JPH04135287A - Pattern recognition method - Google Patents

Abstract

PURPOSE:To accurately recognize an object pattern without being affected by shading conditions by thinning a line after binalizing a picture that includes the object pattern and calculating the ratio of the area of an extracted object picture element obtained after thinning a line to the total area of the picture element. CONSTITUTION:A marking device 5 for marking a car body number on a car body which is carried into a station 1 and a marking inspection unit 10 for automatically reading and inspecting the marked car body number are provided. After the image of a picture picked up when the image of an object pattern is picked up under shaded conditions is binarized, line thinning is conducted. Then, the ratio of the area of an extracted object picture element to the area of each partition is calculated. Thus, the object pattern is recognized on the basis of the combination of respective partitions having either of plural signals obtained by classifying the ratio according to a specified threshold level. In this way, after the influence of the shading conditions of the object pattern is excluded, the combinations of respective partitions are obtained. Thus, the object pattern can be recognized accurately.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a pattern recognition method for recognizing two-dimensional patterns by image processing.

[Prior Art] Conventionally, as a pattern recognition method for recognizing a two-dimensional target pattern, a method of performing pattern recognition by performing image processing on an image obtained by capturing an image of the target pattern to be recognized is generally well known. known (for example, Tokuko Sho 6)
(See Publication No. 1-9562).

By the way, when capturing an image of the target pattern to be recognized, by capturing the target pattern in a state where it is shaded by illumination light, it is possible to obtain an image of the pattern in which the brightness and darkness are further emphasized, which improves the recognition accuracy. You can improve your performance.

[Problems to be Solved by the Invention] However, when capturing an image of a target pattern with shading as described above, the mutual positional relationship between the illuminator, the camera, and the target object, the brightness of the illumination light, and the surrounding Due to various conditions such as the influence of ambient light caused by lighting conditions, etc., the way in which shading is applied to the target pattern is not necessarily consistent, and this poses a problem in that it is difficult to sufficiently improve the recognition accuracy of the target pattern. Ta.

The present invention has been made in view of the above problems, and an object of the present invention is to provide a pattern recognition method that can accurately recognize a target pattern without being affected by shading.

[Means for Solving the Problems] Therefore, the present invention recognizes a two-dimensional pattern by capturing an image of a target pattern to be recognized in a shaded state using illumination light and performing image processing on the obtained image. In a pattern recognition method, an image containing the target pattern is binarized and then thinned, and for each divided area obtained by dividing the image, line thinning is performed in the total pixel area of each divided area. calculate the ratio of the extraction target pixel area, classify the ratio into a plurality of signals according to a predetermined threshold, and recognize the target pattern based on the combination of each section each having one of these signals. This is how it was done.

[Effects of the Invention] According to the present invention, the image obtained by capturing the target pattern in a shaded state is binarized and then thinned, and then the ratio of the extraction target pixel area to the area of each section is calculated. Since the above ratio is calculated, the above ratio can be calculated without being affected by the way shading is applied. and,
Since the target pattern is recognized based on the combination of each section having one of a plurality of signals obtained by classifying the ratio using a predetermined threshold, the method of shading of the target pattern can be A combination of the above sections can be obtained after eliminating the influence, and the target pattern can be recognized accurately.

[Example 1] Hereinafter, a case in which an embodiment of the present invention is applied to, for example, a method of automatically reading and inspecting a vehicle body number engraved on a vehicle body on an automobile production line will be described with reference to the accompanying drawings. do.

FIG. 1 shows the automobile manufacturing line according to this embodiment.
This is a block configuration diagram schematically showing a stamping station that stamps a unique vehicle body number on each vehicle body. A stamping bag [5] for stamping a vehicle body number on a vehicle body brought into the factory, and a stamp inspection device IO for automatically reading and inspecting the stamped vehicle body number are provided.

The engraving bag [5] includes a stamping machine 6 that can automatically stamp a vehicle body number on a predetermined part of the vehicle body (for example, a cowl box), and a stamping controller 7 that controls the stamping operation of the stamping machine 6. The controller 7 receives basic data such as a commit number on the production line of the vehicle body and a vehicle body number corresponding to this number from the host computer 3 of the factory, and stores this data. The marking controller 7 is configured, for example, with a microcomputer as its main part, and is connected to the line controller 2 which controls the entire vehicle body assembly line including the marking station 1.

Further, as shown in FIG. 2, the marking inspection device 10 has the following features:
An illuminator 1 (for example, a high-frequency fluorescent lamp) that shines illumination light on the stamped part of the car body to shade the engraved characters, and the illuminator 11
For example, a plurality of industrial video cameras 12 photograph the illuminated vehicle body number with shading, and as will be explained in detail later, from the image information photographed by the cameras 12, it is possible to determine whether the vehicle body number is actually engraved on the vehicle body. The image processing device I3 is provided with an image processing device 13 that performs pattern recognition of a vehicle body number, and a display device 14, a printer device (not shown), etc. are attached to the image processing device I3.

Hereinafter, an outline of the car body number stamping and inspection process at the stamping station 1 will be described with reference to the flowchart of FIG. 3.

When the line is started and control by the line controller 2 begins, first, a car body conveyor (not shown) is started and the car body is transported to the marking station, where the car body is clamped and fixed to the marking machine 6. (Step #1
to step #3).

Next, the stamping machine 6 is activated and the vehicle body number of the vehicle concerned is stamped on a predetermined location (cowl box) of the vehicle body, and then,
For example, the stamped area is cleaned by air blowing (
Step #4 to Step #6).

Next, the engraved inspection bag! ! 10 is activated, the embossed characters shaded by illumination are imaged by the video camera 12 as a target pattern to be recognized, and the image is taken into the image processing device 13 (step #7). Then, as will be explained in detail later, predetermined image processing is performed on the captured image, and pattern recognition of the engraved characters is performed (step #8).

Then, in step #9, it is collated and determined whether the stamped characters read by pattern recognition based on the image processing described above match the vehicle body number of the vehicle body that has been input and stored in advance, and this determination result is If YES (match), O
The K times signal is output (step #10), and the read vehicle body number is printed along with the commit number and the like by a printer (step #II). When this printing is completed, the marking process for the next workpiece is started (step #12).

On the other hand, the determination result in step #9 above is No (mismatch).
In this case, an NG signal is output and an inspector is called (step #13 and step #14), and the recognized characters read by the pattern recognition and displayed on the monitor screen 14 match the original vehicle body number. It is checked and determined whether or not it is true (step #15). If the determination result is YES (coincidence), in step #16, an OK double signal is output by pressing the manual button (step #!0).

Also, the determination result in step #15 is N.

(In the case of mismatch), the vehicle body number is corrected and stamped on the vehicle concerned (step #17), and abnormality countermeasures are taken for the system according to a predetermined abnormality handling manual. It looks like this.

Next, a character recognition system (
The outline of steps #8 and #9) will be explained with reference to the signal system diagram (FIG. 4) between the marking controller 7 and the image processing device 13 and the flowchart of FIG. 5.

When the image processing system starts, first, preparatory steps are executed, such as selecting which image from the plurality of video cameras I2 is to be input, printing the start time, and checking the presence or absence of a test signal (step #21. #22 and #
23) When these preparations are completed, a READY ON signal is output to the marking controller 7 (step #
24). The test signal is a signal outputted from the marking controller 7 in order to notify the user of the trial stamping. Also, an image processing bag! ! 13, a data retransmission request signal is output to the marking controller 7 during the preparation step.

Next, a signal is received that instructs whether the image processing system is executed in conjunction with the operation of the stamping machine 6 or independently (step #25).

Then, in step #26, it is determined whether or not the stamping machine 6 is to be operated in conjunction with the marking machine 6. If the result of this determination is YES, the system is activated after waiting for input of a activation signal from the stamping machine. When the activation signal is received, basic data such as a commit number and a vehicle body number are subsequently received (steps #27 and #28).

Then, based on the image capture command, the video camera 12
When the image input is completed, an image input completion signal is output to the engraving controller 7 (steps #29, #30 and #31). Note that if the determination result in step #26 is No, that is, if the image processing system is executed independently, step #2
Basic data (see step #28) is received at 8°, and image input in step #30 is performed without waiting for the activation signal (step #27).

Then, the stamped characters (vehicle number) are recognized based on image processing, and the recognized vehicle number is compared with the basic data that has been input and stored in advance, and it is determined whether the two match. is performed (steps #32 and #33).

After this judgment is completed, a reading completion signal is output to the marking controller 7, and the judgment result and commit NO.
9/The vehicle body number is displayed on the monitor screen of the display device l4, and these are printed on the printer (
Step #35. #36 and #37).

If the image processing system is interlocked with the operation of the stamping machine 6, the output signal is reset after waiting for the operation completion signal of the stamping machine 6, and if the image processing system is not interlocked, the output signal is reset (step 43B). , #39 and #40), and then returns to step #23 to repeat the same steps.

Next, a specific example of image processing and pattern recognition by the image processing device 13 will be described with reference to the flowchart of FIG. 6.

First, before inputting the image, read characters for the previous work, store the same characters, and initialize the stored vehicle body number, etc. (Steps #51, #52 and #53)
. Next, basic data for the workpiece is read from a memory means such as a floppy disk, and an image taken by the video camera 12 is input (step #54).
and #55).

Then, in step #56, for the input image,
Window processing is performed to limit the area to be processed within the screen, and this limited area is subjected to binarization processing (step #57) using a predetermined threshold. After that, noise removal processing (step #57) is performed. #58).

In this example, in order to reduce the influence that this shading has on the recognition accuracy of pattern recognition, for the above-mentioned image obtained by adding shading with illumination light after performing the above-mentioned noise removal, A so-called thinning process is performed to narrow the line width of the figure or character displayed (step #59).

Note that the windowing, binarization, noise removal, and line thinning processes described above can all be performed using methods similar to those well known in the past.

Next, in step #60, a series of characters of the target pattern is labeled to determine the order.

In the case of this example, the numbers 0 to 9 and A
For example, a maximum of 19 digits of letters and numbers can be engraved by combining alphabetic characters from Z to Z, and a 14 to 19 digit vehicle body number is assigned depending on the vehicle type.

After that, the area (number of pixels) of each labeled recognized character is calculated, and it is determined whether this area is less than or equal to a predetermined value (for example, 15) (steps #61 and #62)
. If this judgment result is YES, the recognized character is
This is due to noise caused by dust etc. that could not be removed by the noise removal process described above, and it is determined that the label is not actually a letter or number, and the process moves to the next label (Step #6).
3).

On the other hand, if the determination result in step #62 is YES, feature amounts are first extracted in step #64 in order to perform pattern recognition of the character. That is, in the image obtained by the window processing, the area to be pattern-recognized as a character is further limited, and the start coordinates of the image processing are determined. Then, this area is divided into each divided area at a predetermined level of detail (for example, 5 vertical and horizontal divisions) (step #
65). Then, the number of pixels in the 5-divided area is counted, and the coordinates after the 5-divided area are calculated (step #66
and #67). In addition, FIG. 9 shows an example of an image after the above-mentioned thinning, taking the number "2" as an example. The two-dot chain lines in FIG. 9 indicate dividing lines used to divide the image into five parts vertically and horizontally.

In this embodiment, when extracting the feature amounts and performing image segmentation, it is more preferable to set the size of the area to be segmented as a constant size in advance.
The area is determined for each recognized character according to the range recognized as a character, and divided into five areas according to the size of the area corresponding to each character. Therefore, the number of pixels in the area to be divided (see step #66) differs depending on the character. In this way, by varying the area to be divided according to the range recognized as a character,
Image processing can be performed more efficiently than when processing is performed at a constant size regardless of the size, shape, etc. of characters. Further, since the influence of character displacement within the area can be eliminated, more accurate pattern recognition can be performed.

Furthermore, in this embodiment, at this stage (step #6
5) When dividing the above area that was recognized as a character, it was divided into 5 vertically and horizontally, but this was the result of a comparison test with dividing by other numbers (such as 4.6 or IO). , in the case of this example, this five division is
This is because the highest recognition accuracy was generally obtained. Therefore, if the conditions are different, it may be divided by another number.

Next, after performing the above-mentioned 5-division, a window is created to take out a certain arbitrary section of the 5-divided divided area (step #68), and the number of pixels indicating the character part in the divided area, that is, thinning is performed. Extraction target pixel area S
o, the total number of pixels in the divided area, that is, the total pixel area of the divided area St
, and the extraction target pixel area So for the total pixel area St
The ratio Rs is calculated sequentially (steps #69, #70, and #71).

Then, this ratio Rs is set to a predetermined threshold (for example, 10%
) or more, the 2
Assign the signal "2" to the array position corresponding to the section in the first array of No. 5 (Steps #72 and #73)
. Further, if the ratio Rs is smaller than the threshold value and larger than a second threshold value smaller than this (0% for PJ) (0%<Rs<10%), the signal rlJ is further set to If the ratio is 0%, a signal "0" is assigned to the array (steps #74, #75 and #76).

That is, among the 25 arrays for the recognized character, a plurality of signals (o, 1, or 2
) is assigned.

When the signal assignment (assignment) for the above section is completed, the process proceeds to the next section (step #77), the above step #68
Step #76 is then executed. By sequentially repeating this for each section, one of the above signals is applied to all 25 sections for the character, and an arrangement pattern for the character is obtained. and,
By comparing this obtained sequence pattern with the standard data (standard sequence pattern) stored in advance for each character (step #78), it becomes possible to recognize what the character is. ing. FIG. 10 shows an example of the reference arrangement pattern stored in advance, taking the number "2" as an example.

After finishing the pattern recognition for the above characters, step #
At step 79, pattern recognition of the characters of the next label is performed (that is, the steps from step #60 onward are repeated). By repeating this for the total number of digits, pattern recognition of all characters is completed (Step #80), the result is compared with the vehicle body number in the basic data, and the matching result is displayed on the display device and printed on the printer. (Steps #81, #82 and #83).

In this embodiment, the characters to be recognized are hyphen (r-J
), prior to executing step #68, a predetermined number (for example, 4
After adding pixels by the number of pixels) and giving a specific signal (for example, signal "6") to this added part, the above-mentioned steps #68 to #76 are executed to obtain the array pattern. There is. By performing the above processing, it is possible to perform highly accurate pattern recognition even when the Haiphong line has an inclination.

Further, in this embodiment, more preferably, a re-verification program for performing re-verification of each recognized character is provided for each character, and the characters obtained as a result of pattern recognition are processed by this re-verification program. It is now possible to retest and obtain more accurate pattern recognition results.

In particular, for example, the numbers “0” and a, the letter r
For DJs, if there are multiple letters or numbers that have a similar arrangement pattern, put them into one group (set) in advance.
, and when characters with similar patterns within this group are recognized, for a specific partition area,
Image processing is performed with settings different from the previous one, and re-verification is performed by comparing the image processing with different settings with previously stored reference data.

Hereinafter, the re-verification will be described with reference to the flowchart of FIG. 7, taking as an example the case where the number rOJ is recognized.

In the character recognition described above, when all 25 squares of the 5-division array match the standard array pattern and the number "0" is recognized, the "0" re-verification program is started, and first, step At #91, repeat the previous 5 minutes IIJ (step #
65), the thinned image is divided into six vertically and horizontally. It should be noted that each of the 36 first rows in the array obtained by dividing the area to be divided into six vertically and horizontally is assigned a section number from 0 to 35, as shown in FIG. 1L. In addition, the number “0” and the alphabet letter rD have similar patterns.
J means the 30th section and the 35th section in the above 6-division arrangement.
Each section has its own characteristics, and by comparing the two sections, it is possible to clearly distinguish between the two sections.

Next, in steps #92 to #95, for the 30th section of the array divided into six, all pixels of the section are The ratio Rss of the extraction target pixel area So to the area S1. Calculate. Similarly, the ratio R81 for the 35th section is calculated (step #96). It has been previously confirmed that when the number is divided into six, the sum of both ratios is smaller than 40% for the number "0".

Then, in step #97, the sum of both the above ratios (R6,.

+R55s) is smaller than 401%. If YES, it is confirmed that the number is "0", so this rOJ is stored as a recognized character (step #99). On the other hand 1. If the determination result in step #97 is No, NG is displayed (step #99).

In addition, if the last recognized character was the alphabet rDJ, the same steps as steps #91 to #96 above are executed using the rDJ re-verification program, and the sum of the above ratios (R83°+R11ss) is calculated. 40%
or more, the recognition character is rDJ of the alphabet.
It is verified that

Incidentally, examples of other groups with similar patterns include the groups of the number ``8'' and the alphabets rBJ and rVJ, which have the ratio RS f of the 22nd section and the ratio Rs of the first section in the 6-division.

The difference can be clearly distinguished depending on the value of .

Further, as a result of character recognition, a target pattern of a certain character may match a plurality of reference patterns. Therefore, this - scattering is III! If not, NG is displayed. That is, as shown in the flowchart of FIG. 8, character recognition is executed, characters whose target pattern matches the reference pattern are stored, and the number of matches obtained is counted and stored (step # 101 to #Summer 03). Next, it is determined whether or not this -scattering is 1 (step #104), and if No, the character cannot be specified, so NG is displayed (step #105).
), if YES, it is compared with the basic data (step #106), and NG or OK is displayed (step #106).
107 or #108).

In the above embodiment, in the re-verification of recognized characters, when characters in a group with similar arrangement patterns are recognized, the number of image divisions is changed when performing image processing with different settings from the previous one. However, instead of this, for example, the threshold value when performing binarization processing, or the ratio of the extraction target pixel area to the total pixel area Other setting conditions, such as threshold values for determining a plurality of signal values (0, 1, 2) depending on Rs, may be made different.

As described above, according to this embodiment, the image obtained by capturing the target pattern to be recognized (engraved characters of the vehicle body number) with shading is binarized and then thinned,
After that, the ratio Rs of the extraction target pixel area SO to each section area St is calculated, and the subsequent image processing is performed, so accurate pattern recognition is performed without being affected by shading. It is possible.

[Brief Description of the Drawings] The drawings are for detailed explanation of the present invention.
Figure 1 is a block diagram showing the general structure of the stamping station, Figure 2 is a perspective view of the stamping inspection device, Figure 3 is a flowchart explaining the stamping and inspection process at the stamping station, and Figure 4. is a signal system diagram between the engraving controller and the image processing device, and FIG. 5 is a flowchart for explaining the outline of character recognition by the image processing device.
FIG. 6 is a flowchart for explaining a specific example of image processing and pattern recognition by the image processing device, FIG. 7 is a flowchart for explaining re-verification of recognized characters,
FIG. 8 is a flowchart for explaining the process in the case of product matching; FIG. 9 is an explanatory diagram showing an example of an image after thinning; FIG. FIG. 11 is an explanatory diagram showing the arrangement of the eyes in six divisions. 11...Illuminator, I2...Video camera, 13...Image processing device, Rs...Ratio, So pixel area to be extracted, St...Total pixel area of the section. Figure 4 Figure Figure Return to stencil #23 Figure Figure 9 Figure 10 Figure 11

Claims (1)

[Claims] (1) In a pattern recognition method that recognizes a two-dimensional pattern by capturing an image of the target pattern to be recognized in a shaded state using illumination light and performing image processing on the obtained image, an image containing the target pattern is After binarization, thinning is performed, and for each divided area obtained by dividing the image, the ratio of the thinned extraction target pixel area in the divided area to the total pixel area of each divided area is calculated. A pattern recognition method characterized in that the ratio is classified into a plurality of signals according to a predetermined threshold, and the target pattern is recognized based on a combination of sections each having one of these signals.