JPS6136882A - Pattern discrimination device - Google Patents

Abstract

PURPOSE:To integrate the result of discrimination of each screen and plural results of discrimination of each screen to obtain the final result of discrimination, by storing previously the information on the form of a window and its position within a screen viewfield and the deciding conditions to the picture feature quantity in response to the number (areas) of divided screens and then extracting selectively them to give the decision to each corresponding screen. CONSTITUTION:The image pickup signals sent from a TV camera 2 are stored to the screen dividing and memory circuits 30 set opposite to each channel. A memory control circuit MC sends the window information successively to a data latch circuit L3. Then the windows are produced from a window generating part G2 by the coordinate data of the circuit L3 and the data given from a coordinate data generating circuit G1 which is normalized by the position correcting signal PD. At the same time, the binary coded channel information is sent to a latch circuit L2. One of prescribed binary coded picture data is selected via a binary coded signal selecting circuit SE. Hereafter an AND of signals is secured via an AND gate 36. Thus the features are extracted for each window. Such operations are carried out with each subject pattern and can be applied even to plural subject patterns.

Description

[Detailed Description of the Invention] [Technical Field to Which the Invention Pertains] The present invention provides a multi-window pattern discrimination device that is an extension of the template matching method that discriminates patterns based on the degree of matching of an unknown pattern with a standard pattern (tengrate). , especially regarding its improvements.

[Prior art and its problems]

As is clear from the various proposals that have been made for multi-window pattern discrimination devices, an observation area of a target pattern, that is, a window area, is set within one screen field of view, and predetermined image features are determined within this window. The quality of the pattern is determined by measuring the feature amount for each window and comparing it with a predetermined threshold (judgment standard), and further, for the combination between one predetermined window, correlation calculation of the feature amount is performed, The quality of the pattern is determined by comparing the correlation calculation value with another predetermined threshold (judgment standard). For this purpose, an imaging device (for example, an industrial television (ITV)) that images the target pattern is required.
) The signal from the camera) is binarized, the image features are extracted from this zgi information, the position of the target pattern image within the image field of view is detected, and the area to be windowed is determined based on this position information. After the correction (position normalization) is performed, predetermined image features are measured within the set window. Regarding the multi-window type pattern discrimination device, if necessary, please refer to the patent application filed in 1973.
Please refer to No. 7-71499, No. 71500, etc.

In the above-mentioned apparatus, in cases where the object has a pattern with a complicated shape, the entire object pattern is displayed on the TV.
When capturing images with a camera, the resolution of existing TV cameras is insufficient. To deal with this lack of resolution, it is necessary to image the entire area of the object by enlarging and imaging a part of the object, and repeating this partial enlargement multiple times by changing the area of the object. (abbreviated as segmented imaging method). By the way, when complex table patterns are photographed using this divided imaging method, the image patterns taken with the TV camera generally differ depending on the timing of imaging (that is, the area of the object). In cases like this, where the same object is captured by a TV camera as completely different patterns, almost all the judgment conditions associated with pass/fail judgment usually change depending on the target image pattern. It is.

Therefore, when the same object has completely different patterns depending on its position (area) K, it is clear that the conventional method, which targets one image pattern, cannot cope with the situation. In addition, high-speed processing (1 to 10 screens/second) is generally required for inspection equipment, and this is not an exception when the same object is captured as an anisotropic image pattern as described in 5 above. Faster processing is required.

[Purpose of the invention]

This invention was made to solve the above-mentioned problems, and it is possible to perform high-resolution pattern discrimination at low cost and easily, and also to discriminate patterns of a wide variety of objects at high speed. The purpose of this invention is to provide a pattern discrimination device that is capable of

[Key points of the invention]

The present invention provides means for detecting the position of an object using a binarized image pattern based on threshold information for binarizing an imaging signal from a TV camera, and a means for detecting the position of an object in a multi-window pattern discrimination device. In addition to providing a means for normalizing the window occurrence position by
Various judgment conditions are set according to the number of divided screens (areas), including information on the shape of the window and its position within the screen's field of view, as well as judgment threshold information that serves as a criterion for determining pass/fail and sorting based on image features within the window. At the time of judgment, various judgment conditions for the corresponding screen are selectively retrieved from the memory and judged for each corresponding screen, and the judgment results for each screen and multiple judgment results for each Hiro screen are stored in advance. The judgment results are obtained by summing up the results.

[Embodiments of the invention]

FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is a discrimination target object pattern, 2 is an imaging device such as an industrial television (ITV) camera, 3 is a pattern discrimination circuit, 4 is a panel setting section, 5 is an information selection control circuit,
61 to 64 are information selection circuits, and 71 to 74 are memories. The pattern discrimination circuit 3 includes a binarization circuit 31, a screen division circuit 32, an image memory 33, an image position detection/correction circuit 34, a window generation circuit 35, an AND gate 36, and a feature extraction surface IT! 137, a feature data memory circuit 38, a discrimination circuit 39, and the like.

The target pattern 1 cane is transported by a transport device (not shown) within the field of view of the TV right camera. The TV right camera sequentially scans the optical image information of the target pattern and converts it into an electrical signal. This signal is amplified to an appropriate level by a binarization circuit 31 and then binarized. Furthermore, this signal is divided into a predetermined number of pixels by the screen division circuit 32.

The binarized image pattern signal is stored in the image memo 1J33, and is also processed by the image position detection/correction circuit 34.
The relative position of the image pattern within the camera field of view is detected in two-dimensional coordinates. The position data obtained here is given to the window generation circuit 35, from which the window generation timing is corrected.

I/i2M image memory 3 with AND gate 36
The 3-ta is ANDed with the corrected position and area. The feature extraction circuit 37 is a circuit for extracting predetermined graphical features, and is composed of a two-dimensional local memory, a two-dimensional local logic operation circuit, a counter, etc., as is well known. Since the input to the feature extraction circuit 37 is controlled by the window area, a small amount of graphic data within the window area is counted.

The partial data of is stored in the feature data memory circuit 38. The discrimination circuit 39 compares and discriminates the data in the feature data memory 38 with the discrimination threshold data (one of Ih to Dn) selected from the information selection circuits 61 to 64, and outputs the discrimination result.

As a result, the panel setting section 4 sets the target pattern 1 (Pa,
Binarization threshold information (A
l to An), image position detection area information (B1 to Bn), window information (C1 to Cn), and discrimination threshold information (Ih to Dn), for example, as in case ① in the following table. If the information is stored in the information memories 71 to 74 in advance,
During discrimination processing, from external target pattern information SK,
The information selection control circuit 5 operates, and only one piece of corresponding information is selected through the information selection circuits 61 to 64. The window generating circuit 35 is applied to the discriminating circuit 39. That is, the information selection control circuit 5 controls the information selection circuits 61 to 64 in response to the selection signal S corresponding to the target pattern 1, thereby combining the control data of each circuit and the operation circuit.

The processing in the pattern discrimination circuit 3 is carried out by the information selection circuit 61.
Data specified by ~64 (A s B , Cp
D) can be used as appropriate using data corresponding to the target pattern. Note that as the selection information S, for example, an output from a position sensor provided near the transport device can be used.

Here, the shape of the identification target pattern 1 (Pa) is as shown in FIG. 2, for example, and each region P11. is divided by vertical and horizontal broken lines. Assuming that the target pattern 1 is imaged every P12·4·pmn, the relative position with respect to the 1 target pattern is determined so that the field of view of the TV camera 2 has the size of each area. For example, when the image pattern portion Fil is imaged by the TV camera 2, the information selection control circuit 5 can be notified that the image pattern is the area Fil using the selection signal S. 5 is a large number of pieces of information (A1-An, B1-Bn, cl-cn, D1-Dn) in the information memories 71-74.
) to the corresponding information (e.g. Al p B1 p C1p
Di) t'' information selection circuits 61 to 64 can select only one and provide it to the pattern discrimination circuit 3, whereby the pattern discrimination circuit 3 can select the information selection circuits 61 to 64.
Pattern discrimination processing can be performed using the information data specified by . Next, object pattern 1
f! , use aTVTV camera 2 to set the field of view to area P12.
At the same time, the selection signal S notifies the information selection control circuit 5 that the image signal is related to the area P12. The information selection control circuit 5 selects corresponding information from the information memories 71 to 74 (example A2 p B 2 p C2p D 2
) through the information selection circuits 61 to 64,
The pattern discrimination circuit 3 uses the information data specified by the information selection circuits 61 to 64 to perform pattern discrimination processing.

In this way, each part P11yP12...
・・°° Imaging order of horse rests and information data from information memories 71 to 74 (A1.BlpClyDxL(A4°B2
．． C27D2L =-・CkntBnp CnyDn)
By combining the reading orders in advance, comprehensive pattern discrimination processing for one screen becomes possible.

In addition, the target pattern pH ~ T'mn in the above case
The correspondence relationship between the information and the control information (A-D) can be summarized as shown in case (2) in the table.

By doing this, the target pattern 1 can be
In comparison with the case where the entire area (the entire area) is placed within the field of view of the TV camera and the pattern is discriminated at once, as in this example, the object pattern 1 (Pa) is divided into multiple parts (P11 to Pmn
) Imaging each part means increasing the resolution of each part, and therefore brings about the advantage that pattern discrimination accuracy can be improved. Furthermore, compared to the conventional system where several simple Ki units are installed together, the device has the advantage of being smaller and cheaper.

Next, a window superimposition method will be described with reference to FIG.

Now, assuming that the combination of the target pattern and the discrimination information (A-D) is as shown in case ① in the table, first, the image Px of one target pattern is binarized using the binarization threshold A7, Part 21 [! A position correction process B7 is performed based on the image formation. Note that the image pattern at this time is stored in the image memory 33 in FIG. Next, window shape C7
The feature extraction data for each window and the discrimination threshold value D7 are used to perform pattern discrimination. At this time, windows Wl, W2. Pattern discrimination processing is performed only on W3. Continuing on, using the data in the image memory 33, the processing as in case (2) in the table is performed using the window shape C8 and the discrimination threshold D8. In this case, pattern discrimination processing is performed only on windows W4, W5, and W6 as shown in FIG. 3(b), for example. By performing these processes continuously using a single discrimination start signal (selection signal) from the outside, pattern discrimination can be performed by superimposing two different windows on each other, as shown in the composite screen shown in Figure 3 (C). has been done. Note that the determination result is a combination of pattern determination for each window of C7 and CB.

In the above, the window information was changed, but it is also possible to change only the 2-bit conversion threshold A8, as in cases ■ and V in the table, or, as in cases V and ■ in the table. It is also possible to create a combination in which only the position correction data B8 is changed. In other words, by doing cases like ■ to ■ in the table, a large amount of pattern discrimination information exists in one image pattern Px, and as a result, pattern discrimination processing is diversified and its accuracy is improved. (This has the advantage of improving reliability. It goes without saying that these various types of discrimination are performed sequentially in a time-sharing manner based on the contents of the image memory 33 in FIG. 1.) .

By the way, the level of the image pickup signal from the TV left camera usually differs from part to part depending on the surface condition of the object, lighting conditions, and the like. However, in the past, the target pattern was binarized using a single binarization threshold, which made it impossible to deal with partial changes in the imaging signal within the field of view of the screen. This has a considerable impact on the feature extraction process used in the process. In other words, there is a problem in that it is not always possible to obtain stable image information for one target pattern by simply setting one conversion level.

FIG. 4 is a block diagram showing another embodiment of the present invention based on this viewpoint. In the same figure, 301
is the binarization circuit 31 and the 2m level data latch circuit L
2G consisting of 1! 302 is a channel data selection section consisting of an image division and storage circuit 30, a binarized signal selection circuit SE and a binarized channel data latch circuit L2; 303 is a coordinate data generation circuit Gl, a window generation section G2 and a window A window forming section consisting of a data latch circuit L3, 304 a window area memory circuit M
M1 is a 21 level memory circuit, MC is a memory control circuit, 2 is a television camera, and 36 is an AND gate. That is, in this embodiment, it is possible to obtain stable image information by simply providing a plurality of channel binarization circuits and selecting the desired one for each window. This is what I did.

The operation shown in FIG. 4 will be described below with reference also to FIGS. 5 and 6. Here, FIG. 5 is an explanatory diagram for explaining how a window is generated, and FIG. 6 is an explanatory diagram for explaining window information.

An image signal from the TV left camera is given to a binarization section 301. The binarization unit 301 has a plurality of binarization circuits 31, and the binarization threshold value is converted to 2IIi level for each channel, and includes a memory circuit M1 and a binarization level data latch circuit L1.
configured via. Therefore, the image signal from the TV left camera is binarized for each channel, and this binarized image information is stored in the screen division and storage circuit 30 provided for each channel.

Here, for example, a window W7 having a shape as shown in FIG.
To form W8, window area memory M3
6 (b) and (c), the horizontal coordinate position data Xu ~ Xu at which the window should be opened on the screen
+, same (coordinate position data to be closed Xdl~Xdi
It is necessary to store the information for each raster R1, R2, . . . of the television signal. Note that PM in FIG. 5 is a monitor screen, and the "○" mark in the screen PM indicates the point OP at which the window should be opened, and the same <"x" mark indicates the point CL at which the window should be closed. Then, by giving the binary channel information y1 to yj of the memory circuit M2 to these window information as shown in FIG.
A binarization threshold level can be specified for each. Note that the data z1 to Zr shown in FIG. 6 are data for switching in the vertical direction, and the window information is horizontal in the television scan.

It is generated in synchronization with the vertical signal, but at this time, the memory control circuit MC generates window information (window open coordinate data Xul to Xui, window closed coordinate data Xdl to
at) is sequentially sent to the window data latch circuit L3, and a window is generated via the window generation section G2 using the coordinate data of and the data from the coordinate data generation circuit G1 normalized by the position correction signal PD. . At the same time, the memory control circuit MC also sends the binarized channel information y1 to y to the binarized channel data latch circuit L2 in the same manner as the window information, and sends the binarized channel information y1 to y to the binarized channel data latch circuit L2 via the binarized signal selection circuit SE. Select one of the converted image data. Thereafter, features are extracted for each window in the same manner as in FIG. 1 by performing a logical product of these signals in the computer game 36. By doing so, stable feature extraction can be performed regardless of the surface state of the object or the illumination conditions.

In addition, although we considered one target pattern above,
By performing such an operation for each target pattern, it becomes possible to apply it to a plurality of target patterns as well.

〔Effect of the invention〕

According to this invention, the following effects can be expected.

From K This allows for smaller size and higher speed.

(2) By storing a large amount of pattern discrimination information (control information) in a single tube turn discriminator, it is possible to handle a wide variety of products by simply supplying information data corresponding to the target pattern from the outside.
It becomes possible to perform discrimination processing on a wide variety of patterns, making it easy to change product types and adapt to lines that mix a variety of products.

(3) By storing a large number of pattern discrimination information in a predetermined order in i pattern discrimination devices, it is possible to sequentially discriminate corresponding patterns simply by receiving a discrimination start signal from the outside. This makes it possible to subdivide one target pattern and capture a solid object from multiple directions, and as a result, pattern discrimination accuracy can be improved.

(4) A single pattern discrimination device stores a large amount of pattern discrimination information, processing information (binarization level, position correction, etc.).

By storing the information for each window shape (9), it becomes possible to perform pattern discrimination by sequentially switching over one target pattern, and to superimpose windows. In addition, by combining the combination of each processing information and the relationship between the previous and subsequent pattern discrimination processing, such as changing only the binarization level multiple times or using the previous correction data for position correction, , stable and highly accurate pattern discrimination processing is possible.

(5) By providing a binarization circuit with multiple channels,
Because it is possible to select the channel of the binarized image data for each window for one target pattern, it is possible to absorb fluctuations in the imaging signal level for objects of the same shape due to the shape and illumination of the object. This results in the advantages of improved accuracy and improved stability.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram for explaining how a target pattern is divided, FIG. 3 is an explanatory diagram for explaining a window superimposition method, and FIG. 5 is a block diagram showing another embodiment of the present invention, FIG. 5 is a ninth explanatory diagram for explaining how a window is generated, and FIG. 6 is an explanatory diagram for explaining window information. Code explanation 1...Target butter/, 2...Television camera, 3...Pattern discrimination circuit, 4...
・Panel setting section, 5 --- Information selection control circuit, 3
1... Binarization circuit, 32... Screen division circuit, 33... Curved surface image memory, 34... Image position detection/correction circuit, 35...・Window generation circuit, 36・Song・AND gate, 37...A・Feature extraction circuit, 38...Feature data memory circuit, 39...
...Discrimination circuit, 3o.Song/image division and storage circuit, 61-64...Information selection circuit, 71-74
. . . Information memory, 301 . . . Binarization section, 302 . . . Channel data selection section, 303.
...Window forming section, 304...Memory section, L1-L3...Music switch LMI-M3...Song memory circuit, Mc/Memory control section, Gl... ...
Coordinate data generation circuit, G2... field window generation section. Agent Patent attorney Akio Namiki Agent Patent attorney Kiyoshi Matsuzaki + Figure 2 Figure 3 (') ρN

Claims (1)

[Claims] 1) Imaging means for dividing one target pattern into a plurality of parts and imaging the image signal at a predetermined threshold level.
A binarization means for converting into values, a position correction means for detecting the relative position of the target pattern portion on the imaging screen and calculating the correction amount from a predetermined reference position, and correcting the predetermined window information based on the correction amount. and the observation area of the binarized image (
a window forming means for forming a window), a discriminating means for extracting an image feature amount of a target pattern portion within the window and comparing it with a predetermined criterion for discrimination, and the binarization threshold and the criterion. storage means for storing a plurality of types of each of position, window, and determination criteria information; information selection means for reading each piece of stored information based on a predetermined instruction and providing the corresponding means; and information selection means. 1. A pattern discriminating device, comprising: instruction means for giving a predetermined instruction to the means, and for discriminating a target pattern by associating various pieces of information with each of the target pattern portions. 2) The target pattern parts are determined in a predetermined order by instructing the imaging order of each target pattern part through the instruction means.
The pattern discrimination device described in Section. 3) The pattern discrimination device according to claim 1 or 2, wherein the window information is appropriately combined to form mutually overlapping window areas. 4) By providing a plurality of channels of the binarization means,
Claims 1 to 3 are characterized in that for one target pattern, different binarization levels are set for at least each window to enable extraction of a binarized image.
The pattern discrimination device according to any one of paragraphs.