JPH06347235A - Pattern identifying apparatus - Google Patents

Abstract

PURPOSE:To provide the highly universal apparatus, which can perform high- speed processing and can be applied on various kinds of objects in the pattern identifying apparatus for identifying the object on the basis of the images obtained by picking up the images of the object in a plurality of directions. CONSTITUTION:An image synthesizing part 1 receives the images for individual areas, which are obtained by picking up the images of an object in a plurality of directions and forms the image, wherein the above described images are combined. A position correcting part 2 corrects the position of the image of the object in the image in the individual area within the combined image into the specified position. An arrangement correcting part 3 corrects the mutual arrangements of the images in the areas in the combined image into the specified arrangement. A pattern identifying part 4 receives the combined image, wherein the position is corrected and the arrangement is corrected, handles the entire combined image as one image and performs the pattern recognition.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention generally relates to a pattern identification device, and more particularly, it identifies an object such as a three-dimensional object or a two-dimensional object which requires imaging from a plurality of directions to identify the object. The present invention relates to a pattern identification device suitable for

[0002]

2. Description of the Related Art In recent years, an identification system for identifying various objects has been developed by making full use of computer technology. A typical one is an OCR (optical character reading device) for identifying characters such as characters and symbols written on a secondary surface such as a paper surface. As a further development, there is a system for identifying three-dimensional objects such as ordinary industrial products and agricultural products.

An example of this three-dimensional identification system is an appearance inspection system for automating a visual appearance inspection step for inspecting the quality of a product which is the final step of a production line in a factory. This appearance inspection system has already been put to practical use in IC substrate manufacturing lines and the like.

Regarding the general automation technology relating to this appearance inspection system, "Automatic Visual Inspection Inspection Cases by Image Processing" (published by New Technology Communications) edited by the Visual Inspection Inspection Automation Technology Research Committee and "Computer by Hideyuki Tamura" Image Processing: Applied Practice Edition 1 ”(Soken Shuppan).

[0005]

In the conventional identification technique described in the above publications, the features required for identification are extracted from the image obtained by picking up the object from one direction.
Therefore, when identifying an object that requires observation from multiple directions, such as a three-dimensional object, first feature extraction is performed from each image captured from various directions, and then the extracted features in various directions are extracted. Pattern identification is performed by using it in an integrated manner.

However, it is not always easy to use the features of various directions in an integrated manner in the pattern identification process so as to enable highly accurate identification, and the identification process becomes very complicated. In some cases, a satisfactory processing speed cannot be obtained.

Further, the optimum method for integrating the characteristics of each direction varies depending on the shape of the target object, the variation of the posture at the time of supply, etc. Therefore, it is necessary to design a unique pattern identification program for each system. Therefore, it is difficult to construct a highly versatile system applicable to various objects.

Therefore, an object of the present invention is to enable high-speed processing in a pattern identification apparatus for identifying an object such as a three-dimensional object on the basis of picked-up images from a plurality of directions, and to be applicable to various objects. To provide a highly versatile device.

[0009]

The pattern identification apparatus of the present invention is an image combining means for combining a plurality of direction-specific images obtained by imaging an object from a plurality of directions to generate a combined image, and for each direction. Position correction means for correcting the relative position of the object image in the image to a predetermined position, arrangement correction means for correcting the arrangement of the plurality of direction-specific images in the combined image into a predetermined arrangement pattern, and the position correction means for the target And a pattern identification unit that performs pattern identification by treating the combined image as a single image for the combined image in which the object image is corrected and the arrangement of the direction-specific images is corrected by the arrangement correcting unit. Is characterized by.

[0010]

In the apparatus of the present invention, a plurality of direction-specific images obtained by picking up an image of an object from a plurality of directions are combined to form one combined image. In this combined image, the position of the object image in each direction-specific image is corrected to a predetermined position (for example, the center position in each direction image), and the direction-specific image is arranged in a predetermined arrangement pattern ( For example, the arrangement is modified so that the arrangement is in the order according to the area ratio of the black and white area. As a result, in the combined image, the images of the object viewed from a plurality of directions are arranged at a constant position in a regular array. After that, the entire combined image is treated as one image, and pattern identification is performed.

In the pattern identification processing, since the object images of different directions are regularly arranged in the combined image, the feature quantity extracted from these object images is generally compared with the prepared reference feature quantity. It is possible to obtain a highly accurate identification result by various identification methods. In other words, in the pattern identification processing, the work of integrating the feature quantities that has been performed conventionally is not required, so the processing content is simple and a high processing speed can be obtained. Further, since this pattern identification processing is general, it is highly versatile and applicable to various objects.

In the apparatus of the present invention, in addition to the conventional integration work, additional processing such as combining direction-specific images, correcting the position of the object image, and correcting the direction-specific image arrangement is required. . However, it is expected for many objects that these additional processes are simpler than the conventional integration work and that a highly versatile method that is not so sensitive to the shape of the object can be adopted. It Therefore, high-speed processing capability and high versatility can be provided as the entire system that combines the additional processing and the pattern identification processing.

The above expectations may also be disapproved for certain objects. However, even in such a case, the above-mentioned additional processing can be executed as a processing separate from the pattern identification processing. Therefore, if a method such as parallel processing is used, it is complicated in the conventional pattern identification processing. High-speed processing is possible as compared with the case where the integrated processing is performed.

[0014]

1 is a block diagram showing the overall structure of a pattern identification apparatus according to an embodiment of the present invention.

The pattern identification device according to this embodiment is
As is clear with reference to FIG. 1, the image combining unit 1, the position correcting unit 2, the arrangement correcting unit 3, and the pattern identifying unit 4 are provided.
These processing units 1, 2, 3, 4 can actually be realized by one or more programmed computers.

The image synthesizing section 1 inputs a plurality of analog images obtained by picking up images of an object to be identified from a plurality of directions using a video camera, for example. In this embodiment, it is assumed that images obtained by picking up two surfaces, that is, the front surface and the back surface of the identification object are used. 2 shows an example of an image of the front surface of the object, and FIG. 3 shows an example of an image of the back surface.

The image synthesizing section 1 generates a single image 7 as shown in FIG. 4 in which the two input images 5 and 6 are combined, and the combined image 7 is internally formed in the form of digital data. Store in memory.

Here, several variations can be considered for the method of generating the digital combined image. For example, a method of picking up an object from a plurality of directions with a plurality of cameras and electrically connecting the plurality of analog images and then AD-converting the plurality of analog images, or memory-converting the plurality of analog images separately A method of combining the above, or a method of optically combining the images of the objects viewed from a plurality of directions using a mirror and the like, and capturing the combined image with one TV camera and performing AD conversion, etc. Conceivable.

A combined image 7 as shown in FIG. 4 is sent from the image synthesizing section 1 to the position correcting section 2. The position correction unit 2 corrects the relative position of each of the front surface image 5 and the back surface image 6 in the combined image 7 with respect to each of the front surface image areas of the object images 8 and 9 included therein. As the position correcting unit 2, for example, the image rotating unit in the "pattern identifying device" disclosed in Japanese Patent Application No. 4-345791 can be adopted.

FIG. 5 shows the flow of processing of the position correcting section 2. First, the entire area of each of the direction images 5 and 6 is binarized by using a threshold value suitable for distinguishing the object images 8 and 9 from other areas, and the object image is converted from the binarized image. 8 and 9 are extracted (S1). The binarization method is described, for example, in "Image Analysis Handbook" (1991, published by The University of Tokyo Press), supervised by Mikio Takagi and Yohisa Shimoda.

Next, the barycentric positions of the extracted object images 8 and 9 are calculated (S2).

Further, the extracted object images 8 and 9 are binarized by using a threshold value suitable for distinguishing the characters 10 and 11 from other regions, and the binarized object is obtained. Characters 10 and 11 are extracted from the images 8 and 9 (S
3). Then, the second moment of each of the characters 10 and 11 is calculated (S4).

Finally, the affine transformation is used to move the object images 8 and 9 in parallel so that the center of gravity of the object images 8 and 9 is located at the center of the direction images 5 and 6, and The object images 8 and 9 are rotationally moved in the directions in which the second moments of the characters 10 and 11 are maximized.

By this position correction, the combined image 7 of FIG.
Is modified into an image 12 as shown in FIG.

The position correction image 12 as shown in FIG. 6 is sent from the position correction unit 2 to the arrangement correction unit 3. Placement correction unit 3
Corrects the two direction images 5 and 6 in the position correction image 12 so that the two direction images 5 and 6 always have a constant arrangement. For example, in the position-corrected image 12 of FIG. 6, the back surface image 5 is arranged on the left side and the front surface image 6 is arranged on the right side, but this arrangement depends on the posture when the object is supplied to the camera, which is the opposite. It may be arranged. Therefore, the placement correction unit 3
Gives the position-corrected image 12 to the surface image 6 for example.
Is fixed on the left side and the back side image 5 is on the right side.

FIG. 7 shows a processing flow of the arrangement correcting section. First, all the regions in the images 5 and 6 for each direction are binarized, and the object images 8 and 9 are extracted from them (S11).
Next, the extracted object images 8 and 9 are binarized to extract the characters 10 and 11 (S12).

After that, the area ratio of the characters (that is, black pixel areas) 10 and 11 to the remaining white pixel areas (or the entire object area) in each of the object images 8 and 9 is calculated, and the area ratio is calculated. The arrangement of the direction-specific images 5 and 6 is corrected so that is in a predetermined order (for example, large order or small order) (S13).

For example, when the correction is performed such that the areas are arranged from left to right in the ascending order of the area ratio, the layout corrected image 12 of FIG. 6 is corrected like the image 13 of FIG. In the layout corrected image 13 of FIG. 8, the front side image 6 is on the left side and the back side image 5
Is placed on the right side. In this layout corrected image 13, the above-described image layout is always obtained regardless of the posture when the object is supplied to the camera. Therefore, this layout correction image 13 is an image unique to the target object, and one type of layout correction image 13 corresponds to only one type of target object. Therefore, the layout correction image 13 is used to identify the target object. Can be done.

The layout correction image 13 is sent from the layout correction unit 3 to the pattern identification unit 4. The pattern identification unit 4
The entire layout-corrected image 13 that has been input is treated as one image, and the target is identified by pattern-identifying the pattern consisting of the two characters 10 and 11 included in this image.

In this pattern identification, a general pattern identification technique can be used. For example, it is possible to use the "pattern identification device" disclosed in Japanese Patent Application No. 4-345791 mentioned above.

The embodiment described above identifies an object having an attribute different from that of the object used in the description, for example, an object that needs to be imaged from three directions, or a character in the object image. Instead, it can be applied to the identification performed using the outer shape of the object image as a pattern without changing the basics of the processing content, which is highly versatile.

The preferred embodiment of the present invention has been described above, but the present invention is not limited to this embodiment and can be implemented in various different modes. For example, there is no pattern character in the object image, or there is no need to use such a character,
When the outer shape of the object image is used as a pattern for identification, the binarization process (S in FIG. 5 and FIG. 6) in the object image is performed.
3, S12) becomes unnecessary. Then, in the second moment calculation process (S4) of FIG. 5, the second moment of the object image itself is calculated, and in the area ratio calculation process (S13) of FIG.
It suffices to calculate the area ratio of the object image to the direction-specific image.

[0033]

As described above, according to the present invention,
Since the images obtained by picking up the object from a plurality of directions are combined and the position and arrangement of the object image in each direction are corrected to a regular one, pattern identification is performed using the combined image. , The above-mentioned correction processing and pattern identification processing can be performed separately, and since general pattern identification technology can be used without being much influenced by the attributes of the target object, it is suitable for high-speed processing. A highly versatile identification device can be provided.

[Brief description of drawings]

FIG. 1 is a block diagram showing an overall configuration of a pattern identification device according to an embodiment of the present invention.

FIG. 2 is an explanatory view showing an image of a surface of an identification object used in the same embodiment.

FIG. 3 is an explanatory diagram showing an image of the back surface of the identification target.

FIG. 4 is an explanatory view showing one combined image obtained by combining images of the front surface and the back surface of the same identification object.

FIG. 5 is a flowchart showing a processing flow of a position correction unit of the embodiment.

FIG. 6 is an explanatory diagram showing an image in which the position of an object image is corrected by a position correction unit.

FIG. 7 is a flowchart showing a processing flow of a layout correction unit of the embodiment.

FIG. 8 is an explanatory view showing an image in which the arrangement of the object image is corrected by the arrangement correcting unit.

[Explanation of symbols]

 1 Image Compositing Section 2 Position Correcting Section 3 Arrangement Correcting Section 4 Pattern Identifying Section 5 Back Side Image 6 Front Side Image 7 Combined Image 8, 9 Object Image 10, 11 Character 12 Position Corrected Image 13 Arrangement Corrected Image

Claims (5)

[Claims] 1. A pattern identification device for pattern-recognizing the identification target object based on a plurality of direction-specific images obtained by imaging the identification target object from a plurality of directions, and combining and combining the plurality of direction-specific images. An image combining means for generating an image, a position correcting means for correcting the relative position of the object images in the respective direction-specific images to a predetermined position, and a predetermined mutual arrangement of the plurality of object images in the combined image. And a pattern identification means for performing pattern identification by treating the entire combined image as one image for the combined image corrected by the position correcting means and the arrangement correcting means. A pattern identification device comprising: 2. The apparatus according to claim 1, wherein the image combining means includes an optical system that combines and obtains a plurality of images of the object observed from a plurality of directions. 3. The apparatus according to claim 1, wherein the position correction means includes means for moving the object image in parallel to the predetermined position and means for rotating the object image in a predetermined direction. A pattern identification device. 4. The apparatus according to claim 1, wherein the means for calculating an area ratio of the object image in the direction-specific image and the object images in an order based on the calculated area ratio are mutually calculated. A pattern identification device comprising: means for correcting the arrangement. 5. The apparatus according to claim 1, wherein the layout correction means calculates an area ratio of a character existing in the object image in the object image, and the calculated area ratio. And a means for correcting the mutual arrangement of the object images in a sequence based on the pattern identification apparatus.