JPS5816837B2 - pattern detection device - Google Patents

Description

[Detailed Description of the Invention] The present invention detects scratches, holes, etc. on the surface of objects such as engraved characters and solder overlays, and other surface parts that have different properties from those on the ground surface. The present invention relates to a pattern detection device that extracts a pattern on a surface portion.

Conventionally, in order to detect scratches and holes existing on the surface of an object, a binary pattern is obtained by thresholding the pattern detected from the surface of the object to be detected, and by processing this binary pattern appropriately. Detection of scratches and holes has been carried out, but since the resulting binary pattern often contains unnecessary patterns in addition to the desired pattern, it is difficult to accurately detect scratches and holes. The disadvantage is that it cannot be done.

An object of the present invention is to detect patterns caused by small holes etc. from among the detection patterns even when unnecessary patterns are included in the detection pattern obtained from the detection target, and to reliably detect small hole defects. Another object of the present invention is to take a step further and obtain a pattern detection device capable of extracting patterns caused by defects such as small holes.

For this purpose, the present invention provides contrast information as a binary signal obtained by thresholding an input image signal, and thresholding the input image signal after spatially differentiating it. It is characterized by a configuration in which defects such as small holes are detected by logically multiplying the obtained differential information.

Defects here are broadly defined to include the small holes and cracks mentioned above.Generally, there is a sudden change in brightness in areas where defects occur, so if we detect this sudden change by differential processing, This means that defects can be detected reliably.

For example, since small holes and cracks have a concave cross section, they suddenly turn from a bright state to a dark state and then return to a bright state, so if this is detected, small holes and cracks can be detected.

Further, the present invention reads binary pattern images including isolated patterns labeled in the order of pattern occurrence from the memory means in the order in which they are stored, and in synchronization with this readout, performs differential processing on the input image signal and then performs threshold processing. The present invention is characterized by a configuration in which the differential information is obtained by using the isolated pattern, and when the differential information is obtained while the isolated pattern is being read out, the isolated pattern is extracted as a pattern due to a defect.

The present invention will be explained below with reference to the accompanying drawings.

First, the peripheral technology will be explained according to FIGS. 1 to 3.

FIG. 1 shows the shape of a soldering surface after a component is inserted into a printed circuit board to be detected.

As shown in the figure, a solder overlay 3 exists as an electrical connection medium around the pins 1 and 2 of the component inserted into the printed board, but small holes 4 and cracks 5 occur in this solder overlay 3. If there are any defects, they affect the reliability of the solder build-up portion 3, so it is necessary to detect them as defects in advance as much as possible.

To detect these small holes 4 and cracks 5, for example, the second
This is done as shown in the figure.

That is, the surface of the object to be detected 80 is imaged by the TV camera 6 as a detector through the imaging lens 7, and the imaged signal is subjected to threshold processing in the signal processing circuit 11 to form a second pattern, and this binary pattern is This is done by displaying on the TV monitor 12.

In this case, it is preferable that the surface of the detection target 8 be illuminated with uniform brightness.

This is because even if the object to be detected 8 has a glossy surface like the solder overlay portion 3, if it is illuminated in this way, a relatively homogeneous pattern can be detected.

For example, uniform illumination can be achieved by using a ring-shaped light source 7 and a diffuser plate 1 as shown in the figure.
By 0.

When determining the quality of a pattern detected in this manner by performing counting processing, threshold processing is performed.

The threshold processing here includes those using a fixed threshold, a floating threshold, or a variable threshold based on histogram modification, and by such threshold processing, a binary pattern of the detection target 8 is obtained. It is something.

FIG. 3 shows an example of a binary pattern obtained in this manner, in which portions where the level of the imaging signal is high are shown as blank areas, and portions where the level of the imaging signal is low are shown as hatched areas.

When the detection target 8 is the soldering surface shown in FIG. 1, the solder build-up portion 3 is generally obtained as a blank pattern, and the small holes 4 and cracks 5 are obtained as diagonal line patterns 15 and 16, as shown.

However, if the tops of the pins 1 and 2 are smooth horizontal surfaces, or if the solder mound at the center between the pins 1 and 2 is a horizontal surface, those portions can also be obtained as the diagonal line pattern 13° 14.17. .

This is because the soldering surface is illuminated with uniform brightness, but since the imaging lens 7 is located above the pins 1 and 2, it is difficult to see from the direction of the imaging lens 7. This is because there is no illumination light.

In this example, the lighting method shown in the diagram is used.
When an object with a glossy surface such as a solder build-up part is to be targeted, a similar problem arises in that even with other illumination methods, a portion originally obtained as a blank pattern is obtained as a diagonal line pattern.

That is, the hatched pattern 1 corresponds to the small hole 4 and the crack 5.
5.16 is detected as it should be, but the hatched patterns 13, 14, and 17 are unnecessary ones that were detected inadvertently, so unnecessary hatched patterns are replaced with small holes or There is a risk that it may be mistaken for cracks, etc.

Next, embodiments of the present invention will be described with reference to FIGS. 4 to 16.

First, referring to FIG. 4, this shows the configuration of an example of a pattern detection device used to detect whether or not a pattern due to a defect exists in a detection pattern.

In this figure, the signal delay circuit for one sweep is 18.4 x 4.
The picture element detection window 19, the difference circuits 20 and 20', the absolute value circuits 21 and 21', the maximum value extraction circuit 22, and the threshold processing circuit 23 are for differentially processing the input image signal and detecting the gradient. The threshold G value processing circuit 22 outputs a pulse signal only when the absolute value of the differential output is constant and has an upper magnitude.

However, this pulse signal is not output only when a pattern due to a defect is detected, and brightness information should also be taken into consideration.

The threshold processing circuit 24 performs threshold processing on the input image signal to obtain binary contrast information, and the AND circuit 25 performs a logical product of this binary contrast information and the pulse signal described above. This makes it possible to detect patterns based on defects for the first time.

Here, each main circuit part will be explained in detail as follows.

First, as mentioned above, the threshold processing circuit 24 is for converting the input image signal into binary brightness information, and the signal level corresponding to a pattern due to a defect to be detected corresponds to another pattern. If the level of the input image signal is higher than the threshold value vT1, the output logic state is set to II, II (so-called bi-level state), and when it is lower than the threshold value VTt, the output logic state is set to 0. (so-called low level state).

Further, when the signal level corresponding to the pattern due to the defect to be detected is lower than that corresponding to other patterns, the above output logic state is reversed.

Next, explaining the differential processing of the input image signal, the input image signal is delayed by exactly one sweep at the output of the delay circuit 18 by one sweep, but this signal is delayed by the current input signal. It is input to the 4×4 pixel detection window 19 together with the image signal.

This 4×4 pixel detection window 19 has 4 pixels as shown in FIG.
The input image signal related to the delay and the currently input image signal are sequentially shifted through the corresponding registers.
The signal level of each picture element corresponding to 93 is V.

−v3, lV2 Vol, lvt VS2 calculation processing is performed by difference circuits 20, 20' and absolute value circuits 21.2
1'.

This process is a differential process, and after the maximum value extracting circuit 22 extracts the dog out of the two differential values output from the absolute value circuits 21 and 21', the threshold value processing circuit 23 uses a threshold value vT2. It is something to be processed.

The maximum value extraction circuit 22 compares two differential values using a comparison circuit 26, as shown in FIG. By doing, the OR circuit 29
is configured so that a dog differential value is always output.

The output from the maximum value extraction circuit 22 is subjected to threshold processing in the threshold processing circuit 23, and the threshold processing here is such that, for example, if the threshold value VT2T2 or more, the output state is logic 1.
Furthermore, if the threshold value VT2 is smaller than the threshold value VT2, the output logic state is set to O.

Therefore, the AND circuit 25
, 24, the AND circuit 25 outputs a logic signal only when both outputs are logic 1.
Since an output of 1 is obtained, the existence of a pattern due to a defect can be detected using this output logic state.

By the way, the differential processing described above is the simplest spatial differential processing, and the differential value V related to maximum value extraction is v=Max(1v
o-V21, IVl-VS2) However, in addition to the above, there are various difference methods and edge operators for spatial differentiation, and it goes without saying that any of them may be used.

Further, in the threshold processing circuits 23 and 24, the threshold value V
Although T2 t VT□ is fixedly given from the outside, it is not limited to this, but it can be performed by using floating threshold processing that determines the threshold value from the input image signal itself, or by obtaining the histogram of the input screen, and from this. A threshold value may also be determined.

Furthermore, the threshold processing may be performed using two types of thresholds instead of only one type.

For example, instead of the threshold value vT1, the threshold value VTIL +
When VTtH is prepared and the level of the input image signal is V, if vT1L<v<vTlH, the output logic state is set to II.
In other cases, threshold processing may be performed such that the threshold value is set to I II, and II OI+ in other cases.

Now, the present invention will be explained in more detail with reference to FIGS. 7a to 7c.

First, FIG. 7a shows the case where the small hole 4 exists as a hole in the solder build-up part 3, and the curve 30 crosses the small hole 4.
This is the position where the image signal for one sweep is detected, and FIG. 7 shows a cross section at the curve 30 position.

In this case, the image signal a for the detection position of the curve 30 is the first
As shown in Figure C, the upper surface of pins 1 and 2, the center between pins 1 and 2, and the small hole 4 are obtained with a small level
Especially in the small hole 4 part, the brightness changes rapidly.
The level of image signal a will also change rapidly.

A portion where the brightness changes rapidly can also be seen near pin 1, so if the image signal a is differentiated and the spatially differentiated signal C is calculated, threshold processing is performed using the threshold value TT2, the threshold processing circuit 23 The output d at is as shown in the figure.

That is, for the small hole 4 near pin 1, the logic II
The output state of f can be obtained.

Among these, the output state of logic II I If for small hole 4 was obtained as expected, but that for the vicinity of pin 1 was obtained inadvertently.
It will be necessary to erase it.

On the other hand, since the output of the threshold processing circuit 24 for the image signal a is as shown in the figure, it is possible to use this to erase the output state of the logic II If for the vicinity of pin 1.

On the premise that the image signal a has a low level relative to the small hole 4 and that the level change near the small hole 4 is large, the outputs d and b of the threshold processing circuits 23 and 24 are processed by the AND circuit 25.
If the AND circuit 25 performs logical product, unnecessary logic "1"
Since the output state with the output state '' erased is output, if the AND circuit 25 outputs the state of logic 1111', the presence of the small hole 4 can be detected immediately.

FIGS. 8a and 8b are for two-dimensionally explaining the above explanation.

The pattern to be detected in this case is the same as that shown in FIG. 3, and the boundaries of the binary patterns are indicated by broken lines.

As shown in FIG. 8a, of the output of the threshold processing circuit 23, the portion whose logic state is 1"° is used as the black pattern 31 to 3.
4 (in this example, it is assumed that there is also a portion near pin 2 where the brightness changes rapidly).

As can be seen, in addition to the unique color pattern 32.31 corresponding to the crack 5 and small hole 4, the area near the pins 1 and 2 is also detected as a unique color pattern 33.34, which is shown in Fig. 8a.
If we extract the matching part of the pattern shown in Figure 3, we get the 8th
As shown in the figure, only the black patterns corresponding to the target cracks 5 and small holes 4 can be extracted.

As is clear from Figure 8a, by enlarging the binary pattern after the spatial differentiation process, we can extract patterns that are detected as binary patterns like pins and whose gradients change greatly in the vicinity. However, in the present invention, all of these types of enlargement processing, noise removal processing, etc. are considered to be part of gradient detection.

As explained above, the presence or absence of defects such as cracks and small holes can be detected, but the detection is limited to the presence or absence of defects.

However, if there is any defect, it is convenient to be able to grasp the pattern shape of the defect.

FIG. 9 to FIG. 16 show an example of extracting the pattern shape of a defect.

First, referring to FIG. 9, this shows the configuration of an example of a pattern detection device used when extracting a pattern due to a defect existing in a detection pattern.

As shown in the figure, the configuration shown in Figure 4 is used almost as is,
It is constructed by newly adding a signal switching circuit 35, a multi-level memory 36, a labeling circuit 3γ, an overlap detection circuit 38, a label holding register 39, and a pattern extraction circuit 40 as shown in the figure. A sequence control circuit for controlling the circuits is connected to each circuit, but is not shown in the figure for simplicity.

Here, to explain the outline of the circuit operation on the assumption that the input image signal has reproducibility, the input image signal is passed through the signal switching circuit 35 and the threshold processing circuit 24 to the multi-value memory so that it can be stored as a binarized pattern image.

After that, the labeling circuit 39 detects the connectivity of each pattern included in the binarized pattern image, and each isolated pattern is given a pattern occurrence order, that is, a label.
The result is stored in the multilevel memory 36.

In this state, the signal switching circuit 35 is switched, and the differential information of the input image signal is extracted from the threshold processing circuit 23 as described above, and in synchronization with this, the binarized pattern is extracted from the multilevel memory 36. Images are read out in the order in which they are stored, and if picture elements labeled with differential information are output at the same time, an overlap detection circuit 38
is detected, the label is stored in label holding register 3.
Hold it at 9.

A plurality of label holding registers 39 are actually prepared as shown in FIG. The matching circuit 42 detects a match or mismatch with the contents of the label holding register 39, and if the label matches any of the contents of the label holding register 39, the match detection output becomes a set input to the output register 45 via the OR circuit 43. Therefore, the label that is the content of the input register 41 is set in the output register 45, and is output as an extracted pattern.

However, if there is no match with any of the values, the mismatch detection output is input as a reset input to the output register 45 via the inverter circuit 44, so that the contents of the output register 45 are cleared.

Although it is possible to extract defects as a pattern in this way, as is clear from the above explanation, the present invention can be conceived on the premise that the input image signal has reproducibility. .

Now, the labeling circuit 37 will be explained in detail.

The labeling circuit 37 is for attaching a unique value, that is, a label, to each isolated pattern by detecting the connectivity of binary patterns.

This labeling circuit can be realized by a well-known combination of either a dedicated logic circuit or a general-purpose logic circuit, and there is no need to specifically illustrate its circuit configuration, although it is necessary to clarify its function. It is something.

The processing by the labeling circuit is roughly divided into two as shown in FIG. 11, and each requires sweeping the entire surface of the multilevel memory 36 once.

The first process is partial labeling and creation of a label table, but prior to this, the label table is cleared to zero and the label table pointer PNT is set to zero.

Note that partial labeling here means dividing each isolated pattern into one or more parts depending on its shape and attaching different labels to the parts.

In addition, a label table is used because multiple different labels are attached to one isolated pattern by partial labeling.
This is a table showing the connection relationship between these labels.

Furthermore, regarding the pointer PNT, when scanning a binary image sequentially and performing processing 1 in order, that is, when labeling parts in the order of detected patterns such as 1, 2, 3, etc., in the middle of sequential processing. The pointer PNT has the value of the label to be attached to the pattern at that moment.

The operation contents will be described below with respect to the specific pattern example shown in the lower left of FIG. 14.

First, FIG. 13a shows an example of a binary pattern in the multivalued memory 36, in which the pattern existing portion is displayed as II II II, and the other portions are displayed as blank.

Further, FIG. 13 shows a label table, and before the first process, the column L(i) indicating the connection relationship of labels in the label table is cleared to zero as an initial setting.

Detection of pattern connectivity is performed from 0,2 to 0, as shown in Figure 12.
This is carried out using a detection window consisting of picture elements numbered 5, and the neighboring picture element patterns of picture elements 2 to 5 are determined using the picture element 0 as the picture element of interest.

The detection window sequentially scans within the binary pattern image.

That is, the entire screen is scanned stepwise from the top to the bottom while sweeping the screen in the multilevel memory from left to right.

Then, at each moment during scanning, the 14th
New value ■ to the pixel of interest according to the decision logic of the tree structure shown in the figure.

(VQ; new value of the picture element of interest) is written, entries are made in the table, and the pointer PNT is updated as necessary.

Here, to briefly explain the decision logic shown in FIG. 14, ■0゜■2 to ■5 are the values of picture elements 0.2 to 5 of the detection window shown in FIG. 12, respectively, and V□ is As already mentioned, the new value of the picture element of interest, that is, the label of that picture element, is expressed as r
TABLEJ shows the writing process to the label table at the bottom left of FIG. 14.

[TABLEj processing is shown in FIG.
V5) or TABLE (V2.V4), but this is expressed as TABLE (Vl) in Figure 15.
, V2) means that ■1 and ■2 are used as arguments.

In addition, in Figures 14 and 15, ellipses indicate judgments, rectangular frames indicate various processes, arrows extending from the ellipse to the lower left indicate the case when the judgment conditions are met, and arrows extending to the lower right indicate the judgment conditions. If this does not hold true, a small leftward arrow during various processes, for example A+-B, indicates that the value of A is updated to the value of B.

In the pattern example shown in FIG. 13a, the portions where no pattern exists are V as they are.

becomes O. However, if you change the detection window from left to right,
Scan from the upper side to the lower side and select “1” at the top of the cross-shaped pattern 46.
When the ``partial position'' is reached, the values of VO, V2 to ``5'' are all 0, so the process shown in the lower left of FIG.
is updated to 1, and vo becomes the value of pointer PNT, that is, 1.

Furthermore, when the scanning by the detection window continues and the position reaches the 1" portion of the left end of the cross-shaped pattern 46, ■2=1, V3=
Since V4=V5=O, from FIG. 14, ■0←■2, and the value of vo becomes 1.

If such processing continues, all the cross-shaped patterns 46 will be labeled as 1.

After this, the scanning is further continued and when it reaches the 1" portion position at the upper left end of the concave pattern 47, all of ■2 to ■5 are 0, so the process shown in the lower left of FIG. 14, that is, the pointer PNT is updated and becomes 2. , v(1 becomes 2.

Further, when the pointer PNT reaches the 11" position at the upper right end of the concave pattern 47, the pointer PNT is updated and v□ becomes 3.

When the same scanning is continued and one picture element is left from the lower right corner of the concave pattern 47, V5=2 and V2=3, and at this time, the process [PABLE
J operates, label table L(3) becomes 2, and ■o becomes 2.If the same scanning is continued, the first process ends, but at this time, the pattern in the multi-level memory 36 becomes 2. Partial labeling has been completed as shown in Figure 13C, and the table is <L(3)=2 as shown in Figure 13D, which means that label 3 is connected to 2. It shows that there is.

Next, in the second process, the process shown in FIG. 16 is performed.

That is, when the pixel of interest is not O, its label table is referred to, and when L(VO) is O, it is V.

The value of is made to be the value of ■0.

That is, the value does not change.

Also, if L(VO) is not O, ■ the value of 0 is L(VO)
Then, it is determined whether L(VO) is O again.

If it is 0, do nothing; if it is not 0, set the value to ■0.

This process is repeated.

For example, in FIG. 13C, when ■0=3, refer to the label table shown in FIG. 13D and find that L(3)=2.

Change the value to 2. Since L(2)-0, no further changes are made.

If such processing is performed on the entire screen, labeling will be completed as shown in FIG. 13e.

As explained above, the present invention provides brightness information of a binary signal obtained by thresholding an input image signal,
Defects such as small holes existing on the surface of the detection target are detected by logically multiplying the input image signal with the differential information obtained by spatially differentiating the input image signal. The binarized pattern images are read out from the memory means in the order in which they are stored as converted pattern images, and differential information is obtained from the input image signal in synchronization with this readout, and the time point at which the differential information is obtained. At the same time, when the binarized pattern image being read out is an isolated pattern, the isolated pattern is extracted as a pattern due to a defect such as a small hole.

Since the present invention takes into account not only brightness information but also differential information obtained by spatial differential processing of input image signals, it is possible to detect defects and their shapes on the surface of the detection target more reliably than before. Detection or extraction becomes possible.

[Brief explanation of the drawing]

The attached drawings are diagrams for explaining the present invention, in which Fig. 1 is a diagram showing the shape of a soldered surface on a printed board as an example of a detection target, and Fig. 2 is an overall diagram of an apparatus for detecting an image of a detection target. Fig. 3 is a diagram showing an example of a binary pattern image obtained by thresholding the input image for the soldering surface shape in Fig. 1, and Fig. 4 is a diagram showing the detection of the presence or absence of defects. FIG. 5 is a detailed configuration diagram of a maximum value extraction circuit that constitutes a part of the pattern detection device, and FIG. Explanatory diagrams of the detection window constituting a part of the device, Figures T a - c specifically show the input and output waveforms of the main parts of the device shown in Figure 4 when the soldering surface shape is an example of the detection target. Figures 8a and 8b are diagrams for two-dimensionally explaining defect detection when the soldering surface shape is an example of the detection target, and Figure 9 is a diagram for explaining the existence of defects. FIG. 10 is a block diagram of an example of a pattern detection device according to the present invention for extracting defects as a pattern, and FIG. 10 is a detailed block diagram of a pattern extraction circuit that constitutes a part of the pattern detection device, and FIGS. FIG. 16 is a diagram for explaining processing in a labeling circuit that also constitutes a part of the pattern detection device. 18...1-swipe delay circuit, 19...
...Detection window, 20, 20'...Differential circuit,
21.21'... Absolute value circuit, 22...
・Maximum value extraction circuit, 23゜24...Threshold processing circuit, 25...AND circuit, 35...
・Signal switching circuit, 36...Multi-value memory, 37.
... Labeling circuit, 38 ... Overlap detection circuit, 39 ... Label holding register,
40...Pattern extraction circuit.

Claims (1)

[Claims] 1. A pattern configured to detect a pattern portion that uniquely exists as a defect on the surface of a detection target by subjecting an analog image signal obtained by imaging the detection target to threshold processing. In the detection device, the image signal is processed by a threshold processing circuit to obtain binary brightness information, and the image signal is spatially differentiated by a gradient detection circuit and then processed by a threshold processing circuit. A pattern detection device characterized by a configuration that detects the presence or absence of a defect by obtaining differential information and performing an AND operation on the information and the binarized brightness information using an AND circuit. 2. In a pattern detection device configured to detect a pattern portion uniquely existing as a defect on the surface of a detection target by thresholding an analog image signal obtained by imaging the detection target, the image signal After the binarized brightness information obtained by processing with a threshold processing circuit is previously stored in a memory means as a binarized pattern image, a labeling circuit labels isolated patterns included in the image in the order of pattern occurrence. After reading out the binarized pattern images from the memory means in the order in which they were stored, the image signal obtained again in synchronization with the reading is spatially differentiated by a gradient detection circuit and then processed by a threshold processing circuit. differential information is obtained by the processing, and when the isolated pattern is simultaneously read out from the memory means at the time when the information is obtained, the label attached to the isolated pattern is detected, and the label associated with the detection is detected. 1. A pattern detection device characterized by a structure for extracting an isolated pattern marked with a symbol as a pattern due to a defect from the memory means.