JPS60107190A - Graphic recognizing system - Google Patents

Abstract

PURPOSE:To attain ease of recognition of a new graphic whose shape is not clear by storing a dictionary graphic/character in terms of vector data and normalizing a graphic/character to be recognized and its size by means of the vector data of dictionary graphic and character. CONSTITUTION:The graphic/character 12 to be recognized is a dot data having height YW and width XW, and a dictionary graphic/character 13 is a candidate graphic/character and stored in terms of basic sizes XW0XYW0, XW1XYW1,.... The candidate graphic/character is a graphic/character group classified by a characteristic amount extracted from the graphic/character to be recognized. The ratios of size between the graphic/character 12 to be recognized and the candidate graphic/character 13, m0, m1,..., n0, n1,..., are obtained and the candidate graphic/character 12 is formed in a vector data 15 being the same size as the graphic/character to be recognized. The vector data 15 is converted into a dot data 17 and subjected to pattern matching with the graphic/character 12 to be recognized (dot data).

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention stores dictionary figures and characters as vector data, normalizes them to the size of figures and characters to be recognized, and then converts them into dots.
This invention relates to a figure recognition method that performs pattern matching of dot images between dictionary figures and characters and figures and characters to be recognized.

[Technology background]

Various types of drawings, such as electrical circuit drawings, mechanical drawings, plant drawings, and architectural drawings, are a means of transmitting information that humans can understand very clearly and intuitively. Obtaining various types of meaningful information from such drawings was left up to humans' excellent pattern recognition ability, but in recent years, image processing technology (shape recognition)
There is a strong social demand for mechanization and automation, and various shape recognition processing methods have been proposed.

[Conventional technology and problems]

FIG. 1 is a diagram illustrating an example of the figure recognition method of Hashira. In Figure 1, 1-1 to 1-4 are figures to be recognized.
Characters, 2 is the standardization processing unit, 3-1 to 3-4 are the figures φ characters after normalization, 4-1 to 4- are the figures, character dots, and data of the dictionary, and 5 is the pattern matching process unit. show.

When considering the information written in various drawings, in order to include meaningful figures and characters in addition to simple line segments, it is necessary to enter the information into a computer via an image input device (FAX, drum scanner, TV camera, etc.). It is necessary to perform processing such as extraction of line figures, extraction and recognition of figures, and extraction and recognition of characters from the digital image data obtained. However, extraction and recognition processing of these figures and characters is extremely difficult because each person's unique distortions are added to the figures and characters freely drawn by humans. - Many processing devices are designed for drawings where humans have made concessions to machines, such as by setting strict restrictions on the size and position of characters.

Next, we will discuss in detail the problems with figure/character recognition technology. As is well known, conventional figure recognition methods can be broadly divided into two types: feature matching and pattern matching, each of which has advantages and disadvantages. Usually, the figures drawn in drawings are many types of figures of different sizes, and are rarely drawn in a fixed size like letters, so figure recognition is done by feature matching, and letters The recognition is done by pattern matching.

The reason for this is that pattern matching is susceptible to size distortion. As shown in Figure 1, pattern matching involves matching the extracted dot data to the figure to be recognized.
The characters 1-1 to 1-4 are standardized to the dictionary size in the standardization processing unit 2, and the standardized characters are extracted from among the figures, character dots, and data 4-1 to 4-n in the dictionary in the pattern matching processing unit 5. The one closest to figure/character 3-1 or 3-4 is selected. In this way, when the shapes and characters to be recognized are different in size from the shapes and characters in the dictionary, in pattern matching, the shapes and characters to be recognized are matched with the shapes and characters in the dictionary.
It must be standardized to the same size as the letters. In the standardization process, lines are enlarged or reduced while preserving the line width as if they were written with the same writing instrument, but the standardization of dot data is complex and cannot be perfected, and in reality, lines may be thickened or drawn. Crashes may occur. Therefore, if the sizes differ too much, shapes that are far different from the original shapes are likely to occur, which will have a negative impact on recognition. Well, in feature matching, the fact that the shapes have different sizes is one of the valid feature data, but this only applies when the shape of the shape to be recognized is known, and when the shape is unknown. When adding a new figure as a recognition target or considering shape distortion, it is not expected to obtain a very good recognition rate.

[Purpose of the invention]

The present invention is based on the above consideration, and it is possible to easily recognize figures containing handwritten distortion that appear in drawings drawn in a free size, and to recognize new figures whose shapes are unknown. The purpose of this is to provide a figure recognition method that can be used.

[Structure of the invention]

To this end, the figure recognition method of the present invention forms the core of an automatic drawing input device that automatically extracts design information written on drawings, as shown in FIG. A drawing reading device b converts the dot data into dot data, a dot data holding device c1 holds the read dot data, and the dot data is loaded from the dot data holding device to cut out the figure/character area to be recognized. A figure/character/line segment extraction device d1 that extracts line segments; a line segment holding device e1 that holds the line segments extracted by the figure/character/line segment extraction device; and a line segment extraction device e1 that holds the line segments extracted by the figure/character/line segment extraction device. The present invention is used in an automatic drawing input device comprising a figure recognition processing section 1 according to the present invention, which includes a figure/character holding device f to be recognized which holds the figures/characters that have been recognized.

The figure recognition method of the present invention is based on the figure to be recognized written on the drawing.
A figure/character dictionary storage device for storing a character area to be recognized, a figure/character dictionary storage device for storing dictionary figures/characters, a recognition result storage device for storing recognition processing results for the figure/characters to be recognized, and a graphic display device. , and an arithmetic control device, a figure recognition method in which a figure or character closest to the figure or character to be recognized is selected from among the dictionary figures or characters by the arithmetic control device and stored in the recognition result storage device. Therefore, the above-mentioned figure φ character dictionary storage device stores the types of figure e characters that are classified according to the characteristics of characters in figures, the basic figure and character size that defines the size of figures and characters, and the various types of figures and characters. The arithmetic and control device is configured to store the dictionary figure/character by vector data defined by a vector, and the arithmetic and control unit extracts a feature quantity from the figure to be recognized φ character held in the figure to be recognized/character holding device. death,
A dictionary figure/character of the type of figure/character classified by the feature amount is loaded from the figure/character dictionary storage device and normalized to the same size as the figure/character to be recognized, and then the vector fist data is Convert it to dot data with the standardized size, perform pattern matching with the dot data and the dot association data of the figure/character to be recognized, and select the dictionary figure e-character closest to the figure/character to be recognized. The present invention is characterized in that it is configured to perform processing for storing recognition results in a storage device.

[Embodiments of the invention]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 3 is a diagram showing an example of a system configuration to which the present invention is applied;
FIG. 4 is a diagram for explaining one embodiment of the present invention, FIG. 5 is a diagram for explaining the flow of figure recognition processing of the present invention, and FIG. 6 is a diagram for explaining vector data. In the figure, 7 is a figure/character holding device to be recognized, 8 is an arithmetic control unit, and 9 is a figure/character holding device.
10 is a recognition result storage device, 11 is a graphic display device, 12 is a figure/character to be recognized, 13 is a candidate figure/character, 14 is a standardization processing unit, 15
is vector data, 16 is a vector/dot converter,
17 is dot data, 18 is a pattern-matching processing section) 12.

FIG. 3 shows a specific configuration example of the figure recognition processing section in the diagonal frame of the automatic drawing input device shown in FIG.・It is held as data, and the arithmetic and control unit 8 is
Pattern matching is performed by converting the vector data in the character dictionary storage device 9 into dot data of the same size as the size of the dot data of the figure/character to be recognized held in the figure/character holding device 7 to be recognized. It performs processing.

The recognition result storage device 10 stores the results, and the graphic display device 11 is used to register vector data in the graphic/character dictionary storage device 9.

FIG. 4 shows the flow of data processing up to pattern matching of figures and characters to be recognized with dictionary figures and characters. In FIG. 4, a figure/character 12 to be recognized is dot data having a size of Yw in the vertical direction and Xw in the horizontal direction, and is held in the figure/character holding device 7 shown in FIG. The dictionary figure/character 13 is the figure/character shown in FIG.
These are candidate figures and characters selected from the character dictionary storage device 9, and each has a basic size (the size is XWO x YWO).
1XW1×YW1, etc.). The candidate figure/character is the figure/character to be recognized.
Figures/places classified by feature quantities extracted from characters 9
loaded from. Therefore, the standardization processing unit 14 calculates the size ratios m01m1, . . ., ru0, rLl, . . .
.... Then, using this ratio, the candidate figure/character 12 is converted into vector data 15 having the same size as the figure/character to be recognized. Note that the normalization of circles, arcs, etc. is performed, for example, after approximating the lines. The vector/dot converter 16 converts the vector fist data 15 into dot data 17, and the pattern matching processor 18 converts the figure/character 12 (dot data) to be recognized and the vector/dot converter 1.
The pattern by dot data 17 which is the output of 6.
Matching is performed and the figure closest to the figure/character 12 to be recognized is selected from among the dot data 17. As mentioned above,
The present invention stores dictionary figures and characters to be referenced and compared during pattern matching as vector data, and stores figures and characters to be recognized as vector data.
Standardization of the sizes of characters and dictionary figures/characters is performed using vector data of dictionary figures/characters, and then pattern matching is performed.

Next, the flow of processing by the figure recognition method of the present invention will be explained with reference to FIG. Note that in the following explanation, the term "figure" has the same meaning as the above-mentioned figure/character.

■ Load the area for each figure/character from the figure/character holding device to be recognized. Next, perform the process (■).

■ Extract the features of the figure to be recognized. Next, perform the process (■).

■ Classify shapes into groups with similar features based on the extracted features and determine candidate shapes for pattern matching. Next, perform the process (■).

■ Find out how many candidates there are to pattern match.

If there is only one candidate figure and it is uniquely determined as 1, pattern matching is not performed, but if there are two or more N candidate figures, the process (■) (pattern matching process) is performed.

■ Load N candidate shapes from the dictionary and normalize them to the size of the figure to be recognized (Xw, Yw). Next, perform the process (■).

■ Convert the standardized vector fist data to dot data to obtain dot image data. Next, perform the process (■).

■ Perform pattern matching and select the figure whose dot data obtained from the dictionary is closest to the dot data of the figure to be recognized.

FIG. 5 shows vector data of dictionary figures and characters. In Figure 5, for example, the types of straight line vectors are displayed as O1 for the start point and end point (open), 1 for the continuation point (bend), and 2 for the branch (two branches), and θl and θn of the circular arc. The center angle of the arc and the angle of the radius from the reference line are displayed.

FIG. 7 is a diagram explaining the definition of strain temperature. There are various pattern matching processing methods that can be applied in the present invention, and one of them, which has already been proposed by the inventor of the present application, will be outlined below. This method is an algorithm that compares the shape of a figure/character (figure to be recognized/character shape) with a standard form (dictionary figure/character) and selects the figure/character with the smallest distortion as the recognized figure/character. In this algorithm, as shown in Figure 7, it is assumed that the sampled (to be recognized) figure/character X is generated by adding a distortion d to the standard pattern (dictionary figure/character) S. Then, the degree of distortion is defined as an evaluation function. This evaluation function is applied to all basic patterns (dictionary shapes,
The figure/character pattern with the smallest amount of distortion is determined to be the sample figure-contesting character. The distortion measure is defined as D=Σd" (X: (L, j,), S: (L, L)). Here, (L, 7), (4, 1) Jt
The coordinates of the black pixel near 4, d indicates the distance between the two points.

(For example, [Study of recognition of characters written on circuit diagrams] Mitsuo Ishii and Michiko Iwasaki (Fujitsuki) January 25, 1980 Institute of Electronics and Communication Engineers PRL79-83) Next,
A method for extracting figures, characters, and line segments by the figure/character/line segment extraction device d in the automatic drawing input device shown in FIG. 2 will be briefly described. FIG. 8 is a diagram for explaining the character extraction method, and FIG. 9 is a diagram for explaining the figure and line segment extraction method.

Character extraction is performed by finding a black pixel from two images (O: white pixel [background], 1: black pixel [character, figure, line segment]), and then finding a black pixel connected to it by 8 connections as shown in Figure 8 (h). A group of pixels is extracted, and a rectangular frame surrounding it whose size (χ·, 1−) resembles a character is extracted. To do this, we need rules regarding the size of characters and graphics. For example, the letters A and H shown in Figure 8(a) are 5 mm x 5 my+ or less, and the shapes that are not connected to line segments are 5 mm x 511! The above terms correspond to the terms.

Since the extraction of figures and line segments differs depending on the notation of each drawing, the key extraction method differs depending on the type of drawing, but as an example, the extraction method of figures and line segments will be explained below. For example, ■ Separate and extract lines based on patterns of contact points between line segments and figures. In the case of a circuit diagram, separation and extraction can be performed by determining patterns such as T, Y1, . . . as shown in FIGS. 9(α) and (h). Also, as shown in Figure 9 (C), the end point 01 of the Kotobuki oven is
As shown in FIG. 9(d), those having a closed force loop L are extracted as figures. Note that the rules in this case are that line segments do not have open endpoints, and line segments do not form closed loops.

A lattice like the one shown in FIG. 9 (=) is considered to be a line segment even if it is a closed loop if there are line segments coming out of the four corners.

〔Effect of the invention〕

As is clear from the above explanation, according to the present invention, the size of figures with handwritten distortion can be easily standardized by simply using the ratio of vector data, and even unknown figures can be simply registered in a dictionary using vector data. can be easily applied to recognition processing.

Furthermore, since it is vector data, the storage area of the dictionary can also be reduced.

[Brief explanation of the drawing]

Figure 1 is a diagram explaining an example of a conventional figure recognition method, Figure 2
3 is a diagram showing an example of a system configuration to which the present invention is applied; FIG. 4 is a diagram illustrating an embodiment of the present invention; FIG. 5 is a diagram illustrating an embodiment of the present invention. Figure 6 is a diagram explaining the flow of figure recognition processing, Figure 6 is a diagram explaining vector data, Figure 7 is a diagram explaining the definition of distortion measure, Figure 8 is a diagram explaining the character extraction method, and Figure 9 is a diagram explaining the character extraction method. The figure is a diagram illustrating a method for extracting figures and line segments. 1-1 to 1-4 and 12... figure/character to be recognized, 2
and 14... standardization processing section, 3-1 to 3-4...
Figures/characters after standardization, 4-1 to 4-n... Dictionary figures/character dots/data, 5 and 18... Pattern matching processing section, 7 and f... Figures/characters to be recognized Holding device, 8... Arithmetic control device, 9... Graphic/character dictionary storage device, 10... Recognition result storage device, 11...
・Graphic display device, 13... Candidate figure e character, 15... Vector data, 16... Vector/dot conversion unit, 17... Dot data, α
...Drawing, b...Drawing reading device, C...Dot...
Data holding device, d... figure/character/line segment extraction device,
e...Line segment holding device, 9...Graphic recognition processing unit. Patent Applicant Fujitsu Ltd. Representative Patent Attorney Kyotani Shibegu 1m 73 Figure (Japanese!l:) Pu S Pencil to Figure 1

Claims (1)

[Claims] A figure/character storage device to be recognized holds the area of figures/characters to be recognized written on the drawing, a figure/character dictionary memory 1A to store the dictionary figures/characters, and stores the recognition processing results of the figures/characters to be recognized. A recognition result storage device, a graph ink display device, and an arithmetic control device are provided to select the figure e character closest to the figure/character to be recognized from among the dictionary figure Q characters by the arithmetic and control device. It is a figure recognition method that stores the recognition results in a memory device, and the figure/character dictionary memory device stores the types of figures/characters that are classified based on the characteristics of figures and characters, and the shapes/characters that define the sizes of figures and characters. The dictionary figures and characters are stored in accordance with the basic character size and vector data that defines figures and characters using various vectors, and the arithmetic and control unit stores the figures and characters to be recognized in the figure and character holding device. A feature quantity is extracted from the figure/character to be recognized, and a dictionary figure/character of the type of figure/character classified by the feature quantity is loaded from the figure/character dictionary storage device and is the same as the figure/character to be recognized. After that, the vector data is converted to TonoF data with the standardized size, and pattern matching is performed using the dot data and the dot data of the figure/character to be recognized. A figure recognition method, characterized in that it is configured to perform a process of storing a dictionary figure ψ character closest to the recognized figure Φ character in the recognition result storage device.