JPH0233671A - Binarizing device - Google Patents

Abstract

PURPOSE:To detect even a pattern having a small gradation difference at a high speed by obtaining the difference between the gradation value of a noticed picture element of an arithmetic gradation picture and the gradation value of a picture element near the noticed one to compare the obtained difference with the threshold value and deciding the binarizing level of the noticed picture element based on the result of the comparison. CONSTITUTION:A TV camera 4 photographs a work (object) 1 which is carried on a convayor 2. A position sensor 3A detects that the work 1 arrived at a prescribed image pickup position and sends the detection output to a control circuit 18. The sensor 3A also sends a picture fetching signal to an analog switch 6. The signal received from the camera 4 divides one screen into 512X512 picture elements by a picture dividing circuit 8 via the switch 6 and an A/D converter 7 and stores the gradation information on each picture element into a 2-dimensional local memory 9. The output of the memory 9 is inputted to an arithmetic circuit 10 for calculation of the gradation value of each picture element. Then the difference of gradation value is obtained between a noticed picture element and its peripheral one by the difference circuits 13A-13N via a memory 9A and the gradation value memories 12A-12N. This difference of gradation value is compared with the threshold value given from a circuit 15. When said difference exceeds the threshold value, logic 1 is outputted and a logic circuit 17 secures an OR to detect the flaws.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a binarization device that converts an output from an image sensor such as a television camera into a grayscale signal and then binarizes the signal.

[Conventional technology]

Conventionally, this type of 2fiff conversion method includes a fixed 2fiff conversion method that compares with a predetermined fixed threshold, and a differential 2fiff conversion method that compares the image signal with a fixed threshold after differentiating it. It is known as a typical example.

[Problem to be solved by the invention]

However, although these methods are excellent in terms of simple configuration and high-speed processing, they have a fatal drawback in that they cannot detect small differences in shading.

For example, considering the detection of dirt and scratches, the conventional method described above can effectively detect dark black spots, dirt, etc. on white paper, even if they are quite small. However, it is impossible to detect stains with a small density difference, such as pale ink stains and yellow stains as f3'lJ, even if the defect size is very large.

That is, when detecting defects, it is natural to detect dark and light shading defects from large to small, but even light and light shading defects that are large in size should naturally be detected. However, with the above-mentioned conventional method, it is almost impossible to detect defects with light and light shading. In other words, the background to the general use of the conventional methods described above is that, from the perspective of defect inspection, etc., the workpieces to be lost are conveyed continuously and at high speed on a conveyor, etc. Time is several to 20 pieces per second
(It can be said that this purpose has been met, especially since it requires very high speed, at a B+ level.

Therefore, it is an object of the present invention to provide a binarization device that can detect patterns that are large in size but have a small difference in density at high speed.

[Means to solve the problem]

an image capturing means for capturing an image of a moving object; a first image extracting means for obtaining an original grayscale image by at least analog/digital conversion of the image signal for each pixel; The second step is to obtain a calculated grayscale image by calculating the sum or sum of products of the grayscale values and using this as the calculated grayscale value for the focused search for all pixels of the original grayscale image.
and a calculation means for calculating the difference between the grayscale value of the pixel of interest and the grayscale value of each neighboring pixel in the calculated grayscale image, and comparing the difference value with a predetermined threshold value. First, the binarization level of the pixel of interest is determined.

[Effect]

Now, if the noise level of the imaging signal is ±Δn and the true signal of the defect to be detected is δ, then the observed value is δ±Δn. Therefore, when the defective part is larger than the two-dimensional local space of ITI I X m 1, if a simple sum without weighting is calculated for N pixels in the space, the signal of the defective part can be approximated by Nδ, while The noise level is (±Δn
/N'/'').

As a result, the S/N (signal-to-noise ratio) ratio is improved by N3/2, and an image of very good quality can be obtained. Furthermore, since the same calculation is performed on the defect-free portion, a high-quality image can also be obtained from the defect-free portion. If you detect flaws by calculating using a calculated grayscale image with a high S/N ratio, the S/N
The ratio improves in proportion to Ns/x. For example, N = 9
(rn 1-5), 27 times, N = 25
When (rn 1 = 5), the detection sensitivity is improved by 125 times. Therefore, compared to a method in which A/D-converted pixels are compared and binarized, the sensitivity of the binarization is significantly improved, and even a pattern with a slight difference in shading can be detected.

[Real IAf]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a time chart showing the image capture timing ff1j, and FIG. 3 is a diagram for explaining an example of the relationship between a target pixel of an original grayscale image and its surrounding pixels. Figure 4 is an explanatory diagram for explaining the relationship between the original grayscale image and the calculated grayscale image, and Figure 5 is an explanatory diagram for explaining the relationship between the pixel and its neighboring pixels. The explanatory diagram, FIG. 6, is a circuit diagram showing a specific embodiment 1 of the logic circuit shown in FIG. 1. In addition, in Fig. 1, 1 is a workpiece (object),
2 is a conveyor, 3 and 6A are position sensors, and 4 is a television (T
V) Camera, 5 is an amplifier, 6 is an analog switch, 7 is an analog/digital (A/D) converter, 8 is a screen division circuit, 9 and 9A are two-dimensional local memories, 10 is a product-sum calculation circuit, 11. 12A to 12N are calculation gray value memories, 13
A to 13N are differential circuits, 14A to 14N are comparison circuits, 1
5 is a threshold storage circuit, 16 is a calculation result storage circuit, 17
is a logic circuit, and 18 is a control circuit.

A two-dimensional imaging device 4 such as a TV camera constantly images a predetermined position of the conveyor 20. Position sensor 6°3A is work 1
reaches a predetermined position within the field of view of the TV camera 4, and sends its output to the control circuit 18. The control circuit 18 creates an image capture signal as shown in FIG. 2(c) from the TV camera synchronization signal as shown in FIG. 2(a) and the position sensor signal as shown in FIG. 2(b). TV
The signal from camera 4 is amplified to an appropriate level by amplifier 5. The analog switch 6 outputs 1Dl! according to the input signal shown in FIG. 2 (c). Only the signal on the i1 side is sent to the subsequent stage A.
/ Output to D converter 7. In Komome, there is only one T screen.
Since the V right camera number is treated as source information, it can also be applied to moving objects.

A/D conversion converts the 7V left camera g into, for example, an 8-bit digital signal, divides this into, for example, 512 pixels horizontally x 512 pixels vertically, and then divides it into two as the a-light information of each pixel. Send to dimensional local memory 9. The two-dimensional local memory 9 is 8X (512X rn,
) is constructed by cascading m4 stages of shift registers, and simultaneously outputs the gradation information of the pixel of interest and the gradation information of its surrounding pixels. In Figure 6 (a) m1=6,
(b) shows the relationship between the pixel of interest ro and the surrounding pixel "1" when ml-5 is set.

The output of the two-dimensional local memory 9 is given to the sum-of-products calculation circuit 10, which includes the grayscale value Sro of the pixel of interest and the grayscale value Sr1 of the surrounding pixels.
Therefore, for example, for Figure 5 (a), the following equation (1),
In addition, for the same figure (b), the following equation (2) is calculated, and the calculated gray value S is obtained. Find ′.

1 to So'- 2 to Sro+ (s4.+sr2+s, 5+
sr4” Sr5” Sr6” Sr7” Sr
8) ,,,,,, (1)+5r14+5r15
+5r16+5r17+5r18+5r19+5r20
) -- (2) Since such operations are performed on the entire area of the original immersion image, as in the fourth prisoner, the original image security is required. A new calculated grayscale image IC is obtained by multiplying by the function F. Image artist. and engineering. The X, y coordinate relationship is one-to-one, and the shading value is S. GaraS. '. Note that in order to simplify the calculations of equations (1) and (,2), k1=k2−k5=1 may be used. In any case, the calculated grayscale image I is output from the product-sum calculation circuit 10. The gray value of each pixel in is output.

The output of the product-sum calculation circuit 10 is a second two-dimensional local memo 'J
Given to 9A. The two-dimensional local memory 9A is the calculation density 1
tTj image■. It is assumed that the local memory is lj, the local space is lT12 (for example, rn2 = 5), and the gray value is approximately 12 pits. From the output of the two-dimensional local memory 9A, for example, as shown in FIG. 5, the gray value S1 of the pixel of interest Ro. The gray value S81 of the neighboring pixel Ri (R, to Ra) is output. The gray value S of the pixel of interest R6. is stored in the memory 11, and the gray level nur of the neighboring pixel Ri is stored in the memo IJ12A to 12N.

The difference circuits 15A to 13N are connected to each near E cal [! Difference SR, -S□ for II element. The comparison circuits 14A to 1
4N compares each difference with the threshold α given from the threshold storage circuit 15, and outputs logic ``1#'' when, for example, 5RI-8RO>''. The result is memory circuit 1
6, -d is stored. The logic circuit 17 performs a logical sum on the output of the memory circuit 16, for example, and binarizes the output with a logic "1" as a flaw. Instead of the logic circuit 17 that takes the simple OR, it is also possible to take the AND (AN1 to AN4) for each direction and then take the OR (OR) as shown in FIG. In this way, even simple logical sums can be made to work effectively.

Since the above processing is performed on the entire area of the calculated grayscale image, flaws can be detected with high S/N and high sensitivity.

FIG. 7 is a block diagram showing another embodiment of the invention.

This is an image memory 21. instead of the two-dimensional local memory 9 shown in FIG. Two-dimensional local gray value memory 22. Memory control circuit 23. By providing the address calculation circuit 24 and the point-of-interest coordinate generation circuit 25, the two-dimensional local memory 9
A calculation image memory 22A, a memory control circuit 25A, an address calculation circuit 24A, and a point-of-interest coordinate generation circuit 25A are provided in place of A, but the rest is the same as that shown in FIG. In addition, 26 is an arithmetic control circuit,
As a result, calculations such as υ difference and comparison are performed sequentially (serially) for each pixel.

〔Effect of the invention〕

b) For the original grayscale image obtained from the TV left camera, the product sum (or sum) of the grayscale values of the pixel of interest and its two-dimensional surrounding N pixels
Since the operation of determining the calculated gray value of the pixel of interest as the calculated gray value of the pixel of interest is performed for all lff1l elements of the original gray image, it is possible to obtain a calculated aS gray image with a high S/N ratio.

(mouth) Since we have made it possible to perform double conversion based on the densities, the sensitivity has been significantly increased compared to the conventional method (for example, 27 to 1
25 times).

C) Calculation α Regarding the light and bright image area, we calculated the difference in light and brightness between the focused wJ element and the neighboring pixels, and binarized this using the threshold value as a ratio, so the gray value of the normal part Even if there is a gradual change in the value, this can be ignored and normal parts will not be erroneously binarized.

2) Since only one screen immediately after the workpiece arrives within the field of view of the TV camera is targeted for processing, there is no effect even if the workpiece moves.

e) Processing can be performed in a pipeline system, which increases speed.

As a result, it is now possible to detect defects with extremely high speed and accuracy, even defects that are large in size and have faint shading, which was previously considered difficult.

[Brief Description of the Drawings] Fig. 1 is a block diagram showing an embodiment of the present invention, Fig. 2 is a time chart showing an example of image capture timing, and Fig. 3 is a block diagram showing an embodiment of the present invention.
The figure shows the original gradation image. An explanatory diagram for explaining an example of the relationship between the A element and its surrounding pixels, an explanatory diagram for explaining an example of the relationship between the fourth and neighboring fg pixels, and FIG. 6 shows a specific example of the logic circuit shown in FIG. 1. The circuit diagram, FIG. 7, is a block diagram showing another embodiment of the present invention. Code explanation 1...Work, 2...Conveyor, 5,
3A...Position sensor, 4...TV camera, 5...Amplifier, 6...Analog switch, 7...A/D conversion Vessel, 8...
Screen division circuit, 9, 9A... Two-dimensional local memory, 10... Product-sum calculation: circuit, 11, 12.12
A~12N・・・・・・Calculated density (direct memory, 13.1
3A... Calculation results 1. is a circuit, 17...music logic circuit, 21...image memory, 22...music two-dimensional local density memory, 22A...concave calculation image memory,
23, 23A... Memory control circuit, 24, 24
A. Address calculation circuit, 25, 25A.. Track point coordinate generation circuit, 26.. Calculation control circuit. Agent Patent Attorney Akio Namiki

Claims (1)

[Scope of Claims] Imaging means for capturing an image of a moving object; first image extraction means for obtaining an original grayscale image by at least analog/digital conversion of the image signal for each pixel; and a pixel of interest from the original grayscale image. and a second image extracting means for obtaining a calculated gradation image by performing an operation for all pixels of the original gradation image to obtain the sum or product sum of the gradation values of a predetermined number of surrounding pixels and to use this as the calculated gradation value of the pixel of interest. , calculating means for calculating the difference between the gray value of the pixel of interest and the gray value of each neighboring pixel in the calculated gray image and comparing the difference value with a predetermined threshold; A binarization device that determines a binarization level of.