JPH01184583A - System for deciding binarization threshold value - Google Patents

Abstract

PURPOSE:To decide the optimum binarization threshold value at high speed by deciding the binarizing threshold value based on the histogram of the number of boundary image elements in all areas of an image and that of the number of boundary image elements in an edge image area. CONSTITUTION:An original image 10 is inputted to an edge detection part 11, a sequence filter 12, and an accumulative calculation part 1. The output of the edge detection part 11 and the sequence filter 12 are inputted to the accumulative calculation part 1. The accumulative calculation part 1 calculates the accumulative histogram of the light-and-shade value of the image in the edge image area. Also, a boundary area image element number calculation part 2 calculates the histogram of the number of boundary image elements in all areas of the image. An optimum threshold value detection part 19 calculates the degree of coincidence of the boundary image element of the binary image at the time of performing binarization with an edge image group detected at the edge detection part 11 setting the light-and-shade value as the threshold value at every light-and-shade value by using those histograms, and decides the light-and-shade value with the highest degree of coincidence as the optimum threshold value.

Description

[Detailed Description of the Invention] [Summary] Regarding the binarization threshold determination method, there is provided a binarization threshold determination method that can quickly determine an optimal threshold without setting a temporary threshold. The purpose of the present invention is to provide an edge detection unit that inputs an original image and detects candidate points for the contour of an object in the image, and an edge detection unit that inputs the original image and detects candidate points for the outline of an object in the image, and an order filter that replaces the gray value of a pixel with a gray value of a predetermined rank when the gray value of a pixel in a window consisting of neighboring pixels including the pixel is arranged in order of size; For two predetermined images among the images output from the ordered filter, an edge image occupied by the cumulative histogram of the grayscale values of the two images in the entire area of the image and the edge pixel group detected by the edge detection section. a cumulative calculation unit that calculates a cumulative histogram of gray values of the two images in the area; and a total area of the image obtained by the difference for each gray value of the two cumulative histograms in the entire area of the image calculated by the cumulative calculation unit. a boundary area pixel number calculation unit that calculates a boundary pixel number histogram in the edge image area obtained by the difference for each gray value between the two cumulative histograms in the edge image area calculated by the cumulative calculation unit; Then, using the histogram of the number of boundary pixels in the entire area of the image and the histogram of the number of boundary pixels in the edge image area calculated by the boundary area pixel number calculation unit, binarize each gray value using the gray value as a threshold value. an optimal threshold determination unit that calculates the degree of coincidence between the boundary pixels of the binary image and the edge pixel group detected by the edge detection unit when The output of the optimum threshold value determining section is configured to be a threshold value for binarization.

[Industrial application field]

TECHNICAL FIELD The present invention relates to image processing for processing a specific figure in a grayscale image, and particularly to a threshold value determination method for binarization.

A digital image created by AD converting an analog image input from an imaging camera etc. is a two-dimensional still image.
It is expressed as a grayscale image in which pixels are arranged on a grid. The gray value of each pixel corresponds to any level from 0 to 25, for example, and is expressed in 8 bits in binary representation. Digital image processing, which performs high-speed recognition of specific objects and figures in a given grayscale image, is used in factory automation, such as when a visual robot recognizes and picks up an object carried on a conveyor belt. Becomes extremely important for technology. When recognizing objects in grayscale images, binary figures obtained by binarizing the original image using threshold values are used, but in this case, it is required to quickly determine the optimal threshold value. Ru.

[Conventional technology]

The conventional binarization threshold determination method uses an edge image in which candidates for the outline of the object, that is, edges are extracted from the original image;
The degree of coincidence of the boundary image extracted from the boundary of the binary figure obtained by binarizing the original image using the threshold values is calculated for all threshold values, and the threshold value at which the -political change is maximum is adopted. was.

FIG. 2 shows the processing flow of the conventional method. In the same figure,
A gray value gradient calculation unit 20 calculates the magnitude of the differential at each pixel, an edge detection unit 21 detects pixels for which the magnitude of the differential is greater than or equal to a predetermined threshold value, and a threshold temporary setting unit 22 For example, the threshold value processing unit 23 is one that sets one of 256 integer values from 0 to 255 as a temporary threshold value, the threshold processing unit 23 is one that binarizes the input image according to the temporarily set threshold value, and the boundary inspection The output unit 24 detects a boundary of a binarized binary image, that is, a boundary image consisting of pixels in which the pixel value is 1 and the pixel value of at least one pixel in the 8 neighboring pixels is 0. - The growth calculation unit 25 calculates the degree of coincidence between the edge image from the edge detection unit 21 and the boundary image from the boundary point detection unit 24. - The political change table 26 calculates the degree of coincidence between the edge image from the edge detection unit 21 and the boundary image from the boundary point detection unit 24. The optimum threshold determining unit 27 stores the matching degree of the boundary image for all threshold values from 0 to 255, and selects the threshold value that maximizes the matching degree from the matching degree stored in the matching degree table 26. It calculates the value and outputs that value as the optimal threshold value. In addition, in the -growth calculation unit 25, -political change is calculated by the following equation (1) or (2).

[Problem to be solved by the invention]

Therefore, in this method, for example, all temporary threshold values from 0 to 255 are subjected to binarization processing in the threshold processing unit 23 and the boundary point detection unit 24, and -value calculation unit 25
Since the matching degree calculation process is required, a problem arises in that a huge amount of processing time is required.

An object of the present invention is to provide a binarization threshold determination method that can quickly determine an optimal threshold without setting a temporary threshold.

[Means to solve the problem]

FIG. 1 is a block diagram of the present invention, in which an image processing system for processing a grayscale image includes an edge detection section 11 that inputs an original image 10 and detects candidate points for the outline of an object in the image; An ordering filter that replaces the gray value of each pixel with a gray value of a predetermined rank when the gray values of pixels in a window consisting of neighboring pixels including that pixel are arranged in order of size. 1
2, for two predetermined images among the original image 10 and the image output from the order filter 12, a cumulative histogram of the grayscale values of the two images in the entire area of the image is detected by the edge detection unit 11. edge pixel group 1
a cumulative calculation unit 1 that calculates a cumulative histogram of gray values of the two images in an edge image area occupied by a histogram of the number of boundary pixels in the entire area of the image obtained by the difference, and a histogram of the number of boundary pixels in the edge image region obtained by the difference for each gray value of the two cumulative histograms in the edge image region calculated by the cumulative calculation unit 1; Boundary area pixel number calculation unit 2 that calculates
Then, using the boundary pixel number histogram in the entire area of the image and the boundary pixel number histogram in the edge image area calculated by the boundary area pixel number calculation unit 2, a binary value is calculated for each gray value using the gray value as a threshold value. an optimal threshold determining unit 19 that calculates the degree of coincidence between the boundary pixels of the binary image when converted into a binary image and the edge pixel group detected by the edge detection unit 11 and outputs a gray value having the maximum degree;
This shows a threshold value determination method for binarization, characterized in that the output of the optimum threshold value determination unit is used as the threshold value for binarization.

[For production]

In the present invention, for example, if a minimum value filter is used as the order filter 12, the number of boundary pixels of the original image or the image of a specific area can be calculated using a histogram (each gray value is binarized with that gray value as a dark value). A histogram that correlates the number of boundary pixels of a binary image obtained by using a minimum value filter), a cumulative histogram of gray values of that image, and a cumulative histogram of gray values of an image created by processing that image with a minimum value filter. There is.

〔Example〕

Next, embodiments of the present invention will be described with reference to the drawings.

First, the gray value f (xi) at the discrete coordinates Xi of the pixel
While defining some words using the one-dimensional curve of
First, some properties of image processing will be described. Since the image is two-dimensional, the grayscale value at the position (X+, yi) should be expressed as f(xt, yi), but here, for simplicity, it will be considered one-dimensional.

Figure 1 (C) shows 2 for the gradation plane vAf(x) of the original figure.
It is a conceptual diagram of value conversion processing. The gradation value f(Xt) at the discrete coordinate x has any value between O and 255, but depending on the threshold value T, when f(x,)≧Th, B1)=1 f(xt ) <'l'h when B (xt)
=0, then in the figure, for x, <X, B (xt) =Ox1 ≦X≦
B (xt) = 1x2 for x2 B (
xl)xQ, and the binarized sequence (B (Xt)) of the original figure is O
will be shown.

The gray value at the discrete coordinate Xi is f(Xt) and its neighborhood X1
-1 and X,. The shading values at 1 are '(X
s-1), f (Xt++ ), when the original figure is passed through a minimum value filter, the gray value f(x
i) is m i x (f (Xt-+), f(xυ, f (x
t, +)). Here, min is a symbol that means to select the smallest one. In addition, when the original figure is passed through a maximum value filter, the gradation value f(xL
) is max (f(x,,), f(x,), f(xz
−+)). Here, max is a symbol that means to select the largest one. Now, if the discrete curve f (x,) of the original figure in FIG. 1(C) is passed through a minimum value filter, the figure after passing through the minimum value filter is represented by a dotted discrete curve g (Xt ). This curve g (Xi
) is binarized using the same threshold value T1, the binarized series (B-(xi)) of the figure after passing through the minimum value filter becomes the one shown in (2).

Binarized series of the original figure shown in ■ (B (Xi))
, the bits (B (xi−+), B
(xt), B (xt, )) is 0, B (Xt) = 0 is called "contraction j," and when at least one of them is 1, B (Xi) = 1 is called "expansion." Further, a filter that performs contraction processing is referred to as a logical contraction filter, and a filter that performs expansion processing is referred to as an expansion logical filter. Figure 1 (
The binarized series (B (xt
)) through a shrinkage filter, the binary series C(B(xt)) shown in ■ is obtained, and this series is clearly the binarized series (B□ (Xi )
) is equivalent to

That is, (B, (xi)) = C(Bt (xt)
).

Also, in the binarized series of the original figure (BOl)), b (Xt) = B (Xi-+)
Binarized sequence (b
(xt)) is a boundary binary figure. Here, 0 is exclusive OR.

In accordance with this, in Fig. 1(C), the binarized series of ■ (
B(xt)), a series of boundary binary figures (b(Xi)) shown in ■ is obtained. Bit “1” in binary sequence
The number □ is called "area". A series of boundary binary figures (b (
The number of bits "1" in Xt ) ), that is, the area, is referred to as the "number of boundary pixels." In FIG. 1(C), the series (b (X
The number of boundary pixels for i)) is 2, and it is clear that the number of boundary pixels = the binarized series of the original figure (B (Xt))
Binarized series of figures after passing through the face-minimum filter (B
, (Xi)) or, ・Number of boundary pixels = Area of (B(x,)) of the binarized series of the original figure - Binary series C(B (Xi)) contracted after the binarization of the original figure
') ) holds true.

Therefore, the present invention binarizes all temporary threshold values, -
Unlike the conventional method that performs the process of calculating the image quality, this threshold determination method for binarization is fast and can obtain the same results as the conventional method. be done.

ratio! After upper minimum filter processing, the figure obtained by binarization (B,
(Xt)) and the figure C(B(Xi)) obtained by applying a shrinkage logic filter after binarization are the same as -.

The number of boundary pixels of a 5I construction figure can be calculated from the equation: number of boundary pixels = area of original figure - area of figure after contraction.

Further, in the present invention, property 3.4.5, which will be described later, is utilized, but before describing these properties, a histogram of gray values and a cumulative histogram of gray values will be explained.

FIG. 1(d) is a histogram of the gradation values f(Xs), and shows the frequency (number of pixels) of each gradation value. FIG. 1(e) is a cumulative histogram of the gray value f(Xt), and shows the frequency of gray values exceeding the threshold value T7. For example, in the histogram of FIG. 1(d), the threshold value T
The frequency of gray values exceeding h is indicated by the area S, and is shown in Figure 1 (e
), the value at the Th0 point is S. Considering on Figure 1 (C), this S is the interval (Xt
, 'Kg), it is equal to the area (number of "1") of the binarized series (B (Xi) ) of the original figure of ■.

Therefore, the following property holds.

Although the area of the figure obtained by binarizing the gray scale image changes depending on the threshold value, the relationship between the threshold value and the area can be expressed by the cumulative histogram of the original image, which is equivalent to the area. .

The number of boundary pixels of a figure obtained by binarizing a one-step gray image varies depending on the threshold value, but can be calculated as follows: Number of boundary pixels = cumulative histogram of original image - cumulative histogram after minimum filter processing.

In other words, to summarize the above, number of boundary pixels = area of original figure - area of figure after contraction Area of original figure = cumulative histogram of original image Area of figure after contraction - cumulative histogram boundary of image after minimum value filter processing Number of pixels = cumulative histogram of original image - cumulative histogram after minimum value filter processing.

Furthermore, the following property also holds true.

According to property 4, the number of boundary pixels within a preset image area can be calculated by calculating a cumulative histogram within the area, that is, by masking areas other than the area.

In addition, the definition of the degree of coincidence between the edge image and the boundary image is (11
As shown in Part 2, it is given by .

From property 4 and property 5, the numerator in the above equation is equivalent to the number of boundary pixels of the original image in the edge image area, so the total number of pixels that are edges and boundary points = cumulative histogram in the edge image area of the original image - to the original image On the other hand, all the threshold values can be calculated simultaneously using the cumulative histogram in the edge image area for the image after the minimum value filter processing.

Also, referring to the diagram showing the set relationship in Figure 1(f), the denominator of equation (1) is the total number of pixels that are edges or boundary points (AUB) = number of pixels that are edges. Total number (
A) Number of ten boundary points (B) - Total number of pixels that are edges and boundary points (AnB). Here U and n are
These symbols represent the sum and intersection of sets, respectively.

Note that when formula (2) is followed, the denominator is simply the number of boundary pixels, so it can be similarly calculated from: Number of boundary pixels = Cumulative histogram of original image - Cumulative histogram after minimum value filter processing.

FIG. 1(b) shows a configuration diagram of this embodiment. In this figure, the same symbols are used for the same parts as in FIG. 1(a). Image data representing the gray scale value of each pixel of the original image is input by lO. The gray value gradient calculation unit 110 calculates each pixel (Xt
, Fl ), each pixel (xi , y!
) is calculated.

For example, the magnitude ΔH of the gradient of the gray value can be calculated more easily. Here, for example, the coordinates (X =, yz)
If the gradation value at is f (Xi ,'/= ), then
The difference in direction Δr (Xi, yt ) is Δf(xl.

)' i ) =f (Xt*+ ,
)' i ) −f (Xt, )'
i ), and the sum Σ represents the sum of Δf for a 3×3 window.

The comparator 111 detects pixels whose density gradient magnitude ΔH is larger than a predetermined threshold value TΔH. Therefore, the edge detection section 11 can detect an edge image. Minimum filter 1
2 scans a 3×3 window, calculates the minimum gray value of the 9 pixels in the window for each pixel, and calculates each pixel (xt
, yt ) is replaced with the minimum gradation value g (Xi , V5) within the window (corresponding to g(Xi) in FIG. 1 (cl) in one dimension).

13, 14, 15, and 16 correspond to the cumulative calculation unit in FIG.
A cumulative histogram is calculated by scanning the original image input from 0. FIG. 1(g) shows the cumulative histogram calculation unit 1.
FIG.

Specifically, the histogram memory 130, which has addresses 1 to 256 corresponding to the gray level, is first cleared to all zeros, and the input image 10 is scanned to obtain each pixel (xi
, yi) ``1 is added to the contents of the address corresponding to (X4, yl).

Histogram memory 1 when all images have been scanned
The contents of 30 represent a histogram of gray values. In other words, the contents of address N represent the total number of pixels whose gray value is level N. That is, the gradation value histogram calculation control unit 131 accesses the histogram memory 130 using the input gradation value N, adds 1 to the read content, and executes control to write the read content to address N again. Next, the cumulative histogram calculation unit 132 calculates a cumulative histogram by accumulating the gray value histograms from the highest gray value. Specifically, the contents of address 256 of the histogram memory 130 remain as they are, starting from address 255 and continuing up to address 1.Content of address N=Content of address N Content of address N+1・--
--------------------(4) is obtained by sequentially performing the calculation.

14 cumulative histogram calculation unit 2 is a minimum value filter 12
The same processing as that of the cumulative histogram calculation unit 1 of 13 is performed on the output of , and the cumulative histogram of the image after the minimum value filter processing, that is, the area of the figure after contraction is calculated.

15, the cumulative histogram calculation unit 3 inputs the grayscale values of the original image 10 at the same pixels as the edge pixels detected by the edge detection unit 11, performs the same processing as the cumulative histogram calculation unit 1, and calculates the edge image area. Calculate the cumulative histogram of the original image at .

16 cumulative histogram calculation unit 4 calculates the minimum gray value g at the same pixel as the edge pixel detected by the edge detection unit 11.
(Xi % Y*) is input from the output of the minimum value filter 12, and the same processing as that of the cumulative histogram calculation unit 1 is performed to obtain the cumulative histogram of the image after minimum value filter processing in the edge image area. calculate.

The 17 boundary pixel number histogram calculation unit 1 has a histogram memory in the 13 cumulative histogram calculation unit 1 and the 14
The cumulative histogram calculation unit 2 calculates the difference between each address of the histogram memory, and calculates the number of boundary pixels when the original image is binarized with a certain threshold value according to Property 4. Similarly, the 18 boundary pixel number histogram calculation units 2 calculate the difference for each address between the histogram memory in the 15 cumulative histogram calculation units 3 and the histogram memory in the 16 cumulative histogram calculation units 4, and according to property 4, calculate the original image. The number of boundary pixels in the edge image area when binarized using a certain threshold value is calculated. That is, 17 and 18 are the first
This corresponds to the boundary area pixel number calculation unit 2 in Figure 8).

In the optimum threshold determination section 19, the frequency of each gray value (boundary pixel number) in the output of the boundary pixel number histogram calculation section 18 is used as the numerator, and the boundary point which is the output of the boundary pixel number histogram calculation section 17 is used as the numerator. The total number of pixels (B) and 18
Input the total number of pixels that are edges and boundaries (AnB), which is the output of the boundary pixel number histogram calculation unit 2, and the total number of pixels that are edges and edges (A), and calculate (A) + (B) - (
AnB) is used as the denominator, the ratio is calculated for each gray value, and the gray value with the maximum value is output as the optimal threshold value.

In this embodiment, for example, the minimum value filter 12 and the cumulative histogram calculation unit (112, 3, 4) are realized by dedicated hardware, and the boundary point number histogram calculation unit 1.2 of 17.18 and the optimal threshold The determining unit 19 can be realized by an algorithm using a microprocessor.

Expanding and generalizing the above-mentioned properties 1 and 2.3.4, respectively, result in the following.

Property 1: The figure obtained by binarizing after maximum value (minimum value) filter processing is the same as the figure obtained by applying a contraction logical filter after binarization.

Property 2 The number of boundary points of a figure can be calculated using the following three methods.

(1) Area of the original figure - Area of the figure after contraction (2) Area of the figure after expansion - Area of the original figure (3) (Area of the figure after expansion - Area of the figure after contraction) / 2 Property 3 The area of a figure obtained by binarizing a grayscale image changes depending on the threshold value, but the relationship between the threshold value and the area can be expressed by the cumulative histogram of the original image, which is equivalent to the area.

Property 4 The number of boundary pixels of a figure obtained by binarizing a grayscale image changes depending on the threshold value, but the relationship between the threshold value and the number of boundary pixels is as follows.
According to the three methods described in the section of Property 2, (1) Cumulative histogram of the original image - Cumulative histogram after minimum value filter processing (2) Cumulative histogram after maximum value filter processing - Cumulative histogram of the original image (3) It can be calculated using the histogram after maximum value filtering and the cumulative histogram after minimum value filtering. ``Based on property 4(2), the minimum value filter 12 in this embodiment may be replaced with a maximum value filter. Also, based on property 4(3),
Two filters, a maximum value filter and a minimum value filter, may be provided, and the output of the maximum value filter may be sent to the cumulative histogram calculation section 1.3, and the output of the minimum value filter may be sent to the cumulative histogram calculation section 2.4.

Note that instead of the maximum value filter or the minimum value filter, a filter that outputs the second largest pixel value or second smallest pixel value within the window may be used. In this case,
It is possible to obtain results that are not affected by isolated noise.

Further, as for the shape of the window, other than 3×3, a window consisting of a central pixel and pixels connected to that pixel in the sense of 4-connection may be used, or the window shape may be made variable.

Further, the edge detection unit 11 may detect a pixel in which the magnitude of the gradient of the gradation value is maximum along the direction of the gradient of the gradation value. Further, the edge detection unit 11 may detect pixels in which the value of a Gaussian Brassian, which detects the second-order differential value of the gradation value, changes from positive to negative or from negative to positive.

Furthermore, as another implementation form, functions such as spatial filter UtM gradient calculation unit), threshold processing (edge detection unit), order filter (minimum value filter), and histogram calculation (cumulative histogram calculation unit) are dedicated. This can also be realized by adding the functions of a boundary point histogram calculation section and an optimum threshold value determination section to an image processing apparatus provided as hardware or software on a general-purpose computer, and further providing software that controls the entire system.

〔Effect of the invention〕

According to the present invention, the optimal threshold value can be obtained several tens to hundreds of times faster than conventional techniques.

[Brief explanation of the drawing]

Figure 1 (a) is a block diagram of the present invention; Figure 1 (bl is a block diagram of the present embodiment; Figure 1 (C1 is a conceptual diagram of binarization processing for the gradation curve f (x) of the original figure); FIG. 1(d) is a histogram of gray values, FIG. 1(e) is a cumulative histogram of gray values, FIG. 1(f) is a diagram showing set relationships, Configuration diagram, 2nd
The figure is a processing flow diagram of a conventional method. DESCRIPTION OF SYMBOLS 1... Accumulation calculating unit, 2... Boundary area pixel number calculating unit, 11... Edge detecting unit, 12... Ordinal filter, 19... Optimal threshold determining unit.

Claims (1)

[Scope of Claims] 1) An image processing system that processes a grayscale image, comprising: an edge detection unit (11) that inputs an original image (10) and detects candidate points for the outline of an object in the image; 10)
An ordering filter that replaces the gray value of each pixel with a gray value of a predetermined rank when the gray values of pixels in a window consisting of neighboring pixels including that pixel are arranged in order of size. (12), and a cumulative histogram of gray values of the two images in the entire area of the image, and a cumulative calculation unit (1) that calculates a cumulative histogram of gray values of the two images in an edge image area occupied by the edge pixel group (110) detected by the edge detection unit (11); ), the boundary pixel number histogram in the entire area of the image obtained by the difference between the two cumulative histograms for each gray level in the entire area of the image, and the two in the edge image area calculated by the cumulative calculation unit (1) Boundary area pixel number calculation unit (2
), and using the histogram of the number of boundary pixels in the entire area of the image and the histogram of the number of boundary pixels in the edge image area calculated by the boundary area pixel number calculation unit (2), the gray value is set as a threshold value for each gray value. The optimum algorithm calculates the degree of coincidence between the boundary pixels of the binary image when binarized as , and the edge pixel group detected by the edge detection unit (11), and outputs the gray value with the maximum degree of coincidence. A threshold value determination method for binarization, comprising a value determination unit (19), and an output of the optimal threshold determination unit is used as a threshold value for binarization. 2) The threshold value determination method for binarization according to claim 1, wherein the order filter (12) uses a maximum value filter having a predetermined first rank. 3) The threshold value determination method for binarization according to claim 1, characterized in that said order filter (12) uses a minimum value filter whose predetermined rank is last. 4) The ordered filter (12) includes both a first-ranked filter, that is, a maximum value filter, and a lowest-value filter, that is, a minimum value filter. threshold determination method for binarization. 5) The binarization threshold determination method according to claim 1, wherein the order filter (12) has a variable order. 6) Claim 1, 2, 3, 4, or 5, wherein the ordered filter (12) has a variable window shape.
Threshold value determination method for binarization described. 7) The binarization threshold determination method according to claim 1, wherein the cumulative calculation unit (1) uses an original image and an image after sequential filter processing as the predetermined images. 8) The cumulative calculation unit (1) uses, as a predetermined image, the image after the sequential filter processing in which the predetermined ranking is the first, and the image after the sequential filter processing in which the predetermined ranking is the last. The binarization threshold determination method according to claim 1, characterized in that: 9) The threshold value determination method for binarization according to claim 1, wherein the cumulative calculation unit (1) uses, as the predetermined images, images resulting from two processes processed by changing the order in the order filter. . 10) The cumulative calculation unit (1) scans the input image with a histogram memory having an address corresponding to each level of gray value, and increments the contents of the histogram memory at the address corresponding to the gray value of each pixel. 2. The binarization system according to claim 1, further comprising: a control section for calculating a gray value histogram, and a cumulative histogram calculating section for accumulating the contents of the gray value histogram stored in the histogram memory when the increment operation ends. threshold determination method. 11) The optimal threshold determining unit (19) determines the gray value having the maximum frequency as the optimal threshold in the boundary pixel number histogram in the edge image area occupied by the pixel group detected by the edge detecting unit (11). 2. The binarization threshold determination method according to claim 1, wherein: 12) The optimum threshold determining unit (19) uses the frequency for each gray value of the boundary pixel number histogram in the edge image area occupied by the pixel group detected by the edge detecting unit (11) as a numerator, and calculates the frequency of each gray value in the entire area of the image. 2. The binarization threshold determination method according to claim 1, wherein a gray value having a maximum ratio such that the denominator is a frequency of each gray value in a boundary pixel number histogram is output as the optimal threshold value. . 13) The optimum threshold value determination unit (19) uses the frequency for each gray value of the boundary pixel number histogram in the edge image area occupied by the pixel group detected by the edge detection unit (11) as a numerator, and calculates the frequency of each gray value in the entire area of the image. In the edge image area occupied by the pixel group detected by the edge detection section (11), from the sum of the frequency for each gray value of the boundary pixel number histogram in and the total number of pixels detected as an edge by the edge detection section (11). 2. The binarization process according to claim 1, wherein the gradation value having the maximum ratio such that the denominator is a value obtained by subtracting the frequency of each gradation value of the boundary pixel number histogram is output as the optimum threshold value. Value determination method. 14) The binarization threshold determination method according to claim 1, wherein the edge detection unit (11) detects a pixel whose gray value changes rapidly in the vicinity of each pixel. 15) The binarization method according to claim 1, wherein the edge detection unit (11) detects a pixel in which the magnitude of the gradient of the gradation value is maximum along the direction of the gradient of the gradation value. Threshold value determination method. 16) The edge detection unit (11) is characterized in that it detects pixels in which the value of a Gaussian Laplacian (a means for detecting a second-order differential value of a gray value) changes from positive to negative or from negative to positive. The binarization threshold determination method according to claim 1.