JP3844765B2 - Character recognition device - Google Patents

Description

  The present invention relates to a character recognition device for recognizing characters existing in a document image when the document image is transmitted from a facsimile.

  FIG. 20 is a block diagram showing a conventional character recognition device disclosed in the following Patent Document 1. In FIG. 20, when a document image is transmitted from a facsimile, 1 detects vertical line noise present in the document image. A vertical line noise detection unit 2 for removing vertical line noise detected by the vertical line noise detection unit 1, and 3 for character pixel estimation interpolation for correcting character loss due to vertical line noise removal And 4 are an isolated point noise removing unit that removes isolated point noise existing in the document image, and 5 is a character recognition unit that recognizes characters existing in the document image.

Next, the operation will be described.
A conventional character recognition device detects and removes vertical line noise and isolated point noise present in a document image transmitted from a facsimile machine. The vertical line noise referred to here is a facsimile receiver. A black vertical line generated on the document image due to dust adhering to it, and isolated point noise refers to a small cluster of black pixels on the document image.

Specifically, first, when a document image is transmitted from a facsimile, the vertical line noise detection unit 1 detects vertical line noise present in the document image.
The detection of vertical line noise is performed according to the following three determination conditions.
(1) If the number of black pixels on the vertical line is equal to or greater than a certain value, the black pixel on the vertical line is determined as vertical line noise.
(2) If the number of black pixels in the upper and lower a pixels of a certain black pixel is greater than or equal to a certain value, it is determined that the black pixel is vertical line noise. Here, “a” is 1.5 times or more the character height entered in the document image.
(3) A black pixel that is continuous by b pixels or more on the vertical line is obtained, and the black pixel is determined to be vertical line noise. Here, “b” is 1.5 times or more the character height entered in the document image.

  However, the vertical line noise detection unit 1 detects the vertical line noise when the area that should be white pixels on the document image (for example, the margin areas at the upper and lower ends and the left and right ends of the document image) is known in advance. In order to increase the accuracy, the above determination process is executed by regarding one black pixel detected in the region as e black pixels (e> 1).

  Then, when the vertical line noise detection unit 1 detects the vertical line noise, the vertical line noise removal unit 2 replaces the black pixels on the vertical line determined to be the vertical line noise with white pixels, and generates the vertical line noise. Remove.

Next, when the vertical line noise removal unit 2 removes the vertical line noise, the character pixel estimation interpolation unit 3 may lose a character along with the removal of the vertical line noise. Correct deficiencies in
Specifically, the position of the interrupted character line is estimated from the directionality of the edge of the black pixel cluster located on the left and right of the vertical line noise, and the white pixel on the estimated character line is replaced with the black pixel.

Next, when the character pixel estimation interpolation unit 3 corrects the character defect, the isolated point noise removal unit 4 obtains the area of the connected black pixels on the document image, and when the area of the connected black pixels is equal to or smaller than a certain value, The connected black pixel is determined as isolated point noise.
Then, the connected black pixels determined to be isolated point noise are replaced with white pixels, and the isolated point noise is removed.

  When the vertical line noise and isolated point noise are removed in this way, the character recognition unit 5 performs document analysis processing and character recognition processing on the document image from which the noise has been removed, and obtains the character recognition result. Output to the outside.

JP-A-9-238208

  Since the conventional character recognition apparatus is configured as described above, vertical line noise and isolated point noise existing in the document image can be removed, but the shadow of the paper feed roller of the facsimile is located near both ends of the document image. Roller noise that appears black, line noise that causes all pixels on the horizontal line to be white pixels due to electrical noise on the line, and a receiver that causes all pixels on the vertical line to be white pixels because the facsimile receiver is defective. There was a problem that noise and the like could not be removed.

  In addition, when a region that should be a white pixel on the document image (for example, margin regions at the upper and lower ends and the left and right ends of the document image) is known in advance, one black pixel detected in the region is determined. By considering e black pixels (e> 1), the detection accuracy of vertical line noise can be improved. However, if the region that should be a white pixel on the document image is not known in advance, the vertical line noise is detected. There has been a problem that the detection accuracy of can not be improved.

Further, when the area of the connected black pixels on the document image is equal to or smaller than a certain value, the connected black pixel is determined as isolated point noise, but the area of the connected black pixels is obtained for all the black pixels on the document image. Has a problem that it cannot quickly remove isolated point noise.
Further, since the position of the roller noise cannot be obtained accurately, there is a problem that a table or a character string on the document image may be accidentally lost.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a character recognition device that can eliminate the problem of accidentally losing a table or character string on a document image.

  The character recognition device according to the present invention counts the number of black pixels present in a document image for each vertical line, and sets the count value of each vertical line as a threshold value from the left and right ends of the document image toward the center. In comparison, a roller noise removal unit is provided that determines a roller noise region up to a position where the count value is smaller than a threshold value, and replaces a black pixel included in the roller noise region with a white pixel.

  According to this invention, by counting the number of black pixels present in the document image for each vertical line, the count value of each vertical line is compared with the threshold value from the left and right ends of the document image toward the center, Since the roller noise area is determined up to a position where the count value becomes smaller than the threshold value, and a black pixel included in the roller noise area is replaced with a white pixel, the position of the roller noise is determined. As a result, there is an effect that it is possible to eliminate a problem that a table or character string on a document image is mistakenly lost.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing a character recognition apparatus according to Embodiment 1 of the present invention. In FIG. 1, reference numeral 11 denotes a noise removal parameter (specified number) used by the noise removal unit 12 when a document image is transmitted from a facsimile. A noise removal parameter setting unit (parameter setting unit) 12 for setting the noise, a noise removing unit (noise removing unit) 12 for detecting and removing noise existing in the document image, and a noise removing unit 12 for removing the noise. A character recognition unit (character recognition means) for recognizing characters existing in the document image.

  14 counts the number of black pixels present in the document image for each horizontal line, and assigns points to each black pixel according to the counting result, and the points given by the score granting unit 14a. A vertical line noise determination unit 14b that determines vertical line noise as a vertical line noise that is totaled for each vertical line and the total number of points is greater than a specified number, and a vertical line that is determined as vertical line noise by the vertical line noise determination unit 14b. 2 is a vertical line noise processing unit (see FIG. 2).

  15 is a black pixel located at the center of the local area when the number of black pixels included in the local area of the document image is smaller than the specified number and the distribution shape of the black pixels included in the local area has no directionality. An isolated point noise comprising an isolated point noise determining unit 15a for determining as isolated point noise, and an isolated point noise removing unit 15b for replacing a black pixel determined to be an isolated point noise by the isolated point noise determining unit 15a with a white pixel. It is a processing unit (see FIG. 4).

  Reference numeral 16 denotes a roller noise processing unit including a roller noise removing unit that replaces black pixels included in roller noise regions located at both left and right ends of the document image with white pixels, and 17 denotes the number of black pixels present in the document image. A line noise determination unit 17a that determines a line group of horizontal lines in which the number of black pixels sandwiched between horizontal lines having one or more black pixels is counted as line noise is counted for each horizontal line by the line noise determination unit 17a. Of the white pixels in the horizontal line group determined to be line noise, the line noise removing unit 17b replaces the white pixels, which are both black pixels, in the pixels located above and below the horizontal line group with black pixels. A line noise processing unit (see FIG. 8).

  Reference numeral 18 counts the number of black pixels existing in the document image for each vertical line, and determines that the vertical line group between the vertical lines with the number of black pixels being 1 or more and zero is the receiver noise. And the white pixels in the vertical line group determined as the photoreceiver noise by the photoreceiver noise determination unit 18a are white pixels in which both pixels located on the left and right of the vertical line group are black pixels. This is a photoreceiver noise processing unit including a photoreceiver noise removing unit 18b that replaces pixels with black pixels (see FIG. 10).

Next, the operation will be described.
First, an outline of the operation of the character recognition apparatus will be briefly described.
First, when a document image is transmitted from a facsimile, the noise removal parameter setting unit 11 sets a noise removal parameter used by the noise removal unit 12.
The noise removal parameter is a parameter used by each noise processing unit constituting the noise removal unit 12, and has a function of adjusting the amount of noise removal.

  Then, when the noise removal parameter setting unit 11 sets the noise removal parameter, the noise removal unit 12 detects and removes various types of noise from the document image according to the noise removal parameter. Detailed operation of the noise removing unit 12 will be described later.

  When the noise removing unit 12 removes various types of noise, finally, the character recognizing unit 13 performs document analysis processing (character string and chart extraction processing) on the document image from which noise has been removed, The character recognition process of the extracted character string is executed, and the character recognition result is output to the outside.

Next, the detailed operation of the noise removing unit 12 will be described.
First, the vertical line noise processing unit 14 of the noise removal unit 12 detects vertical line noise on the document image and removes the vertical line noise from the document image. The assigning unit 14a counts the number of black pixels present in the document image for each horizontal line, and executes a process for obtaining the number of black pixels present on each horizontal line. And the score provision part 14a provides a score to the black pixel which exists on each horizontal line according to the count result.

Specifically, the number of points given to each black pixel is given according to the counting result, but is set to a larger value as the number of black pixels existing on the horizontal line is smaller.
Accordingly, since the number of black pixels is small in the blank areas at the upper and lower ends of the document image, the number of points assigned to the black pixels in that area is high.
On the contrary, since there are many black pixels in the horizontal line crossing the table and the character string area, the number of points given to the black pixels on the horizontal line is low.

As a result, a high score is given to the black pixels having a high possibility of vertical line noise. For example, in the case of the document image shown in FIG. 12, most of the black pixels in the blank area etc. (Area A is an area where there are almost no black pixels on the horizontal line), and a high score is given to the black pixels included in area A.
On the other hand, most of the black pixels constituting the character and the table are included in the region B (the region B is a region where many black pixels exist on the horizontal line), and the black pixels included in the region B have a low score. Is granted.

Note that the score giving unit 14a has a normal vertical line noise width of about several dots, and it is highly possible that both the left and right sides of each black pixel constituting the vertical line noise are white pixels. When both the left and right pixels are white pixels, points are added to the black pixels. As a result, a high score is given to a black pixel having a high possibility of vertical line noise.
For example, in the case of the document image shown in FIG. 12, the left and right of the black pixels on the vertical line noise N1, N3 having a line width of 1 pixel are mostly white pixels, so that the vertical line noises N1, N3 are formed. Points are added to the black pixels.

Then, when the score assigning unit 14a gives a score to each black pixel, the vertical line noise determination unit 14b adds up the scores of the black pixels existing on the vertical line for each vertical line of the document image, and the total score is A vertical line larger than the parameter TH is determined as vertical line noise.
However, since the width of the vertical line noise is usually about several lines, when n or more vertical lines having a total score greater than the parameter TH are adjacent, it is determined that those vertical lines are not vertical line noise.

In the case of the document image shown in FIG. 12, since the total number of vertical lines N1, N2, N3, and N4 is larger than the parameter TH, the vertical lines N1 to N4 are determined as vertical line noise.
Here, the parameters TH and n are noise removal parameters, and are values set by the noise removal parameter setting unit 11.

Then, when the vertical line noise determination unit 14b determines vertical line noise, the vertical line noise removal unit 14c replaces black pixels on the vertical line determined to be vertical line noise with white pixels.
However, if both of the pixels that touch the left and right of the vertical line noise are black pixels on the same horizontal line as the black pixels on the vertical line determined as vertical line noise, the black pixels on the vertical line are Think of it as part of a character or ruled line and do not replace it with white pixels.

FIG. 3 shows an example of processing of a document image by the vertical line noise processing unit 14. In FIG. 3, 301 is a document image before vertical line noise removal, and 302 is a document image after vertical line noise removal.
In the document image 302 after removing the vertical line noise, the vertical line noise is removed without loss of characters and table ruled lines.

  Next, the isolated point noise processing unit 15 of the noise removing unit 12 detects isolated point noise on the document image and removes the isolated point noise from the document image. The isolated point noise determination unit 15a examines each black pixel present in the document image and determines whether each black pixel is isolated point noise.

Specifically, the number of black pixels included in a local region (for example, a 5 × 5 pixel region) centered on the black pixel is checked, and when the number of black pixels is M or less, the central black pixel Is isolated noise.
Here, M is a noise removal parameter, and is a value set by the noise removal parameter setting unit 11.

However, since the black pixels constituting the characters and ruled lines are known to have local directionality as shown in FIG. 5, the distribution of the black pixels in the local region is one of those shown in FIG. When the number of black pixels included in the local region is M or less, the central black pixel is determined not to be isolated point noise.
In FIG. 5, 501 is a distribution of black pixels having a vertical directionality, 502 is a distribution of black pixels having a horizontal directionality, and 503 is a distribution of black pixels having an oblique directionality.

Then, when the isolated point noise determination unit 15a determines the isolated point noise, the isolated point noise removing unit 15b replaces the black pixel determined to be the isolated point noise with the white pixel.
FIG. 6 shows an example of processing of a document image by the isolated point noise processing unit 15. In the figure, reference numeral 601 denotes a document image before removing isolated point noise, and 602 denotes a document image after removing isolated point noise.
In the document image 602 after the isolated point noise is removed, the isolated point noise is removed without loss of characters and table ruled lines.

Next, the roller noise processing unit 16 of the noise removing unit 12 removes the roller noise on the document image from the document image. Specifically, the X pixel from the left and right ends of the document image toward the center. The inner area is considered as an area including roller noise (hereinafter referred to as roller noise area), and black pixels included in the roller noise area are replaced with white pixels.
Here, X is a noise removal parameter, which is a value set by the noise removal parameter setting unit 11.

FIG. 7 shows an example of processing of a document image by the roller noise processing unit 16. In FIG. 7, reference numeral 701 denotes a document image before roller noise removal, and reference numeral 702 denotes a document image after roller noise removal.
In the document image 702 after removing the roller noise, the roller noise at both ends of the document image is removed.

  Next, the line noise processing unit 17 of the noise removing unit 12 detects line noise on the document image and removes the line noise from the document image. 17a executes a process of checking the number of black pixels on the horizontal line for each horizontal line of the document image.

Then, the line noise determination unit 17a determines that the horizontal line sandwiched between the horizontal lines having the number of black pixels of 1 or more among the horizontal lines having the number of black pixels of zero is the line noise.
However, since the width of the line noise is usually about several lines, when Y or more horizontal lines with zero black pixels are adjacent to each other, it is determined that the horizontal lines are not line noise.

In the case of the document image shown in FIG. 13, a plurality of horizontal lines with zero black pixels are adjacent to each other. However, since the number of adjacent lines is less than Y, the horizontal line group C is determined as line noise. ing.
Here, Y is a noise removal parameter, which is a value set by the noise removal parameter setting unit 11.

Then, the line noise removing unit 17b removes the line noise from the document image when the line noise determining unit 17a determines the line noise.
Specifically, if both pixels that are in contact with the line noise are black pixels, it is considered that part of the characters and ruled lines are lost due to the line noise, and the black pixels that are in contact with the line noise are connected. The white pixel on the line noise is replaced with the black pixel.

FIG. 9 shows an example of document image processing by the line noise processing unit 17, in which 901 is a document image before line noise removal, and 902 is a document image after line noise removal.
In the document image 902 after removal of the line noise, the loss of characters and table ruled lines due to the line noise is corrected.

  Next, the receiver noise processing unit 18 of the noise removing unit 12 detects the receiver noise on the document image and removes the receiver noise from the document image. The photoreceiver noise determination unit 18a performs a process of checking the number of black pixels on the vertical line for each vertical line of the document image.

Then, the light receiver noise determination unit 18a determines a vertical line sandwiched between vertical lines with one or more black pixels among vertical lines with zero black pixels as light receiver noise.
However, since the width of the photoreceiver noise is usually about several lines, when there are Z or more vertical lines with zero black pixels, it is determined that these vertical lines are not photoreceiver noise.

In the case of the document image shown in FIG. 14, a plurality of vertical lines with zero black pixels are adjacent to each other, but the number of adjacent lines is less than Z, so that the vertical line group D is determined to be receiver noise. is doing.
Here, Z is a noise removal parameter, which is a value set by the noise removal parameter setting unit 11.

The photoreceiver noise removing unit 18b removes the photoreceiver noise from the document image when the photoreceiver noise determining unit 18a determines the photoreceiver noise.
Specifically, if both pixels that touch the left and right sides of the receiver noise are black pixels, it is considered that part of the characters and ruled lines are lost due to the receiver noise, and the black pixels that touch the left and right sides of the receiver noise are The white pixels on the receiver noise are replaced with black pixels so that they are connected.

FIG. 11 shows an example of document image processing by the photoreceiver noise processing unit 18. In the figure, reference numeral 1101 denotes a document image before the photoreceiver noise is removed, and 1102 denotes a document image after the photoreceiver noise is removed.
In the document image 1102 after the light receiver noise is removed, the loss of characters and ruled lines due to the light receiver noise is corrected.

  As apparent from the above, according to the first embodiment, the number of black pixels present in the document image is counted for each horizontal line, and a score is assigned to each black pixel according to the counting result. When the total number of points for each vertical line is summed up and the vertical line whose total score is greater than the parameter TH is determined to be vertical line noise, the area that should be white pixels on the document image is not known in advance. However, there is an effect that the detection accuracy of vertical line noise can be improved.

  Further, according to the first embodiment, when the number of black pixels included in the local area of the document image is smaller than M and the distribution shape of the black pixels included in the local area has no directionality, Since the black pixel positioned at the center of the region is determined to be isolated point noise, it is possible to determine isolated point noise without obtaining a connected black pixel on the document image, and as a result, There is an effect that the isolated point noise can be quickly removed.

  Furthermore, according to the first embodiment, since the roller noise processing unit 16, the line noise processing unit 17, and the light receiver noise processing unit 18 are provided, not only vertical line noise and isolated point noise but also roller noise. , Line noise and light receiver noise can be removed.

Embodiment 2. FIG.
FIG. 15 is a block diagram showing a detailed configuration of the roller noise processing unit 16 of the character recognition apparatus according to the second embodiment of the present invention. In the figure, 16a indicates the number of black pixels existing in the document image for each vertical line. A roller noise determination unit that counts and determines a roller noise region based on the counting result, and 16b is a roller noise removal unit that replaces black pixels included in the roller noise regions located at the left and right ends of the document image with white pixels. .

Next, the operation will be described.
In the first embodiment, a method of removing roller noise by considering an area within the X pixel from the left and right ends of the document image to the center as a roller noise region has been described. In this method, the position of the roller noise is accurately determined. Since it is not known, there is a possibility of removing a part of characters and a table.
Therefore, in the second embodiment, the roller noise area is determined in advance to remove the roller noise.

  Specifically, as shown in FIG. 16, the roller noise determination unit 16a of the roller noise processing unit 16 has an area in the G pixel that can be a roller noise area from the left and right ends of the document image toward the center. Considering this, a process for checking the number of black pixels on the vertical line in the area is executed.

However, since a change in the number of black pixels in a vertical line is captured globally, a smoothing process is applied to the number of black pixels so that the number of black pixels in adjacent vertical lines does not vary greatly.
For example, 1301 in FIG. 16 is a document image before roller noise removal, and 1302 is a smoothed black pixel distribution created from this document image.

Then, the roller noise determination unit 16a investigates the black pixel distribution 1302 after smoothing from the left and right ends of the document image toward the center, and until the position where the number of black pixels on the vertical line is smaller than the threshold W, the roller noise is detected. It is determined as an area.
As a result, the area 1303 in FIG. 16 is determined as the roller noise area.
Here, G and W are noise removal parameters, which are values set by the noise removal parameter setting unit 11.

  Then, when the roller noise determination unit 16a determines the roller noise region, the roller noise removal unit 16b replaces black pixels included in the roller noise region with white pixels and removes roller noise from the document image.

  As is apparent from the above, according to the second embodiment, the number of black pixels existing in the document image is counted for each vertical line, and the roller noise area is determined based on the counting result. As a result, the position of the roller noise can be accurately obtained, and as a result, it is possible to eliminate the problem of erroneously losing a table or character string on the document image.

Embodiment 3 FIG.
FIG. 17 is a block diagram showing a character recognition apparatus according to Embodiment 3 of the present invention. In the figure, the same reference numerals as those in FIG.
Reference numeral 19 denotes a reprocessing determination unit (reprocessing determination unit) that instructs the noise removal parameter setting unit 11 to change the noise removal parameter when the character recognition unit 13 fails in character recognition.

Next, the operation will be described.
In the first and second embodiments, since the noise removal process is applied only once to the document image, the noise may not be completely removed from the document image having a lot of noise.
When a document image including a large amount of noise is given to the character recognition unit 13 in this manner, the character recognition unit 13 may determine that the document image cannot be processed and return an error.

  For example, when the character recognition unit 13 processes a certain standard document, the document image including noise and the standard document are different in shape, so that it is determined that the document image is not a desired standard document and an error is returned. In this case, there is a problem that the result of character recognition cannot be output.

  Therefore, in the third embodiment, when an error is returned from the character recognition unit 13 so that a character can be recognized even from a document image including a large amount of noise and a character recognition result can be output. In this case, the noise removal parameter setting unit 11 is instructed to change the noise removal parameter.

  Specifically, when the character recognition unit 13 executes character recognition processing and outputs a character recognition result, the reprocessing determination unit 19 outputs the recognition result to the outside. When the error is output upon failure, the noise removal parameter setting unit 11 is instructed to change the noise removal parameter.

Thereby, the noise removal parameter setting unit 11 changes the value of the noise removal parameter so that more noise can be removed, and the noise removal unit 12 executes the noise removal process again.
The above processing is repeatedly executed a maximum of R times until no error occurs in the character recognition unit 13.

  As apparent from the above, according to the third embodiment, when the character recognition unit 13 fails in character recognition, the noise removal parameter setting unit 11 is instructed to change the noise removal parameter. There is an effect that a character recognition result can be obtained from a document image including a large amount of noise.

Embodiment 4 FIG.
18 is a block diagram showing a character recognition apparatus according to Embodiment 4 of the present invention. In the figure, the same reference numerals as those in FIG.
In the figure, reference numeral 20 denotes a buffer (registration means) for registering the telephone number of the transmission destination of the document image and the noise removal parameter set by the noise removal parameter setting unit 11 when the character recognition unit 13 succeeds in character recognition.

Next, the operation will be described.
In the third embodiment, the reprocessing determination unit 19 outputs the character recognition result when the character recognition unit 13 succeeds in the character recognition processing. However, the reprocessing determination unit 19 performs the character recognition result. And the noise removal parameter set by the noise removal parameter setting unit 11 in association with the destination telephone number of the document image is registered in the buffer 20.

  Subsequently, when another document image is transmitted from the facsimile, the noise removal parameter setting unit 11 reads the noise removal parameter corresponding to the telephone number of the transmission source of the document image from the buffer 20, and sets the noise removal parameter. Output to the noise removal unit 12.

  As a result, the noise removal unit 12 uses the noise removal parameter when the character recognition unit 13 succeeds in the character recognition process. However, the noise characteristics (amount and type of noise) included in the document image are as follows. Since this often depends on the facsimile of the document image transmission source, this is considered to be optimization of noise removal.

  As apparent from the above, according to the fourth embodiment, when another document image is transmitted from the facsimile, the noise removal parameter corresponding to the telephone number of the document image transmission source is read from the buffer 20, and the Since the noise removal parameter is configured to be output to the noise removal unit 12, it is possible to optimize the noise removal, and as a result, it is possible to effectively remove the noise present in the document image. .

Embodiment 5 FIG.
19 is a block diagram showing a character recognition apparatus according to Embodiment 5 of the present invention. In the figure, the same reference numerals as those in FIG.
In the figure, 21 is a noise evaluation unit (noise evaluation means) that calculates a noise evaluation value indicating the amount of noise removed by the noise removal unit 12, and 22 is a noise evaluation value calculated by the noise evaluation unit 21 that is greater than a threshold value. Then, a noise warning unit (message return unit, warning unit) that returns a message prompting maintenance of the facsimile to the transmission source of the document image and presenting a message prompting confirmation of the recognition result by the character recognition unit 13.

Next, the operation will be described.
In the first to fourth embodiments, the character recognition result obtained by the character recognition unit 13 is output externally. However, the amount of noise present in the document image is detected, and various messages are returned. May be.

Specifically, the noise evaluation unit 21 calculates a noise evaluation value indicating the amount of noise removed from the document image by the noise removal unit 12.
For example, the noise evaluation value includes the number of vertical line noises, the number of black pixels determined as isolated point noise, the area of the area determined as roller noise, the number of line noises, and the number of receiver noises. It can be expressed as a linear sum of Note that the noise evaluation value increases as the amount of noise increases.

  Then, when the noise evaluation unit 21 calculates the noise evaluation value, the noise warning unit 22 compares the noise evaluation value with a predetermined threshold, and when the noise evaluation value is larger than the threshold, the transmission source of the document image In response, a message prompting the user to clean or repair the facsimile is returned.

  In addition, when the noise evaluation value is larger than the threshold value, the noise warning unit 22 has a high possibility that there is an error in character recognition by the character recognition unit 13, so that a message that prompts inspection and correction of the recognition result is recognized. Output to the operator of the device.

  As is apparent from the above, according to the fifth embodiment, when the noise evaluation value calculated by the noise evaluation unit 21 is larger than the threshold value, a message for prompting maintenance of the facsimile is returned to the transmission source of the document image. Thus, the maintenance of the transmission source facsimile to which a large amount of noise is applied is promoted, and when the maintenance is executed, the amount of noise is reduced.

  Further, according to the fifth embodiment, when the noise evaluation value calculated by the noise evaluation unit 21 is larger than the threshold value, the message that prompts the character recognition unit 13 to confirm the recognition result is presented. The operator of the recognition apparatus can effectively correct the recognition result by the character recognition unit 13 as necessary.

It is a block diagram which shows the character recognition apparatus by Embodiment 1 of this invention. It is a block diagram which shows the detailed structure of a vertical line noise process part. It is explanatory drawing which shows the removal result of vertical line noise. It is a block diagram which shows the detailed structure of an isolated point noise process part. It is explanatory drawing explaining the directivity of black pixel distribution. It is explanatory drawing which shows the removal result of an isolated point noise. It is explanatory drawing which shows the removal result of roller noise. It is a block diagram which shows the detailed structure of a line noise process part. It is explanatory drawing which shows the removal result of a line noise. It is a block diagram which shows the detailed structure of a light receiver noise process part. It is explanatory drawing which shows the removal result of light receiver noise. It is explanatory drawing explaining operation | movement of a vertical line noise process part. It is explanatory drawing explaining operation | movement of a line noise process part. It is explanatory drawing explaining operation | movement of a light receiver noise process part. It is a block diagram which shows the detailed structure of the roller noise process part of the character recognition apparatus by Embodiment 2 of this invention. It is explanatory drawing which shows the removal result of roller noise. It is a block diagram which shows the character recognition apparatus by Embodiment 3 of this invention. It is a block diagram which shows the character recognition apparatus by Embodiment 4 of this invention. It is a block diagram which shows the character recognition apparatus by Embodiment 5 of this invention. It is a block diagram which shows the conventional character recognition apparatus.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 11 Noise removal parameter setting part (parameter setting means), 12 Noise removal part (noise removal means), 13 Character recognition part (character recognition means), 14a Score assignment part, 14b Vertical line noise determination part, 14c Vertical line noise removal part 15a Isolated point noise determination unit, 15b Isolated point noise removal unit, 16a Roller noise determination unit, 16b Roller noise removal unit, 17a Line noise determination unit, 17b Line noise removal unit, 18a Receiver noise determination unit, 18b Receiver noise Removal unit, 19 reprocessing determination unit (reprocessing determination unit), 20 buffer (registration unit), 21 noise evaluation unit (noise evaluation unit), 22 noise warning unit (message return unit, warning unit).

Claims (4)

  In a character recognition device comprising noise removal means for detecting noise present in a document image and removing the noise, and character recognition means for recognizing a character present in the document image from which noise has been removed by the noise removal means The noise removing means counts the number of black pixels present in the document image for each vertical line, and compares the count value of each vertical line with a threshold value from the left and right ends of the document image toward the center. Character recognition comprising: a roller noise removal unit that determines a roller noise area up to a position where the count value is smaller than a threshold value, and replaces black pixels included in the roller noise area with white pixels apparatus.   2. The character recognition device according to claim 1, further comprising a noise evaluation unit that calculates a noise evaluation value indicating an amount of noise removed by the noise removal unit.   3. The character return device according to claim 2, further comprising: a message return means for returning a message prompting maintenance of the facsimile to the sender of the document image when the noise evaluation value calculated by the noise evaluation means exceeds a predetermined value. Recognition device.   3. The character recognition apparatus according to claim 2, further comprising warning means for presenting a message prompting confirmation of the recognition result by the character recognition means when the noise evaluation value calculated by the noise evaluation means exceeds a predetermined value.

Images