JP2001045303A - Image thresholding method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve readability of a low contrast part in an image thresholding method for thresholding an analog or multi-gradation digital image to be processed by level-discriminating it according to a threshold. SOLUTION: A concentration histogram of each component of R, and B of a picture and maximum values Rmax, Gmax, and Bmax and minimum Rmin, Gmin, and Bmin of concentration histogram distribution of each R, G, and B are calculated (s8), and the concentration histogram distribution is extended so that the values of both edges of each data value area can be obtained (s9-s11). Thus, correction of brightenss and the correction of contrast shortage and the correction of color tone can be operated. As a result, satisfactory binary pictures suitable for contents judgment can be obtained. Also, a contrast and the color tone can be simultaneously corrected, and processing speed can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for binarizing a color still image obtained by a digital still camera, an image scanner or a video capture circuit into black and white.

[0002]

2. Description of the Related Art In converting a color still image into a black-and-white binary image, the obtained color image signals and data are level-discriminated with a predetermined threshold value.
Various methods have been known for creating the threshold. For example, a method of simply setting the average luminance value of all pixels as a threshold for binarization, or a density histogram indicated by reference numeral α1 in FIG. Bit, that is, 256 gradations from 0 to 255), and a luminance value corresponding to the valley portion of the histogram indicated by reference numeral α2 is used as a threshold value.

Further, another prior art, Japanese Patent Laid-Open No.
No. 728 proposes that the background and the character component are separated by a neural network that has been trained in advance without using a threshold value in binarization on the premise of application to character recognition and the like. I have.

[0004]

However, in the above-described threshold value determination method, either one of a case where one threshold value is used for the entire image and a case where the image is divided and a threshold value is obtained for each minute area are used. Also in this case, as shown by reference numeral α3 in FIG. 11, when the contrast is low and the density histogram is biased, the threshold value is set to an extremely high or low value, and the binarized image is visually inspected. However, there is a problem that an image is difficult to determine. For example, in the case of a digital image obtained by photographing a black-and-white newspaper photograph with a digital camera using an image scanner, a CCD, a C-MOS area sensor, etc., the contrast of the newspaper photograph portion is low. A minute change included in the digital image is also binarized to 1 or 0, respectively, resulting in a binary image whose content is difficult to determine. Similarly, in the case where a natural image is binarized for the effect of giving an impact, and when binarization is performed as preprocessing such as character recognition, the threshold component determination method according to the above-described conventional technique causes the noise component to be reduced. Increased and undesirable.

On the other hand, in the case of binarization using the neural network, the result depends on the degree of learning of the net. Therefore, if the learning time is short, it is difficult to deal with both natural images and characters. There are problems such as a large overhead before practical use.

An object of the present invention is to provide an image binarization method capable of improving the readability of a low-contrast portion of an image to be binarized and reducing noise.

[0007]

According to the image binarization method of the present invention, an analog or multi-tone digital image to be processed is level-discriminated by a predetermined threshold value.
In the image binarization method in which the binarization process is performed, the density histogram of each of R, G, and B components of the analog or multi-tone digital image to be processed and R,
Maximum value Rma of density histogram distribution of each of G and B
x, Gmax, Bmax and minimum values Rmin, Gmi
n and Bmin, and the maximum values Rmax, Gmax,
The brightness correction, the lack of contrast correction, and the color tone correction are performed by extending the density histogram distribution so that Bmax and the minimum values Rmin, Gmin, and Bmin become values at both ends of the data value range.

According to the above arrangement, the brightness and contrast of the image are corrected by expanding the density histogram distribution before the binarization process, and the accuracy of the binarization using the threshold is improved. can do.
If there is a difference between the distributions of the density histograms of R, G, and B, the expansion process converts the values so that the value ranges of the respective colors become equal, so that the color tone is such that a white portion is originally colored. Can be corrected at the same time.

Further, by adopting the color tone correction, when the color tone is deviated, the luminance largely changes in the white portion, and when the color tone is deviated to blue, for example, the luminance largely decreases in the white portion, and the threshold value is lowered. In spite of the fact that extraction is not performed properly, such a problem can be prevented, and as a result, a good binary image suitable for content discrimination can be obtained. Further, the contrast and the color tone can be simultaneously corrected, and the processing speed can be improved.

In the image binarizing method according to the present invention, the decompression processing may be performed such that the density values before decompression are R, G, B, the density values after decompression are R ', G', B ', The minimum value of 0,
When the maximum value is MAX, R ′ = {MAX / (Rmax−Rmin)} × (R−R
min) G ′ = {MAX / (Gmax−Gmin)} × (GG
min) B ′ = {MAX / (Bmax−Bmin)} × (BB
min) according to the linear equation.

According to the above arrangement, the maximum value Rma
x, Gmax, Bmax and minimum values Rmin, Gmi
n and Bmin are values MA at both ends of the data value range, respectively.
X and 0 can be expanded by simple processing, and the density histogram distribution is expanded for each of R, G, and B. Therefore, even an image that cannot be corrected by single-color tone correction is high. A converted image with contrast can be obtained.

Still further, the image binarization method of the present invention comprises:
A pixel considered to be a white pixel is extracted from the image as a white pixel, and average values Wr, Wg, Wb of respective density values of R, G, B of the extracted white pixel are obtained, and the average values Wr, Wg, W
Conversion coefficients Kr, Kg, Kb such that b becomes equal to each other
And multiplying each of the R, G, and B density values of all pixels of the original image by the conversion coefficients Kr, Kg, and Kb, respectively,
Further, white balance correction is performed.

According to the above configuration, for example, taking advantage of the feature that white pixels have extremely low saturation, pixels which are originally considered to be white pixels are extracted as white pixels, and the extracted white pixels R, G, B. Average density values Wr, W obtained by adding the respective density values of B and dividing by the number.
From the g and Wb, assuming that the average luminance of the original image is Wy, Kr = Wy / Wr, Kg = Wy / Wg, and Kb = Wy such that Wr · Kr = Wg · Kg = Wb · Kb = Wy holds. / W
By multiplying each of the R, G, and B density values of all the pixels of the original image by the conversion coefficients Kr, Kg, and Kb obtained in b, the color tone of the entire image that is originally biased toward the color attached to the white portion Can be corrected.

Therefore, for example, in an image in which a highlight area having a large area is present in the original image, the color temperature of the light source or a malfunction of the white balance inside the digital camera may cause the white area to be originally attached as described above. Even in the case of an image in which the color tone of the entire image is biased toward the resulting color and the color correction effect based on the density histogram according to claim 1 is small, the color tone of the entire image is corrected, and then a binarization process is performed after a natural color tone is obtained. This makes it possible to increase the number of target images that can be subjected to color tone correction.

The image binarization method of the present invention compares the maximum values Rmax, Gmax, and Bmax with a predetermined threshold value Yth, and determines that Rmax> Yth, Gmax.
If the three conditions of> Yth and Bmax> Yth are satisfied at the same time, it is determined that the image is difficult to correct the color tone by the expansion of the density histogram according to the first aspect, and the white balance correction according to the third aspect is performed.

According to the above arrangement, it is possible to determine an image for which color correction by expansion of the density histogram according to claim 1 is difficult, because there is a highlight portion having a large area in the original image. White balance correction can be appropriately performed.

Still further, the image binarization method of the present invention comprises:
The entire area of the image is divided into minute areas, and for the minute areas, a threshold value Bth for binarization is calculated from the average value Yave of the luminance values of the pixels whose luminance is equal to or more than the constant value Yp, and the threshold value is equal to or more than the constant value Yp If the number m of pixels and the arbitrarily provided reference value M satisfy m <M, it is determined that most of the minute area is a dark black solid portion, and an arbitrary fixed threshold value is used. If not, it is determined that the minute area is not the black solid portion but the luminance is appropriately distributed within the value range, and the threshold value Bth is used.

According to the above configuration, as a result of binarization in a minute area, luminance information of only a portion where a pixel value is desirably 1 such as a sky in a natural image and a background in a character is obtained. Extraction, and the threshold value Bth for binarization is determined.
The result of the binarization process is, for example, that the dark background is definitely 0,
The bright part is definitely 1 and the dark background is 1 at random
Or, noise such as binarization to 0 does not occur, and is effective as preprocessing for character recognition.

Further, according to the image binarization method of the present invention, when any one minute area is set as a noticed minute area, interpolation is performed between the arbitrary small area and a threshold value of several neighboring areas including the noticed minute area. , A new threshold value Ba of the noted minute area
ve.

According to the above configuration, a digital image obtained by photographing a black-and-white document partially reflects some shade, and the black pixel value of a black character is approximated by an adjacent pixel.
Since the pixel value of the white portion of the background is largely different and the threshold value Bth of the binarization is greatly different between the adjacent minute areas, a good binarized image can be obtained.

[0021]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described.
The following is a description based on FIGS. 1 to 10.

FIG. 1 is a block diagram showing an example of the configuration of an image processing apparatus to which the image binarization method according to the present invention is applied. The image processing device includes a processing device 1 realized by a microcomputer or the like, a display device 2 realized by a cathode ray tube or a liquid crystal display device, an image processing memory 3 realized by a RAM or the like, a hard disk device, and the like. , A command input device 5 implemented by a mouse or a keyboard, a program storage memory 6 implemented by a RAM or the like, and an external interface circuit 7. .

In response to a user operation from the command input device 5, the processing device 1 fetches an image processing program and data of an image to be processed, performs black and white binarization processing as described in detail later, It performs input / output control and arithmetic processing of the entire system such as storage of processing data and creation of display images. The display device 2 includes a display unit, a display memory, an interface circuit for the processing device 1, and the like, and displays a display image created by the processing device 1, such as a command or an image to be processed. Is displayed.

The image processing memory 3 stores image value data of an image to be processed and is used as a work area for image processing calculations. The main storage device 4 stores and stores an image to be processed, an image after processing, an image processing program, and the like. The command input device 5 is operated by a user to input execution of a program, stop, designation of an image to be processed, designation of a storage destination of a processed image, and the like.

The program storage memory 6 is used to read and execute the image processing program stored in the main storage device 4 before executing the program. The external interface circuit 7 captures data of a color still image to be processed obtained by the digital still camera, image scanner, video capture circuit, or the like, and outputs image data after binarization processing to a printing apparatus. Used for output.

The image binarization method according to the present invention generally corrects the brightness and contrast of the image to be processed before the binarization processing, and furthermore, the white part is originally colored. To correct the color shift as if Then, the image is divided into arbitrary small areas, and binarization processing is performed using threshold values obtained for each area.

A color still image file to be processed is read out from an original image file which is fetched via the external interface circuit 7 and stored in the main storage device 4, and is designated by a JPEG or the like. The data of the image to be processed in the format is expanded into, for example, an 8-bit bitmap image for each of R, G, and B colors for each pixel by expansion processing or the like, and stored in the image processing memory 3. .

In order to further predict what kind of subject the image is in the binarization process and what method to perform image correction, FIG. 2 shows. In this manner, the entire screen is divided into small areas (in FIG. 2, the vertical and horizontal directions of the developed bitmap are each divided into five, that is, the entire screen is divided into 25 small areas). That is, the brightness of each small area, R, G, B
By calculating an image attribute such as an average value, an IQ value, or a color difference signal such as BY, RY, or the like, it is possible to determine a method of estimating the subject or correcting an image.

For example, as shown in FIG. 2, the entire screen is divided into the aforementioned 25 small areas, and further divided into 9 central areas and 16 peripheral areas. The nine central areas are area numbers 6, 7, 8, 11, 12, 13, 16,
It is a rectangular area constituted by 17, 18 and the surrounding 16 areas are areas excluding the central 9 area from the entire screen.

First, the average luminance Yc of the nine central areas is determined. The average luminance Yc of the central nine areas is obtained, for example, by calculating the R, G, and B components of the pixels one pixel at a time according to the following equation in accordance with the human luminosity characteristics.
It is obtained by integrating all the pixels in the area and dividing by the number of pixels.

Y = 0.3R + 0.59G + 0.11B (1) Next, similarly, the average luminance Ya of the surrounding 16 areas is obtained. Subsequently, the average luminance Yc and Ya are compared, and Y
If c is smaller than Ya, it can be determined that the subject at the center of the screen is in backlight, and the following correction is preferentially performed on the gradation in the central part darker than the surroundings. On the other hand, if Yc is greater than or equal to Ya, it can be determined that the image such as a landscape or a group photo of a person should be emphasized on the entire screen, and correction is performed on all the pixels in the 25 areas of the entire screen. To do. In this manner, the method of estimating the subject and correcting the image can be determined.

FIGS. 3 and 4 are graphs for explaining the density correction and the color tone correction. In the correction, first, as shown in FIG. 3, the density histograms of the R, G, and B components, and the maximum values Rmax, Gmax, Bmax and the minimum values Rmin, Gmin, Bmin of the density histogram distributions of R, G, and B, respectively. Is required.
That is, in FIG. 3, reference numeral β1 is an example of a density histogram of an arbitrary color among R, G, and B obtained from the original image. The minimum value of the distribution range is represented by min, and the maximum value is represented by max. Is shown.

Next, the correction is performed by setting the minimum value min and the maximum value max to 8 bits, that is, 2 to 0 to 255.
This is realized by extending to a value range of 56 gradations and making a distribution as indicated by reference numeral β2. That is, by the expansion processing, the pixels of the color are distributed from the minimum value to the maximum value of the value range as an image, and the brightness is corrected and the distribution is concentrated only at a specific density value. The lack of contrast can be corrected.

Also, as shown in FIG. 4A, when the distribution of the density histograms of R, G, and B in the original image is different (FIG. 4A illustrates a reddish image). )
Is converted by the expansion of the density histogram so that the value range of each color becomes equal, and as shown in FIG. 4B, the color tone is also corrected at the same time. Assuming that the density values before expansion are R, G, B and the density values after expansion are R ', G', B ', the equation used for the above expansion is obtained by a linear equation as shown in the following equation, for example. The density values R and R ′ are also shown in FIG. R ′ = {255 / (Rmax−Rmin)} × (R−Rmin) (2) G ′ = {255 / (Gmax−Gmin)} × (G−Gmin) (3) B ′ = {255 / (Bmax−Bmin)｝ × (B−Bmin) (4) In this manner, the density correction and the color tone correction can be performed simultaneously, and the processing speed can be improved.

However, in the case of an image having a small color correction effect based on the density histogram, the white balance correction is further performed. First, in order to determine whether or not to perform correction, the maximum values Rmax, Gmax, and Bmax of the R, G, and B density histograms and a predetermined threshold Y
Th is compared with Rmax> Yth, Gmax>
If the three conditions of Yth and Bmax> Yth are satisfied at the same time, it is determined that the image is an image for which color correction by expansion of the density histogram is difficult, and white balance correction is performed.
That is, the threshold value Yth is set to R, G, B as described above.
In the range of 8 bits for each color, that is, in the range of 0 to 255, for example, 245 is selected. For an image in which the above three conditions are satisfied at the same time, the maximum value of the density histogram of all R, G, and B is 255. An image with similar values, such as a large area highlight in the original image,
This is because the color correction effect by the density histogram is reduced.

Next, in the white balance correction, a pixel which is originally white in an image is extracted, and R and R of the extracted white pixel are extracted.
Average density values Wr, Wg, Wb obtained by adding the respective density values of G and B and dividing by the number of white pixels.
, Conversion coefficients Kr, Kg, and Kb for whitening white pixels for each of R, G, and B colors are obtained, and R,
This is performed by multiplying each of the density values of G and B by the conversion coefficients Kr, Kg, and Kb, respectively. The conversion coefficients Kr, K
Assuming that the average luminance of the original image is Wy, the results obtained by multiplying the average density values Wr, Wg, and Wb are equal to each other, and that the luminance is not changed. Wr · Kr = Wg Kg = Wb · Kb = Wy (5) Kr = Wy / Wr (6) Kg = Wy / Wg (7) Kb = Wy / Wb (8)

The extraction of the white pixels takes advantage of the fact that the saturation is extremely low.
10, are represented by two values of I and Q. As shown in Expression 11, both absolute values of I and Q are both defined values Wth.
In the case of being smaller, it can be performed as a pixel having extremely small saturation, that is, a white pixel. I = 0.599R-0.277G-0.322B (9) Q = 0.213R-0.525G-0.312B (10) | I | <Wth AND | Q | <Wth (11) As a result, since the R value, B value, and G value of a pixel that is originally white are corrected to equal values, that is, white (accurately, gray), the color tone of the entire image can be corrected and a natural color tone can be obtained. Thus, it is possible to increase the number of target images for which the color tone can be corrected.
Further, for example, when the color tone of the image is biased toward blue, it is possible to eliminate the problem that the luminance value of the white pixel is reduced from Expression 1 and the threshold value extraction accuracy described later is reduced.

When the contrast and the color tone of the original image are corrected as described above, each pixel is binarized by a threshold using the corrected data. First, the screen is divided into minute areas as preparation for obtaining a threshold for binarization. Next, a threshold value for binarization is determined for each of the divided micro areas based on the average of high luminance pixels. Subsequently, all pixels in each minute area are binarized by the threshold value.

The division into the minute areas is performed, for example, by using the 1024 × size shown in FIG.
Assuming that an image of 768 pixels is an original image, the image is divided by 64 × 48. In this case, each of the minute areas a1, a2,... Shown in FIG. 6B is a square area composed of 16 × 16 pixels. The number of divisions is 64 × 4 in this example.
Although it is set to 8, it may be changed appropriately according to the image size.

The determination of the threshold value for binarization is based on the above average of high luminance pixels. First, the average luminance Yave of the image is obtained, and the luminance of the pixels satisfying the following condition is obtained for all the pixels in the small area of interest. Yp is integrated, and the total value Yw is obtained. Yp ≧ k × Yave (12) Next, the number m of pixels that satisfies the condition of the expression 12 is obtained, and the number m of pixels and a reference value M of an arbitrary number, for example, 10
And m <10, that is, within 16 × 16 = 256 pixels, a predetermined coefficient k times (k>) of the average luminance Yave
1) When the number of pixels with the luminance Yp of the above is less than 10, it is determined that most of the minute area is a dark black solid portion, and it is desirable that the binarized result is set to 0, and that area is determined. Is the median value of the range (128 when the 8-bit resolution is set for each of R, G, and B colors). On the other hand, when m ≧ 10, the minute area is not a solid black portion,
It is determined that the luminance is appropriately distributed within the value range, and the average luminance Yave is set as a threshold Bth.

Thus, for example, in the case of a digital image obtained by scanning a newspaper photograph or a headline, such a solid black portion has a lower pixel value than the median of the range, and as described above, By binarizing with a value, it can be converted to a value 0. On the other hand, if the average luminance Yave is set as a threshold value in this case, the average luminance Yave becomes a low value, the black portion is randomly converted to 1 or 0, and the black portion of the original image is not accurately converted to 0 and cannot be converted. I will.

Then, for all the micro areas in the original image, the threshold value obtained as described above is used as a provisional threshold value, and interpolation is performed with the provisional threshold value of a nearby micro area including the attention micro area. , A new threshold value Bave of the minute area of interest is calculated, and the new threshold value Bave is multiplied by a predetermined coefficient n to obtain an actual threshold value Bt, and a binarization process is performed.

Therefore, for example, in a digital image obtained by photographing a black and white document, some shade is partially reflected, and the pixel value of a black character is close to that of an adjacent pixel, but the white portion of the background Even if there is a partial shading change in the original image in which the pixel value of is significantly different, and the provisional threshold value greatly changes between the minute area of interest and the adjacent minute area, Can be eased.

However, there is an extreme change in lightness and darkness between the adjacent minute areas. For example, the noticeable minute area c3 is located in the bright part, and any minute area c2, c4, b3 or d3 is located in the dark part. In such a case, in order to prevent the influence of the area, if there is a large difference between the two thresholds as shown in the following equation, the threshold of the area is not used for the interpolation calculation. Threshold value of attention area−threshold value of arbitrary adjacent area> reference value (13) In this way, the binarization threshold value and the binarization process using the threshold value can be performed.

FIG. 7 is a flowchart for explaining the image binarization method according to the embodiment of the present invention as described above. At step s1, a file of an image to be processed is read in a pre-designated format such as the JPEG from the original image file taken in through the external interface circuit 7 and stored in the main storage device 4. , Is developed as a bit map on the image processing memory 3. In step s2, for the purpose of estimating the subject and determining the correction method, for example, as shown in FIG.
The entire screen is divided into 25 small areas.

In step s3, the luminance Y for each pixel in the central nine areas is obtained according to the above equation 1, and the luminance Y is integrated for all the pixels in the central nine areas and divided by the number of pixels to obtain the average luminance. Yc is required. In step s4, the average luminance Ya of the surrounding 16 areas is similarly obtained.

In step s5, the average luminances Yc and Yc
is compared with Y. If Yc is smaller than Ya, it is determined that the subject at the center of the screen is in backlight, and step s6 is performed.
To preferentially correct the gradation at the center, which is darker than the surroundings,
The calculation of the density histogram and the extraction of white pixels are set for the pixels in the central nine areas. Conversely, when Yc is greater than or equal to Ya, it is determined that the image such as a landscape or group photo of a person should be emphasized. Then, in step s7, all 25 pixels in the entire screen are set as targets. In Step s8, Step s6 or s7
, The density histograms of the R, G, and B components and the maximum values Rmax, Gm
ax, Bmax and minimum values Rmin, Gmin, Bm
in is required. In steps s9, s10, and s11, the density histograms of R, G, and B obtained in step s8 are expanded according to the expressions 2, 3, and 4,
Contrast correction and color tone correction are performed.

In step s12, the aforementioned R, G, B
Values Rmax, Gmax, B of each density histogram
The magnitude of max and the threshold Yth is determined, and Rmax>
If the three conditions of Yth, Gmax> Yth, and Bmax> Yth are simultaneously satisfied, it is determined that the image is an image for which color correction by the expansion of the density histogram in steps s9 to s11 is difficult, and the second color in steps s13 to s15 described below. Add a correction step to the process.

In step s13, pixels which are originally white in the image are extracted from the expressions 9 to 11, and the R, G, and B density values of the extracted white pixels are added, respectively. By dividing by the number of white pixels, the average density values Wr, Wg, Wb of the R, G, B colors of the white pixels are obtained. In step s14, the density average value W
From r, Wg, and Wb, R,
A conversion coefficient Kr for whitening a white pixel for each of G and B colors,
Kg and Kb are determined. In step s15, the R, G, and B density values of all the pixels of the original image are multiplied by the conversion coefficients Kr, Kg, and Kb of Expressions 6, 7, and 8, respectively. Thus, the R value, the G value, and the B value of the pixel that is originally white are corrected to the same value, so that the color tone of the entire image is corrected and a natural color tone is obtained.

After the above steps s2 to s15, the contrast and the tone of the original image are corrected in order to obtain a good binarized image. In step s16, the preparation for obtaining the binarization threshold is performed. The image is 64x
The image is divided into minute areas of 48 images. In step s17, a threshold for binarization is determined for each micro area divided in s16 based on the average of high-luminance pixels, and all pixels in each micro area are binarized using the threshold. In step s18, the binarized image is stored in a specified image format, and a binarized image file is created.

FIG. 8 is a flowchart for explaining the details of the binarization process in step s17.
In step s21, the average luminance Yave of the image is obtained. In step s22, the luminance Yp of the pixels satisfying the condition of the expression 12 is integrated for all the pixels in the small area of interest, and the total value Yw is obtained. In step s23, the number m of pixels that satisfies the condition of Expression 12 is obtained.

In step s24, the reference number M of the number of pixels m
When m <M, it is determined that most of the minute area is a dark black solid portion and it is desirable to set the binarized result to 0, and in step s25, that area is determined. Is set as the median value of the value range, and if m ≧ 10, it is determined that the small area is not a solid black portion but the luminance is appropriately distributed within the value range. The luminance Yave is set to a threshold Bth. The process moves from step s25 and s26 to step s27. In the above description, the processing of steps s22 to s26 is for one minute area in the image.
It is determined whether or not the process has been performed for all the small areas in the image. If not, the process returns to step s22.
If so, since the provisional threshold values have been determined for all the small areas in the image, the process proceeds to step s28.

In step s28, there is a partial change in shading in the original image, and the provisional binarization threshold Bth obtained in steps s22 to s28 greatly changes between the minute area of interest and the minute area adjacent thereto. In order to alleviate this, a temporary threshold value B of a 3 × 3 minute area near the focused minute area is excluded except for the area corresponding to Expression 13.
Interpolation is performed between th and a new temporary threshold value Bave is obtained.

In steps s29 to s31,
The provisional threshold value Bave after interpolation obtained in step s28 is multiplied by a coefficient n to obtain a threshold value Bt for actual binarization.
Using the threshold value Bt, all pixels in the minute area are binarized. When the binarization processing is completed for a certain small area, for example, all the pixels in c3, the next minute area of interest becomes d3, and the adjacent minute area to be interpolated is also shifted by one section. Thus, in step s32, the process returns to step s28 until the binarization process is completed for all the small areas, and when completed, returns to step s18.

FIGS. 9 and 10 show the results of the binarization process. 9A and 10A are the original images, FIG. 9A is a business card image, and FIG. 10A is a landscape image. In FIG. 9A, the background has a gradation, and in FIG. 10A, the sky has a gradation. FIG.
(B) and FIG. 10 (b) both show 2
FIG. 9 (c) and FIG.
0 (c) indicates the binarization processing result according to the present invention.

In FIG. 9B, noise is present around the background portion and around the character, which is inappropriate as an input image for character recognition or the like. Further, in FIG. 10B, the empty gradation is expressed as a noise dot pattern as it is. On the other hand, in FIG. 9C, it is understood that there is little noise around the background and the characters. FIG.
At 0 (c), the sky portion is distributed to 1 over almost the entire region, and it is understood that the image is a binary image with high contrast.

In this manner, the readability of the low-contrast portion of the image to be binarized can be improved, and the noise can be reduced.

[0058]

As described above, according to the image binarization method of the present invention, the analog or multi-tone digital image to be processed is level-discriminated with a predetermined threshold value.
In an image binarization method in which a binarization process is performed,
Before the binarization processing, the density histogram of each component of R, G, and B of the image to be processed and the maximum values Rmax, Gmax, and Bmax of the density histogram distributions of R, G, and B, respectively.
x and minimum values Rmin, Gmin, Bmin are obtained,
The brightness correction, the lack of contrast correction, and the color tone correction are performed by expanding the density histogram distribution so that the maximum values Rmax, Gmax, Bmax and the minimum values Rmin, Gmin, Bmin become values at both ends of the data value range. .

Therefore, the accuracy of the binarization process can be improved by correcting the lack of brightness and contrast. In addition, there is a difference in the distribution of the density histograms of R, G, and B, and it is possible to simultaneously correct a color tone shift such that a white portion is originally colored. Furthermore, by adopting the color tone correction, it is possible to prevent a problem that when the color tone is biased, the luminance largely changes in the white portion, and the threshold value is not properly extracted, and the content determination is performed. A suitable good binary image can be obtained. Further, the contrast and the color tone can be simultaneously corrected, and the processing speed can be improved.

Further, in the image binarization method of the present invention, as described above, the decompression processing is performed by a linear expression for each of R, G, and B colors.

Therefore, the image can be expanded by simple processing, and the density histogram distribution is expanded for each of R, G, and B. Therefore, even for an image that cannot be corrected by a single color tone correction, a high contrast image can be obtained. A converted image can be obtained.

Further, the image binarization method of the present invention
As described above, a pixel considered to be a white pixel is extracted from the image as a white pixel, and the average values Wr, Wg, Wb of the R, G, B density values of the extracted white pixel are obtained, and the average value is obtained. Conversion coefficient K such that Wr, Wg, Wb are mutually equal
r, Kg, and Kb are obtained, and the white balance correction is further performed by multiplying the R, G, and B density values of all the pixels of the original image by the conversion coefficients Kr, Kg, and Kb, respectively.

Therefore, the color tone of the entire image is biased toward the color originally attached to the white portion, and even in the case of an image having a small color correction effect based on the density histogram according to the first aspect, the color tone of the entire image is corrected. The binarization process can be performed after the color tone is adjusted, and the number of target images for which the color tone can be corrected can be increased.

As described above, the image binarization method of the present invention uses the density histogram expansion according to claim 1 when all of the maximum values Rmax, Gmax, and Bmax are larger than a predetermined threshold Yth. It is determined that the image is an image for which color correction is difficult, and the white balance correction according to claim 3 is performed.

Therefore, it is possible to determine an image in which there is a highlight portion having a large area in the original image and it is difficult to perform color tone correction by expanding the density histogram according to the first aspect. Can be done properly.

Further, the image binarizing method of the present invention
As described above, the entire area of the image is divided into small areas,
For the small area, the binarization threshold Bth is calculated from the average value Yave of the luminance values of the pixels whose luminance is equal to or more than the constant value Yp.
When the number m of pixels that is equal to or more than the predetermined value Yp is less than an arbitrary reference value M, the small area is determined to be a dark black solid portion, and an arbitrary fixed threshold value is used.
If the value is equal to or larger than the reference value M, it is determined that the minute area is not the solid black portion but the luminance is appropriately distributed within the value range, and the threshold value Bth is used.

Therefore, luminance information of only pixels having a pixel value equal to or more than the predetermined value Yp, such as a sky in the case of a natural image and a background in the case of a character, is preferably extracted, and a threshold value for binarization is obtained. Since Bth is determined, noise such that a dark background is randomly binarized to 1 or 0 does not occur, and is effective as preprocessing for character recognition.

Further, according to the image binarization method of the present invention, as described above, when any one minute area is set as the noticeable minute area, the difference between the arbitrary minute area and the threshold value of several neighboring areas including the noticeable minute area is determined. By taking the interpolation, a new threshold value Bave of the noted minute area is calculated.

Therefore, a large difference in the threshold value Bth for binarization between adjacent minute areas can be mitigated.
It is possible to obtain a binarized image.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an example of a configuration of an image processing apparatus to which an image binarization method according to the present invention is applied.

FIG. 2 is a diagram for explaining an area of interest in an image at the time of binarization processing.

FIG. 3 is a graph for explaining a density histogram expansion process for realizing density correction.

FIG. 4 is a density histogram for explaining color tone correction.

FIG. 5 is a graph for explaining a conversion formula used for the density histogram expansion.

FIG. 6 is a diagram for describing division of an image to be subjected to a binarization process into minute areas.

FIG. 7 is a flowchart illustrating an image binarization method according to an embodiment of the present invention.

FIG. 8 is a flowchart illustrating details of a binarization process in FIG. 7;

FIG. 9 is a diagram illustrating an example of a binarization processing result;

FIG. 10 is a diagram illustrating another example of the binarization processing result.

FIG. 11 is a density histogram for explaining how to obtain a conventional binarization threshold.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 processing device 2 display device 3 image processing memory 4 main storage device 5 command input device 6 program storage memory 7 external interface circuit

Claims (6)

[Claims] An image binarization method for performing a binarization process by level-discriminating a digital image of an analog or a plurality of gradations to be processed with a predetermined threshold value, wherein: The density histogram of each of the R, G, and B components of the gradation digital image,
Maximum value Rma of density histogram distribution of each of G and B
x, Gmax, Bmax and minimum values Rmin, Gmi
n and Bmin are obtained, and the density histogram distribution is expanded so that the maximum values Rmax, Gmax, and Bmax and the minimum values Rmin, Gmin, and Bmin become values at both ends of the data value range. An image binarization method comprising performing correction and color tone correction. 2. The method according to claim 1, wherein the density value before expansion is R,
G, B, the density values after expansion are R ′, G ′, B ′, the minimum value of the data range is 0, and the maximum value is MAX, R ′ = {MAX / (Rmax−Rmin)} × (R -R
min) G ′ = {MAX / (Gmax−Gmin)} × (GG
min) B ′ = {MAX / (Bmax−Bmin)} × (BB
2. The image binarization method according to claim 1, wherein the image is binarized by a linear expression. 3. Extracting a pixel considered to be a white pixel from an image as a white pixel, and averaging Wr, Rr, G, and B density values of the extracted white pixel.
Wg and Wb are calculated, and conversion coefficients Kr, Kg and Kb are determined so that their average values Wr, Wg and Wb are equal to each other. The conversion coefficients are converted to the density values of R, G and B of all pixels of the original image. 3. A white balance correction is further performed by multiplying Kr, Kg, and Kb, respectively.
The image binarization method described in the above. 4. The maximum value Rmax, Gmax, Bmax.
Is compared with a predetermined threshold value Yth, and Rmax
If the three conditions of> Yth, Gmax> Yth, Bmax> Yth are satisfied at the same time, it is determined that the image is difficult to perform color correction by the density histogram expansion according to claim 1,
4. The image binarization method according to claim 3, wherein the white balance correction according to claim 3 is performed. 5. An entire area of an image is divided into minute areas, and a threshold value Bth for binarization of the minute areas from an average luminance value Yave of pixels having a luminance equal to or more than a predetermined value Yp.
If the number m of pixels that is equal to or more than the predetermined value Yp and the arbitrarily provided number of reference values M satisfy m <M, an arbitrary fixed threshold is used. Is the threshold Bth
The image binarization method according to any one of claims 1 to 4, wherein 6. A new threshold value Bave for an attention minute area, when an arbitrary one minute area is set as an attention minute area, interpolation is performed between the arbitrary minute area and a threshold value of several neighboring areas including the attention minute area. The image binarization method according to claim 1, wherein the image is calculated as: