JPS63186384A - Pattern recognizing device - Google Patents

Abstract

PURPOSE:To improve the pattern recognizing factor by processing integratedly two partial patterns approximate to each other as a single partial pattern. CONSTITUTION:A partial template TP dictionary 2 decomposes a standard pattern SP into the TP and stores the identification symbols. An SP dictionary 3 stores the SP in the form of a gathering of the identification symbols of plural TP. An input processing part 4 decomposes an unknown pattern 1 into partial patterns 10 and produces the connected state information 13. An integration processing part 5 integrates two patterns 10 adjacent to each other and outputs a single processed pattern 11. A partial comparison part 6 compares the pattern 11 with a TP12 and sends the identification symbol of the TP12 to a pattern producing part 7 when the coincidence is obtained from said comparison. If no coincidence is obtained, the identification symbol of the TP12 which is similar within a prescribed range is outputted to the part 7. The unknown pattern 15 is produced from the information 13 and the output of the part 6. A comparison part 8 compares the pattern 15 with an SP14, i.e., the output of the dictionary 3 and outputs the identification result 9. Thus the pattern recognizing factor is improved.

Description

[Detailed Description of the Invention] [Summary] The present invention relates to a means for reading and recognizing characters and figures, and in particular, in order to reduce the amount of learning in figure recognition, all figures are treated as a set of partial patterns, and the partial patterns are separated by a predetermined distance. A pattern recognition method that reduces the amount of learning and improves the recognition rate by integrating shapes that exist in close proximity to each other [Industrial application field] Pattern recognition method as a data input method for computers As technology has progressed, reading devices that previously could only handle characters and graphics in a certain format are now able to read freely, for example, handwritten characters and graphics.

However, as the processing power of computers increases, further improvements in reading performance are required.

[Prior Art] There are various ways to recognize figures using a computer, but in order to recognize complex figures, it is necessary to decompose the figure into simple figures that make up the figure, so-called "partial templates", and then create various figures and their constituent parts. Prepare partial templates as dictionaries. To recognize an unknown figure, a method is used in which the unknown figure is decomposed into partial patterns, the partial patterns are compared with partial templates, matching partial templates are recognized, and all figures are recognized as a set of partial templates. Conventional examples of this method are shown in FIGS. 6 to 9, and the processing method will be explained.

FIG. 6 shows the procedure for creating a partial template.

Various figures to be recognized are selected and a dictionary is created from them.Here, as an example, a procedure for processing a diode figure, which is one of the symbol figures used in a circuit diagram, will be explained.

To create a partial template from the prepared standard pattern (A), focus on bending points, end points, and intersections of the line segments of the figure, and cut out partial templates as shown by T1 to T4 in Figure (B). Give an identification symbol.

Here, those of the same shape, namely TI and T in the figure
When the same unit figure is rotated as shown in 4, the same identification symbol is given to both sides. If this is written out in detail, it will be a unit figure with a simple shape as shown in Figure (C), and the length of each school will be standardized as a unit length.

Next, this partial template is converted into a dictionary to create a partial template dictionary as shown in FIG. In this dictionary, corresponding to the "identification symbol", the "angle" made by the technique and the identification symbols S1 to Sn of the "standard pattern" in which the partial template is used are entered.

Since this partial template dictionary is commonly used for all standard patterns, there is no need to change it even if a standard pattern is added unless a special figure is handled. When adding a standard pattern, it is sufficient to add the identification symbol of the standard pattern to the "Standard Pattern" section of the partial template dictionary that constitutes the standard pattern.

Next, the generation of the standard pattern dictionary will be explained with reference to FIGS. 8(^) and (13). When cutting out a partial template as shown in FIG. 6(B) from the standard pattern shown in FIG. 6(A), mark each end point of the standard pattern with a number ■ to ■ as shown in FIG. 8(B). From this, a standard pattern dictionary shown in FIG. 8(A) is created. In the "Template Identification Symbol" field, enter the identification symbols T1 to T4 of the partial templates shown in FIG. 7, which are used for each of the endpoints (■) to (■).

The "direction" term is the direction that the partial template is facing. For example, if it is T1, the direction of the figure drawn in the "shape" column of the partial template dictionary shown in Fig. 7 is set to 0 degrees, and the direction is counterclockwise. Enter the angle of rotation in the direction.

"Angle error" is the allowable error between the angle of the branch entered in the partial template dictionary and the angle of the branch of the partial pattern obtained from the unknown figure.

Next, the mutual connection state of each end point ■～■ is called “end point connection”.
Enter the distance between the connected end points in the "bond distance" section. This allows you to specify a specific dimension, such as the height of a shape by "1".
Use the length standardized as ``.'' Enter the tolerance for bonding distance in the ``Distance Error'' section.

These two types of dictionaries are created for all standard patterns required as recognition targets. This work is extremely easy to create because the partial template is a simple figure, and the standard pattern dictionary can also be created almost mechanically.

FIG. 9 shows an example of a figure recognition device that generates a dictionary in this manner and performs figure recognition. The generation of the dictionary described above is executed by the pattern processing section 22. Next, a procedure for performing figure recognition using this device will be explained.

The unknown figure 1 is usually scanned by a photoelectric conversion device or the like, decomposed into pixels, read, and stored in a figure storage circuit built into the input processing section 4. Descriptions of these devices will be omitted since many devices based on known techniques can be found on the market.

Recognizing shapes is similar to the procedure for creating a dictionary.

The dimensions of the unknown figure 1 are first standardized in the input processing section 4. That is, set the scale factor so that the height and width of the figure match the outer dimensions of the figure used in the standard pattern dictionary 3, perform the reduction or enlargement operation, and then perform the same operation as shown in FIG. Decompose it into partial patterns using the following steps, and at the same time find the bonding state and bonding distance of the end points. In creating the dictionary, it was decomposed into partial templates, but here the difference is that it is decomposed into partial patterns that are compared with the partial templates. The decomposed partial patterns are sequentially compared with the partial templates that are the contents of the partial template dictionary 2 in the partial comparison unit 6, and matching partial templates are determined, and the connection state and connection distance of the end points determined by the input processing unit 4 are determined. The pattern generating section 7 generates an unknown graphic pattern in the same format as the standard pattern dictionary shown in FIG. 8.

The unknown graphic pattern is compared with the standard pattern shown in FIG. 8 (Δ), which is the content of the standard pattern dictionary 3, in the pattern comparing section 8. If a standard pattern with matching content is detected, the name of the standard pattern is output as recognition result 9.

In this way, the so-called "bottom-up" method is widely used in which complex figures such as circuit symbols are recognized by being broken down into partial patterns.

[Problem that the invention seeks to solve]

In the conventional method, a partial template dictionary is created by determining partial templates as described above, but when handwritten figures are recognized, it is also necessary to deal with unknown figures that contain some ambiguity. For this reason, for example, when it is necessary to recognize a figure as shown in FIG. 3 as an unknown figure, special partial templates are required for each part of t1 to t2 and t3 to t4 shown in FIG. When considering various cases, the number of types of partial templates increases rapidly, and when partial patterns that have not been taken into account appear, there are problems such as a decrease in the recognition rate. The present invention provides a means to ameliorate these problems.

[Means for solving problems]

In order to solve the above problems, the present invention uses a figure recognition method whose principle is shown in FIG.

The partial template dictionary (2) decomposes a standard pattern into a plurality of partial templates that are simple figures, describes the geometric characteristics of each partial template, attaches an identification symbol, and stores the partial templates.

The standard pattern dictionary (3) stores standard patterns as a set of identification symbols of a plurality of partial templates constituting the standard patterns, together with their combination states.

The input processing unit (4) decomposes an unknown figure (1) input as a recognition target into a plurality of partial patterns (1o) and creates connection state information (13) thereof.

The integrated processing unit (5) processes two adjacent partial patterns (10
) are within a predetermined distance, they are integrated to form and output one processed pattern (11), and when they are outside the predetermined distance, they are each output as one processed pattern (11).

The first comparison unit (6) compares the processed pattern (11) which is the output of the integration processing unit (5) and the partial template dictionary (2).
The partial templates (12) that are the outputs of are compared, and if they match, the identification symbol of the partial templates (12) is output to the pattern creation section (7).

In addition, when the first comparison unit (6) cannot detect a match between the processed pattern (11) and the partial template 1-(12), the first comparison unit (6) compares the partial template (1-1) which is similar within a predetermined error range.
2) and outputs its identification symbol to the pattern creation section (7) as a substitute for the matching partial template.

A pattern creation section (7) creates an unknown graphic pattern (15) from the combination state information (13) that is the output of the input processing section (4) and the output of the first comparison section (6).

The second comparison section (8) compares the unknown figure pattern (15) with the standard pattern (14) which is the output of the standard pattern dictionary (3).
and outputs the recognition result (9).

[Effect]

By integrating two adjacent partial patterns in the integration processing unit and processing them as one partial pattern, it is possible to prevent the collapse of shapes such as the corners of the pattern of unknown shapes.
By correcting the ambiguity and forcibly matching the contents of the partial template dictionary with the unknown geometric pattern,
Improve recognition rate.

〔Example〕

FIG. 2 shows an embodiment of the present invention, and FIGS. 3 to 5 show a system for integrating adjacent partial patterns, which is the gist of the present invention.

The configuration of the figure recognition device shown in FIG. 2 is not much different from that of the conventional system, and is similar to the conventional system shown in FIG. 9 with an integrated processing section added. Since the processing method for figure recognition in the present invention is substantially the same as the procedure described in the conventional method, the details will be omitted, and the gist of the present invention, which is the process of integrating figures including ambiguity, will be explained in detail.

In the method of the present invention, the contents and creation procedures of the standard pattern 21, pattern processing unit 22, partial template dictionary 2, and standard pattern dictionary 3 follow those of the conventional method. Furthermore, although the procedure for recognizing unknown figures is the same as in the conventional example, the present invention is characterized by a method for processing figures that are distorted or ambiguous.

The processing contents of the input processing unit 4, partial comparison unit 6, and integration processing unit 5 when recognizing a slightly distorted handwritten circuit symbol pattern as shown in FIG. Add an explanation about the integration process.

■ As an unknown figure, assume a slightly distorted figure with a handwritten diode symbol as shown in Figure 3. The input processing unit 4 focuses on the bending points and end points of the line, extracts each point tl-t8, and cuts out a figure with a constant radius around each point. The meaning of this circle is to remove small distortions of the recognition figure, such as the portions tl-t2 and L3 to t4 in FIG. 4. In other words,
From the partial pattern, 2 within a range smaller than the diameter of this circle.
The radius of the circle is determined so as to eliminate figures that have more than 2 bends. Here, as partial patterns, t5 to t8 can be used as they are because the cutting circles do not overlap, but tl, t2 and t3. Since the cutout circles for the pair t4 overlap, the integration process described below is performed.

■As shown in FIG. 5, overlapping partial patterns of cutout circles are subjected to integration processing. As shown in the figure, this is a process that integrates overlapping circles into one circle, expands the range, and synthesizes them as one partial pattern.It focuses only on lines that extend outside the overlapping circles, and Ignore small bends. As shown in the figure, the portion of tLt2 is corrected as a result of integration by ignoring the missing corner portion. Also t3.
In the t4 part, the upper and lower lines that appear outside the circle are shifted,
At this time, a certain tolerance range is set, and within this range "
The "deviation" is integrated and processed as being on one line.

■In the part t7, the upper and lower lines deviate from the straight line, but this is because a certain tolerance range is set when comparing with the partial template 12 in the partial comparison section 6, and the "deviation" within this range is 1
By processing it as if it were on a straight line of the book, it would be processed as matching the partial pattern that is a straight line.

■Next, use the partial pattern obtained by processing ■ and ■,
The pattern creation section 7 creates an unknown figure pattern in the same way as in the conventional example, and the Tsumaturn comparison section 8 compares it with the standard pattern 14 that is the content of the standard pattern dictionary 3, and reads out and recognizes the standard pattern name from the matching standard patterns. Output as result 9.

As described above, by performing the integration process and the error absorption process by setting the allowable error range, it is possible to perform correction processing on a figure that has some degree of distortion or ambiguity, and to recognize the figure. It is preferable to use a microprocessor for each of these processing units, and faster processing may be achieved if each comparison unit uses hardware logic.

〔Effect of the invention〕

As described above, by using the integration process and the error absorption process together, it is possible to perform figure recognition by correcting figures with small distortions, and it is possible to improve the recognition rate.

[Brief explanation of the drawing]

Fig. 1 is a diagram of the principle of the figure recognition device according to the present invention, Fig. 2 is an embodiment of the figure recognition device according to the present invention, Fig. 3 is an example of an unknown figure, Fig. 4 is a partial pattern extraction, and Fig. 5 is An example of partial pattern integration, Figure 6 shows creation of a partial template, Figure 7 shows an example of a partial template dictionary, Figure 8 shows generation of a standard pattern dictionary, and Figure 9 shows an example of a conventional figure recognition 883i device. ing. In Figures 1, 2, and 9, 1 is an unknown figure, 2 is a partial template dictionary, 3 is a standard pattern dictionary, 4 is an input processing section, 5 is an integration processing section, 6 is the first comparison 7 is a pattern creation section, 8 is a second comparison section, 9 is a recognition result, 10 is a partial pattern, 11 is a processed pattern, 12 is a partial template, 13 is combination state information, 14 is a standard pattern, 15 is a Indicates an unknown geometric pattern. Figure 2 (,A) Example of standard pattern
(B) Continuation of the 4th page of cutting out the partial template

Claims (1)

[Claims] A partial template dictionary (2
), and a standard pattern dictionary (3) that describes and stores the standard pattern as a set of the identification symbols of the plurality of partial templates constituting the standard pattern, together with the combination state thereof. An unknown figure (1) input as a target is decomposed into a plurality of partial patterns (10),
Input processing unit (4) that creates the connection state information (13)
, an integrated processing unit (5) that processes the plurality of partial patterns (10), and a processed pattern (11) that is an output of the integrated processing unit (5).
) and a partial template (12) that is the output of the partial template dictionary (2), and if they match, a first comparison unit (6) that outputs an identification symbol of the partial template (12); Connection state information (13) which is the output of the input processing unit (4)
and a pattern creation unit (7) that creates an unknown figure pattern (15) from the identification symbol output from the first comparison unit (6); and the unknown figure pattern (15) and the standard pattern dictionary. (
3) and outputs a recognition result (9); When the partial patterns (10) are within a predetermined distance, they are integrated to form and output one processed pattern (11), and when they are outside the predetermined distance, each processed pattern (11) is ), and when it is not possible to detect a match between the processed pattern (11) and the partial template (12), the first comparison unit (6) outputs a partial template () that is similar within a predetermined error range. 12)
A figure recognition device characterized in that a partial template (12) is detected as a matching partial template.