JPH0713836B2 - Binarization circuit - Google Patents

Description

The present invention relates to an automatic visual inspection apparatus for inspecting scratches and shapes of various parts (workpieces) by utilizing pattern recognition technology, and more particularly to a binarization circuit thereof. Is.

[Conventional technology]

Conventionally, as a circuit of this type, a two-dimensional imaging device (for example,
A fixed binarization circuit that compares the signal from the TV camera with a fixed threshold value after appropriately amplifying the output of the TV camera),
A floating binarization circuit which compares a signal from a camera with a camera signal with a signal obtained through a CR incomplete integration circuit as a threshold value is known as a typical one.

[Problems to be Solved by the Invention]

By the way, in a scratch inspection or the like, a work is generally conveyed continuously by a belt conveyor or the like, and in many cases, the defective portion is small and the contrast between the defective portion and the normal portion is small. Therefore, the signal obtained from the TV camera has a small contrast between the normal portion and the defective portion, and there is a big problem that it is often difficult to detect these scratches by the conventional binarizing circuit. It should be noted that, as a typical defect that is difficult to detect from the past, there is hair attached to a food or drug container, or scratches caused by metal or the like.

Therefore, it is an object of the present invention to provide a binarization circuit that can detect a scratch, which has been conventionally difficult to detect, with high accuracy even when a work continuously moves.

[Means for Solving the Problems]

Since mainly moving objects are targeted here, only the signal of one frame when the work enters the field of view of the TV camera is valid and is the target of judgment. That is, the signal of the TV camera is divided into pixels, the signal of the TV camera is A / D converted (for example, 8 bits) for each pixel, and is captured as a grayscale image. Etc. to emphasize the contrast. As a signal enhancement method, attention is paid to the fact that typical scratches that have been difficult to detect in the past have linear characteristics, that is, they have directionality, and if there are linear shades, this method is emphasized. I am trying. Therefore, the pixel of interest
The shade value S _{0 of} ij and the shade value S ₁ of pixels around the secondary neighborhood of m × m
~ Sm is divided into a combination of four (linear) pixels in four directions, and the product-sum operation is performed for each direction, and the value with the smallest grayscale value (or the largest) is calculated among the calculated values in the four directions. It is detected and the value is emphasized as the shade of the point of interest ij. This operation is executed for all the images to obtain contrast-enhanced images. In this two-dimensional local surface m × m, the product sum of the shades of a plurality of (linear) pixels in four directions is calculated, and the minimum shade (in the case of black flaws) or the maximum shade (in the case of white flaws) in the four directions is calculated. ). Therefore, when a flaw is present, it has the effect of integrating the flaw concentration along the direction of the flaw, and has the effect of reducing noise for the defect-free portion, improving S / N and enhancing contrast. Done.

This contrast-enhanced signal is used as the density of the target pixel ij, and the density is calculated with surrounding neighboring pixels to detect flaws. At this time, the neighboring neighboring pixels are used as the density of the image before contrast enhancement at the same pixel address. Use grayscale data. This is because in order to detect flaw detection with high precision, the contrast-enhanced image needs only the point of interest ij,
This is because if the peripheral neighboring pixels with respect to the point of interest are not subjected to contrast enhancement, the effect of enlarging the pattern is not obtained and a good effect is obtained. In particular, it has a remarkable effect when the pattern of the non-defective product is a complicated shade pattern. For this reason, the grayscale operation is performed on the n × n two-dimensional local space between different memories. As a method, the difference density is obtained for each direction by the focus point and the grayscale values on both sides of the focus point, and the direction is focused. So that the scratches are binarized. Furthermore, it is desirable to confirm that a valley (in the case of a black flaw) or a mountain (in the case of a white flaw) is formed in the light and shade on both sides of the point of interest.

[Action]

(1) For a grayscale image obtained via the imaging device, m
The two-dimensional local space of the × m two-dimensional local memory is divided into a plurality of directions around the point of interest, and the sum of the gray values of the plurality of pixels N is calculated in each direction. Therefore, if there are any scratches, the contrast of the scratches will be contrast-enhanced at a depth of N times, and if there is no defect, noise will be reduced. , And noise is removed at the same time as signal enhancement, resulting in high S / N.
Becomes

(2) Since the sum of products of a plurality of pixels is obtained for each direction, the contrast is expanded along a linear flaw that has been difficult to detect in the past, and the detection sensitivity is significantly improved as compared with the conventional method that relies only on the density of a single pixel. .

(3) Since the pixels are arranged in a plurality of directions (four directions), the contrast is enhanced regardless of the direction of the scratch.

(4) As the data of the target pixel, the gray value of the target point ij is the gray value that gives the minimum gray value (in the case of black flaws) or the maximum gray value (in the case of white flaws) of the product-sum gray value obtained for each direction. Since it is a value, it is possible to enhance the contrast of scratches with high accuracy.

(5) Since the above contrast enhancement is performed using the grayscale two-dimensional local memory, the contrast enhancement can be performed over the entire effective screen area.

(6) The contrast-enhanced image is set as the shade of the point of interest ij, and ij of the image before the peripheral-enhanced pixels are subjected to the contrast enhancement
Since the density difference is detected by using the densities of the peripheral neighboring pixels, the effect of enlarging the density pattern by the contrast enhancement is not exerted, and the flaw detection can be stabilized.

(7) Since the processing of item (6) is performed using the two-dimensional local memory, stable detection can be performed over the entire effective screen.

(8) Since the density difference detection is also performed for each direction in the flaw detection calculation, it is possible to detect a linear flaw with high accuracy.

(9) Since valleys or ridges are detected based on the relationship with the pixels on both sides of the point of interest, remarkably highly accurate detection can be performed.

(10) Since the above processing is performed only for the signal of one frame in which the work enters the TV camera, the same effect can be obtained for a moving object.

(11) High-speed processing becomes possible.

[Embodiment] FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is a conveyor, 2 is a work (article), 3 is a position sensor, 4 is a determination timing circuit, 5 is a TV camera, and 6
Is an amplifier, 7 is a screen division circuit, 8 is an analog / digital (A / D) converter, 9 and 9A are grayscale two-dimensional local memories, 10 is a product-sum calculation circuit for each direction, and 11 is a minimum / maximum grayscale value detection circuit. ,
Reference numeral 12 is a two-dimensional spatial density calculation circuit, 13 is a comparison circuit, 14 is a threshold value setting circuit, 15 is a theoretical circuit, and 16 is a recognition circuit.

First, the TV camera 5 constantly images the conveyor 1 and the position sensor 3 detects that the work (article) 2 has arrived at an appropriate position within the field of view of the TV camera. The signal of the TV camera (imaging signal) is always amplified to an appropriate level via the amplifier 6. In the screen division circuit 7, the TV camera signal is spatially discretized into a large number of pixels (for example, 512 × 512) by a scanning line of the TV camera and a high-frequency clock of, for example, 12 MHz synchronized with the scanning line, and at the same time, an analog signal corresponding to each pixel is obtained. Sample and hold the signal. The judgment timing circuit 4 receives the signal from the position sensor 3 and knows that the work has entered the field of view of the TV camera.
It is output as a valid period signal only during the frame period. Circuit 7
In response to this command, outputs a signal only for one frame period. The A / D conversion circuit 8 multi-values the gradation of the TV camera into, for example, 8 bits to obtain a gradation image. This grayscale image data is m
The density of the two-dimensional local surface is output in parallel by inputting it to the density two-dimensional local memory 9 having a density of 8 bits of xm (ex.3x3). FIG. 2 shows an example of the local space in the case of 3 × 3. It should be noted that S _{0 to} S ₈ in the figure indicate the grayscale value of each pixel.

Next, by the direction-wise product-sum calculation circuit 10 and the minimum / maximum gray value detection circuit 11, the density S _{0 of the} point of interest ij and the density of the neighboring pixels
A predetermined calculation is performed from S _{1 to} S ₈ to obtain a contrast-enhanced image Ic (i, j). This operation is performed for the entire effective screen area. This is done automatically by using the known two-dimensional local memory. This idea is illustrated in Fig. 3 (a) and (b), which shows the original image I ₀ (i, j).
On the other hand, the target point ij of I ₀ and the neighboring pixels (I
The density of ₀ (illustrated as i ± Δx, j ± Δy) is multiplied by the function F ₁ to obtain the contrast-enhanced image Ic (i,
j) is shown.

In the present invention, as a concrete function of F ₁ , first, an m × m two-dimensional local space is shown in FIG. 4 as AA ′, BB ′, CC ′, D−.
As indicated by D ′, the pixel is divided into a plurality of (linear) pixels in four directions centering on the point of interest. A plurality of pixels for each direction of the four-direction (linear) pixels (3 pixels for 3 × 3 pixels)
Calculate the sum of products for the concentration of. Examples of the product-sum formula include the following formulas.

A-A ': S ₁ + kS ₀ + S ₅ = Sij _A BB-: S ₂ + kS ₀ + S ₆ = Sij _B C-C': S ₃ + kS ₀ + S ₇ = Sij _C D-D ': S ₄ + kS ₀ + S ₈ = Sij _D (1) Note that k is a weighting coefficient, and for example, 2 is adopted. Since the direction is predetermined, the pixel combination is also uniquely determined. Therefore, as shown in FIG. 5, for example, the density values of the grayscale two-dimensional local memory 9 are latched by the grayscale extraction units 21A to 21D for each direction, and the sum of products calculation units 22A to 22D calculate the sum of products for each direction. The calculation of the equation (1) can be easily performed. The circuit 11 detects the minimum density (in the case of a black defect) or the maximum density (in the case of a white defect) of the four sets of density values calculated in this way, and sets it as the contrast-enhanced image of the point of interest ij. In this way, by multiplying and summing a plurality of (linear) pixels (N pixels) for each direction, if there is a flaw, the density is integrated along the flaw, so that the flaw density increases and the noise in the defect-free portion is reduced. Decrease. Therefore, the contrast is enhanced, and at the same time, a low noise screen is formed. Contrast enhancement can be performed irrespective of the directionality of scratches by performing the sum of products of the (pixel) pixels for each direction in four directions and taking the minimum (or maximum) density from the four sets. In the above, 3 × 3 is illustrated as m × m, but 5 × 5 may be used, and an example of the direction division is shown in FIG.

As described above, the circuit 11 outputs the contrast-enhanced image Ic (i, j).

Next, a flaw detection method will be described. In order to detect flaws, the presence or absence of flaws is detected based on the difference in density between the point of interest and its neighboring pixels. At this time, in the present invention, the gray value of the point of interest ij is set to the gray value of Ic (i, j), and the neighboring pixels i ± Δx ′, of the original image I ₀ (i, j) at the same address are set as the neighboring pixels. The density value of j ± Δy ′ is adopted. The concept is shown in Fig. 3 (c), where the gray level calculation function for flaw detection is F ₂ .

As a concrete function, first, consider a 5 × 5 space as shown in FIG. 7 as an n × n two-dimensional local space for flaw detection,
8 pixels in the vicinity of the inside are taken discretely as shown in FIG. The two-dimensional local calculation is applied to the original image I ₀ (i, j). In FIG. 1, pixels other than the pixel of interest are input from the memory circuit 9 to the two-dimensional local memory circuit (II) 9A, and
For 8 discrete pixels from the output of the dimensional local memory 9A,
The density of peripheral neighboring pixels is input to the two-dimensional space calculation circuit 12 for flaw detection. On the other hand, the gray value detection circuit 11 to the circuit 12
The density of the target pixel ij is input to. It should be noted that the positional relationship between the target point of the circuit 12 and the neighboring pixels is adjusted so that the correspondence relationship does not change like the relationship between the target point of one image and the peripheral pixels. Then, in the two-dimensional spatial density calculation circuit 12, the neighboring points are combined in four directions for each direction, and the grayscale difference calculation (ex.
Perform S ₁ ′ + S ₅ ′ −Ic (i, j)). The comparison circuit 13 compares the gradation differences of these four groups with the threshold value α, and when the difference is larger than α, it is output as a logic “1”. This is checked in the logic circuit 15 in four directions, and if it is established in either direction, ij
Pixels of interest are output as scratched. Therefore, the circuit 15 is called an OR circuit, and the output of the logic circuit 15 is sent to the recognition circuit 16 as a binarized signal, and the scratches are recognized.

As another example of the two-dimensional space calculation for detecting the flaw, for example, as shown in FIG. 9, two pixels on both sides of the target point are provided.
For S ₁ ′ and S ₅ ′, a method of detecting a flaw when the density of the point of interest is higher than the density of two pixels on both sides by a predetermined threshold value α or more depending on the direction may be considered. By this operation, it is checked that the point of interest is a valley (black scratch) for each direction. This valley detection is executed in four directions, and if the condition is satisfied in either direction, the pixel ij is output as a flaw.

〔The invention's effect〕

(1) For the original image, a product-sum operation focusing on four (direction) pixels in each direction is performed, the defect density is expanded along the direction of the defect, and noise in the defect-free portion is reduced. Since the contrast enhancement calculation is performed, it becomes easy to detect a flaw which has been difficult in the past.

(2) In the two-dimensional local space calculation for flaw detection, the image of the point of interest ij uses the data of the point ij of the contrast-enhanced image, and the neighboring pixels include the image I ₀ (i, Since the pixel data corresponding to j) is used and the effect of expanding (expanding) the low density difference part by the contrast emphasis is not in the peripheral neighboring pixels, it is possible to detect with high accuracy even for a pattern having a complicated shade state. it can.

(3) In the density calculation for detecting scratches, the calculation is performed by giving directionality to the two-dimensional local surface.
Binarization along the direction of scratches is possible.

(4) From the above, it is possible to detect hair and scratches that have been difficult in the past with high accuracy. Further, it has been confirmed that it also works effectively for detecting low-contrast shapes such as engraved characters.

(5) Since the signal of only one frame of the TV camera is used, it is possible to reliably and highly accurately detect a moving object.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 is a block diagram.
FIG. 3 is an explanatory view for explaining a two-dimensional local space of × 3, FIG. 3 is an explanatory view for explaining the concept of the contrast enhancement calculation and the two-dimensional spatial density calculation in the present invention, and FIG. 4 is shown in FIG. FIG. 5 is an explanatory view for explaining an example of direction allocation of the two-dimensional local space shown in FIG. 5, FIG. 5 is a block diagram showing a concrete example of the product-sum calculation circuit shown in FIG. 1, and FIG. 6 is a two-dimensional local space of 5 × 5. And FIG. 7 is an explanatory diagram for explaining an example of direction allocation, FIG. 7 is an explanatory diagram for explaining an example of a two-dimensional local space used for flaw detection, and FIG. 8 is a conceptual diagram showing an outline of a two-dimensional spatial density calculation. FIG. 9 is a conceptual diagram showing the outline of another example of the two-dimensional spatial density calculation. Reference numeral 1 ... Conveyor, 2 ... Work, 3 ... Position sensor, 4
...... Judgment timing circuit, 5 …… TV camera, 6 …… Amplifier, 7 …… Screen division circuit, 8 …… Analog / digital (A / D) converter, 9,9A …… 2D local memory, 10 ・ ・ ・… Side-by-direction product-sum operation circuit, 11 …… Minimum / maximum gray value detection circuit,
12 …… two-dimensional spatial concentration calculation circuit, 13 …… comparison circuit, 14 ・ ・ ・
... Threshold setting circuit, 15 ... Logic circuit, 16 ... Recognition circuit, 21A-21D ... Gray level extraction section, 22A-22D ... Sum of products operation section.

Claims (2)

[Claims] 1. An image pickup means for picking up an image of an object, a timing control means for making an image pickup signal effective when the object enters the field of view of the image pickup means, and an analog / digital conversion of the image pickup signal in pixel units. A two-dimensional local space is divided into a plurality of directions from the original image extraction means that obtains the original grayscale image, and the grayscale information of the point of interest and the grayscale information of pixels around the point of interest for the original grayscale image, and the sum of products operation is performed for each. Emphasizing image extraction means that obtains a contrast-enhanced image by performing processing for all pixels of the original gray-scale image by using the density that is the smallest or the largest of the direction-wise product sum values as the contrast-enhancing density. And a calculation means for calculating density for each two-dimensional local space using corresponding pixels of the original gray image as peripheral neighboring pixels using an image, and comparing the calculated value with a predetermined threshold value. Binarizing circuit, characterized in that binarization. 2. The binarization circuit according to claim 1, wherein:
The calculating means has a density difference calculating means for obtaining a density difference for each direction focusing on at least a plurality of directionalities, and is significant when there is at least one direction in which the density difference is equal to or more than a threshold value. Binarization circuit to do.