DE3303841C2 - Pattern checking device and method for checking a pattern - Google Patents

Description

The present invention relates to a pattern testing device and a method for testing a Pattern according to the preamble of claim 1 or of claim 2.

An Ent A decision must be made as to whether a Pattern can be classified as good or bad.

First, with reference to FIGS. 1 (a) and 1 (b), it will be described how a conventional pattern checking device works according to the so-called template method. The drawing shows two samples to be tested 1 and 1 ', and sampling points 2 to 8 are marked with a +.

In the case shown in Fig. 1 (a), it is assumed that the checked pattern 1 is the number "8". As can be seen from the drawing, the sampling points 2 to 8 are assessed using a template, not shown, in order to check the presence of a sample at each point. If a piece of the pattern is present at such a point, this corresponds to a logical "1"; if there is no pattern, this corresponds to a logical "0". The test results are coded for each of the points 2 to 8 , and this code is used to determine which known pattern the tested pattern is to be assigned to.

In the example shown in Fig. 1 (a), the test leads to the result that there is a piece of pattern at all points 2 to 8 (logical "1"). It can be determined here that the pattern 1 is the number "8". In the example shown in FIG. 1 (b), there is no pattern only at the sampling point 8 (logical "0"). In this case, a predefined decision algorithm is available which can be used to determine that this is the character "0".

The advantages of such, according to the stencil process working sample test system are known: Due to the Compression, d. H. of the existence of a pattern only at the sampling point, needs little information amount to be examined; the test results allow easy classification of the pattern; it can be effectively in a pattern recognition device, such as B. use an optical character reader, in which classifies an unknown tested sample will determine which (known) standard pattern it matches.

If you use the template method for a sample test advises, there are some problems. A problem is that when checking whether a pattern is good or is bad because the pattern in an enlarged one Window area corresponding to the sampling point is present or not, a similar but bad pattern err is usually judged as a good pattern. This is a he considerable deficiency, for the following reason: at the Alternative decision "pattern" / "no pattern" will, if there is no pattern in the window area, an appraisal division "bad". However, if a pattern exists, but this is the size of a good one  Pattern differs, so it cannot be considered bad be judged.

Generally, some vary in printed patterns Area of a pattern section covering the window area according to the checked position, and each window area has its characteristic (in this case large and small area or area sizes) in the sample cut. Therefore, such a characteristic is not extracted and measured, it is impossible to get the pattern correct to consider.

A sample testing device according to the preamble of the patent Claims 1 is known from DE 30 12 958 A1. Will be at the state of the art in scanning for example one Number only part of this number from the field of view of the photoelectric the converter, the data from the recorded part stored in a first buffer memory, and after moving the converter, the data from the adjacent area in a second buffer memory This may result in an overlap. By processing the data in both buffers generated the "right" pattern. This can be done with help of two buffer memories the field of view of a photoelectric Increase converter.

In DE-OS 17 74 812 there is a sample testing device for detection NEN described by characters in which a light pen one Screens over, which is formed in matrix form and an array of columns and rows of submatrices contains. For example, each submatrix can have four points that either receive light from the light pen or Not. If more than two points get light, the corresponding submatrix a "1" signal, otherwise a "0" signal down. The signals are aligned in rows and columns  cached and then rated. The rated matrix is compared with pre-stored template values. The Sign is recognized when there is a certain largest match the checked character is correct with one of the stencils chen results.

It is often desirable to use a sample test facility for ver to be able to determine whether any optical pattern has a certain configuration or not. Would, for example, according to DE-OS 17 74 812 instead of one Use a camera to light the entire pattern depicts, the sign under consideration could be, for example move sideways, and then with the known one Establishment can no longer be clearly determined whether it is is a specific character or not.

The general advantage of stenciling is that that due to the presence of a pattern on certain Points only a small amount of information can be processed got to. Devices of this type are therefore suitable for optical signs readers, but not for the purpose, practically any Check optical patterns for their configuration.

The object of the invention is to provide a sample testing device and to specify a procedure for the type test with which based on the stencil process a general nes pattern can be checked correctly.

According to the invention, this object is achieved by a sample test direction according to claim 1 or a method according to Claim 2 solved.

According to this solution, a correlation operation is provided hen. This makes it possible to have an exact check even then if the relative location of the pattern under consideration  to the receiving photoelectric converter, for example a camera that is not precisely defined. The usual thing and expensive exact positioning of the pattern with respect to the scanning device can be omitted.

It can be practically anywhere in the eye field of the photoelectric converter set window areas len, it is up to the user, number, form, Select the size and position of the individual windows. Thereby practically any pattern recognition is possible.

In a preferred embodiment the device according to the invention is a decision whether a pattern is good or bad which ver a changing electrical signal will work, which you get by that with a two-dimensional sequential scanning photoelectric Converter (e.g. an industrial television camera) an image of a sample to be checked. On elec trical paths become more in the field of view of the television camera rere (generally many) window areas set, for size characteristics of each window area to extract the seen pattern section. With every fen ster becomes a threshold range of the upper and the un lower limit set. Is the size of the extract  characteristics within the area, it becomes a logical Assuming "1", if not, a logical "0" is displayed taken (this is a primary decision). Subsequently some of the window areas are grouped together sums up. One set of primary decision results for each the included window area is set in advance the primary decision table and with compared this to decide in this way whether the pattern is classified as good or bad (this is a secondary decision). With the table mentioned is a set of primary decision makers results for each group, which means for every area is formed that the field of view of a good pattern is divided into window areas in the same way. The Se customer decision is made for each group, and whether the pattern is classified as good or bad, is assessed based on the composition of the results.

If necessary, for the group described by the above secondary decision was classified as bad, the next step is a correlation decision guided. The correlation decision should follow are explained in more detail.

Even if the tested pattern is good, it will hold Size of the extra through the predetermined window areas these features are sometimes not within one agreed threshold range if the pattern in the Field of view of the camera from an assigned position reached. This leads to the result being the primary ent divorce is "bad". However, if the pattern is only ge deviates slightly from the assigned position, he says May range through correlation between that through the certain window area recorded feature size and feature size captured by the next window area  (in this case the correlation means addition, sub traction and the like) that the preserved correla tion size within the predetermined threshold range remains, which leads to the result that the primary ent divorce leads to the result "good". Starting from the This basic idea is a combination of suitable ter related window areas determined to a arithmetic processing (e.g. addition) of each char size between these window areas The processing result is then held back and it becomes the primary decision (this is called the Correction lations primary decision called). Commitment of correlation primary decision results a pre-formed primary correlation decision compared and compared with this table (the This table is created in the same way as the one above explained primary decision table) to a correla to make and assess secondary decisions, whether the pattern is classified as good or bad.

It follows from the above that the decision whether a checked pattern is good or bad is classified, is thereby brought about according to the invention, that in addition to a primary and a secondary entity a correlation decision if necessary follows. The shape, size and other properties of a Window areas are arbitrary and according to the ge tested pattern set in a suitable manner. The fen sterform is not limited to a rectangle. Yourself understandable in addition to the area of a Mu the circumferential length in the window area of the pattern section or sizes other than by one Window area extracted feature sizes used become.  

The following are exemplary embodiments of the invention hand of the drawing explained in more detail. Show it:

Fig. 1 (a) and (b) are each a diagram for illustrating the operation of a conventional directional Musterprüfein,

Fig. 2 (a) illustrates an outline ranging from the setting Fensterbe,

Fig. 2 (b), the shape of a pattern,

Fig. 2 (c) an observed through window areas of pattern,

Fig. 3 shows another example of setting of Fen die rich,

Fig. 4 is a block diagram of an embodiment of the dung OF INVENTION,

Fig. 5 is a primary decision table,

Fig. 6 shows an example of a correlation primary decision table,

Fig. 7 is a flow chart illustrating the operation of the embodiment illustrated in Fig. 4 embodiment,

Fig. 8 is a flowchart illustrating NEN this way of working in the individual, and

Fig. 9 is a flowchart showing another operation.

Fig. 2 (a) shows an example of setting window areas, Fig. 2 (b) the shape of a pattern to be checked, and Fig. 2 (c) the pattern viewed through the window areas.

In these sketches, the reference numerals 21 to 27 mean rectangular window areas, and the reference numeral 20 denotes a pattern to be checked, in this case the number “8”. As shown in Fig. 2 (c), each part of the pattern 20 is viewed through the window areas 21 to 27 .

Fig. 3 shows another example of the setting of window areas. Outside of the window areas 21 to 27 finally shown in FIG. 2 (a), there are window areas 31 to 36 , and inside the window areas 21 to 27 , window areas 41 to 48 are set. As a result, if there is a positional deviation of the pattern, it is possible to create a correlation between the middle window areas 21 to 27 and the outside or inside window areas.

Fig. 4 shows a block diagram representation of an embodiment of the invention. The block diagram shows a test of the pattern 50 , an industrial television camera 51 , a splitting circuit 52 for binary coded picture elements, feature extracting circuits 53 and 53 ', counters 54 and 54 ', a count value memory 55 , a primary decision circuit 56 , a secondary decision circuit 57 , one Correlation formation decision circuit 58 which decides whether correlation formation is necessary or not, a correlation calculation circuit 59 , a correlation primary decision circuit 60 , a correlation secondary decision circuit 61 , a synthesis decision circuit 62 , a window formation circuit 63 , a control circuit 64 , a set point memory 65 and a keyboard 66 .

This circuit works as follows: The television camera 51 takes an image of the test pattern 50 and generates an electrical signal that changes over time. The electrical signal is binary coded in the binary coded pixel splitting circuit 52 so that it is split into pixels (dots). In general, the electrical signals for an image are split into 320 points in the horizontal X-axis direction and 240 points in the vertical Y-axis direction, for a total of almost 77,000 points. Upon receiving a point signal from the split circuit 52 , the feature extracting circuit 53 extracts the feature from the signal. If you take z. B. that the feature is "large or small area", he holds the area by looking at white point signals or black point signals and counting the length of a section in which the white point signals are present from the counter 54 are (the number of points is counted) while a lateral scan is carried out in the horizontal direction (ie: the area is determined by a set of length values). The circuit 53 'is used to extract features other than the surface size. Do you want z. B. detect the length of a transition or a boundary of white and black, the circuit extracts a boundary point, and the length of the boundary is determined by counting the number of boundary points by the counter 54 '. The counters 54 and 54 'operate in the counting mode under the control of a window area signal which is generated by the window formation circuit 63 . The counting result thus represents the feature size extracted from each window area, which is identified by the window area number and the feature type and is stored in the memory 55 . The window formation circuit 63 generates output signals, which are indicative of a set of window areas of any shape, which are prepared on the basis of data entered into the setpoint memory 65 via the keyboard 66 .

When the television camera has finished scanning a field, the measurement and storage of the feature data for each window area are completed, so that, based on this feature data, the test pattern is judged by the subsequent circuit parts 56 to 65 whether it is as is good or bad.

The feature data read out from the memory 55 for each window area is compared with the upper limit set point α _ÿ and lower limit set point β _ÿ set for each window area in the primary decision circuit 56 (i is the type of a feature, j is the number of a window area) to make a primary decision as to whether the data is in the range between the upper and lower limit threshold or not. The following description deals with the processing of feature type (i) performed for decision making.

The primary decision circuit 56 makes a primary decision by checking whether the data D _{jj of} the window area j regarding the feature type (i) (in the case of the area is the area size) within the through the upper limit setting point α _ÿ and the lower limit setting point β _ÿ defined range or not. This check is carried out for each window area. The result of the primary decision is a logic "1" if the data is within the range between the upper and lower limit threshold, and it is a logic "0" if the data is out of the range. The upper and lower limit thresholds can be selected according to each window area and feature type. This threshold data is entered via the keyboard 66 and stored in the set point memory 65 so that the primary decision circuit 56 reads and processes this data.

Often a pattern to be tested is made up of N small ones Sample pieces together (where N is any gan ze number). Therefore, each window area becomes N Groups formed that he ent N small sample pieces speaks. For example, if a pattern to be checked Letter sequence is "ABCD", this pattern continues composed of four small sample pieces (each character forms a small sample). Sometimes everyone passes small sample piece not of a letter or one Number, but from one formed by several characters unit.

The combination of the grouped window area numbers is entered via the keyboard and stored in the setting point memory 65 . The secondary decision circuit 57 reads the combination of the grouped window areas from the memory 65 , and depending on the combination, the primary decision results generated by the primary decision circuit 56 for each window area are divided into groups. Subsequently, the secondary decision circuit 57, the clustered primary decision results of the memory from the set point 65 read primary decision table ge genüber and compares it with this, to thereby separate ent to whether the results are good or bad as a group. The primary decision table was previously determined for each pattern to be checked and stored in the set point memory 65 .

FIG. 5 shows a primary decision table in which the digits "8" (see FIG. 2 (b)) and "7" represent patterns which are checked by using the window areas set in FIG. 3. From Fig. 3 it can be seen that the number "8" as the test pattern, the primary decision table contains the coded data (11111110 ... 0), which are arranged according to the numbering sequence of the window areas, while the primary decision table for the number "7" contains the encoded data (1110000 ... 0).

The secondary decision circuit 57 makes a secondary decision for each group. If the secondary decision regarding a group is negative (NG), it is decided in the correlation forming decision circuit 58 whether the correlation decision is necessary or not. The correlation forming decision circuit 58 decides in some cases that a correlation decision is necessary, while in other cases it decides that such a decision is not necessary; this depends on the conditions of various types described above. If the correlation decision is required, the correlation calculation circuit 59 calculates the extracted feature data (it adds the data of the window area 21 to the data of the window area 31 ) in accordance with the prescribed combination of window areas. The data of this combination is read from the memory 65 . The results are output as new data DC _jk .

The new data DC _jk is given to the correlation primary decision circuit 60 where the primary correlation decision is made as to whether or not the data is in the range between the upper limit set point αC _jk and the lower limit set point βC _jk . If they are within this range, a logical "1" is output. If the data are outside the range, a logical "0" is output. Such correlation primary decision processing is done for all associated combinations of window areas.

In the correlation secondary decision circuit 61 , the correlation primary decision results of the circuit 60 are compared with the correlation primary decision table read out from the set point memory 65 to make the correlation secondary decision. If there is a match, the decision regarding the types of characteristics of the group is made.

Fig. 6 shows an example of the correlation primary decision table obtained when the number "8" is slightly shifted upward with respect to the relative position of the window areas shown in Fig. 2 (c) and the window areas shown in Fig. 3 be used. It can be seen from Fig. 6 that if the pattern "8" is shifted slightly upward from its position, the window area 21 has only a small value of the extracted feature and that the data cannot be a logical "1" , but are a logical "0", and that the data can assume the logical value "1" precisely because the sum of the feature sizes of the window areas 21 and 31 is formed. The same applies to the window areas 24 and 27 . The correlation primary decision table is determined in advance and stored in the set point memory 65 via the keyboard 66 . This process is similar to the process described above for the primary decision table.

As described above, the secondary decision processing is based on the feature quantity data measured directly through the window area, and the secondary correlation decision is based on correlation data obtained by processing the measured data. In this way, if the result is saved in accordance with one of the two decision-making processes, by the Syntheseent discriminating circuit 62, the output result that the decision is secured.

Fig. 7 illustrates the operation of the embodiment shown in Fig. 4 Darge. Fig. 8 shows in a flow chart the details of the operation. In Fig. 8, each window area is simply called a "window".

In the embodiment described above has ever the window area rectangular. However, it can any other shape, e.g. B. a square, a pentagon, a polygon with n corners, a round shape or a ring shape. Especially when to check that en pattern is not a letter but a number it is important to have a shape of the fen that matches the pattern range.

In the above-described embodiment, only the primary correlation decision results are used for the secondary correlation decision in the correlation decision processing, but in some cases it may be appropriate to use the primary decision results in addition to the primary correlation decision results. If the addition was given as the correlation above, in some applications the finding of the remainder of the part or the ratio will lead to good results if there are appropriate patterns. While in the above exemplary embodiment the correlation only took place with regard to the type (i) of features, depending on the nature of the pattern, very good results will be achieved in some cases by the correlation between several feature types. Fig. 9 shows a flow chart for the operation in such cases.

While the area size and the length of a limit was specified as characteristic types the most diverse local Area information is used, for example se an intersection, an end point, an arc point, a oblique line and the like.

It can be seen that the invention provides a means when using a keyboard input in a memory according to the nature of the sample, information regarding Shape, location, number, etc. of a window area, ver different types of thresholds, correlation conditions different types of decision conditions etc. can be freely set. With conventional Sample test equipment depends on the circuit arrangement or the algorithm very strongly from the Mu mortality, so that such Einrich only in very limited numbers let put. The device according to the invention carries the differences of a pattern to be checked in ho Hem dimensions account so that they can be mass-produced is common.

The advantages of the invention are summarized below to get nabbed.

• 1) A window area can be set even where there is no pattern, as indicated by the reference drawing 31 to 48 in Fig. 3, whereas a window area can only be set in a conventional arrangement where there is a sample . Therefore, if a pattern to be checked is good but offset, which cannot be judged in the conventional arrangement, the pattern can be easily classified as bad.
• 2) A decision is made in each window area, where the upper and lower threshold thresholds are sepa advice for each type of feature and each window area are adjustable so that not only the presence of a Patterns in the window area, but also a sizes change of pattern or a change of shape of the pattern assessed quantitatively and thereby the accuracy of the Decision on a slight change in the pattern can be increased noticeably. Even if the density of the tested sample fluctuates partially, the test area accuracy by controlling the threshold accordingly the pattern can be obtained.
• 3) Since the window areas are grouped, about which allows a decision to be made separately the number of misjudgments of a good Mu decrease sters as bad pattern even then if scattering occurs in parts of a good pattern (a very slight slippage or a change in Density).
• 4) In addition to the measured data, the data the correlation between multiple panes for deciding whether a pattern is good or bad, used, which makes the pattern high and steady Accuracy can be checked. Especially with a gu the sample, the absolute measured value of the device changes Device and depending on various conditions  equally common. With such a system are the Compare measured values with the upper and lower limit values , a mistaken pattern is often mistaken for one bad pattern held or the test factor for one bad pattern lowered. However, also under such Conditions show the measured values of different window names submit some connections or connections, such as e.g. B. "the relative ratio of measured values of different Window area is constant "," the sum of measured values different window areas is constant ", and" one Pattern change is classified by the fact that the rest of measured values from different window areas becomes ". This enables the test to be carried out with high accuracy perform using the above conditions ren
• 5) When performing the correlation secondary decision Not only the result of the primary correlation divorce compared to that of a good pattern like it was carried out above, but to carry out the corrections Lation decision also becomes the primary decision result of a good sample for checking the sample calculated which resulted in an even more accurate decision results. Even if the pattern to be checked ver changes, it can be checked with high accuracy without that a good pattern was considered a bad pattern
• 6) By correlating certain feature data with an their characteristic data can be used in certain cases achieve respectable results. One correlates for example like a surface and the rest of the circumferential length similar like a contamination test, you get clear about the difference between good and bad.  
• 7) The shape of the window area is essentially arbitrary, so that the window area with ver different patterns is applicable, and the number of Pattern areas can be reduced to a quick to get le decision.
• 8) Local surface information is suitable as special Target type of measured characteristics, so that even a change change like a very small interruption of a pattern can be extracted with high accuracy.

Claims (2)

1. sample testing facility,
with a two-dimensionally sequential scanning photoelectric converter ( 51 ) which scans an optical image of a pattern to be tested ( 50 ) and emits a chronological sequence of electrical signals, and
with a splitting circuit ( 52 ) which splits the electrical signal received from the converter ( 51 ) into picture elements in order to output binary-coded output data as characterized by,
that an adjusting device ( 63 ) is provided for adjusting a plurality of window areas ( 21 - 48 ) in the field of view of the photoelectric converter ( 51 ),
that means ( 53 , 54 ) are provided which, based on the binary data corresponding to each window area, measure a size of a certain feature of the pattern to be checked in each window area,
that a primary decision means ( 56 ) compares the data from each window area from a group of window areas with an upper and a lower threshold value and outputs a primary decision result signal when the data falls within the threshold value area between the upper and lower threshold values,
that a secondary decision device is provided which compares the primary decision result signals obtained from the data of the group of window areas with previously stored decision data for that group to output a pattern check signal indicating whether this is in the group of window areas seeming pattern is a good pattern or not according to the stored decision data, and
that correlation forming decision means ( 58 ) is provided having a primary correlation decision means that compares the data from predetermined combinations of window areas with upper and lower threshold values and outputs a primary correlation decision result signal when the combined data is within of the threshold value range and has a secondary correlation decision device which compares the result signals from the primary correlation decision device from the data of the predefined window areas in order to output a correlation pattern test signal which indicates whether a pattern occurring in the group of window areas is a good pattern based on pre-stored correlation primary decision data. 2. A method for testing a pattern which is scanned with a two-dimensional scanning photoelectric converter, the scanning data being converted into binary image element data and the binary data being evaluated for testing the pattern, characterized by the following steps:
Setting multiple window areas within a field of view of the photoelectric converter and defining combinations of each window area,
Measuring the size of a certain feature of the pattern in each window area in which the binary data from the corresponding window area is processed,
Performing an addition or subtraction operation of the measured feature size of the pattern for all predefined combinations of window areas,
Making a primary correlation decision to decide whether the results of the operation are within the range between an upper and a lower limit,
Performing a secondary correlation decision to compare a set of the results of the primary correlation decision with pre-stored primary correlation decision data, and
Make a decision about the suitability of the pattern by comparing the result of the secondary correlation decision with a reference value.