JPH0737085A - Reference pattern preparation device - Google Patents

Abstract

PURPOSE:To eliminate the need of alignment among plural patterns by supplying a typical initial reference pattern first and adding correction by the pattern matched with it. CONSTITUTION:Pictures are inputted to a picture storage part 12 by a camera 10 and matching with the initial reference pattern stored in a reference pattern storage part 24 is performed by a matching part 26. The matching part 26 matches an initial storage pattern and the partial pictures of a size and a shape same as the initial reference pattern at respective positions in the pictures and sends the position data of the partial pictures whose matching degree is equal to or more than a threshold value stored in a threshold value storage part 28 to a picture segmenting part 18. Thus, the patterns similar to the initial reference pattern of equal to or more than a fixed matching degree can be segmented from the pictures for learning. The patterns are provided with the variation of density to a degree approximately the same as the pictures at the time of detecting the patterns, the reference pattern close to the patterns in the actual pictures can be obtained by averaging them and the appropriate threshold value of the matching degree can be recognized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to an image obtained by image input means such as an industrial television camera (ITV).
The present invention relates to a reference pattern creation device that creates a reference pattern used for pattern matching for identifying and positioning an object in an image by detecting a specific pattern.

[0002]

2. Description of the Related Art Conventionally, even when the shape of a pattern to be detected is known in advance, it is possible to predict what kind of pattern it will be in an image input through a camera or the like under specific optical conditions. Is difficult, and it is common to cut out the reference pattern from the image input under the same conditions as the actual detection target.

At this time, in order to cut out a plurality of patterns for averaging, it is necessary to align the positions of the plurality of patterns.

As a method of defining the position of the pattern,
For example, JP-A-62-200252 and JP-A-4-1134
As disclosed in 87, a method based on the center of gravity is known.

The method of defining the pattern position by the center of gravity can accurately specify the position only when the pixel value of each part of the pattern is always constant, and is generally applied to a binary image.

On the other hand, as a method of pattern matching, not only a binary image is targeted, but also Japanese Patent Laid-Open No.
A method based on normalized cross-correlation is also known, as disclosed in 283881. Here, the normalization has the effect of compensating for variations in which the same pattern appears in an image at various intermediate densities due to variations in illumination conditions, etc., assuming the existence of intermediate densities. The pattern matching method using an image having an intermediate density, which is represented by the normalized cross-correlation method, is capable of compensating for density variations due to variations in illumination conditions and the like, which is a characteristic compared to binary image pattern matching. It is premised that the target image has a variation in density.

[0007]

However, when attempting to align the positions of the reference patterns, which are cut out from images having variously varying intermediate densities, the use of the center of gravity causes the influence of the variation in the densities in the patterns. There is a problem that the position of the center of gravity fluctuates greatly.

Therefore, when trying to obtain a highly reliable reference pattern by averaging a plurality of reference patterns, it is difficult to align the plurality of patterns.

The present invention has been made to solve the above-mentioned conventional problems, and an object of the present invention is to provide a reference pattern creating apparatus capable of creating a highly accurate reference pattern without manually cutting out many patterns. And

[0010]

According to the present invention, in a reference pattern creating apparatus for creating a reference pattern used when matching an image and a reference pattern, the means for creating an initial reference pattern is the same as the matching processing target. Matching means for matching the learning image input under the conditions with the initial reference pattern, and all the partial images having the same size and shape as the initial reference pattern, the matching degree of which in the matching exceeds a certain value. An image cutting-out unit that cuts out from the learning image and a unit that averages pixel values of all partial images for each pixel position of the cut-out partial image group, and averages each pixel position The above object is achieved by creating a reference pattern in which the converted pixel values are used as pixel values at the same pixel position.

[0011]

According to the present invention, an initial reference pattern is provided for pattern matching, and the position of the reference pattern in the learning pattern for cutting out the reference pattern is set to the initial reference pattern regardless of the means such as the center of gravity. It is designed to find the position where the degree of matching is a certain value or more.

That is, first, a constant value for determining the degree of matching with the initial reference pattern is given, and then the learning image input under the same conditions as the matching processing target is matched with the initial reference pattern. All the partial images of the same size and shape as the initial reference pattern, where the degree of matching exceeds the constant value given above, are cut out from the learning image, and for each pixel position of the obtained partial image group. In addition, a pattern having a pixel value that is the average of the pixel values of all partial images is created, and this is used as the reference pattern.

[0013]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention.

In FIG. 1, 10 is a camera, 12 is an image storage unit, 14 is an image display unit, 16 is an initial pattern creation operation unit, 18 is an image cutout unit, 20 is a pattern storage unit,
22 is an averaging calculation unit, 24 is a reference pattern storage unit, 26
Is a matching unit, 28 is a threshold value storage unit, 30 is a counter,
32 is a result output unit.

The operation of this embodiment will be described below.

The image input by the camera 10 is stored in the image storage unit 12 and displayed on the image display unit 14.

First, when the initial reference pattern is created, the initial pattern creation operation unit 16 sends the coordinate data in the image of the initial pattern to the image cutout unit 18, so that the initial reference pattern is stored in the pattern storage unit 20 and the averaging operation. Part 22
It is sent to the reference pattern storage unit 24 via.

Next, the reference pattern is learned.

The camera 10 inputs an image to the image storage unit 12, and the matching unit 26 performs matching with the initial reference pattern stored in the reference pattern storage unit 24. As shown in FIG. 2, the matching unit 26 scans the image while matching a partial image having the same size and shape as the initial reference pattern at each position in the image with the initial storage pattern, and the threshold storage unit 28. The position data of the partial image having the matching degree equal to or higher than the threshold value stored in is sent to the image cutout unit 18.

The processing performed at each position of the image as matching is performed by using f (x, y) for the input image and g (x, y) for the reference pattern.
y), the position of the reference pattern is (u, v), the shape of the reference pattern is rectangular, and the dimensions are (w, h). For example, in the normalized cross-correlation method, the matching degree is given by the following equation (1). To be

[0022]

[Equation 1]

The SSDA method (Sequential Similarity)
According to the detection algorithm (residual sequential test method), the reciprocal of the matching degree is calculated by the following equation (2).

[0024]

[Equation 2]

However, even during the calculation, the calculation is terminated when the value of the degree of match (u, v) in this equation exceeds a certain threshold value.

Matching is performed by the above processing.

Based on this data, the image cutout unit 18 sends a partial image from the image in the image storage unit 12 to the cutout pattern storage unit 20 and, at the same time, updates the counter 30.

The pattern storage unit 24 stores a plurality of partial images, and the counter 30 stores the number of partial images. When the cutout of the partial images is completed, the averaging calculation unit 22 averages the partial images in the pattern storage unit 20 using the value of the counter 30, and the average of the partial images in the reference pattern storage unit 24 is calculated for each pixel. make.

Since the matching section 26 also sends the position data to the result output section 32, the pattern position can be detected by taking this out. Also, during that time, the number of partial images in the pattern storage unit 20 can be increased to improve the reference pattern.

Therefore, according to this embodiment, it is possible to cut out a pattern similar to the initial reference pattern with a certain degree of matching or more from the learning image. Since these patterns are cut out from the learning image input under the same conditions as in the actual pattern detection, they contain the same density variations as in the pattern detection image. A reference pattern close to the pattern in the actual image can be obtained, and an appropriate threshold value of the matching degree can be known.

[0031]

As described above, according to the present invention,
The standard pattern used for pattern matching is not created by averaging many patterns cut out from the actual image at once, but first, a typical initial reference pattern is given and correction is performed using the pattern that matches this. Since it is added, the alignment between a plurality of patterns becomes unnecessary.

Further, since the creation of the reference pattern and the actual matching are the same processing, the reference pattern can be improved to a more reliable one while actually operating the matching system, and many patterns are cut out manually. This has the effect that a highly accurate reference pattern can be created.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of a reference pattern creating apparatus according to this embodiment.

FIG. 2 is an explanatory diagram showing image matching according to the present embodiment.

[Explanation of symbols]

 16 ... Initial pattern creation operation unit 18 ... Image cutout unit 20 ... Pattern storage unit 22 ... Averaging calculation unit 24 ... Reference pattern storage unit 26 ... Matching unit 28 ... Threshold storage unit

Claims (1)

[Claims] 1. A reference pattern creating apparatus for creating a reference pattern used when matching an image with a reference pattern, a means for creating an initial reference pattern, and a learning image input under the same conditions as a matching processing target. And a matching unit for matching the initial reference pattern with the initial reference pattern, and an image cutout that cuts out all partial images having the same size and shape as the initial reference pattern, the matching degree of which exceeds a certain value in the matching, from the learning image. And a means for averaging the pixel values of all the partial images for each pixel position of the clipped partial image group, wherein the pixel values averaged for each pixel position are the same. A reference pattern creating apparatus, which creates a reference pattern having pixel values at pixel positions.