for recognizing an input paper image containing a lot of noise and a large amount of noise and performing confirmation correction increases, and the processing time of the operator becomes longer than that of observing the input paper without inputting characters. For this reason, noise removal has conventionally been performed by filter processing or mask processing. However, it was not possible to remove horizontal and vertical dotted lines, broken lines, and the like with irregular intervals over the entire screen. In addition, there is a problem that even if the horizontal and vertical linear noise that is considered to be removable is deleted only because it is long, the ruled lines and characters on the input form are erased.

[0003]

SUMMARY OF THE INVENTION It is an object of the present invention, the irregular dotted or dashed noise in the horizontal and vertical input paper image inputted from a facsimile or the like, the image quality or noise due to the removal of the linear noise magnitude It is intended to improve the accuracy of character recognition and shorten the processing time of the operator by determining that the image quality is bad due to the image quality.

[0004]

A noise processing method of the present invention scans an input paper image input by a scanner including a facsimile in horizontal and vertical directions, and a black pixel column in which black pixels exist at a certain interval or less. of detecting the length and width, said
Length is set in advance, exceeds a threshold for the length of the black pixel row, and the width is set in advance, and black pixel row to be removed a black-pixel rows when more than the threshold to the width of the black pixel columns Pixel row judgment stage to judge and black pixels judged to be removed
On both sides in the width direction of the black pixel area, which is an area composed of
A peripheral area that is an area within a predetermined distance from the periphery of the area
Of the black pixels in each width direction in the black pixel area
Located in the width direction of the black pixel in the peripheral area
Image that converts black pixels that are all white pixels to white pixels
It has a string conversion stage . Further, the noise processing device of the present invention
Is an input sheet entered by a scanner including a facsimile
Scans the image horizontally and vertically with constant black pixels
Detecting the length and width of the black pixel row existing below the interval of
Set the threshold for the length of the black pixel row, the length of which is preset.
Over the width of the black pixel column, which width is preset.
When the black pixel row is equal to or less than a threshold value,
Pixel column determination means for determination, and black image determined to be removed
Both sides in the width direction of the black pixel area, which is an area composed of elements
, Which is an area within a predetermined distance from the area surrounding the area
Inspect the pixels in the area and check the black in each width direction within the black pixel area.
Of the pixels, the position in the width direction of the black pixel in the peripheral area
Converts black pixels, which are all white pixels, to white pixels
And a pixel column conversion means for converting the pixel column.

Further, the preset threshold value for the length of the black pixel row is determined according to the inclination angle of the input paper image.
, For example, if there is no skew in the input paper image,
The input paper image has a length corresponding to the width of the input
If the paper image is slanted, the length exceeds the character input field.
That's it.

[0006] Furthermore, the pixel row determination step, exceeds a threshold for the length of the preset black pixel row, and detected
Width is determined at the input paper image was e Yue threshold to the width of the black pixel row image quality poor.

The present invention scans an input paper image input by a scanner such as a facsimile in the horizontal and vertical directions, and regards a black pixel row in which black pixels are present at a certain interval or less as a straight line, thereby forming a dotted line or an indefinite space. It is characterized in that the broken line is detected as the length and width of the straight line, the inclination angle of the input paper image is detected, and noise removal or image quality determination can be performed based on the length and the inclination angle of the straight line . As a result, it is possible to remove horizontal and vertical dotted lines or broken line noise with irregular intervals that could not be removed conventionally, and it is possible to distinguish the ruled lines and noise in the input paper by using the inclination angle of the input paper image. It is possible to remove noise without erasing the ruled line indicating the character entry field. Furthermore, since noise that erases ruled lines and character information can be determined, in the case of such noise, it is significantly different from the conventional technology in that unnecessary recognition is not performed as defective image quality.

[0008] In the present invention, first horizontal and vertical dotted lines in the input paper image, broken lines, and detects a straight line length and width, if the length or width of the character entry column corresponding following line segments that no noise determination However, if there is a line segment having a large length or width and a size large enough to hide ruled lines or character information, it is determined that the image quality is poor and the subsequent processing is not performed. . When there is a line segment whose length is equal to or greater than the specified value and whose width is equal to or less than the specified value, the inclination angle of the input paper image is detected. For example, when the inclination is not the line segment to have a length of the horizontal vertical width corresponding input sheet image to remove border width corresponding following segment, if there is slope greater than or equal to the length exceeding the character entry column Remove line segments that are less than or equal to the ruled line width. As described above, the characteristic peculiar to large noise and the inclination angle of the input paper can be used to enable accurate noise removal of the input paper image and determine the image quality defect of the input paper image.

[0009]

Embodiments of the present invention will now be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing a method for detecting horizontal and vertical line segments, and FIG. 3 is a diagram showing processing for removing horizontal and vertical line segments. is there.
FIG. 4 is a diagram showing a method of removing horizontal and vertical line segments.
FIG. 4A is a diagram showing a black pixel column to be removed, and FIG. 4B is a diagram showing a black pixel column to be removed.

As shown in FIG. 1, 1 is an image memory unit for storing an input paper image, 2 is a horizontal and vertical scanning of the image memory unit 1 to detect the length and width of a black pixel column, and the black to be removed is detected. A horizontal / vertical line segment detection unit 3 for determining a pixel row and an image quality defect, 3 is a paper inclination detection unit for detecting an inclination angle of an input paper image stored in the image memory unit 1, and 4 is a black pixel row to be removed. The horizontal and vertical line segment removing units 5 and 5 inspect the pixels on both sides of the sandwiched width direction and convert the black pixels to be removed into white pixels if all the pixels on both sides are white pixels. The input paper error detected by the inclination detector 3
A sheet inclination correction unit that performs inclination correction of the image memory unit 1 according to the inclination angle of the image, 6 is a character recognition unit that recognizes the characters of the input paper image stored in the image memory unit 1, and 7 is while looking at the input image. It is a confirmation and correction unit that corrects the recognition result.

Next, the operation of the embodiment of the present invention will be described in detail with reference to FIGS. First, the horizontal / vertical line segment detection unit 2 scans the image memory unit in the horizontal direction and the vertical direction as shown in FIG. 2 to detect the length and width of the black pixel row existing at a constant interval (S) or less (see FIG. P1 of 3
And P2). Then, the length of the detected horizontal and vertical black pixel row is determined (P3 in FIG. 3, hereinafter FIG. 3 is omitted), the detected line segment is less than the specified value La, and the width of the black pixel row is the specified value. B
If less than, it is determined that there is no large linear noise (P4),
The horizontal / vertical line segment removing unit 4 is notified that there is no removal line segment, and the sheet inclination detecting unit 3 is instructed to detect the inclination of the input sheet image (P5).

When the length of the detected horizontal and vertical black pixel row is the specified value La or more and the width of the black pixel row is the specified value B or more (P
In 6), the noise is such that the characters and ruled lines are hidden, so it is determined that the image quality is poor, and the confirmation / correction unit 7 is notified of the poor image quality (P7), and the operator determines. Of the other line segments, the length of the detected horizontal and vertical black pixel row is the specified value L.
When the width of the black pixel row is a or more and less than the specified value B (P8)
Notifies the horizontal and vertical line segment removing unit 4 of the length, width, and position information of the detected line segment, and instructs the sheet inclination detecting unit 3 to detect the inclination (P5). Upon receiving the inclination detection instruction, the sheet inclination detection unit 3 inclines at an arbitrary angle, collects the appearance frequencies of pixels having the same vertical distance from the arbitrary coordinates on the reference coordinate axis, and determines the appearance frequency as the maximum. The tilt angle of the coordinate axis is obtained, the tilt amount at the time of input of the input paper is detected by this tilt angle, and the tilt amount is notified to the horizontal and vertical line segment removing unit 4 and the paper tilt detecting unit 5.

[0014] In the input paper image tilt angle received taking <br/> was horizontal vertical line removal section 4, when the inclination angle is 0 degrees (P9) black pixels detected in the horizontal and vertical line detection unit 2 When the line length is equal to or greater than the specified value Lb and the width is less than the specified value B, the removal processing is performed (P11), and if there is an inclination (P11).
In 10), the removal processing is performed on the line segment whose black pixel column length is the specified value La or more and whose width is less than the specified value B (P11).

In the line segment removal processing , peripheral pixels on both sides in the width direction of the designated black pixel column are examined as shown in FIG. 4, and when the pixel pixel is sandwiched between white pixel columns, the designated black pixel is changed to a white pixel. The black pixel row is removed by conversion (P12). Then, when the removal of the instructed line segment is finished, the sheet inclination detection unit 5 is notified of the inclination correction instruction (P13). Upon receipt of the notification of the tilt correction instruction, the paper tilt correction unit 5 performs tilt correction of the input paper image based on the tilt angle notified from the paper tilt detection unit 3 (P7). When the inclination correction is completed, the sheet inclination correction unit 5 notifies the character recognition unit 6 of a character recognition instruction (P14).
Upon receiving the notification of the character recognition instruction, the character recognition unit 6 analyzes the input paper image, cuts out the characters, recognizes the characters, and notifies the confirmation correction unit 7 of the recognition result (P7). The confirmation correction unit 7 that has received the notification of the recognition result recognizes the recognition result and the image memory unit l.
The input paper image of is displayed on the display, and the operator confirms and corrects it. However, the confirmation correction unit 7 which has received the notification of the poor image quality displays the input paper image of the image memory unit 1 on the display, and the operator inputs the character string from the keyboard while looking at the input paper image to create the character data. The specified value in the above description is 0 <La <L.
b and Lb≈W.

FIG. 2 is a diagram showing a method of detecting horizontal and vertical line segments, 10 is an input paper image stored in the image memory unit 1, 11 is a ruled line representing a character entry field, and 12 is a ruled line.
Horizontal dotted line noise interval S2 are wide b2, 13 is a horizontal dashed line noise interval S 3 with a width b2, 14 is a horizontal dashed line noise 13
Horizontal straight line noise of width b2 on the extension line of, and 15 is width b
3 horizontal straight line noise, 16 vertical dotted line noise with width bl and interval S1, 17 vertical vertical line noise with width b1, 18 width b
3, vertical straight line noise 3, 19 is the length of the horizontal broken line noise 12, 20 is the total length of the horizontal broken line noise 13 and horizontal straight line noise 14, 21 is the length of the horizontal straight line noise 15, and 22 is the vertical dotted line noise 16. Length, 23 is vertical linear noise 17
, 24 is the length of the vertical linear noise 18, W is the maximum scanning width and maximum scanning height of the input paper image 10 (in this example, it is a square area), B is the noise line width, and La is noise. The noise threshold of length Lb is a scanning width threshold for determining that the scanning width is reached.

Next, a method of detecting horizontal and vertical line segments will be described with reference to FIG. First, the image memory unit 1 moves to the lower end while scanning horizontally from the upper end. Then, the presence of black pixels on each scan line is investigated, and if black pixels are detected,
Check whether the next black pixel exists within the range of the distance indicated by the specified value . When black pixels are detected, the distance between black pixels is added as the number of black pixels. If it cannot be detected, the next black pixel is searched for without adding it as the number of black pixels. When the horizontal dotted line noise 12 having a width of b2 pixels and an interval of S2 is examined by such a method, since the black pixels are less than the specified value S, it is considered that the black pixels are continuous and the length 19 of the dotted line noise is obtained. Since the interval between the horizontal broken line noises 13 having the width b2 and the interval S3 is less than the specified value S, and the interval S4 with the linear noise 14 is not less than the specified value S, the length from the end of the broken line to the end is horizontal.
A total length 20 of the broken line noise and the straight line noise obtained by adding the length of the straight line noise 14 having a width of 2 pixels is obtained. For the linear noise 15 having a width of b3 pixels, the length 21 of the linear noise is obtained. Also in the vertical direction, the scanning proceeds from the left end to the right end while scanning in the vertical direction, and the same process as the detection of the horizontal line segment is performed.
The vertical dotted line noise 16 of 1 is examined, and since it is less than the specified value S, it is regarded as a continuous black pixel, and the length 22 of the dotted line noise is obtained. The vertical linear noise 17 having the width b1 is the length 2 of the linear noise.
3, the vertical linear noise 18 of width b3 obtains the length 24 of the linear noise.

In this way, the length of noise of each scanning line is measured, and then, in the case of a horizontal line segment, the vertical line is observed by observing the vertical connection of the horizontal line segments in which black pixels are sandwiched by a specified value or less. In the case of minutes, the width of the vertical horizontal line segment is detected by looking at the horizontal connection of the vertical line segments sandwiched by the number of black pixels equal to or less than a predetermined value. In this example, the vertical dotted line noise 16 and the vertical straight line noise 17 have a width b1, the horizontal dotted line noise 12, the horizontal broken line noise 13 and the horizontal straight line noise 14 have a width b2 and the horizontal straight line noise 1 respectively.
5, the vertical linear noise 18 is detected as having a width b3. In addition,
The relationship of specified values is 0 <S1 <S2 <S3 <S <S4.

As a result, horizontal dotted line noise 12, vertical dotted line noise 16, which is equal to or larger than the noise threshold Lb and smaller than the noise line width B,
Alternatively, in the case of the vertical straight line noise 17, the horizontal and vertical line segment removal processing is performed even if the inclination amount is 0, and in the case of only the horizontal broken line noise 13 and the horizontal straight line noise 14 that are equal to or larger than the scanning width La and less than the noise line width B, the inclination amount When is not 0, the horizontal and vertical line segment removal processing is performed, and when the horizontal linear noise 15 or the vertical linear noise 18 having the noise threshold value La or more and the noise line width B or more, the image quality defect can be determined.

By performing the above processing, the specified value is displayed.
If there are black pixels within the range of the distance that can be detected, even if the intervals are irregular, it can be detected as the same length as the line segment connecting the dotted line and the broken line. It can be detected as a line segment of sufficient length. In the above process, whether the tilt angle is 0 or not is used as a criterion for judgment, but even if the tilt angle is small, there is no problem depending on the situation. Further, the noise length thresholds are La and Lb.
, And B has one threshold value for the pixel width of noise, but it is also possible to make more detailed determinations by increasing the number and combining it with the inclination angle . Further, in this example, the input paper image using ruled lines is used, but even if there are only ruled lines and there are no ruled lines, the communication method PRU94-33 (199) is used.
Similar processing can be performed by obtaining the tilt angle using the black pixel distribution of the character line as shown in 4-09).

FIG. 4 is a diagram showing a method for removing horizontal and vertical line segments, FIG. 4 (a) is a diagram showing a black pixel row to be removed, and FIG.
(B) is a diagram in which the removal target black pixel row is removed, 30 is a removal target black pixel, 31 is a black pixel, 32 is a white pixel changed from a black pixel, and 33 is a white pixel.

Next, a method of removing horizontal and vertical line segments will be described with reference to FIG. First, as shown in FIG. 4A, when the position and width of the removal pair line segment are known, the removal target black pixel 30 is identified.
The white pixels on the upper and lower two scanning lines of the black pixel 30 to be removed are adjusted, and the black pixel 30 to be removed is shown in FIG. 4B only when the black pixel 30 to be removed is sandwiched by the white pixels 33. To white pixels 32. By performing this processing from end to end of the scanning line, the noise line segment can be removed. In this example, two pixels above and below the black pixel 30 to be removed are aligned, but one pixel or more pixels may be aligned. Further, not only the upper and lower sides but also the adjacent peripheral pixels can be adjusted.

[0023]

As described above, according to the present invention ,
Irregular dotted lines and broken lines that are horizontal and vertical to the input paper image can be detected with a simple method, and noise lines are removed by using the inclination of the input paper image. It does not erase line segments such as ruled lines even on the input paper image that has been input, and enables accurate noise removal, and image quality for the input paper image that cannot be clearly recognized depending on the length and width of noise. Since it is possible to determine a defect, there is an advantage that unnecessary recognition is not performed.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a diagram showing a method of detecting horizontal and vertical line segments.

FIG. 3 is a diagram showing a removal process of horizontal and vertical line segments.

4 is a diagram showing a method for removing horizontal and vertical line segments, and FIG.
FIG. 4A is a diagram showing a black pixel column to be removed, and FIG. 4B is a diagram showing a black pixel column to be removed.

[Explanation of symbols]

1 Image Memory Section 2 Horizontal / Vertical Line Segment Detection Section 3 Paper Inclination Detection Section 4 Horizontal / Vertical Line Segment Removal Section 5 Paper Inclination Correction Section 6 Character Recognition Section 7 Confirmation / Correction Section 10 Input Paper Image 11 Ruled Lines 12 for Character Entry Fields 12 Horizontal Dotted Lines Noise 13 Horizontal broken line noise 14 Horizontal straight line noise 15 Horizontal straight line noise 16 Vertical dotted line noise 17 Vertical straight line noise 18 Vertical straight line noise 19 Horizontal broken line noise 12 length 20 Horizontal broken line noise 13 and horizontal straight line noise 14 total length 21 Horizontal Length 22 of linear noise 15 Length 23 of vertical dotted noise 16 Length 24 of vertical linear noise 17 Length 30 of vertical linear noise 18 Black pixel 31 to be removed 31 Black pixel 32 White pixel changed from black pixel 33 White pixel

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Naoko Iwasaki 1-3 1-3 Gotenyama, Musashino-shi, Tokyo NTT Advanced Technology Corporation (58) Fields investigated (Int.Cl. ⁷ , DB name) H04N 1/38-1/409 G06T 5/00 300

Claims (8)

(57) [Claims] 1. A scanning an input paper image input by a scanner comprising a facsimile horizontally and vertically, to detect the length and width of the black pixel array black pixels are present in the following regular intervals, the said length preset, beyond the threshold for the length of the black pixel rows, and the width is set in advance, the pixel determined as a black pixel row to be removed a black-pixel rows when the threshold value or less to the width of the black pixel columns Black, which is the area that consists of the black pixels that are determined to be removed at the column determination stage
Predetermined distance from the periphery of the pixel area on both sides in the width direction
The pixels in the peripheral area, which is the area inside the
Of the black pixels in each width direction in the elementary area,
All the pixels located in the width direction of the black pixel are white pixels.
A noise processing method comprising a pixel column conversion step of converting a black pixel into a white pixel . 2. A threshold for the length of the preset black pixel row, depending on the inclination angle of the input sheet image
The noise processing method according to claim 1 , which is determined . 3. If the preset threshold value for the length of the black pixel row has no inclination of the input paper image,
The input paper image has a length corresponding to the width of the input
If the paper image is slanted, the length exceeds the character input field.
It is the noise processing method according to claim 2, wherein. 4. The pixel string determination step comprises detecting a detected length.
But above the threshold for the length of the preset black pixel row, and the detected width Yue threshold to the width of the black pixel columns
Wherein at was e determines input paper image image quality and poor noise processing method according to claim 1, wherein. 5. A scanning an input paper image input by a scanner comprising a facsimile horizontally and vertically, to detect the length and width of the black pixel array black pixels are present in the following regular intervals, the said length A pixel that is determined to be a black pixel row to be removed when a preset threshold value for the length of the black pixel row is exceeded and the width is equal to or less than a preset threshold value for the width of the black pixel row. Black, which is an area composed of the column determination means and the black pixels determined to be removed
Predetermined distance from the periphery of the pixel area on both sides in the width direction
The pixels in the peripheral area, which is the area inside the
Of the black pixels in each width direction in the elementary area,
All the pixels located in the width direction of the black pixel are white pixels.
A noise processing device having a pixel column conversion means for converting a black pixel into a white pixel . 6. The threshold for the length of the preset black pixel row, depending on the inclination angle of the input sheet image
The noise processing device according to claim 5, which is determined . 7. If the preset threshold value for the length of the black pixel row has no inclination of the input paper image,
The input paper image has a length corresponding to the width of the input
If the paper image is slanted, the length exceeds the character input field.
Is the noise processing apparatus according to claim 6, wherein. 8. The pixel row determination means has a detected length exceeding a preset threshold for the length of a black pixel row, and a detected width exceeding a threshold for the width of a black pixel row.
The noise processing device according to claim 5, wherein the image quality of the input paper is determined to be poor when it is obtained .