JP2000259829A - Image recognition method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To quickly extract the shape of a measured object by calculating the density difference between a part corresponding to the measured object and its peripheral part in regard to the kernel component images of the red, green and blue images respectively and using the kernel component image having the largest density difference to extract the edge of the measured object. SOLUTION: A color image FCR is separated into the component images and the noise components are eliminates. Then the density difference of every component image is calculated along a scanning line L crossing the outer fringe of an image pattern TP that is obtained by photographing a measured object T. In other words, the density differences CR, CG and CB of the R, G and B images are calculated respectively. These differences CR, CG and CB are compared with each other and the component image having the largest density difference is used to extract the edge of the object T. Thus, it is possible to quickly extract even the border line of an object that cannot be extracted with its monochromatic image since the density difference of this image cannot be calculated.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition apparatus used in a two-dimensional measuring device or a three-dimensional measuring device for measuring an object to be measured by imaging the object to be measured placed on a horizontal table from above. About the method.

[0002]

2. Description of the Related Art As a conventional two-dimensional measuring machine, for example, Japanese Patent Laid-Open Publication No. Hei 8-189813 has a horizontal table on which an object to be measured is placed, and a predetermined distance is provided above the object to be measured. Moved in the X-Y direction parallel to the table
A camera equipped with a D camera (imaging means) is known. When measuring an electrode of a panel used for a plasma display, for example, using such a two-dimensional measuring device, an image of the electrode is taken by a CCD camera, an edge of the electrode is extracted from the image data obtained, and coordinates of the edge are extracted. Measure the dimensions of the electrodes from, or extract the alignment marks of the panel, and perform pattern matching to determine whether the shape of the alignment marks matches the shape of the regular master pattern stored in advance. If the shapes match, the alignment mark of the panel to be measured is measured, and the coordinate system of the panel is set. The color images are three primary colors of light: a red image (R image), a green image (G image),
It can be decomposed into three color constituent images of a blue image (B image). When extracting an edge, an edge portion to be measured is specified, and edge detection is performed for each of the three color component images.
When performing pattern matching, the outer edge of the pattern is extracted by image processing. However, in the case of a color image, the pattern shape must be extracted for each of the three color component images. The same pattern shape is not always extracted. Converting a color image into a single-color image composed of density difference data solves such a problem.
As a method of converting a color image into a monochromatic image, for example, according to Japanese Patent Application Laid-Open No. H10-145612, a conversion vector for determining a ratio of three colors constituting a color image so that a density difference of the image is emphasized is obtained. There is known a method of combining three colors based on a conversion vector and converting the three colors into a single-color image.

[0003]

In the above-described conventional method of converting a color image into a single-color image, a conversion vector is obtained from a captured color image, and further, an operation is performed to obtain a composition ratio of each color based on the conversion vector. A problem that time is required occurs.

In view of the above problems, an object of the present invention is to provide an image recognition method capable of extracting a shape of a measured object from a color image in a short time.

[0005]

According to a first aspect of the present invention, an edge, which is a boundary of a pattern of a device under test, is detected from a color image of the device under test captured by an image capturing means. In the method of recognizing an image from the position of the edge, a color image is separated into a red image (R image), a green image (G image), and a blue image (B image). The density difference between the portion corresponding to the device under test and the portion surrounding the portion is determined for the constituent images, and the edge of the device under test is extracted using the component image having the largest density difference among the component images. Features.

According to a second aspect of the present invention, there is provided an image recognition method for extracting a pattern shape of an object to be measured as a master pattern stored in advance from a color image of the object to be measured captured by an imaging means. The master pattern is extracted from the constituent image having the largest density difference among the constituent images of the red image (R image), the green image (G image), and the blue image (B image) constituting It is characterized in that a pattern shape is extracted from a constituent image of the same type as a constituent image from which a master image is extracted, among constituent images constituting a color image.

Since the density difference between the three color component images is often different, if the shape of the object to be measured is extracted using the component image having the largest density difference among the component images, the conventional extraction method can be used. It does not require such complicated calculations. When performing pattern matching, a master pattern is extracted from a constituent image having the largest density difference, and if a pattern shape relating to the device under test is extracted from a constituent image of the same type as the constituent image from which the master pattern is extracted, the master pattern is extracted. Pattern matching can be performed by comparing patterns and extracted pattern shapes under the same conditions.

If the image contains noise, the separated red image (R image), green image (G image), and blue image (B image) are subjected to filter processing to remove the noise contained in the image. It is desirable to remove it.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Referring to FIG. 1, reference numeral 1 denotes a two-dimensional measuring machine provided with a flat table 11 having a semi-permeable property such as ground glass. On the table 11, a printed wiring board T as an object to be measured is placed.
On the table 11, X-
A CCD camera 2, which is an imaging means movable in the Y direction, is attached. The image captured by the CCD camera 2 is a color image. On both sides of the CCD camera 2, epi-illumination devices (not shown) coaxial with a microscope optical system for guiding illumination light, which are not shown, are attached to the imaging position of the CCD camera 2. Also, as shown in FIG.
Is disposed below the light-emitting device 12. When the device T has a light-transmitting property, the surface light-emitting device 12 emits light, and the pattern shape of the device T becomes a silhouette. In a case where the printed circuit board does not have a light-transmitting property as in the case of a printed wiring board, light is irradiated onto the upper surface of the DUT via the optical fiber 21. Image data from the CCD camera 2 is taken into a computer 3 including an image processing device 31 and displayed as a color image on a color monitor 35 installed near the two-dimensional measuring device 1. When a signal of a color image is input from the CCD camera 2 to the computer 3, the computer 3 separates the color image into constituent images of a red image (R image), a green image (G image), and a blue image (B image). I do. When there is an unnecessary pattern that becomes an error factor when extracting a pattern shape, filter processing is performed to remove the unnecessary pattern that is noise. The filter processing may be a conventionally known filter processing such as a median filter, and removes a noise component having a prominent value such as pulse noise from a captured image and removes a signal other than a signal constituting a target pattern shape. I do. When the noise component is removed from each of the constituent images in this manner, the image processing apparatus combines the constituent images again and causes the color monitor 35 to display the color image. Further, as shown in FIG. 3, after separating the color image FCR into the respective constituent images and removing the noise components, the respective components are arranged along the scanning line L straddling the outer edge of the image pattern TP ′ obtained by imaging the device under test T. A density difference is obtained for each image. That is, the density difference CR in the R image and the density difference C in the G image
A density difference CB between the G and B images is obtained. The length of the scanning line L is arbitrarily defined by determining a start point and an end point by operating a mouse. The scanning line L may be defined obliquely. Since the scanning line L can be freely defined in this way, it can be defined at an appropriate position according to the density of the image. Then, the respective density differences are compared, and the constituent image having the largest density difference is specified. In the present embodiment, the density difference CG in the G image is larger than the density differences CR and CB in the other constituent images. Therefore, the G image FG is selected from the constituent images.
Select When there are a plurality of constituent images having the same density difference, priorities are assigned to the constituent images, and a constituent image having a higher priority is selected. When detecting the edge of the image pattern TP ′, the coordinates of a point where the density difference of the G image at a position or range specified on the image of the monitor 35 exceeds a predetermined threshold value is set as the coordinates of the image pattern TP ′.
The pattern width and the pattern interval can be measured by similarly obtaining the coordinates of the edge of another opposing image pattern and comparing the two coordinates. Even when the shape of the image pattern TP ′ includes a curve such as a circle, the coordinates and dimensions of the edge can be similarly obtained. Also, as shown in FIG.
When performing pattern matching, the G image FG is binarized using a predetermined threshold. Based on the binarized G image, the periphery of the image pattern TP 'is obtained by a conventional processing method, and the size and position of the image pattern TP' are obtained. In the embodiment shown in FIG. 3, the pattern shape does not change even if the color image is separated into each component image. However, if the pattern shape portion is not completely one color, the pattern image is separated into each component image. The pattern shape differs for each constituent image. Even in such a case, there is no problem in performing pattern matching. As shown in FIG. 5, a region TD including a pattern to be a master is designated in the color image FCR.
Then, the R image FR, the G image FG, and the B
Image FB. The region T is set in various shapes by a measurer. For example, the area TD may be circular or rectangular, and the area TD may include a plurality of image patterns, or may include only one image pattern. At this time, each range TD appearing in each constituent image
The density difference of each shape in R, TDG, and TDB is different. Assuming that the density difference in the G image FG is the largest, the range TDG including the pattern shape is set to the master image G-MAST.
It is registered in the computer 3 as ER. At that time, the master image F-MASTER and the master image F-
The data that the master is the G image FG is also registered. Next, a color image 4G obtained by capturing another position of the device under test to be subjected to pattern matching is input to the image processing device 31, and the region TD displayed on the monitor image is specified. The image processing device 31 performs the color image 4G
The above-mentioned master image F-MASTER is called from the memory of the computer 3 in order to perform pattern matching. At that time, the master image F-MASTER is G
The data that was the image FG is also called. Then, the image processing device 31 extracts an image TDG ′ having a large density difference from the constituent images constituting the color image 4G, and performs pattern matching between the master image F-MASTER and the image TDG ′. When the pattern shapes are different, pattern matching is performed using the correlation coefficient. For example, the difference from the master pattern is divided into nine steps, for example, 0.9, 0.8, 0.7,... Thus, a criterion for pattern matching is provided. As described above, if pattern matching is performed using the correlation coefficient, inspection can be quickly performed on a plurality of printed wiring boards. In addition, using the correlation coefficient enables not only inspection but also pattern matching and search to be performed quickly, and by combining with the above-described method of obtaining a density difference, a coordinate position of a pattern shape can be obtained. You can quickly know if you are in a position. In the above embodiment, the present invention is applied to a two-dimensional measuring machine. However, the present invention can be applied to a three-dimensional measuring machine.

[0010]

As is apparent from the above description, the present invention does not generate a single-color image from three constituent images constituting a color image as in the conventional method, but instead generates one of the constituent images. Is used, the image processing speed is higher than in the conventional method. Further, in the conventional monochrome image, even if the boundary line cannot be extracted because there is no density difference, the boundary line can be quickly extracted.

[Brief description of the drawings]

FIG. 1 is an external view of a two-dimensional measuring machine to which the present invention is applied.

FIG. 2 is a block diagram illustrating a configuration of an image processing apparatus.

FIG. 3 is a conceptual diagram illustrating a first embodiment of the present invention.

FIG. 4 is a conceptual diagram illustrating an edge extraction step.

FIG. 5 is a conceptual diagram illustrating a second embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 2D measuring machine 2 CCD camera (imaging means) 3 Computer 31 Image processing apparatus

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04N 17/04 G06F 15/70 330Z 5C082 H04N 1/40 101C 5L096 1/46 Z 9A001 F term (Reference) 2F065 AA07 AA12 AA51 BB02 BB13 BB22 BB24 CC18 CC25 DD06 FF01 FF02 FF04 GG18 HH16 JJ03 JJ26 LL02 PP03 PP24 QQ04 QQ24 QQ33 QQ38 QQ41 UU05 5B057 CE02 CE06 DA07 DB06 DC16 DC22 5C06101B03 CC01 CCB5C011 CC1 DA87 MM02 5L096 AA02 EA05 FA14 FA69 GA40 9A001 BB02 BB03 BB04 BB06 DZ08 EE02 EE05 FF03 HH21 HH25 HH28 HH29 HH31 JJ45 KK37 KK54 LL05

Claims (3)

[Claims] 1. A method for detecting an edge, which is a boundary line of a pattern of a device under test, from a color image of the device under test captured by an imaging unit and recognizing the image from the position of the edge, forms a color image. Each of the three component images is separated into a red image (R image), a green image (G image), and a blue image (B image). An image recognition method comprising: obtaining a density difference of a target object; and extracting an edge of an object to be measured using a constituent image having the largest density difference among constituent images. 2. An image recognition method for extracting a pattern shape of an object to be measured as a master pattern stored in advance from a color image of the object to be measured imaged by an imaging means, wherein a red image (R) Image), a green image (G image), and a blue image (B image), the master pattern is extracted from a constituent image having the largest density difference among constituent images, and a constituent image forming a captured color image An image recognition method characterized by extracting a pattern shape from a constituent image of the same type as a constituent image from which a master image has been extracted. 3. A filter process is performed on the separated red image (R image), green image (G image), and blue image (B image) to be included in the image and cause an error when extracting an edge. 2. The method according to claim 1, wherein noise included in the image is removed in order to remove data on an unnecessary pattern.
Alternatively, the image recognition method according to claim 2.