JPH0991431A - Pattern recognition method utilizing color ccd camera - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recognize an object accurately even when a label is in existence at an optional position. SOLUTION: A label 2 whose color differs from a product surface color is stuck to an area of a product 1 where its surface color is a single color, the label 2 is picked up by a color CCD camera 3 by using the product surface color around the label for a background color, an image signal outputted from the color CCD camera 3 is given to an image processing unit 4, in which the image is processed for arithmetic operation. Then a brightness difference image is obtained between the different signal color images, the background image is obtained by inverting the picture element value of the label image equally to that of the background image on the difference image being the binarized luminance difference image, the extracted image between the background image and the difference image is obtained, and the label image is detected by retrieving a small image area on the extracted image.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technique for recognizing a label attached to a product in a manufacturing line or the like, and uses a color CCD camera for accurately recognizing whether a label is correct or not without being affected by a label position error. Pattern recognition method.

[0002]

2. Description of the Related Art In a production line for manufacturing tractors, etc., a nameplate (label) indicating the model etc. is attached to the body of the tractor, and whether or not the type of the label matches the model of the tractor. It is inspected by visual inspection.

It is considered that this inspection process can be automated by using an image processing apparatus. As a method of checking the correctness of the label using the conventional image processing technology,
For example, there is a method shown in FIGS. In the method shown in FIG. 6, the product is arranged in the photographing area of the camera with relatively gentle positioning accuracy, and the label 12 is imaged with the product as the background 11. The captured image is binarized to form the input image IP (a in FIG. 6), and the input image I
The comparison reference image S formed by scanning (searching) the entire screen area of P to detect the image (pattern) of the label 12 and imaging the detected image of the label 12 and the correct label 13 in advance.
The pattern was matched with P1 (b in FIG. 6).

In the method shown in FIG. 7, the product is arranged in the photographing area of the camera with strict positioning accuracy, the label 12 is imaged with the product as the background 11, the imaged image is binarized, and the input image IP ( 7 a) is formed. On the other hand, while the correct label 13 is correctly positioned on the product in advance and the product is placed in the photographing area of the camera with strict positioning accuracy, the label 13 is imaged with the product as the background 11 and the imaged image is binarized. The comparison reference image SP2 (b in FIG. 7) is formed. Then, the input image I
Image calculation is performed between P and the comparison reference image SP2 to calculate the difference in pixel value between the two images to form a difference image DP (c in FIG. 7), and based on the shape of the label image in the difference image DP. The correctness of the label is judged.

[0005]

However, in the method shown in FIG. 6, since the entire screen area of the input image IP is searched, the processing time is long and the shape similar to the background around the label 2 is obtained. If there is, there is a problem that the shape of the label is erroneously recognized. In the method shown in FIG. 7, since the difference between the input image IP and the comparison reference image SP2 is calculated, if there is a position error in the position of the label 2 with respect to the product and the position of the product within the photographing area of the camera,
Since this position error appears as a value indicating the difference between the labels 2 in the difference image DP, there is a problem in that the labels 2 are erroneously recognized.

The present invention solves the above-mentioned problems, and
An object of the present invention is to provide a pattern recognition method using a color CCD camera that can accurately recognize a label to be inspected at an arbitrary position in a captured image.

[0007]

In order to solve the above-mentioned problems, a pattern recognition method using a color CCD camera of the present invention is a color recognition method for a label attached to the surface of a product.
A method of recognizing a label pattern by picking up an image with a CD camera and processing the picked-up image with an image processing apparatus, wherein a surface color different from the product surface color is formed in an area where the surface color of the product is monochromatic. A label is attached, and this label is used as color C with the product surface color around the label as the background color.
An image signal captured by a CD camera and output from a color CCD camera is input to an image processing device, and in the image processing device, a single color image is formed for each of the three primary colors based on the image signal, and a product surface color is obtained. A difference image is formed by performing an image calculation for obtaining a difference in luminance value between a corresponding single color image and a single color image corresponding to the label surface color, and the difference image is binarized to form a difference image. However, by inverting the pixel values of the label image that form a closed area surrounded by the background image on the difference image to the same values as the pixel values of the background image around the label image, the background from which the label image has been removed By forming an image and performing an image calculation to obtain the difference in pixel value between the background image and the difference image, an extracted image in which only the pixels of the label image have different values on the entire screen area is formed. Set the small image area including the label image above is obtained by a structure for detecting the label image by scanning the small image area.

In this structure, in a normal picked-up image picked up by a color CCD camera, the surface color of the member to which the label is attached forms the background color around the label, and this product surface color forms a large area and on the screen. , The surface color of other members of the product exists as a disturbance element. In a monochromatic image corresponding to the product surface color, the background image around the label image has a high brightness value, and the image of the label image and other members on the product that are disturbance elements have a low brightness. On the other hand, in a single-color image corresponding to the label surface color, the background image around the label image has a low brightness value, and the image of the label image and other members of the product having a surface color similar to the label surface color have a high brightness. . Therefore, in the brightness difference image formed by calculating the difference in brightness value between both monochromatic images, the brightness difference between the pixel of the label image and the pixels of the background image around the label image becomes large, and the label image and the background The boundary with the image becomes clear, and the brightness difference between the pixels of the background image becomes relatively smaller than the brightness difference with other images. By this, it is possible to eliminate an element that causes an error in the process of binarizing the brightness difference image to form the difference image,
A label image having an accurate shape can be formed on the difference image.

In the extracted image formed by the difference between the background image and the difference image, only the label image has an inverted value, so setting a small image area including the label image on the entire screen area is a disturbance element. It can be performed easily and accurately without being affected. Since the label image is detected by scanning the inside of this small image area, the label image can be recognized in a short time.

Here, if the detected label image is pattern-matched with a comparison reference image prepared in advance, the correctness of the label can be accurately determined.

[0011]

Embodiments of the present invention will be described below with reference to the drawings. 1 to 5, a product 1 such as a tractor moving on a manufacturing line has a label 2 such as a name plate attached to a coated surface. In this label 2, the surface color of the product 1 is a single color (red in this embodiment)
The label 2 is a single color (green type in the present embodiment) different from the product surface color (red type).

The color CCD camera 3 takes an image of the label 2 with the product surface color around the label as a background color, and an image signal output from the color CCD camera 3 is output to the image processing device 4.
Input to the monitor 5 and perform image processing while displaying on the monitor 5,
The pattern (shape) of the label 2 is recognized and the correctness of the label 2 is determined. The color CCD camera 3 outputs an image signal in which the reflected light from the detection object is decomposed into three components for each hue (R, G, B) of the three primary colors, but a different color CCD camera for each hue is used. It can also be installed. In a normal captured image captured by the color CCD camera 3, the surface color (red) of the member to which the label 2 is attached constitutes the background color around the label, and the surface color (red) of the product 1 forms a large area. The surface colors of other members of the product 1 are present as disturbance elements on the screen.

In the image processing device 4, first, a monochromatic image is formed for each hue (R, G, B) of the three primary colors based on the image signal. Next, as shown in FIG. 2, a monochromatic image corresponding to the product surface color (red), that is, the R image Mr (a in FIG. 2), and a monochromatic image corresponding to the label surface color (green), that is, the G image Mg. (B in FIG. 2), an image calculation for obtaining a difference in luminance value is performed to obtain a luminance difference image Mt (c in FIG. 2).
To form

In the R image Mr (a in FIG. 2) corresponding to the product surface color (red), the background image B around the label image L has a high brightness value, which is the label image L and the disturbance element 1. The image N of the other member has low brightness. However,
The disturbance element O that resembles the product surface color (red) has high brightness. On the other hand, the G image Mg corresponding to the label surface color (green)
In (b of FIG. 2), the background image B around the label image L has a low luminance value, and the image N of the label image L and the other member of the product 1 having a surface color similar to the label surface color (green). Becomes high brightness.

Therefore, both monochromatic images, that is, the R image Mr.
In the brightness difference image Mt (c in FIG. 2) formed by calculating the difference in brightness value between the G image Mg and the G image Mg, the brightness difference between the pixels of the label image L and the pixels of the background image B around the label image is As the size increases, the boundary between the label image L and the background image B becomes clear, and the luminance difference between each pixel of the background image B becomes relatively smaller than the luminance difference between the other images N.

Next, as shown in FIG. 3, the brightness difference image M
The difference image Md is obtained by binarizing t (c in FIG. 2).
(A in FIG. 3) is formed. At this time, the brightness difference image Mt
In (c of FIG. 2), the boundary between the label image L and the background image B is clear, and the difference in luminance between the pixels of the background image B is small, so binarization is performed to form the difference image Md (a in FIG. 3). It is possible to eliminate an element that causes an error in the process of performing, and it is possible to form a label image L having an accurate shape on the difference image Md (a in FIG. 3). Further, the pixel value (0) of the label image L forming a closed area surrounded by the background image B on the difference image Md (a in FIG. 3) is set to the label image L.
The background image M from which the label image L has been removed by inverting it to the same value as the pixel value (1) of the background image B around the
b (b in FIG. 3) is formed. At this time, the area to be inverted is conditioned to be a closed area surrounded by the background image B, so that the image N other than the label image L is not inverted and maintains the original value.

Next, as shown in FIG. 4, the background image Mb
By performing an image calculation for obtaining a difference in pixel value between the difference image Md (a in FIG. 4) and the difference image Md (b in FIG. 4), a different value (1) is obtained only in the pixel of the label image 1 on the entire screen area. An eggplant extracted image Mo is formed.

Next, as shown in FIG. 5, the extracted image Mo
The small image area As including the label image L is set on (a in FIG. 5), and the inside of the small image area As is scanned to detect the label image L. At this time, in the extracted image Mo, since only the label image L has an inverted value, narrowing down the small image area As including the label image L on the entire screen area can be easily performed without being affected by a disturbance element, And it can be done accurately. Further, since the label image L is detected by scanning the inside of the small image area As, the label image L can be recognized in a short time. Furthermore, if the detected label image L is pattern-matched with the comparison reference image Mf (b of FIG. 5) prepared in advance, the correctness of the label 2 can be accurately determined.

[0019]

As described above, according to the present invention, the boundary between the label image and the background image is made clear by obtaining the difference in the brightness value between the monochromatic images of different hues, and the background image is separated. It is possible to reduce the difference in luminance between pixels, eliminate error factors in binarization, and form an accurate label image. Since only the label image has the inverted value in the extracted image, the setting of the small image area including the label image on the entire screen area can be easily and accurately performed without being affected by the disturbance element.

[Brief description of drawings]

FIG. 1 is a schematic diagram showing a configuration of image processing equipment according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an image processing method in the same embodiment.

FIG. 3 is an explanatory diagram showing an image processing method in the same embodiment.

FIG. 4 is an explanatory diagram showing an image processing method in the same embodiment.

FIG. 5 is an explanatory diagram showing an image processing method in the same embodiment.

FIG. 6 is an explanatory diagram showing a conventional image processing method.

FIG. 7 is an explanatory diagram showing a conventional image processing method.

[Explanation of symbols]

 1 product 2 label 3 color CCD camera 4 image processor 5 monitor

Claims (1)

[Claims] 1. A color C attached to a label attached to the surface of a product.
A method of recognizing a label pattern by picking up an image with a CD camera and processing the picked-up image with an image processing apparatus, wherein a surface color different from the product surface color is formed in an area where the surface color of the product is monochromatic. A label is attached, and this label is used as color C with the product surface color around the label as the background color.
An image signal captured by a CD camera and output from a color CCD camera is input to an image processing device. In the image processing device, a monochromatic image is formed for each of the three primary colors based on the image signal, and a product surface color is obtained. A difference image is formed by performing an image calculation for obtaining a difference in luminance value between a corresponding single color image and a single color image corresponding to the label surface color, and the difference image is binarized to form a difference image. Then, by inverting the pixel value of the label image forming a closed area surrounded by the background image on the difference image to the same value as the pixel value of the background image around the label image,
An extracted image in which only the pixels of the label image have different values on the entire screen area by forming a background image from which the label image has been removed and performing an image operation to obtain the pixel value difference between the background image and the difference image. And a small image area including the label image is set on the extracted image, and the label image is detected by scanning in the small image area.
A pattern recognition method using a camera.