JPS6179375A - Binarization system - Google Patents

Abstract

PURPOSE:To apply binarization correctly to a picture of an optional object even with uneven lighting nby dividing the entire picture into rectangular partial regions and obtaining an optimum threshold value giving a ratio of black picture element number to white picture element number in coincidence with the ratio of black region to white region of a test pattern at each rectangualar partial region. CONSTITUTION:A binarization circuit reads the content of a memory 8 storing a threshold value Zkh at each partial region of the test pattern by using an address generating cirlcuit 7 to recognizes the Zkh. A comparison circuit 10 compares the Zkh with a value of all picture elements belonging to the partial region Qkh, outputs logical 1 when the value of picture element is larger than the threshold value and outputs 0 when smaller than the threshold value. An accumulation circuit 9 inputs a data 0 or 1 from the comparator circuit 10 and accumulates the value to a register Rkh corresponding to each partial region Qkh. Since the value of the picture element is compared with the threshold value Zkh at each partial region, the accumulation is executed at each register Rkh. It is possible to read the content of the register Rkh in the accumulation circuit 9 from an operating device.

Description

DETAILED DESCRIPTION OF THE INVENTION (A) Field of Application in Industry-11 This invention can be commonly used in all image processing that targets two-dimensional 21-direction images.

For example, the present invention relates to a binarization method that is necessary for pre-processing of robot visual devices, image-based measuring devices or inspection devices, optical character reading devices, facsimiles, and the like.

Binary means two values, white and black.
It may also be expressed as 0 or 1. The operation of dividing a two-dimensional pixel into pixels and making the degree of brightness of each pixel correspond to either white or black is called binarization. The color tone of the pixels of the original image is a continuous light and shade of gray, with white and black as the extremes.

Although the degree of brightness and darkness is an analog quantity, in the case of processing such as object recognition, shape determination, and character reading, it is convenient to perform image processing after distinguishing into binary values, white or black.

To perform binarization, the brightness value of the pixel is compared with a predetermined threshold value, and the pixel is divided into two values depending on whether it is above or below threshold 1.

How do we give the threshold? That is the most important thing. How should the threshold be determined so that the contours of objects, the lines of letters and inscriptions appear most clearly? This is the problem.

(a) Prior Art Conventional image binarization power formulas can be roughly divided into two types.

The first method is to fix the threshold value or to set it manually. This is called the fixed threshold method.

The second method is to automatically determine the threshold value. This will be called the automatic threshold determination method.

There are various automatic threshold determination methods. For example, (I) a method in which the valleys of the density histogram of an image are used as thresholds; A density level for each pixel is determined for the entire image or a partial image, and a histogram is used that graphically represents the number of pixels belonging to each density level.

The valleys of the histogram are those where the number of pixels belonging to that level is small.

Due to the nature of the original image, there are not necessarily valleys in the histogram, and even if there are valleys, if the valleys are off to one side or the other, there is a drawback that 21-value conversion cannot be performed correctly.

This is called the trough point threshold in the cylinder.

(m) A method in which the value of the brightness level between the maximum and minimum brightness levels is used as the threshold. The maximum and minimum brightness levels are determined for the entire image or a partial image, and half of the brightness level is used as the threshold.

Since this does not concern the density distribution, the threshold value is not necessarily the average value of the luminance levels.

It can be called an intermediate threshold.

Field A method where the threshold is set at a place where the brightness level changes sharply. In the vicinity of the contour line, the brightness level changes rapidly in the direction perpendicular to the line. This change is detected, and the brightness level of the portion where the change occurs strongly is set as a threshold. It can be simply called a differential threshold.

In any of these methods (I) to (sub), there are those that use the same threshold value for the entire image and those that use different threshold values for each partial area.

(1) Problems that the invention attempts to solve The conventional binarization method has the following drawbacks.

The fixed threshold method cannot adequately cope with changes in illuminance. When the intensity of the illumination light increases, the area of the white part increases, regardless of the actual outline of the object, and when the illumination light becomes weaker, the area of the white part decreases. It would be good if the intensity of the illumination light was constant, but this is not always the case.

The automatic threshold value determination method can divide the image into black and white, regardless of changes in illuminance.

However, even in this case, the threshold value considered to be optimal is merely estimated l- based on the brightness information of the image.

In the case of a valley point threshold, valley points may not necessarily exist depending on the pattern of the image. In this case, binarization is difficult. Sometimes there are two or more points. In this case, binarization remains inconvenient.

In the case of an intermediate threshold value, the threshold value is an intermediate value between the highest and lowest brightness levels.

The number of pixels with the highest brightness or the number of pixels with the lowest brightness may be extremely small or large. The intermediate value is often the average value. Unfortunately, intermediate values do not necessarily correspond to density levels that fall on the boundary between white and black.

Although the differential threshold is a method of determining a threshold with high validity, the threshold does not reach a constant level like the trough threshold and the intermediate threshold. The value of the level at which the differential is maximum varies depending on the location within the image, and the threshold value must be determined locally. Furthermore, since the differential is maximized, the brightness and darkness of pixels must be leveled in many stages.

If the number of steps for dividing the bright and dark stages is small, the differential value will contain many errors, and the determination of the threshold value itself will involve a large amount of error. Another drawback is that the calculations are complicated.

Furthermore, in the case of the valley point threshold and the intermediate threshold, if the same threshold is used for the entire image, there is a problem that it is difficult to deal with cases where there is illumination unevenness.

An object of the present invention is to provide a binarization method that can correctly binarize the brightness and darkness values of pixels even if there are changes in illuminance or uneven illumination.

(1) Means for Solving the Problems The present invention uses a test pattern having a partial area in which the ratio of black and white areas is predetermined in order to solve the problems listed in ``-''.

Before capturing a target image, a test pattern is captured, and this test pattern is binarized using two appropriate thresholds regarding brightness. The area of the black part or the area of the white part when binarized is different from the area of the black or white part of the test pattern. Therefore, the image of the same test pattern is binarized by changing the threshold value. The threshold value is changed repeatedly until the area of the black part of the binarized image becomes approximately equal to the area of the black part of the test pattern. In this way, a threshold value that can correctly binarize the test pattern is found.

In this way, the threshold value is changed for each area where a plurality of vertical and horizontal pixels are gathered.

A threshold value is determined for each partial region.

The initial threshold may be any suitable value. By changing the threshold value, a threshold value that can correctly binarize the test pattern is asymptotically determined.

We will calculate the threshold value Z for the partial region, which is obtained by using the test pattern regardless of the target image (the actual image).
This is completely different from those that use intermediate values, maximum differential points, etc. as thresholds. Since the threshold value is determined independently of the target image, this method is close to the fixed threshold method. However, since the test pattern is used and the partial image is determined by a threshold value, if the illumination illuminating the test pattern and the target image are equal, this threshold value will always correctly binarize the partial image. It should be possible.

Actually, the entire image is divided into m×l rectangular partial regions. FIG. 3 shows four of these divisions. Each partial region Q is a collection of many pixels arranged vertically (in yellow).

Therefore, the partial area is labeled with h. The entire image is made up of one partial area arranged horizontally by m1 and vertically. The partial area is called Qkh.

1 (===l, , m, h=l, , l.

A threshold Zkh, which is particularly relevant for sub-areas, should be determined. In the present invention, the value of Z is asymptotically brought closer to the threshold value for correctly binarizing the test pattern.

It is desirable to choose a method that converges quickly. but,
Any method may be used to converge Z to the correct value.

For simplicity, omit h from the suffix of the partial area,
explain.

A test pattern shown in FIG. 3 is used in which the shaded area is black and the other areas are white. It is assumed that the entire test pattern matches the entire image.

It is assumed that the repetition of the partial area and the repetition of black and white of the test pattern are the same.

In other words, the partial area Q has a black part and a white part of the same size at the same position. Here, a test pattern is shown in which a black square is located at the center and a white frame is formed around the periphery, but this is merely an example. It is sufficient that the same pattern is repeated in each partial region, and any fine pattern is acceptable.

Although the same test pattern is repeated, since the intensity of the illumination light differs between the center and the periphery of the object, it is not possible to correctly binarize the images of all partial regions using the same threshold value.

Let U be the total number of pixels included in the partial area Q.

Among all pixels U, the number of pixels that become black when binarized is B1
Let W be the number of pixels that become white.

B + W2U (1) It is.

It is assumed that among the 11 parts corresponding to the partial area Q of the Death 1 pattern, the black part is S/[J. The white part is (1-
5/U).

If a test pattern is imaged and binarized using an appropriate threshold value Zf, the number of black pixels should be 81 and the number of white pixels should be (U-5).

However, in reality, the intensity of the illumination light is not constant, so
An appropriate threshold value Z cannot be determined in advance.

For a certain partial image, the number of black pixels is equal to the threshold Z
is a function of However, it is assumed that the brighter the threshold value Z is, the more positive the threshold value Z becomes. The Z value is determined by the brightness level.

If Z is low, only extremely black parts will be black pixels.

If Z is high, binarization is performed at a high brightness level, so
The number of black pixels increases.

Therefore, it can be said that the number B of black pixels of an image obtained by binarizing a certain partial image captured with a certain illumination light using a certain threshold value Z is a monotonically increasing function of Z.

When the test pattern corresponds to one partial image, S among the U pixels becomes a black pixel. The threshold should be selected so as to generate eight black pixels.

Eventually, the black pixel &B(Z) as a function of the threshold Z becomes B(
Z gives an appropriate threshold value Zf so that Z)=S (2).

Assume that the number B of black pixels is shown as a function of Z as shown in FIG. This functional form is unknown. This is because it differs depending on lighting etc. We want to find Zf that satisfies equation (2).

There are many possible ways. Give an example.

(1) For example, three predetermined fixed threshold values Z
l, Z2, and Z3, the number of black pixels in the partial area of the test pattern binarized using this threshold is B1, B2,
Let's call it B3.

If B(Z) is approximated as a quadratic function, a, b,
With c as an unknown constant, the following four equations (3) hold true. From here, by eliminating alb and c, Zf is obtained. This is obtained as a solution to the determinant. The threshold value Zf obtained in this way is 3
From only the values B4, B2, and B8 of points Z1, Z2, and Z3, (
2) This is the most accurate method for determining the threshold value zf that satisfies the formula. FIG. 4 is a diagram explaining this method, where the horizontal axis is the threshold value and the vertical axis is the number of black pixels.

(2) Similarly, if we know the values at 4 points, the determinant of equation (7) becomes an equation with 5 rows and 5 columns, and (ZlB, Z',
Zl, 1, B1) becomes 5 rows. However, i
= Q, ., 4, and it is assumed that Zo==ZfBo-8.

(3) As the number of points for which the relationship between the threshold Z and the number of black pixels B when binarized is known increases,
An appropriate threshold value Zf that satisfies equation (2) can be accurately determined.

Black pixels & for n thresholds f+1i (z)
(B□), then (n+1) elements (z
in% ”', z,, 1, B, ) more ( n +
1) A determinant is created by superimposing the row vectors (i-0, -1n), and Zf can be found by assuming that the determinant becomes 0.

The way to find Zf of L is to know several relationships between B and Z and then find Zf all at once.

In addition, there may be a method of asymptotically determining zf.

(4) Figure 5 shows the relationship between Z and B. What you are looking for is B (
Z)=S totoru Zf. It is called the simple B2 curve.

Count the black pixels &Bl, B2 at two points Zl, Z2. Zl and Z2 are selected to be smaller and larger than Zf.

Let the point whose coordinates are Zl and B1 be Pl, and the point whose coordinates are Z2 and B2 be B2.

Straight line P, B2 is the first approximation of the B2 curve, but it is considerably off. Let R2 be the point on PI B2 that becomes B2S. Let Z3 be the threshold value corresponding to R2.

The number of black pixels B8 is determined for Z8. This gives point P8 on the B2 curve.

From point P3, draw a straight line parallel to straight line P and p2, and B=
Let the point where it intersects with S be R3. Let the Z coordinate of this point be 74.

B2-B.

It is. The coefficient of (spa) in equation (8) is the straight line P, B
It is the slope of 2 and can be simply written as. In this way, a recurrence formula regarding the threshold Z is obtained. In this way, Z is successively determined, and the appropriate threshold value Zf is found as the limit.

In reality, it is not necessary to perform this calculation infinitely.

A good approximation can be obtained in about 3 to 6 times.

(5) The method of finding (Z,) using a recurrence formula is suitable for calculation by computer. The recurrence formula is not limited to the On) formula. Without fixing the slope of the straight line to P, B2, P, ,
It may be set as the gradient of P. In this case, as Zj++ = Zj+ω3 (5-B5)
A recurrence formula such as (13) can be used.

In this way, the appropriate threshold value of equation (2) is obtained by the methods described in (1) to (5).

The binarization method of the present invention (a) converts the entire image into (m×1 rectangular partial area (Qkh)
(b) Within each rectangular partial area, the area ratio of the white or black part is constant, In addition, a circuit that images a known test pattern, binarizes it using a temporary threshold (Zj), and counts the number of white or black pixels in each rectangular partial area on these two images; is a circuit that calculates the optimum distance so that the number of pixels matches the number of pixels corresponding to the white or black area of the test pattern.

(e) Effect The test pattern is a pattern in which the ratio of black and white is constant and the same partial pattern with the same ratio is repeated vertically and horizontally (by mXl).

The image created by capturing this test pattern (mXn
) into subregions (Qkh ), and set different thresholds (Zkh
) to binarize. Assume that, as a result of binarization, among all the pixels U in a certain partial area, the black pixels are B1 and the white pixels are W.

It is assumed that the area ratio of the point portion of the test pattern is S/U, and the area ratio of the white portion is l#1 (1-5/U).

If the threshold value Z is determined, the number of black pixels B1 and the number of self-pixels W are determined, so B, , W are functions of Z.

Therefore, the value of Z is changed so that the number B of black pixels becomes equal to S. From this, the optimal threshold 1st line Zf is known. Zf
There is a method to find Zf by solving the determinant, and there is also a method to find Zf asymptotically.

Zf is determined for each partial region. This is set as a fixed threshold Zf(k,h) corresponding to the partial area.

The fixed threshold Zf(klh) is a function of the lighting conditions, but as long as the lighting conditions are kept the same, Zf(klh) is an appropriate threshold. After this, give the same illumination to the target object,
Capture and binarize. The binarization threshold is a fixed threshold Zf (k, h) determined in advance.

zf(k, h) provides a threshold value that allows correct restoration of the black and white of the test pattern. Therefore, any target can be illuminated with the same light E! When an image is captured under 11 conditions, the threshold f for binarization is set so that the outline of the object and the black-and-white boundary can be correctly restored.

(force) example FIG. 1 is an overall configuration diagram of the binarization method of the present invention.

1 is a device that captures an image to be processed. This scans the captured image and outputs it as a video signal. This is a good sign for analog quantities.

Reference numeral 2 denotes a control device that digitally converts the video signal output from the imaging device 1 into one pixel (n pixels) and outputs it together with various control signals.

Since it is converted into n bits, this is a result of dividing the brightness change from white to black into 2n levels. Continuous changes in brightness of pixels with continuous changes in brightness and darkness are converted into 2n discontinuous gradations.

3 binarizes an image consisting of a set of two-tone pixels output from the control device 2 using different threshold values for each (mxβ) rectangular partial area, and outputs white pixels as 1 and black pixels as 0. It is something. It is called a binarization circuit 3.

4 is an arithmetic circuit for determining the threshold level Z for binarization in the partial area Q divided into m×1 pieces.

FIG. 2 is a side view of the configuration of the binarization circuit.

5 is an interface circuit between the control device, the arithmetic circuit, and the main circuit.

6 is a timing signal generation circuit that generates timing signals necessary to control each part of the circuit.

7 is an address generation circuit. 8 is a memory having a storage capacity of (mxI!>words).

9 is an accumulation circuit having (m×1 registers).

10 is a comparison circuit. The comparison circuit 10 compares the brightness level of each pixel output from the control device with the threshold value for binarization in each partial area stored in advance in 8. It outputs 1 when the brightness level is smaller than the threshold value, and outputs O when the brightness level is smaller than the threshold value.

Next, a method for determining the optimal binarization threshold for each partial region will be specifically described.

FIG. 3 is a plan view showing an example of a test pattern. This is a test pattern in which the entire image is divided into (mXn) rectangular parts, and each rectangular part has a black area of a predetermined position and a predetermined size. In this example, a square black area exists at the center of each rectangular area, but the invention is not limited to this. It is sufficient as long as the black areas have the same size and position.

The ratio of the black area to the area of the rectangular partial area is S/U. This is the same for all rectangular partial areas.

Image the test pattern. Assume that the rectangular partial area corresponds to U pixels. Then, the brightness of the pixels in the imaged rectangular partial area is binarized using a certain threshold value Z, and the number of pixels classified as black pixels is counted. If there are 8 pixels, this is the optimal threshold value. That's what it means.

Assume that the brightness level is gradated to n bits. One level interval corresponds to the 111th brightness, and this level interval is assumed to be 1. Assume that the minimum brightness value is 0 and the maximum value is 2n.

Consider the brightness levels H and L as appropriate. This is Hikaru.

The optimal threshold value is close to the maximum and minimum values of the degree of I
], ■The value should be between 7 levels. The equation is 2>H>L>0 (+4), and Hl
We start with L as the initial closure and asymptotically approach the optimal threshold.

First, the arithmetic unit 4 writes a threshold value to an address (kh) corresponding to each rectangular partial area Qkh of the memory 8 in the binarization circuit to set an initial threshold value. L may be a common value for all partial regions. However, it may be a different value depending on the partial area.

Then, the test pattern is imaged. A brightness signal of an image is sent from the imaging device 1 to the control device 2 as a video signal. This scans the screen from left to right and outputs the luminance levels of all pixels as continuous shadows in an order that moves down one stage at a time. The control device 2 A/D converts the luminance level of each pixel to obtain a value that is a 2nd gray level.

The binarization circuit has 1 address generation point! )7,
For each partial area of the test pattern, the contents of the memory 8 in which the threshold value Zkh is stored are read out, and Z□h is determined. The comparison circuit 10 compares Zkh with the values of all pixels belonging to the partial region Qkh, and outputs 1 when the pixel value is larger than the threshold value, and outputs 0 when it is smaller than the threshold value.

The accumulating circuit 9 inputs data of 0 or 1 from the comparator circuit 10 and accumulates this value in the register Rkh corresponding to each partial area Qkh. pixel value for each subregion,
Since the comparison with the threshold value Zkh is performed, accumulation is performed sequentially for each register Rkh.

The contents of the register (Rkh) in the accumulation circuit 9 can be read from the arithmetic unit 4.

The first threshold Zkh is, as already mentioned, the threshold value. I departed on the condition that it was a small Zkh-L.

The value accumulated in the register Rkh corresponds to the number of black pixels or the number of white pixels. Let SLi be the number of black pixels in a certain partial area when binarized.

Similarly, the threshold value H is written in the memory 8, the test pattern is imaged, and the partial area Qkh is compared with the threshold value H for each pixel.
Find the number of black pixels SHi in .

In this way, the black pixel numbers SHi and SLl when binarized using the thresholds H and L are determined for the partial region.

i is a suffix representing (kh).

Here, the suffix i or (kh) of the rectangular partial area will be omitted for simplicity.

We use T(1)2L+ω(S-5L) (15) as a first approximation to the optimal threshold. However, ω is a constant defined by the amount of increase in the threshold value required to increase the number of black pixels by one.

Using T(1) as a threshold, pixels included in the same partial region Q are binarized. The number of black pixels at this time is assumed to be 5 (1). Next, the second approximation T(2) is determined as follows.

T(2) −T(1)+ω(S-5(1))
(+7) Similarly, when the k-th closure is T (k) and the number of black pixels is S' (k), the (k+1)-th threshold T (k+1) is T (k+1) - T (k) 10ω(S
−5(k)) (18), the optimal threshold value is successively approximated. In reality, the optimum threshold value is converged after several iterative calculations.

The method described below is the same as the method described in (4) of (-1:). As already mentioned, in addition to this, Zf
There are several ways to find.

(G) Effect The present invention divides the entire image into rectangular partial areas, images a test pattern consisting of repeating a fixed pattern of black areas and white areas corresponding to each rectangular partial area, and , an optimal threshold value is determined that gives a ratio of the number of black pixels to the number of white pixels that matches the ratio of the black area to the white area of the test pattern.

Since a different binarization threshold is determined for each rectangular partial region,
Even if there is uneven illumination, it is possible to obtain a single binarized image in which the image of any object is correctly dinitrated.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the overall configuration of the binarization method. FIG. 2 is a block diagram of a binarization circuit. FIG. 3 is a diagram showing the test pattern corresponding to all images. Figure 4 shows the optimal (
An explanatory diagram for determining mZf. FIG. 5 is an explanatory diagram of how to asymptotically obtain the optimum value Zf from the data of threshold value Z-number of black pixels B. 1 Imaging device 2 Control device 3 Binarization circuit 4... Arithmetic circuit 5 Interface circuit 6 Timing generation circuit 7 Address generation circuit 8 Memory 9 Accumulation circuit 10 Comparison circuit Inventor Kuni Ogi Misaki
1) Yasushi

Claims (1)

[Claims] A two-dimensional image to be processed is captured, converted into a video signal, and then converted into an analog/digital signal.
A first means for obtaining a digital signal of bits, and a second means for binarizing the obtained n-bit digital image of each pixel into white pixels and black pixels using different threshold values for each m×l rectangular partial area. a third means for counting the number of white pixels or black pixels in each partial region; and a fourth means for calculating an optimal binarization threshold based on the value obtained by the third means. A pattern having a predetermined ratio of black areas and white areas is
Image a test pattern that is repeated l times, set a threshold value for each partial region, binarize it, and find a threshold value at which the ratio of the number of white pixels or black pixels is equal to the ratio of the white area and black area of the test pattern. A binarization method that is characterized by