JP2012065230A - Image binarizing method and image binarizing apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image binarizing method and an image binarizing apparatus that provide an appropriate binary image even when ruled lines, characters, or background portions of an input image have various differences in density.SOLUTION: The image binarizing apparatus comprises: a mask image generating unit that generates a binary mask image on the basis of an edge intensity as the absolute value of a differential value of an input image; a pattern extracting unit that generates a pattern-extracted image by extracting pixels corresponding to effective pixels of the mask image from the input image; a labeling unit that divides the effective pixels of the mask image for each of continuous regions; a threshold value determining unit that determines a binarization threshold value for each individual region of the pattern-extracted images corresponding to the continuous regions based on a brightness distribution of the individual region; and a binarization processing unit that binarizes the individual regions using the binarization threshold value for each individual region.

Description

  The present invention relates to an image binarization method and an image binarization apparatus for binarizing an input image such as a form.

  Conventionally, when image processing or pattern recognition is performed on an input image such as a form, binarization processing is often performed as preprocessing. Here, the binarization processing refers to processing for separating the target image such as ruled lines and characters from the background by binarizing the luminance value of the input image using a predetermined threshold.

  For example, in Patent Document 1, a histogram of a character region and a histogram of a background region are created based on a second derivative value of an input image, and an intersection of both histograms is determined as a threshold value. An image binarization method for binarizing the image is disclosed.

  Patent Document 2 discloses an image binarization method in which an input image is divided into a plurality of preset blocks and a threshold value is determined for each block.

Japanese Patent No. 3831797 JP 2005-165925 A

  However, when the input image is binarized using a single threshold as in the image binarization method described in Patent Document 1, there are various shade differences in the ruled lines, characters, or background of the input image. In some cases, it is difficult to obtain an appropriate binary image.

  For example, when a part of a white background includes a background having the same brightness as a ruled line or a character, if the binarization is performed using a single threshold, the background part may remain in the binary image. There is. Further, if the threshold is set to be dark in order to exclude the background portion, the ruled lines and characters may be erased together with the background portion.

  Even if the threshold value is determined for each predetermined block as in the image binarization method described in Patent Document 2, if a background with different luminance is mixed in the block, the background in the binary image Therefore, it is difficult to obtain an appropriate binarized image.

  From these facts, how can an image binarization method or image binarization apparatus that can obtain an appropriate binary image even when there are various shading differences in ruled lines, characters, or background of the input image? The realization is a big issue.

  The present invention has been made to solve the above-described problems of the prior art, and an appropriate binary image can be obtained even when there are various shading differences in ruled lines, characters, or background of an input image. An object is to provide an image binarization method and an image binarization apparatus.

  In order to solve the above-described problems and achieve the object, the present invention is an image binarization method for binarizing an input image, and obtains an edge strength which is an absolute value of a differential value of the input image. A mask image generating step for generating a binary mask image in which a pixel whose edge intensity is equal to or greater than a predetermined edge intensity threshold is an effective pixel and a pixel whose edge intensity is less than the edge intensity threshold is an invalid pixel; and an effective pixel of the mask image A die cutting step for generating a die cut image by extracting corresponding pixels from the input image, a step for dividing effective pixels of the mask image into continuous regions, and the die cut image corresponding to the continuous regions A threshold value determination step for determining a binarization threshold value for each individual region based on a luminance distribution of the individual region, and a binarization value determined in the threshold value determination step for each individual region Characterized in that it includes a binarization step of binarizing the pixels included in the individual areas with the value.

  Further, in the present invention according to the above invention, in the mask image generation step, the edge intensity distribution is generated using the edge intensity of the input image, and the edge based on the second derivative value of the generated edge intensity distribution. An inflection point of the intensity distribution is detected, and an edge strength at the detected inflection point is determined as the edge strength threshold value.

  Further, the present invention is the above invention, wherein a boundary region between a positive edge pixel having a positive differential value and a negative edge pixel having a negative differential value among the pixels having the edge intensity equal to or higher than the edge intensity threshold. A boundary region in which the arrangement order of the positive edge pixels and the negative edge pixels is opposite to the arrangement order of the positive edge pixels and the negative edge pixels obtained from one line segment is extracted as a valley region and extracted. The method further includes a disabling step of replacing the effective pixel of the mask image located in the valley region with the ineffective pixel.

  Further, the present invention is the above invention, wherein a boundary region between a positive edge pixel having a positive differential value and a negative edge pixel having a negative differential value among the pixels having the edge intensity equal to or higher than the edge intensity threshold. A boundary region in which the arrangement order of the positive edge pixels and the negative edge pixels is the same as the arrangement order of the positive edge pixels and the negative edge pixels obtained from one line segment is extracted and extracted as a mountain region The method further includes a validation step of replacing invalid pixels of the mask image located in the mountain area with valid pixels.

  Also, in the present invention according to the above invention, the threshold value determining step is a valley formed by a peak portion located on the highest brightness side in the brightness distribution of the individual region and another peak portion adjacent to the peak portion. The luminance value in the unit is determined as the binarization threshold value.

  The present invention is also an image binarization device that binarizes an input image, and obtains an edge strength that is an absolute value of a differential value of the input image, and the edge strength is equal to or greater than a predetermined edge strength threshold value. A mask image generating means for generating a binary mask image in which a pixel is an effective pixel and a pixel less than the edge intensity threshold is an invalid pixel, and a pixel corresponding to the effective pixel of the mask image is extracted from the input image A die cutting unit for generating a die cut image, a sorting unit for dividing the effective pixels of the mask image into continuous regions, and a luminance distribution of the individual region for each individual region of the die cut image corresponding to the continuous region And a threshold value determining means for determining a binarization threshold value based on the image, and for each individual area, an image included in the individual area using the binarization threshold value determined by the threshold value determining means. Characterized in that a and a binarizing means for binarizing.

  According to the present invention, the edge intensity, which is the absolute value of the differential value of the input image, is obtained, and pixels having the edge intensity equal to or higher than the predetermined edge intensity threshold are effective pixels, and pixels having the edge intensity less than the edge intensity threshold are invalid pixels A mask image of values is generated, a die cut image is generated by extracting pixels corresponding to effective pixels of the mask image from the input image, and effective pixels of the mask image are divided into continuous regions, and corresponding to the continuous regions. For each individual area of the die-cut image, a binarization threshold is determined based on the luminance distribution of the individual area, and the pixels included in the individual area are binarized using the binarization threshold for each individual area. Therefore, there is an effect that an appropriate binary image can be obtained even when there are various shade differences in the ruled lines, characters, or background of the input image.

  Further, according to the present invention, an edge intensity distribution is generated using the edge intensity of the input image, and an inflection point of the edge intensity distribution is detected based on the second derivative value of the generated edge intensity distribution, and the detection is performed. Since the edge strength at the inflection point is determined as the edge strength threshold, the ruled line / character region and the background region can be appropriately separated.

  Further, according to the present invention, a boundary region between a positive edge pixel having a positive differential value and a negative edge pixel having a negative differential value among pixels having an edge strength equal to or greater than an edge strength threshold, the positive edge A boundary region in which the arrangement order of pixels and negative edge pixels is opposite to the arrangement order of positive edge pixels and negative edge pixels obtained from one line segment is extracted as a valley region, and a mask image located in the extracted valley region is extracted. Since the effective pixels are replaced with the invalid pixels, the ruled lines and characters combined in the mask image can be appropriately separated from each other.

  Further, according to the present invention, a boundary region between a positive edge pixel having a positive differential value and a negative edge pixel having a negative differential value among pixels having an edge strength equal to or greater than an edge strength threshold, the positive edge A boundary area in which the arrangement order of pixels and negative edge pixels is the same as the arrangement order of positive edge pixels and negative edge pixels obtained from one line segment is extracted as a mountain area, and the invalid pixel of the mask image located in the extracted mountain area Is replaced with an effective pixel, so that one ruled line or character separated in the mask image can be appropriately combined.

  Further, according to the present invention, the luminance value in the valley formed by the peak located on the highest brightness side in the luminance distribution of the individual area and the other peak adjacent to the peak is used as the binarization threshold. Since the determination is made, the background area that could not be removed in the die cutting process can be surely removed in the binarization process.

FIG. 1 is a diagram showing an overview of an image binarization method according to the present invention. FIG. 2 is a block diagram illustrating the configuration of the image binarization apparatus according to the present embodiment. FIG. 3 is a diagram illustrating a processing procedure of mask image generation processing. FIG. 4 is an explanatory diagram of edge strength threshold value determination processing. FIG. 5 is an explanatory diagram of the mask image correction process. FIG. 6 is an explanatory diagram of the threshold value determination process. FIG. 7 is a flowchart showing a processing procedure executed by the image binarization apparatus. FIG. 8 is a flowchart showing another processing procedure of the mask image correction process.

  Exemplary embodiments of an image binarization method and an image binarization apparatus according to the present invention will be described below in detail with reference to the accompanying drawings.

  First, prior to detailed description of the embodiment, an outline of an image binarization method according to the present invention will be described with reference to FIG. FIG. 1 is a diagram showing an overview of an image binarization method according to the present invention. Note that (A) in the figure shows an example of a binary image obtained by a conventional image binarization technique, and (B) in the figure shows an outline of the image binarization technique according to the present invention. Each is shown.

  As shown in FIG. 1A, in the conventional image binarization method, for example, one binarization threshold is determined based on the luminance distribution of the input image, and the determined binarization threshold is used for the entire input image. To generate a binary image.

  For this reason, in the conventional image binarization method, when there is a difference in density in the background of the input image (see (A-1) in FIG. 1), there is a possibility that a dark background portion may remain in the binary image ( It was difficult to obtain an appropriate binary image (see (A-2) in FIG. 1). Also, if the binarization threshold is set high to exclude such background portions, there is a risk that even the ruled lines and characters will disappear together with the background portions.

  Therefore, in the image binarization method according to the present invention, a ruled line region and a character region (hereinafter referred to as “ruled line / character region”) are extracted from an input image using a predetermined mask image (that is, a background region is extracted). Exclude) Die cutting. Then, after dividing the die-cut image obtained by die-cutting into individual regions, the binarization threshold is determined for each individual region.

  First, in the image binarization method according to the present invention, as shown in FIG. 1B, a mask image is generated based on the edge strength of the input image (see (B-1) in FIG. 1).

  Specifically, in the image binarization method according to the present invention, a differential value of an input image is obtained using a differential filter such as a Sobel filter, and an edge strength which is an absolute value of the differential value is obtained. In the image binarization method according to the present invention, a binary image in which a pixel having an edge strength equal to or higher than a predetermined edge strength threshold is set as an effective pixel (white pixel) and a pixel having an edge strength lower than the edge strength threshold is set as an invalid pixel (black pixel). A mask image is generated.

  This makes it possible to obtain a mask image in which ruled lines / character areas having higher edge strength compared to the background area are represented by white pixels, and background portions having lower edge strength compared to the ruled lines / character areas are represented by black pixels. Can do.

  Such a mask image is an image obtained by expanding ruled lines and characters of the input image. This is because the differential filter obtains the edge strength using the pixel values of the pixel of interest and the surrounding pixels, and the pixel values of the pixels constituting the ruled line and the character influence the edge strength of the surrounding pixels. Because.

  Subsequently, in the image binarization method according to the present invention, the input image is die-cut using the generated mask image (see (B-2) in FIG. 1). Specifically, pixels corresponding to effective pixels (white pixels) of the mask image are extracted from the input image.

  Thereby, a die-cut image obtained by extracting the ruled line / character region from the input image (in other words, the background region is removed from the input image) is obtained. However, as described above, since the ruled lines and characters of the mask image are thicker than the ruled lines and characters of the input image, actually, not only the ruled lines and characters but also the surrounding background are extracted.

  Subsequently, in the image binarization method according to the present invention, the mask image is divided into continuous regions (see (B-3) in FIG. 1). Here, the continuous area is an area where effective pixels (white pixels) are continuous in the mask image. With this process, the mask image is divided into ruled lines and characters.

  Subsequently, in the image binarization method according to the present invention, a binarization threshold value is determined for each individual region of the punched image corresponding to the continuous region of the mask image based on the luminance distribution of the individual region (FIG. 1). (See (B-4)).

  Specifically, in the image binarization method according to the present invention, in the valley formed by the peak located on the highest brightness side in the luminance distribution of the individual area and the other peak adjacent to the peak. The luminance value is determined as a binarization threshold value.

  That is, the background color is usually lighter than the color of the ruled lines and characters (that is, close to white), so that the peak located on the highest luminance side (that is, the side close to white) is removed by the mask image. This is considered to correspond to the distribution of the background area that could not be fully covered. Therefore, the ruled line / character area and the background area can be reliably separated by determining the luminance value at the valley between the peak and the other peak as the binarization threshold.

  In the image binarization method according to the present invention, binarization is performed for each individual region using the binarization threshold values determined for each individual region (see (B-5) in FIG. 1). As a result, unlike the conventional binary image, it is possible to obtain a binary image in which the background area is reliably removed.

  As described above, in the image binarization method according to the present invention, the mask image is generated based on the edge strength of the input image, and the input image is die-cut using the generated mask image. Further, in the image binarization method according to the present invention, the effective pixels of the mask image are divided into continuous regions, and 2 for each individual region of the die-cut image corresponding to the continuous region based on the luminance distribution of the individual region. A value threshold is determined.

  In the image binarization method according to the present invention, each individual area is binarized using the binarization threshold value determined for each individual area. Therefore, according to the image binarization method according to the present invention, an appropriate binary image can be obtained even when there are various shade differences in the ruled lines, characters, or background of the input image.

  In the image binarization method according to the present invention, the edge strength at the inflection point of the edge strength distribution of the input image is calculated by paying attention to the difference in the shape of the edge strength distribution of the background region and the edge strength distribution of the ruled line / character region. The edge strength threshold is determined, but details of this point will be described later.

  Further, as described above, in the mask image, the ruled lines and characters are thicker than the input image, so that adjacent ruled lines and characters may be combined. In addition, since a mask image is generated based on the edge strength, one line may be separated into two.

  Therefore, in the image binarization method according to the present invention, a positive edge in a boundary region between a positive edge pixel having a positive differential value and a negative edge pixel having a negative differential value among pixels having an edge strength equal to or greater than an edge strength threshold. Based on the arrangement order of the pixels and the negative edge pixels, a region to be separated (valley region) and a region to be combined (mountain region) are detected, and the mask image is corrected based on the detected valley region and mountain region. This point will also be described later.

  Hereinafter, an embodiment of an image binarization method and an image binarization apparatus to which the image binarization method described with reference to FIG. 1 is applied will be described in detail. In the following, an image binarization apparatus that acquires an input image from an image reading apparatus (for example, a scanner apparatus) will be described. However, an image binarization apparatus including a scanner apparatus may be configured.

  FIG. 2 is a block diagram illustrating the configuration of the image binarization apparatus 10 according to the present embodiment. In the figure, only components necessary for explaining the characteristics of the image binarization apparatus 10 are shown, and descriptions of general components are omitted.

  As shown in FIG. 2, the image binarization apparatus 10 is connected to an image reading apparatus 20, and acquires an image read by the image reading apparatus 20 (for example, an image obtained by scanning a form). For example, a so-called scanner device can be used as the image reading device 20.

  The image binarization apparatus 10 includes a control unit 11 and a storage unit 12. The control unit 11 includes a mask image generation unit 11a, a mask image correction unit 11b, a die cutting unit 11c, and a labeling unit 11d. And a threshold value determination unit 11e and a binarization processing unit 11f. Then, the storage unit 12 stores label information 12a and threshold information 12b.

  The mask image generation unit 11 a is a processing unit that generates a mask image based on the edge strength of the input image when the input image is input from the image reading device 20. Specifically, the mask image generation unit 11a generates a differential image by extracting edges from the input image using a Sobel filter, and performs threshold processing on the generated differential image using a predetermined edge intensity threshold value. To generate a mask image.

  Here, the processing procedure of the mask image generation process by the mask image generation unit 11a will be described with reference to FIG. FIG. 3 is a diagram illustrating a processing procedure of mask image generation processing.

  As shown in FIG. 3, the mask image generation unit 11a applies a horizontal and vertical Sobel filter to the input image, respectively, so that the horizontal edge image and the vertical direction each having a differential value as a pixel value. The differential image is generated (see (1) in FIG. 3).

  In this embodiment, a case where a Sobel filter is used as an example of a differential filter will be described. However, a filter other than a Sobel filter such as a Previt filter or a Laplacian filter may be used as a differential filter.

  Subsequently, the mask image generation unit 11a determines an edge intensity threshold value for each differential image in the horizontal direction and the vertical method (see (2) in FIG. 3). The specific contents of the edge strength threshold determination process will be described later with reference to FIG.

  Subsequently, the mask image generation unit 11a performs threshold processing of each differential image using the edge intensity threshold determined for each differential image (see (3) in FIG. 3).

  Specifically, the mask image generation unit 11a sets a pixel having a negative differential value as a black pixel (pixel value “−1”) among pixels having an edge strength that is an absolute value of the differential value and an edge strength threshold value or more, A pixel having a positive differential value is defined as a white pixel (pixel value “1”). In addition, the mask image generation unit 11a sets a pixel whose edge strength is less than the edge strength threshold as a gray pixel (pixel value “0”). As a result, an edge image represented by three colors of black, white or gray is obtained. In this way, the differential value image after the threshold processing is an edge image.

  Then, the mask image generation unit 11a converts black pixels or white pixels among the pixels of the edge image into “white pixels” and gray pixels into “black pixels”, and then combines the two edge images. (See (4) in FIG. 3), a black and white binary mask image is generated. Hereinafter, the white pixel of the mask image may be referred to as “effective pixel”, and the black pixel may be referred to as “invalid pixel”.

  Here, the edge strength threshold value determination process shown in (2) of FIG. 3 will be described with reference to FIG. FIG. 4 is an explanatory diagram of edge strength threshold value determination processing. In the figure, as an example, edge strength threshold value determination processing for a differential image in the horizontal direction is shown. Further, (A) in the figure shows the differential value distribution of the differential image in the horizontal direction, and (B) in the figure shows how the edge strength threshold is determined based on the secondary differential value of the edge strength distribution. Respectively. The edge intensity distribution is an edge intensity distribution that is an absolute value of a differential value.

  As shown in FIG. 4A, the differential value distribution of the input image (in other words, the pixel value distribution of the differential image) is in the vicinity of the peak of the gentle distribution peak (near the differential value of “0”). The shape has a steep peak (see (A-1) in the figure). Here, the differential value distribution is composed of the differential value distribution in the background area shown in (A-2) of the figure and the differential value distribution in the ruled line / character area shown in (A-3) of the figure. It is thought that.

  Specifically, the pixels in the background area are overwhelmingly those with a differential value of “0”. This is because the background region usually has no edge. However, since there is a slight shading difference due to dirt or the like, there are a small number of pixels having a differential value other than “0” in the background area. Therefore, the edge intensity distribution in the background region tends to decrease exponentially with the differential value “0” as the apex, as shown in FIG.

  On the other hand, the pixels constituting the ruled lines and characters have various edge strengths depending on the difference in density from the background area. However, since the change in density becomes gentle due to image blurring or the like at the boundary with the background area, the number of pixels having a value near the differential value “0” is relatively large. Accordingly, the differential value distribution in the ruled line / character region tends to decrease gently with the differential value “0” as a vertex, as shown in FIG.

  Therefore, the mask image generation unit 11a detects an inflection point of the edge intensity distribution of the differential image, and uses the edge intensity at the detected inflection point as an edge intensity threshold, thereby separating the background area and the ruled line / character area. It was decided.

  Specifically, as shown in FIG. 4B, the mask image generation unit 11a takes the edge strength as the class value by taking the distribution of the edge strength that is the absolute value of the differential value of the edge image. Generate an edge strength distribution.

  Further, the mask image generation unit 11a obtains a primary differential value of the edge intensity distribution. In addition, the mask image generation unit 11a obtains a secondary differential value of the edge intensity distribution by further performing primary differentiation of the inverted primary differential value. In FIG. 4B, the positive / negative is inverted by multiplying the primary differential value by “−1”.

  The vertical axis on the left side of the edge intensity distribution shown in FIG. 4B is the “number of pixels”, and represents the frequency of the distribution plotted with “●” marks. In addition, the vertical axis on the right side of the edge intensity distribution is “differential value”, and represents the frequency of the distribution plotted with “▲” and the distribution plotted with “■”.

  Subsequently, the mask image generation unit 11a detects, as an inflection point, a point (zero cross point) at which the sign of the secondary differential value of the edge intensity distribution changes. Specifically, the mask image generation unit 11a detects, as an inflection point, an intersection between a straight line connecting the respective secondary differential values (plot values) before and after the change of positive and negative and the differential value “0”. Then, the mask image generation unit 11a determines the edge strength at the detected inflection point as the edge strength threshold.

  As described above, the mask image generation unit 11a generates an edge intensity distribution using the edge intensity of the input image, and detects an inflection point of the edge intensity distribution based on the second derivative value of the generated edge intensity distribution. Then, the edge strength at the detected inflection point is determined as the edge strength threshold. Therefore, it is possible to obtain a mask image in which the ruled line / character area and the background area are appropriately separated.

  In this embodiment, the edge strength threshold is determined by paying attention to the difference in the shape of the edge strength distribution of the background region and the edge strength distribution of the ruled line / character region. Therefore, as shown in FIG. 4A, the background area and the ruled line / character area can be appropriately separated even when the two distributions overlap.

  Although the case where the zero cross point is detected as the inflection point has been described here, the present invention is not limited to this. For example, the plot value immediately before or after the change of positive / negative may be detected as the inflection point. In this way, it is not necessary to calculate the zero cross point, so that the processing load required for detecting the inflection point can be reduced. Further, if the inflection point is detected using a value smoothed by a moving average or the like, the influence of noise is excluded, so that the inflection point can be detected more appropriately.

  Although the edge strength threshold value determination process for the horizontal edge image has been described here, the edge strength threshold value can be determined for the vertical edge image in the same procedure. Further, when the mask image generation unit 11a generates a mask image, the mask image generation unit 11a performs a process of passing the generated mask image together with the edge image to the mask image correction unit 11b.

  Returning to the description of FIG. 2, the mask image correcting unit 11b will be described. The mask image correction unit 11b is a processing unit that corrects a mask image using an edge image.

  Here, the mask image correction process by the mask image correction unit 11b will be described with reference to FIG. FIG. 5 is an explanatory diagram of the mask image correction process. Note that (A) in the figure shows the portion to be corrected of the mask image, (B) in the figure shows an example of the operation of the mask image correction processing, and (C) in the figure shows the valley image and An example of a mask image after correction using a mountain image is shown.

  As shown in FIG. 5A-1, when two lines are close to each other in the input image, in the edge image, a positive edge pixel having a positive edge strength among edge pixels obtained from one line. (White pixel) may be in contact with a negative edge pixel (black pixel) having a negative edge strength among edge pixels obtained from the other line. In such a case, in the mask image, these two lines are expressed as one line.

  Further, as shown in FIG. 5A-2, when a thick line exists in the input image, in the edge image, a positive edge pixel (white pixel) and a negative edge pixel (black pixel) obtained from the line. ) May leave. In such a case, one line is represented as two lines in the mask image.

  Therefore, the mask image correcting unit 11b detects the boundary between the positive edge pixel and the negative edge pixel obtained from different lines (“valley” shown in FIG. 5A-1), and includes such a valley. By replacing the white pixels of the mask image located in the valley area with black pixels, the combined line is corrected to the original two lines.

  Further, the mask image correcting unit 11b detects the boundary between the positive edge pixel and the negative edge pixel obtained from one line (“mountain” shown in FIG. 5A-2), and includes such a mountain. By replacing black pixels on the mask image located in the mountain area with white pixels, the line separated into two lines is corrected to the original single line.

  Specifically, as shown in FIG. 5B, the mask image correcting unit 11b scans the edge image pixel by pixel in a predetermined scanning direction (in this case, the right direction), and white A valley which is a boundary changing from a pixel to a black pixel is detected. And the mask image correction part 11b will extract the area | region containing the white pixel and black pixel which are located in the both sides of the detected trough as a trough area | region, if a trough is detected (refer (B-1) of the figure).

  Similarly, when the mask image correcting unit 11b detects a mountain which is a boundary changing from a black pixel to a white pixel, the mask image correcting unit 11b extracts a region including the black pixel and the white pixel located on both sides of the detected mountain as a mountain region. (Refer to (B-2) in FIG. 5).

  In addition, the mask image correcting unit 11b not only has a region where the black pixel and the white pixel are adjacent to each other, but also the black pixel and the white pixel are close to each other within a predetermined width threshold (here, 8 pixels) via the gray pixel. Are also extracted as valley regions and mountain regions. For example, when the black pixel changes from a black pixel to a white pixel through one pixel, the mask image correcting unit 11b extracts the region including the black pixel, the gray pixel, and the white pixel as a mountain region ( (See (B-3) in FIG. 5).

  Note that the mask image correcting unit 11b does not detect such a region as a valley region or a mountain region when the white pixel and the black pixel are separated by a predetermined width threshold or more via the gray pixel ((B- in FIG. 5). 4)). Further, here, only the case where the valley region and the mountain region are extracted from the edge image in the horizontal direction is shown, but the valley region and the mountain region are similarly extracted from the edge image in the vertical direction.

  Subsequently, the mask image correction unit 11b generates a valley image in which the extracted valley region is a black pixel (invalid pixel) and the other region is a white pixel (effective pixel). Similarly, the mask image correcting unit 11b generates a mountain image in which the extracted mountain area is a white pixel (effective pixel) and the other area is a black pixel (invalid pixel).

  Then, the mask image correcting unit 11b replaces the white pixel (effective pixel) of the mask image corresponding to the black pixel (invalid pixel) of the valley image with the black pixel (invalid pixel). That is, the valley image is superimposed on the mask image, and the white portion of the mask image located in the black portion of the valley image is replaced with black. As a result, as shown in FIG. 5C, the boundary portion between different lines becomes a black pixel, and the ruled lines and characters that have been combined are separated.

  Further, the mask image correcting unit 11b replaces the black pixel (invalid pixel) of the mask image corresponding to the white pixel (effective pixel) of the mountain image with a white pixel (effective pixel). As a result, as shown in FIG. 5C, the boundary portion of the two lines that should be one line becomes a white pixel, and one separated ruled line or character is combined.

  That is, by correcting the mask image using the valley image or the mountain image, it becomes possible to appropriately perform the labeling process described later.

  As described above, the mask image correcting unit 11b is a boundary region between the positive edge pixel (white pixel) and the negative edge pixel (black pixel), and the arrangement order of the positive edge pixel and the negative edge pixel is determined from one line segment. A boundary region that is opposite to the arrangement order of the obtained positive edge pixels and negative edge pixels is extracted as a valley region. Further, the mask image correcting unit 11b replaces the effective pixel (white pixel) of the mask image located in the extracted valley region with the invalid pixel (black pixel). Therefore, ruled lines and characters that have been combined in the mask image can be appropriately separated.

  The mask image correcting unit 11b is a boundary region between the positive edge pixel (white pixel) and the negative edge pixel (black pixel), and the arrangement order of the positive edge pixel and the negative edge pixel can be obtained from one line segment. A boundary region having the same arrangement order of positive edge pixels and negative edge pixels is extracted as a mountain region. In addition, the mask image correcting unit 11b replaces the invalid pixel (black pixel) of the extracted mask image located in the mountain area with the effective pixel (white pixel). Therefore, it is possible to appropriately combine one ruled line or character that has been separated in the mask image.

  The mask image correcting unit 11b also performs a process of passing the corrected mask image to the die cutting unit 11c and the labeling unit 11d.

  Returning to the description of FIG. 2, the die cutting portion 11 c will be described. When the die cutting unit 11c receives a corrected mask image (hereinafter simply referred to as “mask image”) from the mask image correcting unit 11b, the die cutting unit 11c performs die cutting of the input image using the received mask image. Part. Specifically, the die cutting unit 11c performs a process of generating a die cutting image by extracting pixels corresponding to effective pixels (white pixels) of the mask image from the input image.

  Further, when the die cutting unit 11c generates a die cutting image, the die cutting unit 11c also performs a process of passing the generated die cutting image to the threshold value determination unit 11e and the binarization processing unit 11f.

  The labeling unit 11d is a processing unit that performs a labeling process that divides the effective pixels of the received mask image into continuous regions when the mask image is received from the mask image correction unit 11b. Thereby, the mask image is divided for each ruled line or for each character.

  Moreover, the labeling part 11d will memorize | store the positional information on each continuous area | region in the memory | storage part 12 as label information 12a, if a labeling process is completed.

  When the threshold determination unit 11e receives a die-cut image from the die-cutting unit 11c, the threshold determination unit 11e specifies an individual region of the die-cut image corresponding to each continuous region of the mask image based on the label information 12a. Is a processing unit for determining a binarization threshold. Specifically, the threshold determination unit 11e performs a process of generating a luminance distribution of the individual area and determining a binarization threshold based on the generated luminance distribution.

  Here, the threshold value determination process by the threshold value determination unit 11e will be described with reference to FIG. FIG. 6 is an explanatory diagram of the threshold value determination process. Note that (A-1) in the figure shows an example of the input image, (A-2) in the figure shows a mask image generated from the input image, and (A-3) in the figure shows. The figure in which the mask image is superimposed on the input image is shown in (A-4) of FIG. Further, (B) in the figure shows an example of the luminance distribution of the individual area.

  When the input image shown in (A-1) of FIG. 6 is cut using the mask image shown in (A-2) of FIG. 6, it is shown in areas 200a and 200b of (A-3) of FIG. As described above, the surrounding background is extracted together with the ruled lines and characters of the input image. This is because in the mask image, the ruled lines and characters are thicker than the input image, and the background cannot be completely removed. Note that the area 200a shown in (A-3) in the figure is a part of the individual area of the punched image corresponding to the continuous area 100a shown in (A-2) in the figure, and ( A region 200b shown in A-3) is a part of the individual region of the die-cut image corresponding to the continuous region 100b shown in (A-2) of FIG.

  Here, the color of the background area is usually lighter than the color of the ruled line / character area. In other words, the luminance of the background area is usually higher than that of the ruled line / character area. Therefore, as shown in FIG. 6B, the threshold value determination unit 11e generates a luminance distribution of the individual region, and among the generated luminance distribution, the peak portion located on the highest luminance side and the peak portion are related to the peak portion. The luminance value in the valley 400 formed by other adjacent peaks is determined as a binarization threshold. Thereby, the binarization threshold value which more reliably removes the background area that could not be removed in the die cutting process can be obtained for each individual area. As a result, as shown in (A-4) of the figure, it is possible to obtain a binary image in which the background area that could not be removed by the die cutting process is more reliably removed.

  Note that, in a region where the brightness of the background existing outside and inside the ruled line is different, as in the region 200a illustrated in FIG. 6A-3, only the bright (bright) background is removed and the brightness is low. The (dark) background will remain without being removed. For this reason, in the binary image, a part of the binary image may be displayed slightly thicker than the input image, as in an area 300a shown in FIG. However, since most of the background area can be separated from the ruled line / character area when the input image is cut using the mask image, even if the background area remains slightly, Thus, the visibility of the binary image is not impaired.

  Further, when the threshold value determination unit 11e determines the binarization threshold value, the threshold value determination unit 11e also performs a process of storing the determined binarization threshold value in the storage unit 12 as threshold information 12b together with the position information of the corresponding individual area.

  Returning to the description of FIG. 2, the binarization processing unit 11f will be described. When the binarization processing unit 11f receives a die-cut image from the die-cutting unit 11c, the binarization processing unit 11f specifies an individual area of the die-cut image and a binarization threshold to be applied to each individual area based on the threshold information 12b. The processing unit performs binarization processing for each individual area. Note that the binarization processing unit 11f outputs the binarized die-cut image as a binary image when binarization is completed for all the individual areas.

  The storage unit 12 is a storage unit configured by a storage device such as a nonvolatile memory or a hard disk drive, and stores label information 12a and threshold information 12b. The label information 12a is information generated by the labeling unit 11d, and is information in which position information of the corresponding continuous area is associated with each identification information for identifying the continuous area of the mask image. The threshold information 12b is information generated by the threshold determination unit 11e, and is information in which a binarization threshold applied to a corresponding individual area is associated with each piece of position information of the individual area.

  Next, a processing procedure executed by the image binarization apparatus 10 according to the present embodiment will be described with reference to FIG. FIG. 7 is a flowchart showing a processing procedure executed by the image binarization apparatus 10.

  As shown in FIG. 7, in the image binarization apparatus 10, the mask image generation unit 11 a smoothes the input image using an average value filter or the like (step S <b> 101), and then horizontally using a Sobel filter. The differential image is generated by performing the primary differentiation in the vertical direction (step S102).

  Subsequently, the mask image generation unit 11a determines an edge intensity threshold value of each differential image using the edge intensity distribution (step S103), and performs threshold processing on each differential image using the determined edge intensity threshold value, thereby obtaining an edge image. Is generated (step S104). Then, the mask image generation unit 11a generates a mask image by combining the edge images (step S105).

  Subsequently, in the image binarization apparatus 10, the mask image correcting unit 11b generates a valley image and a mountain image using each edge image and corrects the mask image using the generated valley image and mountain image (step In step S106, the die cutting unit 11c generates a die cutting image by performing die cutting of the input image using the corrected mask image (step S107). Further, the labeling unit 11d performs a labeling process on the corrected mask image (step S108).

  Subsequently, in the image binarization apparatus 10, the threshold value determination unit 11e selects one individual area of the punched image (step S109), and determines the binarization threshold value based on the luminance distribution of the selected individual area. (Step S110) The binarization processing unit 11f binarizes the pixels in the individual area using the determined binarization threshold (Step S111).

  Then, the image binarization apparatus 10 determines whether or not all the individual areas have been processed (step S112), and if not processed (step S112, No), steps S109 to S112 for the unprocessed individual areas. Repeat the process. On the other hand, if all the individual areas have been processed (step S112, Yes), the process ends.

  As described above, in this embodiment, the mask image generation unit obtains the edge strength that is the absolute value of the differential value of the input image, and uses the pixel having the edge strength equal to or higher than a predetermined edge strength threshold as an effective pixel. A binary mask image in which pixels less than the intensity threshold are invalid pixels is generated, and a die cutting unit generates a die cutting image by extracting pixels corresponding to effective pixels of the mask image from the input image, and a labeling unit However, the effective pixels of the mask image are divided into continuous regions, and the threshold value determination unit determines a binarization threshold value for each individual region of the punched image corresponding to the continuous region based on the luminance distribution of the individual region. The binarization processing unit binarizes pixels included in the individual area using a binarization threshold for each individual area. Therefore, an appropriate binary image can be obtained even when there are various shade differences in the ruled lines, characters, or background of the input image.

  By the way, in a form or the like, the boundary line of each item may be expressed not by a ruled line but by a background color difference. In such a boundary line, unlike a ruled line, an edge appears only on one side, so that the mask image is likely to be blurred, and an appropriate die-cut image may not be generated.

  Therefore, the mask image correcting unit 11b detects the edge of a relatively long continuous line segment such as the above-described boundary line or ruled line using a condition different from the edge strength threshold, and uses the detection result to mask the mask image. It is good also as eliminating the above-mentioned blur by correcting. Here, the mask image correction processing will be described with reference to FIG. FIG. 8 is a flowchart showing another processing procedure of the mask image correction process.

  As shown in FIG. 8, the mask image correcting unit 11b extracts a region in which pixels having a predetermined edge strength or higher are continuous for a predetermined distance in the horizontal direction from the differential image in the horizontal direction (step S201). Similarly, the mask image correction unit 11b extracts, from the differential image in the vertical direction, a region in which pixels having a predetermined edge strength or higher are continued for a predetermined amount in the vertical direction (step S202). For example, the mask image correction unit 11b extracts from each edge image a region in which 50 or more pixels having an edge strength of “2” or more are continuous.

  Subsequently, the mask image correcting unit 11b replaces the invalid pixel (black pixel) of the mask image corresponding to the region extracted from each differential image with the effective pixel (white pixel) (step S203), and ends the process.

  As a result, blurring in the mask image of the boundary line or ruled line expressed by the difference in the background color can be eliminated, so that a more appropriate die cut image can be obtained.

  In the embodiment described above, both the ruled lines and the characters are extracted from the input image using the mask image. However, only the ruled lines or only the characters can be extracted from the input image.

  In such a case, the die cutting unit 11c specifies a continuous region that matches a predetermined condition from among the continuous regions of the mask image using the label information 12a, and performs die cutting of the input image using only the specified continuous region. Do. For example, the die cutting unit 11c cuts only the ruled line without extracting characters, dotted lines, or noise by performing die cutting of the input image using only a continuous region that is continuous in the horizontal direction or the vertical direction by 50 pixels or more. Can be extracted.

  As described above, the image binarization method and the image binarization apparatus according to the present invention are used to obtain an appropriate binary image even when there is a difference in shading between the ruled lines, characters, or background of the input image. This is useful, and is particularly suitable for binarizing a free format form.

DESCRIPTION OF SYMBOLS 10 Image binarization apparatus 11 Control part 11a Mask image generation part 11b Mask image correction part 11c Die-cut part 11d Labeling part 11e Threshold determination part 11f Binarization process part 12 Storage part 12a Label information 12b Threshold information 20 Image reader

Claims (6)

An image binarization method for binarizing an input image,
A binary mask image in which an edge strength which is an absolute value of a differential value of the input image is obtained, and a pixel whose edge strength is equal to or greater than a predetermined edge strength threshold is an effective pixel and a pixel whose edge strength is less than the edge strength threshold is an invalid pixel A mask image generation step for generating
A die-cutting step of generating a die-cut image by extracting pixels corresponding to effective pixels of the mask image from the input image;
A step of dividing the effective pixels of the mask image into continuous regions;
A threshold value determining step for determining a binarization threshold value for each individual area of the punched image corresponding to the continuous area based on a luminance distribution of the individual area;
A binarization step for binarizing pixels included in the individual region using the binarization threshold value determined in the threshold value determination step for each individual region. . The mask image generation step includes
An edge intensity distribution is generated using the edge intensity of the input image, an inflection point of the edge intensity distribution is detected based on a second derivative value of the generated edge intensity distribution, and the edge intensity at the detected inflection point The image binarization method according to claim 1, wherein the threshold value is determined as the edge intensity threshold value. A boundary region between a positive edge pixel having a positive differential value and a negative edge pixel having a negative differential value among the pixels having the edge strength equal to or higher than the edge strength threshold, the positive edge pixel and the negative edge A boundary region in which the pixel arrangement order is opposite to the arrangement order of the positive edge pixel and the negative edge pixel obtained from one line segment is extracted as a valley region, and the effective mask image located in the extracted valley region is extracted. The image binarization method according to claim 1, further comprising an invalidation step of replacing a pixel with an invalid pixel. A boundary region between a positive edge pixel having a positive differential value and a negative edge pixel having a negative differential value among the pixels having the edge strength equal to or higher than the edge strength threshold, the positive edge pixel and the negative edge A boundary region in which the pixel arrangement order is the same as the arrangement order of the positive edge pixel and the negative edge pixel obtained from one line segment is extracted as a mountain region, and invalid pixels of the mask image located in the extracted mountain region are extracted. The image binarization method according to claim 1, further comprising an enabling step of replacing with effective pixels. The threshold determination step includes
Determining a luminance value in a trough formed by a peak located on the highest luminance side in the luminance distribution of the individual region and another peak adjacent to the peak as the binarization threshold; The image binarization method according to any one of claims 1 to 4. An image binarization device for binarizing an input image,
A binary mask image in which an edge strength which is an absolute value of a differential value of the input image is obtained, and a pixel whose edge strength is equal to or greater than a predetermined edge strength threshold is an effective pixel and a pixel whose edge strength is less than the edge strength threshold is an invalid pixel Mask image generation means for generating
A die-cutting means for generating a die-cut image by extracting pixels corresponding to effective pixels of the mask image from the input image;
Partitioning means for partitioning the effective pixels of the mask image for each continuous region;
Threshold value determining means for determining a binarization threshold value for each individual region of the die-cut image corresponding to the continuous region based on a luminance distribution of the individual region;
An image binarization comprising: binarization means for binarizing pixels included in the individual area using the binarization threshold value determined by the threshold value determination means for each individual area apparatus.

Images