JPH0935058A - Image recognizing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain the human-friendly image recognizing method which is applicable to the recognition of a general image. SOLUTION: Outline data on an original image are extracted (S16) by roughly scanning the entire area of an original image (S12). Prototype image data which are similar to the extracted outline data and previously stored and held are detected (S18). On the basis of the information attached to the detected specific prototype image data, the area part of the outline data corresponding to a feature area of the specific prototype image data is finely scanned (S20). On the basis of the data obtained by this fine scanning, a feature image is extracted to discriminate the original image (S22). The discrimination of the original image by this procedure is performed through the two stages of rough and fine scans and this procedure is similar to focusing on the feature part by hand, so that the image can be discriminated in a short time with a small amount of scanning data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition method,
In particular, it relates to a prototype image data reference type image recognition method for performing feature extraction by referring to the prototype image data.

[0002]

2. Description of the Related Art Conventionally, an image recognition method is generally indispensable in order to perform a more human-like image feature extraction process using a computer and to output an image having a feature that is easily accepted by humans. It is regarded as a method.

In recent years, the evolution and development of hardware such as computers and communication technologies have been remarkable. Further, correspondingly, the evolution of software regarding various data processing methods is also the same. With regard to image data processing technology, multimedia technology has advanced rapidly, and great progress has been made in image compression technology and transmission technology.
Particularly, from the viewpoint of an interface, all of them relate to devices, devices, and software on devices and software on devices.

On the other hand, the technology relating to image data is
It is different from other general data in that it is intended to be finally received by the human retina in addition to the contact point between devices. In short, the requirement for human friendliness is more important. In recent years, many proposals have been made for making the computer human-friendly so that a computer can perform a process of extracting a more human characteristic image and outputting an image having a characteristic that is easily accepted by people.

For example, as a conventional example having a technical scope similar to that of the present invention, Japanese Patent Laid-Open No. 6-3 aims to perform high-speed and highly reliable recognition without being affected by positional deviation and skew deviation.
No. 18245, and Japanese Patent Laid-Open No. 5-35910 for the purpose of making image processing determination without bothering the operator even if characters written vertically and horizontally are mixed.

[0006]

However, the conventional computer system does not have an image recognition function capable of sufficiently performing an image recognition process in capturing a general arbitrary image. The reality is that it has not progressed so much that it can be industrially endured. Therefore, in order to utilize the advanced peripheral technology as described above, it is necessary and required to devise a new computer mechanism that emulates a human image recognition function. A more human image recognition method is an important task to achieve this purpose.

It is an object of the present invention to provide a human-friendly image recognition method applicable to general image recognition.

[0008]

In order to achieve the above object, the image recognition method of the present invention uses a coarse scanning step of performing a coarse scanning over the entire area of an original image and data obtained by the coarse scanning step. A contour data extraction step of extracting the contour data of the original image, a search step of comparing the contour data with the prototype image data stored and held in advance, and detecting specific prototype image data having a predetermined match in the comparison. Prototype image data detecting step, a fine scanning step of densely scanning an area portion of contour data corresponding to a characteristic area of the specific prototype image data based on information attached to the detected specific prototype image data, and a fine scanning step It is characterized in that a characteristic image is extracted based on the data obtained by and the original image is identified.

The above coarse scanning process may be executed with a grid number that is approximately 1/10 or less of the grid number in the fine scanning process. Further, the identification of the original image has a characteristic image search step of comparing the data obtained by the fine scanning step with the characteristic image data stored in advance, and if the original image is identified by this characteristic image search step. Good.

[0010]

Therefore, according to the image recognition method of the present invention,
A rough scan is performed over the entire area of the original image to extract the contour data of the original image. The obtained contour data is compared with the prototype image data stored and held in advance, and in this comparison, specific prototype image data having a predetermined match is detected.
Based on the information associated with the detected specific prototype image data, the region portion of the contour data corresponding to the characteristic region of the specific prototype image data is scanned more densely. The characteristic image is extracted based on the data obtained by this dense scanning to identify the original image.

Therefore, the identification of the original image is carried out by two-step scanning of coarse and fine. The fine scan is performed only on the feature portion required for the identification, and the original image is identified by a procedure similar to that of a person focusing on the feature portion.

[0012]

Embodiments of the image recognition method according to the present invention will now be described in detail with reference to the accompanying drawings. 1 and 2, an embodiment of the image recognition method of the present invention is shown. FIG. 1 is a flow chart for explaining the procedure of the embodiment, and FIG. 2 is a block diagram showing a configuration example of an image recognition apparatus for carrying out the procedure of FIG.

The image recognition apparatus to which the image recognition method of this embodiment is applied is composed of a general image data processing system as shown in FIG. The image recognition apparatus according to the embodiment includes an image data processing apparatus 1 that constitutes a core part of the image recognition apparatus,
An original image data input device 2 that captures or outputs image data of a CCD video camera, a VTR, or the like, stores image data used when the image data processing device 1 processes the image data, and / or stores an image during processing. It is composed of an image data storage device 3 for temporarily storing data, a display device 4 which is a monitor display, and an operation unit 5 for inputting an operation command signal.

FIG. 1 shows the operation of the image recognition method of this embodiment. An image data processing procedure based on the image recognition apparatus of FIG. 2 will be described below with reference to FIG.

First, the image data processing apparatus 1 inputs original image data to be a recognized image from the image data input apparatus 2 (step S10).

The image data processing apparatus 1 performs a rough scan on the entire area of the input original image data (step S12). This rough scan is about 1 of the maximum number of grids.
Scanning is performed with a coarse grid number within 1/0. The main scan is executed for the purpose of extracting a characteristic portion of the original image. In this embodiment, coarse and fine two-step scanning is executed. The main scan is one of the two-stage scans, and is a rough scan for grasping the outline of the image as compared with the other dense scan that is finally required for recognizing the original image. is there.

In step S14, the above step S12 is performed.
The contour data is extracted based on the scan data acquired by the rough scan of. The contour data is image data of a characteristic portion existing in the entire image of the original image data. For example, a contour line of an object, a protruding portion, a depressed portion, a grayscale change point, a color change point, a high brightness point, and a moving object in a still background.

The extracted contour data is processed in step S16.
Used to search for. This "search" refers to comparison with the original image data. Image data serving as prototype image data is previously stored and stored in the image data storage device 3. The image data processing device 1 reads this prototype image data from the image data storage device 3 and sequentially compares it with the extracted contour data to determine specific prototype image data having high coincidence.

Through the search in step S16, prototype image data having a predetermined match is selected and detected (step S18).

There are various methods for speeding up the search and increasing the probability in the search and prototype image data detection in steps S16 and S18 described above. For example, the range of types of the recognized images is limited, the recognized images are classified in advance, the search focusing on the contour shape is performed, the search focusing on color matching is performed, and so on. is there.

In the original image data corresponding to the original image data detected as described above, the accompanying information is registered in advance. This accompanying information is "data relating to the object of the prototype image data". For example, if the image recognition apparatus of the embodiment is intended for use in a specific manufacturing line, it is a product / part or the like on the manufacturing line, and is data relating to each type. The type classification here is a classification including quality of the product, and means necessary and sufficient data for this classification.

The fine scan in step S20 is a fine scan with respect to the coarse scan in step S12. The main scan is locally performed as compared with the above-described coarse scan performed over the entire area of the original image data. The range, position, etc. of this local portion are determined based on the above-mentioned "data relating to the object of the prototype image data". Since the purpose of image recognition in this embodiment is to identify a recognized image, execution of partial scanning is performed in order to acquire data required for identification faster, in more detail, and more accurately. . Therefore, the “data related to the object of the prototype image data” is the related data necessary for identifying the recognized image. Based on the above example of the determination of product quality, detailed data other than the product is unnecessary, and detailed scanning in the unnecessary area is not necessary. Therefore, the detailed data acquired in this step is data corresponding to the related data.

In step S22, the data relating to the characteristic image obtained in step S20 is compared with the "data relating to the object of the prototype image data" to identify and identify the recognized image. The comparison between the data relating to the characteristic image and the "data relating to the object of the prototype image data" will be as follows based on the above example of the determination of the quality of the product. For example, the image data of the characteristic portion relating to the non-defective product and the image data of the characteristic portion relating to the defective product are stored in advance, and the data relating to the characteristic image acquired in step S20 is the image data of any of these characteristic portions. The closeness is compared.

When the above image recognition procedure is put together again, rough raster scanning is performed on the original image to create contour data.
Initially about 10 of the maximum number of grids finally needed
The contour data is promptly extracted by executing the scanning of the original image with a coarse grid number within one-tenth.

Next, by comparing the contour data with a prototype image data group that is internally stored in advance, a specific prototype image data corresponding to the contour data is selected, and subsequently, a characteristic region of the prototype image data is reversed. The characteristic area of the contour data corresponding to can be automatically specified. That is, the position of the characteristic region of the contour data can be quickly recognized by using the prototype image data. In this, the position of the characteristic region of the original image is specified in a short time by utilizing the information related to the prototype image data group which is internally held in advance. When visually recognizing a real space, not all image information has the same value, and the information that is actually important is contour data on a relatively coarse scale, or even characteristic data on the contour data. It is often detailed information near the area.

According to the above embodiment, with a small number of scans,
Moreover, important information of the image can be recognized in a short time. Therefore, it becomes possible to flexibly acquire information that can promptly respond to the actual changing situation.
It is possible to obtain an industrially useful image recognition method that is simple and highly reliable.

Although the above embodiment is one preferred embodiment of the present invention, the present invention is not limited to this embodiment, and various modifications can be made without departing from the spirit of the present invention.

[0028]

As is apparent from the above description, according to the image recognition method of the present invention, rough scanning is performed over the entire area of the original image to extract the contour data of the original image, and the extracted contour data is stored in advance. The held prototype image data is compared, and in this comparison, specific prototype image data having a predetermined match is detected. Based on the information attached to the detected specific prototype image data, the area portion of the contour data corresponding to the characteristic area of the specific prototype image data is densely scanned. The characteristic image is extracted based on the data obtained by this dense scanning to identify the original image.

Therefore, the identification of the original image is executed by the coarse and fine two-step scanning, and first, the characteristic portion of the entire image is extracted by the coarse scanning. The image is identified by the fine scanning of the characteristic portion. This identification procedure is similar to that of a person focusing on a feature, which requires less scan data and allows image identification in a short time. Therefore, it becomes possible to quickly and accurately respond to general image recognition, and the industrial utility value of the image recognition system is improved.

[Brief description of drawings]

FIG. 1 is a flowchart showing an embodiment of an image recognition method of the present invention.

FIG. 2 is a block diagram showing a configuration example of an image recognition system applied to this embodiment.

[Explanation of symbols]

 1 image data processing device 2 image data input device 3 image data storage device 4 display device 5 operation unit

Claims (3)

[Claims] 1. A rough scanning step of performing a rough scanning over the entire area of the original image, a contour data extracting step of extracting contour data of the original image from the data acquired by the rough scanning step, and the contour data. A search step of comparing with the prototype image data stored in advance, a prototype image data detecting step of detecting the specific prototype image data having a predetermined match in the comparison, and the specific prototype image detected A fine scanning step of densely scanning an area portion of the contour data corresponding to a characteristic area of the specific prototype image data based on information accompanying the data, and a characteristic image is extracted based on the data obtained by the fine scanning step. An image recognition method, characterized in that the original image is identified. 2. The image recognition method according to claim 1, wherein the rough scanning step is executed with a grid number that is approximately 1/10 or less of the grid number in the fine scanning step. 3. The identification of the original image further includes a characteristic image search step of comparing data obtained by the fine scanning step with characteristic image data stored in advance, The image recognition method according to claim 1, wherein the original image is identified.