JPH03237571A - Device for calculating binarizing threshold of image - Google Patents

Abstract

PURPOSE:To prevent the generation of adverse influence or the like due to uneven brightness by comparing a difference between the brightness of each picture element of an image in a window and that of a specific picture element with a parameter proportional to the contrast to form a histogram and calculating a threshold based upon the histogram. CONSTITUTION:The brightness information of nine picture elements e.g. of an intra-window image in a many-valued image is latched by latch circuits 70 to 78 and two parameters rH, RU proportional to a contrast to be a difference between the brightness of an image part in an image to be a target and that of its background part are supplied to comparators 80 to 87 and compared with the brightness information d0 of a center picture element latched in the circuit 70. Binary data fi = 1 in the case of -RL < di -d0 < -rH or fi = 0 in the other case is outputted, an illegal pattern is detected, an illegal pattern histogram is formed based upon the detected result and an optimum binarizing threshold is determined in accordance with another necessary formed histogram. Thereby, a threshold prevented from adverse influence to be generated when plural images having respectively different brightness unenvenness or contrasts exist can be obtained.

Description

[Detailed Description of the Invention] <Industrial Application Field> The present invention relates to image processing technology applied to the eyes of intelligent robots and visual inspection devices. The present invention relates to an image binarization threshold calculation device used to calculate a threshold for obtaining the best binary image when performing binarization processing on a multivalued image.

<Prior art> The applicant recently proposed an image binarization threshold calculation device that can calculate an optimal binarization threshold at high speed and stably even if the size of the object is unknown. (Unexamined Japanese Patent Publication No. 1986-5)
No. 1586).

FIG. 6 shows the configuration of this device, in which a microcomputer 10 is the main control unit, and a window scanning section 11.
Illegal pattern detection section 12. Illegal pattern histogram generation section 13. Brightness histogram generation section 149 Image quality histogram generation section 15. Histogram smoothing processing unit 16
．． It is equipped with an optimal threshold value search section 17.

The window scanning unit 11 includes an address generator 19 connected to the microcomputer 10 via a CPU bus 18 and an address bus 21 . Data bus 23. An image memory 24 connected to the control bus 22 via an image bus 20-, and a FIFO connected to the data bus 23,
The first shift register 25 and the first group of latch circuits 26 to 29 of the (t-in first-out) type, and the same FIF connected to this final stage latch circuit 29
It is composed of an O-type second shift register 30 and a second group of latch circuits 31 to 34.

This window scanning unit 11 is for setting a rectangular window 60 as shown in FIG. Predetermined pixel A0~ per image
The brightness data d0 to d7 of A are taken out at the same time. Note that in the figure, x marks are pixels that are not subject to processing.

The first shift register 25 has the number of stages obtained by subtracting the number of stages of the first group of latch circuits 26 to 29 from n (see FIG. 8), where the number of pixels in the horizontal direction of the image memory 24 is n (see FIG. 8).
The four-stage latch circuits 26 to 29 of the first group, which have a delay circuit (n-4) and which performs an operation of delaying the output of brightness data by this number of stages, operate as a shift register as a whole, and the final The output of the stage latch circuit 29 is output to the second shift register 30. This second shift register 30 also has (n-4) stages, and delays the output of the brightness data by the number of stages to delay the output of the brightness data to the four-stage latch circuit 3 of the second group.
Tell 1-34.

Brightness data d, -d of each pixel A0-A7.

is output to the illegal pattern detection section 12, and the brightness data d+ of the specific pixel A is also output to the illegal pattern histogram generation section 13 and the brightness histogram generation section 14.

The illegal pattern detection unit 12 determines whether the image in the window 60 is a legal pattern or an illegal pattern, and sets a flag of "1" if it is an illegal pattern, and sets a flag of "O" if it is a legal pattern. Output each flag.

FIG. 7 shows the configuration of the conventional illegal pattern detection section 12, which includes eight latch circuits 35 to 42 that latch each brightness data d0 to d7, brightness data d of a specific pixel A, and other pixels A, and A2 to A? includes seven comparison circuits 43 to 49 for comparing the brightness data do and d2 to d7 in magnitude. These comparison circuits 43~
49, each pixel A0 and A2 to A of the window 60 is set using the brightness data dI of a specific pixel AI as a threshold value.
, for binarizing the brightness data d0 and d2 to d, of each brightness data d, (however, i・0
, 2.3. ,.

..., 7) is "1" when d ≧ d, d, < d.

In this case, each binary data fi of "0" is output. Note that the binary data f of the specific pixel A.

is always rl.

FIG. 9 shows that the brightness data d0 to d of each pixel A0 to A in the window 60 is binarized to binary data f0 to f.
It shows the process of conversion to 7.

The binary data f! (i=0.2.3...
. , 7) are output to the detection logic unit 50, which uses these binary data fi to determine whether or not the image within the window 60 is an illegal pattern, and if it is an illegal pattern, “ A flag of "1" is output, and a flag of "0" is output if the pattern is legal. Here, a legal pattern means a binary pattern that is suitable as an image of an outline, and an illegal pattern means a binary pattern that is inappropriate as an image of an outline.A legal pattern is a binary pattern that is not suitable as an image of an outline. The pattern becomes smooth, and the illegal pattern takes a non-smooth form as shown in FIG. Note that in Figures 10 and 11, the shaded area indicates a pixel with binary data rl, and in the binary pattern of Figure 11, it is "1".
An rQJ hole 61 is formed inside the pixel region of .

Returning to FIG. 6, illegal pattern histogram generation section 1
3, each time the detection logic unit 50 detects an illegal pattern and outputs a flag of "1", the frequency of occurrence of this flag is counted to generate a histogram of the illegal pattern. The brightness histogram generator 14 uses a window 60
A brightness histogram is generated by extracting the brightness data d of the pixel A at a specific position within. The image quality histogram generation unit 15 generates an image quality histogram from the illegal pattern histogram and the brightness histogram. This image quality histogram provides an evaluation standard for image quality (for example, a standard indicating poor image quality).

The histogram smoothing processing unit 16 is for smoothing the image quality histogram, and as a result, a smoothed image quality histogram as shown in FIG. 12 is generated.

In FIG. 12, the horizontal axis is k (L threshold) and the vertical axis is the poor image quality, where the poor image quality is the threshold value k. It is minimum at t.

The optimal threshold searching section 17 searches for this threshold kopz as the optimal threshold, and notifies the microcomputer 10 of the search result via the CPU bus 18.

In the above configuration, when an image of an object is captured by an image capturing device (not shown), a grayscale image consisting of the object and its back weight is stored in the image memory 24.

This grayscale image is subjected to recognition processing after binarization processing, but prior to the binarization, the optimum binarization threshold value is calculated by the binarization threshold calculation device shown in Fig. 6. Become.

First, after clearing the illegal pattern histogram generation section 13 and the brightness histogram generation section 14 to zero, the window scanning section 11 sets a window 60 on the grayscale image and performs rask scanning.

That is, in the window scanning unit 11, the address generator 19 sequentially generates the address of each pixel in the image memory 24, and outputs the brightness data of the pixel corresponding to the address to the first shift register 25 via the image bus 20. let

These brightness data are transferred to the first shift register 25,
Group latch circuits 26-29. Second shift register 30
．． The brightness data d0 to d2 of the pixels A0 to A8 within the window 60 are sequentially sent to the second group of latch circuits 31 to 34, and the brightness data d0 to d2 of the pixels A0 to A8 within the window 60 are sent from each latch circuit of the first group to the second shift register 30 and Each latch circuit 31 of the second group
From 34 onwards, brightness data d of pixels A, ~A.

~d, are respectively taken out.

The illegal pattern detection unit 12 takes in these brightness values d0 to d, and after binarizing each pixel in the window 60 using the brightness data d+ of a specific pixel A as a threshold value, the binary pattern is Determine whether the pattern corresponds to an illegal pattern.

If it is an illegal pattern, a detection flag "1" is output, and the illegal pattern histogram generation unit 13 counts the number of occurrences of the illegal pattern for each brightness (threshold value) to generate an illegal pattern histogram. be done. Similarly, the brightness histogram generation unit 14 counts the frequency of each brightness and generates a brightness histogram.

When the generation of these histograms is completed, the next image quality histogram generation unit 15 calculates an image quality histogram using both histograms. This image quality histogram is smoothed by a histogram smoothing processing unit 16 as necessary, and then an optimum threshold value search unit 17 searches for an optimum threshold value from this smoothed image quality histogram and uses the search result as a microprocessor. Output to computer 10.

Note that although the optimal binarization threshold is calculated using hardware in the above example, the calculation is not limited to this, and it is also possible to calculate it using software.

<Problems to be Solved by the Invention> In the case of the binarization threshold calculation device having the above configuration, the illegal pattern detection unit 12 uses the brightness data d of the specific pixel A in the window 6o as the threshold, For that specific pixel AI, the binary data is always "1", and for each pixel A= (i・0, 2.3..., 7) in the window 60, the brightness data d is d! If ≧d, the binary data is “1”; if d,<d, the binary data is “0”
It is said that Therefore, even if the pixel is in the same image part as the specific pixel A, the brightness data d may be affected due to uneven brightness.
, is even slightly smaller than the brightness data di of the specific pixel AI, the binary data becomes 'OJ', making it difficult to detect an illegal pattern with high accuracy.

Furthermore, as shown in FIG. 13, when there are multiple image parts 2°3 with different contrasts in one image 1, according to the conventional binarization threshold calculation device described above, both image parts An operation for calculating an optimal binarization threshold value is performed to extract 2.3 as a binary image. For this reason, when one of the image parts is targeted for recognition, etc., the binarization threshold is not optimal for that target image part, making it difficult to obtain the best binary image. .

This invention makes it possible to calculate the optimal binarization threshold without being adversely affected by uneven brightness, and even if there are multiple image parts with different contrasts in one image, the target Optimal 2 for the image part
It is an object of the present invention to provide an image binarization threshold calculation device that enables calculation of a digitization threshold.

<Means for Solving the Problems> The binarization threshold calculation device of the present invention focuses on an arbitrary image part and calculates a threshold for binarizing the entire multivalued image. a scanning means for setting a window and scanning the multivalued image; a first information extraction means for extracting information regarding the brightness of the image within the window; and a first information extraction means for extracting information regarding the brightness of the image within the window; a second information extraction means for extracting information regarding the suitability of the first. Information generating means for generating third information that provides an image quality evaluation standard from each piece of information from the second information extracting means;
and threshold calculation means for calculating an optimal threshold value that provides the best image quality based on the third information.

The second information extraction means compares the difference between the brightness of each pixel within the window and the brightness of a specific pixel with a parameter proportional to the contrast of the image portion of interest, and extracts the image within the window. It includes a binarization circuit section that performs binarization processing, and a judgment circuit that decides whether or not the binary pattern obtained by the binarization circuit section is appropriate as an image pattern of an outline part.

<Operation> The second information extraction means compares the difference between the brightness of each pixel within the window and the brightness of a specific pixel with a parameter proportional to the contrast of the image part of interest, and extracts the image within the window. Since it is binarized,
When there are multiple image parts with different contrasts in one image, by selecting the above parameters according to the contrast of the target image part, the optimal binarization threshold value can be determined for that image part. It can be calculated.

Furthermore, even if there is unevenness in brightness within the same image portion, it is possible to prevent the binary pattern obtained by binarizing the image within the window from being adversely affected by the unevenness.

<Embodiment> FIG. 1 shows a specific example of an illegal pattern detection section 12' which is the bottom of the binarization threshold calculation device of the present invention.

In the illustrated example, when a window 5 of 3×3 pixels in length and width as shown in FIG.
This is to determine whether the pattern is illegal. Note that although the window 5 in the illustrated example has a different shape from the window 60 described above, the shape of the window is not particularly limited.
Shapes other than these two types may be used.

FIG. 3 shows an example of the configuration of a window scanning section 11' for scanning the window 5. As shown in FIG. In the figure, six latch circuits 91 to 96 and two delay circuits 97 and 98 such as shift registers are used to realize rask scanning of the window 5. Brightness data d0 to d8 for pixels A0 to A are simultaneously extracted and provided to the illegal pattern detection section 12'.

It goes without saying that other components of the window scanning section 11' include an image memory 24, an address generator 19, etc., as shown in FIG. Also, 2 of this invention
Similar to the conventional example shown in FIG. 6, the valorization threshold calculation device is controlled by a microcomputer 10, and includes, in addition to the window scanning section 11' and illegal pattern detection section 12',
Illegal pattern histogram generation section 13, brightness histogram main generation section 142, image quality histogram generation section 15. Histogram smoothing processing unit 16. It is equipped with an optimum threshold value search section 17.

Returning to FIG. 1, the illegal pattern detection unit 12' detects brightness data d of each pixel A0 to A within the window 5. -d,
The brightness data d0 of a specific pixel (center pixel) Ao is transmitted from the first latch circuit 70 to the second and subsequent latch circuits 71 to 7.
From 8, brightness data d1 to d of other pixels A1 to As,
are input to eight comparison circuits 80 to 87, respectively.

In addition to the above-mentioned brightness data, these comparison circuits 80 to 87 also contain the contrast (
Two parameters rM and rL (where rH>rL) that are proportional to the difference between the brightness of the image portion and the brightness of the background portion are input. As a result, each comparison circuit 80 to 87 outputs brightness data d for each pixel A=(i・1,2.3...,8) and brightness data d0 for the specific pixel A0.
The difference (dt ao) is -rt <dt -d.

If <-rH, "O" is output as binary data f, and otherwise, "1" is output as binary data f to the detection logic unit 50.

FIG. 4 shows the relationship between the difference in brightness data (di −aO) and the outputs of the comparison circuits 80 to 87, and (d
If the value of i do) is within the region S, the comparison output is rQJ
If it is outside the area S, the comparison output is "1". Note that the difference in brightness data (di-do) is at least -25 because the brightness data takes any value of 256 tones.
It takes a value from 6 to a maximum of 256.

The detection logic section 50 receives the binary data fi given from each of the comparison circuits 80 to 87, and the binary data f regarding the specific pixel A0 (however, fo is always set to "1").

), it is determined whether the image in the window 5 is an illegal pattern or not, and if it is an illegal pattern, a flag of "1" is output, and if it is a legal pattern, a flag of "0" is output.

In this way, when the two parameters rH and rL, which are proportional to the contrast of the target image part, are used in the illegal pattern detection process, the difference in brightness data (at -do) is outside the area S in Figure 4, and the binary image in window 5 becomes an illegal pattern, making it impossible to obtain a binarization threshold that clearly extracts only the target image part. can.

Assuming a grayscale image 1 including an image portion 2 with high contrast and an image portion 3 with low contrast as shown in FIG. 13, the difference in brightness data for the constituent pixels of each image portion 2.3 ( dz da) has a distribution as shown in FIG. 5 (3).

Therefore, when the image portion 2 with high contrast is targeted, the parameter r14+rL is set to a large value as shown in FIG. 5(2). As a result, the difference in brightness data (di-do) at the contour of the image portion 3 with low contrast is outside the area S in FIG. 5 (2), and the binary image within the window 5 becomes an illegal pattern. ,
It is possible to obtain a binarization threshold that clearly extracts only image portion 2.

Furthermore, when the image portion 3 with low contrast is targeted, the above-mentioned parameter rM+rL is set small as shown in FIG. 5(1). As a result, the difference in brightness data (di do) at the contour of image part 2 with large contrast is outside the area S in Figure 5 (1), and the binary image in window 5 becomes an illegal pattern. , it is possible to obtain a binarization threshold that clearly extracts only image portion 3.

Note that the above parameter r)l+rL may be manually inputted as an appropriate value based on the brightness distribution of the pixels, or may be automatically determined and inputted by fuzzy inference or the like.

<Effects of the Invention> As described above, the present invention extracts information regarding the suitability of an outline of an image within a window for scanning a multivalued image by determining the brightness and specific pixel values for each pixel within the window. Since the image within the window is binarized by comparing the difference in brightness with a parameter proportional to the contrast of the specific image part of interest, multiple images with different contrasts can be generated in one image. When a portion exists, by selecting the parameters according to the contrast of the target image portion, it is now possible to calculate the optimal binarization threshold for that image portion.

Furthermore, even if there is unevenness in brightness within the same image area, the binary pattern obtained by binarizing the image within the window can be prevented from being adversely affected by the unevenness, thus achieving the purpose of the invention. It has a remarkable effect.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an example of the configuration of an illegal pattern detection section used in the binarization threshold calculation device of the present invention, and FIG. 2 is an explanatory diagram showing the concept of a window set for a grayscale image. Fig. 3 is a block diagram showing a configuration example of an illegal pattern, Fig. 4 is an explanatory diagram showing the relationship between the difference in brightness data and the comparison output of the comparison circuit, and Fig. 5 is an explanatory diagram showing the parameter setting method. , Fig. 6 is a block diagram showing the configuration of a conventional binarization threshold calculation device, Fig. 7 is a block diagram showing an example of the configuration of a conventional illegal pattern detection section, and Fig. 8 is a block diagram showing the configuration of a conventional illegal pattern detection section. FIG. 9 is an explanatory diagram showing the binarization process of the image in the window. FIG. 10 is an explanatory diagram showing an example of a legal pattern.
FIG. 1 is an explanatory diagram showing an example of an illegal pattern, FIG. 12 is an explanatory diagram showing an image quality histogram, and FIG. 13 is an explanatory diagram showing an example of an image including image parts with different contrasts. 11...Window scanning unit 12...Illegal pattern detection unit 13...Illegal pattern histogram generation unit 14'
... Brightness histogram generation section 15 ... Image quality histogram generation section 17 ... Optimal threshold value search section

Claims (1)

[Claims] An apparatus for calculating a threshold value for binarizing the entire multivalued image by focusing on an arbitrary image part, the apparatus comprising: setting a window for the multivalued image and scanning the multivalued image; a first information extraction means for extracting information regarding the brightness of the image within the window; and a second information extraction means for extracting information regarding the suitability of the contour as an image for the image within the window; information generating means for generating third information giving an image quality evaluation standard from each piece of information from the first and second information extracting means; and calculating an optimal threshold value giving the best image quality based on the third information. threshold value calculation means, the second information extraction means compares the difference between the brightness of each pixel in the window and the brightness of a specific pixel with a parameter proportional to the contrast of the image portion of interest. and a judgment circuit that judges whether or not the binary pattern obtained by the binarization circuit is appropriate as an image pattern of the contour part. Image binarization threshold calculation device.