JPH0477879A - Method for binarizing picture - Google Patents

Abstract

PURPOSE:To calculate an exact binarized threshold value by avoiding dependency to a pattern area distribution proportion not with the density histogram of the picture whole face but with the difference of the histogram of a local area. CONSTITUTION:An SAL and an SAH are the areas of an area AL and an area AH respectively, a window is scanned within a picture, when average density or a sum within the window is high, the SAH is larger than the SAL and when it is low, the SAL is relatively larger than the SAH. Then, when the value of the SAL when the SAL is larger is defined as an SALP and the value of the SAL when the SAH is larger is defined as an SALq, a difference of a density histogram DELTAf becomes '0' when SALq=SALq, since SALpnot equal to SALq can be conditioned easily by window scanning, the value of (x) to be DELTAf=0 becomes x=(dL+dH)/2 and an exact threshold value can be selected. Thus, an exact binarized threshold value can be calculated.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image binarization method for converting a grayscale image into a binarization pattern in an image processing process for pattern matching, pattern inspection, etc.

(Prior Art) Conventionally, regarding the binarization method for pattern matching and pattern inspection using an electronic computer, etc., there are the P-tile method and the mode method as methods based on the production distribution.

The P tile method uses the density distribution of the input image to calculate a given percentage P that has a cumulative density distribution obtained in descending order of density.
In this method, the density is calculated and this density value is used as the threshold value.

The modal method is a method of finding a valley in the concentration distribution and using it as a threshold, on the premise that the concentration distribution exhibits bimodality.

Further, as a method of binarization, there is a discriminant analysis method.

This method calculates the degree of separation when the concentration distribution is divided into two by a certain value, and sets the value at which this degree of separation is maximum as the threshold value.

[Problem to be solved by the invention]

However, in the conventional P-tile method, it is a prerequisite that the ratio P is known, and it cannot be used as a general binarization method, nor can it be used in a pattern matching test. There is a problem that this method is limited to cases where there is a reference pattern to be compared, the positional relationship between the input image and the reference pattern is known, and there are no defects or redundant patterns in the input cover pattern.

In addition, in the conventional mode method, the threshold value can be found when the pattern portion and the background have similar density distribution shapes and the areas occupied by each are approximately equal; however, when the area ratio is 1
As it deviates from =1, the position of the valley in the concentration distribution shifts toward the concentration side with a smaller area, and the valley eventually disappears. In other words, there is a problem that a correct threshold value cannot be obtained except under the above-mentioned special conditions.

Further, in the conventional discriminant analysis method, for example, when the area of the pattern portion is small, separation becomes poor, and as with the mode method, there is a problem in that incorrect threshold values are frequently output.

An object of the present invention is to provide an image binarization method that is less dependent on density distribution.

The above and other objects and novel features of the present invention will become apparent from the description of this specification and the accompanying drawings.

[Means to solve the problem]

In order to achieve the above-mentioned object, the present invention provides an image binarization method in which a small-sized window is scanned with an input grayscale image, and the average value or sum of the image area within the window is calculated as the average value or sum of the image area within the window. The density histogram of the image within the window is calculated as a density level, and the density level is the maximum or larger than a pre-specified limit value, and the second density level is the minimum density level or a pre-specified limit value.
The most important feature is to calculate the difference between the density histogram of the image within the window that is less than the limit value of , and to binarize the input image using the zero-crossing point of the difference histogram as a threshold.

[For production]

According to the above-mentioned means, while scanning the window within the image, the average value or sum of pixels included in the window is determined as the density level of the image within the window, and the image within the window is determined when the density level is high and when the density level is low. Since the cello-crossing point of the difference histogram obtained by taking the difference between the density histograms is used as the threshold for binarization, it is possible to avoid dependence on the pattern area distribution ratio and calculate the correct binarization threshold. can.

In other words, even if the pattern is small and its distribution ratio is small, or even if the pattern portion occupies almost the entire surface, the density histogram of only the pattern portion and the density histogram of only the background portion can be divided into 1 to gram without being affected by noise. Since the separated and extracted difference histogram 1-gram is obtained, the zero crossing point of the difference histogram can be used as a threshold value. It does not depend on pattern area percentage. The only exception is
In cases where the entire surface is only the background or the entire surface is a pattern portion, the area of the difference histogram is zero in that case, and it can be easily determined that the object belongs to this exception. In the conventional method, this determination cannot be made, so it is not easy to determine whether the threshold value is correct or incorrect, whereas the present invention can make this determination easy.

[Embodiments of the invention]

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

Although the image binarization method of the present invention can be implemented by devices such as computers, various image processing devices, and encoders, a computer is used here.

In addition, in all the figures for explaining the embodiment, parts having the same functions are given the same reference numerals, and repeated explanations thereof will be omitted.

1A to 1C are diagrams for explaining the principle of the image binarization method of the present invention.

FIG. i-A is a diagram showing an example of the window 10, and there are no special restrictions on the shape of the window 10. Let the area of this window 10 be SA.

FIG. 1B is a diagram showing the concentration region of the window 10, and the window 1
++ 1 ++ area H where 0 is a high density level of binarization
The area that also serves as the 110++ area is designated as AH, and the area that also serves as the 110++ area is designated as AL.

Figure 1C shows the density histogram 1 of the area around the window 10.
and the density histogram 2 of the area H, where the area is centered at the average level dL and the variance σ
It is assumed that the region H has a Gaussian distribution with a spread of σ2, and the area H has a Gaussian distribution with a spread of variance σ2 centered at the average level d.

This distribution is, respectively. -(x-dL)2/a2-(x-dH)2/σ2.

Now, as shown in FIG. 1B, if the window includes area H and area H, the density histogram f within the window is f=sAL. -(x-dL)2/σ2+5A)le
-(x-d,)2/σ2 However, S AL and S AH are the areas of the area AL and the area AH, respectively.

When the window is scanned in the image and the average density or summation within the window is high, SA1. l is large compared to SAL, and when it is low, SAL is relatively large compared to 5A14.

Therefore, if the value of SAL when SAL is large is SAL, and the value of SAL when SAH is large is SALq, then the difference Δf in the density histogram is Δf=sAL
pe-(x-dL)''-(X-dH)2
102+(SA SAL,) e 3AL, e-(x-dL)"/σ2-(x-dl,)"
/σ2+(SA 5AL-) e = (SAL-5AL9)(e-(x-dL)"/σ2-
(X-dJ2/ff")-〇△f becomes pa0'' when S ALp" S ALQ, but since 5ALp≠5ALq can be easily conditioned by window scanning, Δf=o. The value of X is x=
(dL+dn)/2. That is, the correct threshold value can be selected.

FIGS. 2A to 2C are diagrams showing density histograms based on the principle of the image binarization method of the present invention explained in FIGS. 1A to 1C. FIG. 2A is a histogram within a window of 1 to gram when SAL is large, FIG. 2B is a histogram within a window when SAH is large, and FIG. 2C is a difference histogram. In FIG. 2C, the positive and negative areas are always the same, and the zero crossing point is the correct threshold at which Δf=o.

If there is a possibility that the detection accuracy of this zero-crossing point is low due to noise, the center of gravity of the positive and negative points may be set as the zero-crossing point.

As for the zero cross point, the detection accuracy becomes higher when the area of 1Δf[ is larger. Therefore, ideally, the positions of the two windows when calculating the histogram difference Δf are set to be when the image density within each window is maximum and minimum. However, according to experiments, at the position where SAL or SAH becomes large,
If each of them occupies more than half of the SA, it is easy to calculate the zero-crossing point.

[Example 1] FIG. 3 shows a flowchart of Example 1 of the image binarization method of the present invention.

The image binarization method of this embodiment first involves inputting an image (Step 1.01), moving a window of a preset size and shape within the image (Step 02), and calculating the average value within the window (Step 02). 103). Step 1: Scan the window in a fixed range within the image.
Repeat steps 02 and 103. In step 104, a check is made to see if the scan is complete. When the scanning is completed, search for the position of the window where the average value within the window obtained in step 103 is the maximum or minimum (step 105), calculate the density histogram of the image within the window at each position (step 106), and calculate the histogram. Divide the difference between (step 1o7).

Next, the zero-crossing point of the difference histogram is calculated (step 108), the value is binarized using the threshold value (step 109), and a binary image is output (step 110).
).

[Embodiment 2] FIG. 4 is a flowchart of Embodiment 2 in which the processing time is shortened by not scanning the entire window, but by stopping scanning when threshold calculation becomes possible during scanning. FIG.

The image binarization method of the second embodiment includes inputting the image in step 201), moving the window in step 202), calculating the average value within the window in step 203), and determining the average value in areas where the pattern is excessive, that is, areas A and H. If it is larger than the large limit value th+ (step 204), record the position of this window (step 205); otherwise, if the area with few patterns, that is, the area AL, is smaller than the large limit value thL (step 206). , record the position of this window (step 207), otherwise move the window (step 202)
,'' repeat the process described in 1.

In step 208, it is determined whether the position of a window larger than the limit value thH has already been determined, and if it has been determined, step 204 is passed. In step 209, tJs is determined from the limit value thL.
It is determined whether the position of the window has already been determined, and if it has been determined, step 206 is passed.

In step 210, it is determined whether the positions of the windows are equal to each other, and when they are determined, the density histograms of both positions are calculated (step 211), the difference histogram is calculated (step 212), and the zero crossing point is calculated (step 211). 213
), and binarize using that point as a threshold (step 21
4).

Note that regarding step 211, if the answer is YES in step 204,
The density histogram may be calculated and recorded each time , or each time YES is determined in step 206 .

As can be seen from the above description, according to Examples 1 and 2, the density histogram of the entire image is not used, but the difference between the density histograms of local areas is used, so the dependence on the pattern area distribution ratio is reduced. It is possible to avoid this problem and calculate a correct binarization threshold.

In other words, even if the pattern is small and its distribution ratio is small, or even if the pattern portion occupies almost the entire surface, the density histogram of only the pattern portion and the density histogram of only the background portion can be calculated without being affected by noise. Since a differential histogram is obtained after being separated and extracted, the zero-crossing point of the differential histogram can be used as a threshold value. It does not depend on pattern area percentage. The only exceptions are when the entire surface is only the background or when the entire surface is a pattern part, but in that case, the area of the difference histogram becomes zero, and it can be easily determined that the area belongs to this exception.

In the conventional method, this discrimination cannot be made, so it is not easy to judge whether the threshold value is correct or incorrect, but one of the important effects of the present invention is that this judgment is easy.

The present invention has been specifically explained above based on examples, but
It goes without saying that the present invention is not limited to the embodiments described above, and can be modified in various ways without departing from the spirit thereof.

For example, in this embodiment, the in-window density level is the in-window average value, but it may be the in-window density total value. Furthermore, as a method for finding the zero-crossing point, in addition to the method of finding it by data processing the difference histogram waveform near the zero-crossing point, it goes without saying that it is also possible to find it by calculating the points of symmetry or the center of gravity of the difference histogram.

〔Effect of the invention〕

As explained above, according to the present invention, the density histogram of the entire image is not used, but the difference between the histograms of local regions is used, thereby avoiding dependence on the pattern area distribution ratio and achieving correct binarization. A threshold value can be calculated.

[Brief explanation of drawings]

FIGS. 1A to 1C are diagrams for explaining the principle of the image binarization method of the present invention, and FIGS. 2A to 2C are images 2 of the present invention explained in FIGS. 1A to 1C. Figure 3 is a flowchart of Embodiment 1 of the image binarization method of the present invention, and Figure 4 is a diagram showing a density histogram based on the principle of the digitization method. FIG. 7 is a diagram showing a flowchart of a second embodiment in which processing time is shortened by stopping scanning at the point when value calculation becomes possible. In the figure, 1... the density histogram of the region, 2...
...Concentration histogram of area H, 10...window.

Claims (1)

[Claims] (1) Scan a small window on the input grayscale image, calculate the average value or sum of the image area within the window as the density level of the image within the window, and determine whether the density level is the maximum or greater than a pre-specified limit value. The density histogram of the image within the window is
It is characterized by calculating the difference between the density histogram of the image within the window where the density level is the minimum or below a pre-specified second limit value, and binarizing the input image using the zero-crossing point of the difference histogram as a threshold. Image binarization method.