JPS648870B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a line figure recognition system for extracting straight lines, singular points, and symbols from line figure image data read by a photoelectric conversion device.

CAD (Computer Design)
Currently, inputting drawings of electrical circuits and electronic circuits into computers in the Aided Design (Aided Design) system is done exclusively manually, but if this could be done automatically, manual input would be possible. This can be expected to prevent mistakes and significantly reduce man-hours.

In order to automatically load such drawings,
A machine recognition system that optically reads drawings consisting of line diagrams using a photoelectric conversion device and converts them into image data consisting of binary signals, and further extracts the straight lines, singular points, and symbols that make up the drawings from this image data. It becomes necessary.

An object of the present invention is to provide a novel system capable of extracting straight lines, singular points, and symbols from a binary image of a drawing consisting of a diagram mainly composed of straight lines. According to the line figure recognition system of the present invention, it is possible to separate and identify connecting lines and symbol parts, and it not only extracts line segments but also records connection information for each, so circuit networks etc. can be easily expressed. Furthermore, the system can handle any number of types by registering their size and shape.

In order to achieve the above-mentioned object, the line figure recognition system of the present invention analyzes image data consisting of binary signals obtained from line figures via a photoelectric conversion device, in the upper direction in the traveling direction of the line segment, In a line segment tracking sensor that has three detection sections, front and bottom, if the 3-bit information consisting of the output signal of each detection section is (010), (000), or (other), a line is detected in the traveling direction of the sensor. Find out whether the segments are continuous, the line segment has ended, or a singular point on the line segment, and if the line segment is continuous, trace the line segment in the direction of movement, and if the line segment has ended, a straight line extracting means that determines whether or not it is a true end by scanning several pixels ahead in the advancing direction, and a fixed area containing the singular point found by the straight line extracting means. The peripheral distribution of the image within the area, that is, the pixel distribution on the side when pixels perpendicular to this side are gathered together on a specific side, the direction in which the straight line branches extend, their lengths, and the average slope near the singular point are used to calculate the shape of the figure. a singular point extraction means that measures the features of the geometric shape and also distinguishes symbols from line segments and roughly categorizes symbol shapes by determining the presence or absence of a black circle at the intersection from the area of the intersection; By comparing information on a series of singular points on the contour line and information on a series of distances between feature points on the contour line for the symbol found by the singularity extracting means with a predetermined template. The present invention is characterized by comprising symbol extraction means for identifying and extracting symbols.

Examples will be described below.

In the line figure recognition system of the present invention, a drawing consisting of a line figure accumulated as a binary image via a photoelectric converter such as a scanner is processed according to the following procedure.

Step (Extraction of starting point) When a pixel is encountered by scanning from one end of the screen, it is determined whether this point meets the conditions for a starting point, and if it meets the conditions, it is recorded in the starting point stack. The starting point conditions are to scan a certain range around the point, and if it satisfies the conditions that it has a certain length or more and that both ends of the line are symbols (or intersection points), it is considered to be on the line. is the starting point. The starting point is the starting point for the next screen scan when the process returns to the step after finishing the processing below the step.

Step (straight line section tracing) The starting point coordinates are extracted from the starting point stack, and tracing is performed from that point in the direction in which the line extends. If lines intersect on the way, the intersecting position and intersecting situation are recorded as connection information. If necessary, register it in the starting point stack as well. At this point, a mark is placed on the original screen of the line segments that have been traced up to now as traced, and the step is to prevent tracing the same line segment twice (termination condition).If the termination condition is satisfied, tracing is stopped. Termination conditions are open line and symbol encountered. In this case, if symbol recognition is necessary, the process moves to the next step; otherwise, the process moves to the next step.

Step (symbol recognition) Recognize symbols and record the results.

Step If the starting point stack is not empty, return to the step and trace in another direction. If the starting point stack is empty, the process returns to the step and searches for a starting point again. When the scanning of the entire screen is completed in this manner, recognition is completed and stopped.

Next, specific processing methods in each of the above steps will be explained in detail.

[Line segment tracing]

FIG. 1 is a diagram showing a line segment tracking sensor in an embodiment of the line figure recognition system of the present invention. In the figure, S ₁ , S ₂ , and S ₃ indicate independent detection units, which move along the line segment L in the direction of the arrow. Detection unit
S ₁ , S ₂ , and S ₃ express whether or not there is a pixel in the area covered by each using 3-bit information consisting of the respective output signals. For example, a detector located in the direction of movement with respect to the target point (X, Y)
There are pixels as indicated by diagonal lines in _S2 , and other detectors
When there are no pixels in S ₂ and S ₃ , the output information is (010)
It is expressed as The output information of the line segment tracking sensor includes:
The following three cases may exist: The first case is (010). In this case, the sensor is assumed to be on the sensor line segment, and the entire sensor is advanced in the traveling direction. The second case is (000). In this case, it is treated as an open line. However, in consideration of line breaks due to blurring, the determination is made after scanning several pixels in the advancing direction. The third case is a case other than the above two cases. In this case, it is treated as a singularity. That is, the sensor is in contact with a line intersection, intersection, or system. In this case, the movement of the sensor is stopped and singular point processing described below is performed.

[Singularity processing]

After detecting a singular point and stopping the sensor, a rectangular area of a fixed size is assumed based on the coordinates (X, Y) of the singular point, and singular point processing is performed on this area.

FIG. 2 is an explanatory diagram showing the region of singularity processing. In the figure, l ₁ ×l ₂ indicates a rectangular area in which singularity processing should be performed, and the mark . is the point (coordinates X, Y) where sensor A detected the singularity. The arrow indicates the direction of movement of the sensor. Figure 2 shows the intersection,
The shaded area is the part included in the rectangular area.

In order to perform singularity processing, the following quantities are measured for pixels within the area shown in FIG.

Marginal distribution In other words, pixels in a direction perpendicular to a certain side are gathered together and the distribution of pixels on the side is investigated.

The extending direction and length of the branch, the average slope near the singular point, and the intersection of lines in an electrical circuit diagram, the area of the intersection is measured to determine the presence or absence of a black circle at the intersection. If there is a black circle, it is an intersection,
When there is no black circle, it is just an intersection.

After the above measurements are completed, the singularities are roughly classified. Singular points are roughly classified into three types based on the shape of the figure at the singular point: a simple T-shape, a T-shape with a small circle at the intersection, and a case where it branches into a Y-shape. Most types of symbols can be determined by

In addition, in the case of electronic circuit diagrams, when a singular point is an intersection, its coordinates and branch direction are registered, intersections are ignored, and symbols are divided into three types according to the major classification of singular points described above, depending on their shape. Broadly categorize. Even in drawings other than electronic circuit diagrams, arbitrary connection information can be recorded by changing the registration conditions.

[Symbol recognition]

There are two situations on the screen when starting symbol recognition processing, one of which depends on the system configuration.

(i) After scanning the entire screen, all line segments that satisfy the conditions such as length and number of bends have been extracted, and line segments, symbols, characters, etc. that do not satisfy the conditions remain unprocessed. If you are in a state where

(ii) When one line segment is extracted and symbols remain at its end points. In this case, not all line segments have been extracted yet.

In case (i), contour tracing is performed on the remaining pixel area, and several contour lines are encoded and stored in memory. In case (ii), contour tracing is performed in the same way, but in consideration of the existence of unprocessed line segments, lines that intersect perpendicularly to the contours forming the symbol are ignored. Other line segments that do not satisfy the conditions for symbol outlines are ignored.

FIG. 3 is an explanatory diagram showing contour tracing of a symbol. In the same figure, S indicates a symbol, and the line segments with diagonal lines are processed line segments, and contour tracing is performed in the direction of the arrow from the end point indicated by the . This indicates that line segments that do not have a line segment are unprocessed, so they are ignored and contour tracing is performed.

During contour tracing, not only the contour information is encoded, but also the starting point of contour tracing and the intersection position with the ignored line segment are recorded. Freeman code is used as the code, and the outer contour line is targeted.

FIG. 4 is an explanatory diagram showing an example of a recording method of contour information. The figure shows contour information for one symbol, where 1 is the contour number, 2 is the coordinates (X, Y) of the tracing start point, 3 is the contour length, 4 is the contour code, and 5 is the intersection. The number of line segments present, 6, is the coordinate of the intersecting line segment, and one record is constituted by the information from 1 to 6 above.

Once the contour information is recorded in this way, the contour is then smoothed to extract bending points.
This is done by detecting locations with large changes from a series of contour codes. If such bending points and intersections of line segments are combined as feature points, a series of feature points is created by smoothing. In the case of the symbol shown in Figure 3, if an intersection is represented by T and a bend is represented by L, starting from the end point marked with (T→L→L→T→L→L→T→
(T)). (In this case, turn clockwise,
The first T and the last T are the same. ) The feature point sequence obtained in this way is expressed as 〓 ⁽ⁿ⁾ . Here, n is the number of feature points.

Also, when the distance between adjacent feature points i and j is d _ij , the distance series for a certain symbol is 〓 ⁽ⁿ⁾
It is expressed as In other words, the distance series 〓 ⁽ⁿ⁾ is expressed by the following equation.

〓 ⁽ⁿ⁾ = (d ₁₂ , d ₂₃ , d ₃₄ , ......, d _o1 ) On the other hand, the feature point series and distance series of each symbol are registered in the memory as a template.
Assuming that the number of symbols to be registered now is M, the feature point series and distance series are as follows.

〓 ₁ , 〓 ₂ , ......, 〓 _M : Feature point series 〓 ₁ , 〓 ₂ , ......, 〓 _M : Distance series However, when registering, those with the same number of feature points should be grouped together in the same category. . That is, they are classified as shown below.

Class 1: (〓 _a ⁽¹⁾ , 〓 _a ⁽¹⁾ ), (〓 _b ⁽¹⁾ , 〓 _b ⁽¹⁾ ),...
... Class 2: (〓 _d ⁽²⁾ , 〓 _d ⁽²⁾ ), ...... 〓 Therefore, a symbol whose contour line has k feature points can be matched with a template of class k. .

FIG. 5 is an explanatory diagram showing matching between unknown symbols and templates. In the figure, reference numeral 11 indicates an unknown symbol, which has k feature points and has an extracted feature point series 11 _-1 and a distance series 11 _-2 . 12 indicates the content of the template, 1
2 _-1 is a class, 12 _-2 is a feature point series, 12 _-3 is a distance series, and 12 _-4 is a symbol name.

In FIG. 5, it is shown that an unknown symbol 11 having k feature points is matched with a template 13 of class k. In addition, in template 12, as an example of a template for symbols belonging to class 3, the feature point series is
VVV, an equilateral triangle is exemplified as a symbol with distance series d ₁₂ , d ₂₃ , d ₃₁ , and as an example of a template for a symbol belonging to class 4, the feature point series is LLLL and the distance series is d ₁₂ , d A rectangle is exemplified as a symbol with ₂₃ , d ₃₄ and d ₄₁ .

Next, templates with matching feature point sequences are selected from class k, and the template with the closest distance sequence is selected as the answer. If there is no matching template, the symbol will be rejected.

Note that even for the same symbol, the feature point series and the distance series change depending on the number of connecting lines, so it goes without saying that the template must be created to include all cases.

The line figure recognition system of the present invention extracts straight lines, singular points, and symbols by sequentially performing the various processes described above on line diagrams constituting electric circuit diagrams and electronic circuit diagrams. It is. Hereinafter, an example of the system configuration and the configuration of each part in the line figure recognition system of the present invention will be described in detail.

FIG. 6 is a block diagram showing the system configuration of an embodiment of the line figure recognition system of the present invention. The figure shows an example of a system in which symbol recognition is performed in case (i), that is, after all line segment elements are removed from the entire screen.

The entire system is roughly divided into a memory control section 21 and a determination control section 22, and these sections operate in synchronization with each other. In the memory control unit 21, the external image memory
Original image data consisting of binary signals read from a photoelectric conversion device (not shown) is stored in _M1 . Further, connection information indicating intersections, bends, etc. of line segments is stored in the connection information memory _M2 . Further, the shape information memory _M3 stores information such as the outline of the symbol and the length of one side. On the other hand, in the template memory _M4 , templates representing the shapes of symbols, which are composed of classified feature point series and distance series, are registered in advance. Memory control circuit MC
monitors input/output signals in an external memory group consisting of original image memory M ₁ , connection information memory M ₂ , shape information memory M ₃ and template memory M ₄ .

The determination control unit 22 includes a determination circuit group and a determination control circuit.
It consists of a PC. In the determination circuit group, the line segment determination circuit _PR1 determines whether the sensor is tracing the signal line. Therefore, as long as the sensor is moving on the signal line, only the line segment judgment circuit PR ₁ operates in the judgment circuit group, the feature extraction circuit PR _2.
When the sensor encounters an intersection, an intersection, or a symbol, each quantity shown in ~ above is measured for a predetermined singularity processing area, the features are extracted, and the singularity is roughly classified. In the case of a symbol, the matching circuit _PR3 operates to perform symbol recognition by matching the classified feature point series and distance series with the template. The judgment controller PC is
Input/output signals are monitored in a determination circuit group consisting of three processors: a line segment determination circuit PR ₁ , a feature extraction circuit PR ₂ , and a matching circuit PR ₃ . A bus line 23 is connected between the external memory group and the judgment circuit group.
are tied together.

In performing the above operation, three processors make up the judgment circuit group, namely the line segment judgment circuit.
PR ₁ , feature extraction circuit PR ₂ and matching circuit
PR ₃ selects the memory necessary for the processing from among the four memories in the external memory group, namely, original image memory M ₁ , connection information memory M ₂ , shape information memory M ₃ and template memory M ₄ . , and exchange data between them. In other words, in () line segment tracing processing, the line segment discriminator
PR ₁ reads image data from the original image memory M ₁ and tracks it, and when it reaches a singular point, writes connection information such as the coordinates and direction of the end point of the line segment to the connection information memory M ₂ . Furthermore, the image data corresponding to the path from the starting point of the tracking sensor to the singular point is marked as tracked and returned to the memory _M1 .

() At the singular point, the judgment control circuit PC activates the feature extraction circuit PR ₂ , reads out the coordinates of the singular point from the connection information memory M ₂ and the image data from the original image memory M ₁ , performs processing, and outputs the processing results. The data is written again to the original image memory M ₁ and the connection information memory M ₂ .

() In the case of symbol processing, first the feature extraction circuit
PR ₂ is activated and reads singular point coordinates from connection information memory M ₂ and unprocessed pixels from original image memory M ₁ to create a feature point series. This characteristic information is once read into the shape information memory _M3 , and then the matching circuit _PR3 starts processing. That is, template data is read from the template memory _M4 and unknown symbol data is read from the shape information memory _M3 , matching is performed, and the processing results are read into the shape information memory _M3 . The original image data of the symbol for which matching has been completed is marked as processed and returned to the memory _M1 .

FIG. 7 is a diagram showing the data input/output relationship between each memory and each processor (judgment circuit) in the system shown in FIG. 6, and collectively shows the series of processing steps described above. . In the figure, PR ₁ , PR ₂ and PR ₃ are processors that constitute a line segment determination circuit feature extraction circuit and a matching circuit, respectively, and M ₁ , M ₂ , M ₃ and M ₄
are an original image memory, a connection information memory, a shape information memory, and a template memory, which constitute the external memory, respectively. In the figure, arrows indicate the flow of data, indicating that data is output and input according to the direction of the arrow. Further, t indicates the passage of time.

In the line figure recognition system of the present invention, the system performs sequential processing due to the nature of the algorithm. For example, after processing line segments in a certain area of the screen, the line segment judgment circuit processor PR ₁ moves to another area. It is also possible to have the feature extraction circuit processor PR ₂ and the matching circuit processor PR ₃ operate in parallel while the process is being performed.

Hereinafter, the line segment determination circuit shown in FIG. 6,
Specific configuration examples of the feature extraction circuit and the matching circuit will be explained.

FIG. 8 is a block diagram showing an example of the configuration of a line segment determination circuit. In the figure, 31 is a connection information register, 32 is an array controller, 33 is a judger, 34 is an input/output (I/O) buffer, and 35 is an input/output (I/O) buffer.
O) It is a controller.

In FIG. 8, first, the coordinates of the starting point of line segment tracing are received from the connection information memory into the X and Y of the connection information register 31 via the I/O buffer 34. Next, image data from the original image memory is received through the I/O buffer 34 to the array controller, and the image data received in the array controller 32 is subjected to determination by the tracking sensor shown in FIG. Convert it into the required array format. In this case, due to the structure of the original image memory, for example, if it is stored in the form of a compressed code, it cannot be tracked by the sensor as it is, so it must be decoded to return it to the image form. If the image memory can store images in the form of images, such as dot memory, such conversion is not necessary.

In the determiner 33, determination is made using the tracking sensor shown in FIG. 1, and if the determination result is a line segment, J in the connection information register 31 is set to -1. In this case, a signal is sent to the original image memory via the I/O buffer 34, whereby the next image data is sent from the original image memory, and the next determination is made again. If the determination result in the determiner 33 is a singular point, set J in the connection information register 31 to 0 or 1, edit the coordinates X and Y of the singular point and the determination result J in the connection information register 31, and set the connection information. Send to memory. Note that the I/O controller 35 is
The connection information register 31, array controller 32, determiner 33 and I/O buffer 3 in the above operation
Controls the input and output of signals between the four.

FIG. 9 is a block diagram showing an example of the configuration of the feature extraction circuit. In the figure, 41 is a connection information register, 42 is an array controller, 43 is a peripheral distribution measuring device, 44 is a slope measuring device, 45 is a branch measuring device,
46 is a determiner, 47 is a contour extractor, 48 is a contour encoder, 49 is an input/output (I/O) buffer,
50 is an input/output (I/O) controller.

When a singular point is encountered on a line segment, the image data is transferred from the original image memory to the I/O buffer 49.
is received by the array controller 42 via the array controller 4.
The image data received in step 2 is converted into a rectangular shape corresponding to the region of singularity processing shown in FIG. In this case as well, the array controller 32 in FIG.
For the same reasons mentioned above, such a conversion may not be necessary.

For the image data arranged in a rectangular shape in the array controller 42, the marginal distribution is measured by the marginal distribution measuring device 43, the average slope near the singular point is measured by the inclination measuring device 44, and the branch measuring device 4
5, the direction in which the branch extends from the singular point and its length are measured. The determiner 46 includes a peripheral distribution measuring device 43, a slope measuring device 44, and a branch measuring device 45.
The type of singular point is investigated by combining the data of the respective features extracted in , and determining whether it is an intersection or which of the three major classifications of symbols mentioned above. The determination result in the determiner 46 is written into the connection information register 41, edited together with the pre-written coordinates of the singular point, and sent to the connection information memory via the I/O buffer 49.

When extracting features of a symbol, a contour extractor 47 extracts a contour from the image data in the array controller 42 . Contour extraction can be performed by the well-known method of tracing image contours using a small mask. The method is not limited to this method, and any other contour line extraction method may be used. The extracted contour data is sent to the contour encoder 48 and encoded. In encoding, the slope of the contour line is encoded by numbers 0 to 7 every 45 degrees. This is done using Freeman code.
The contour encoder 48 also smooths the contour at the same time. Smoothing is performed by removing noise components such as small irregularities in the contour line and emphasizing only large bends. The smoothed and encoded contour data is sent to the shape information memory via the I/O buffer 49 as a feature point series.

The I/O controller 50 controls input/output of signals between the connection information register 41, the determiner 46, the contour encoder 48, and the I/O buffer 49.

FIG. 10 is a block diagram showing an example of the configuration of matching information. In the figure, 51 is a code length register, 52 is a code register, 53 is an intersection register, 54 is a template register, and 55 is a code length register.
is an EX-OR circuit, 56 is a distance measuring device, 57 is a comparator, 58 is a determiner, 59 is an input/output (I/O) buffer, and 60 is an input/output (I/O) controller.

When recognizing symbols, the length of the contour line series R _L , the contour line series R _C , and the intersection coordinate series R _CR
are read out from the shape information memory and sent to the code length register 5 via the I/O buffer 59.
1, written to the code register 52 and intersection register 53.

The contents of the code length register 51 are sent to the template memory via the I/O buffer 59, and templates with the same code length, that is, the same type, are selected and read out. The read template data passes through the I/O buffer 59 to the template register 54, and the value of class k, the feature point series, and the distance series are read according to R' _L , R' _C , and R' _CR , respectively. It will be done.

Contour line series Rc in code register 52 and feature point series in template register 54
A mismatch with _R'C is detected in the EX-OR circuit 55. Furthermore, the intersection coordinate series R _CR in the intersection register 53 and the distance series R' _CR in the template register 54 are
The respective distances are calculated in the comparator 57, and the calculated distances are compared in the comparator 57.

Information on the comparison results in the EX-OR circuit 55 and the comparator 57 is sent to the determiner 58. The determiner 58 first detects whether the contour lines match, and if they do not match, reads out a different feature point series with the same sequence length from the template memory and further compares it. If a matching contour line series is found in this way, a match in distance is detected for the coordinate series of the next intersection, and if there is no match, a matching feature point series with a different distance series is found from the template memory. Read to find more matches.

In this way, when both comparison results in the determiner 58 match, it is determined that the symbol is the symbol represented by the template. The result of the judgment is I/O buffer 5
9 and is stored in the shape information memory. Furthermore, I/
The O controller 60 controls the input/output of signals among the code length register 51, code register 52, intersection register 53, determiner 58, and I/O buffer 59 in the above operation.

As described above, according to the line figure recognition system of the present invention, straight lines, singular points, and symbols can be extracted from a binary image of a drawing consisting of line figures mainly consisting of straight lines. At this time, the bond line and the symbol portion can be separated and identified.
Furthermore, in addition to extracting line segments, it is also possible to simultaneously extract and record connection information between line segments. Furthermore, any number of types of symbols can be extracted by registering their shapes and sizes. Therefore, according to the line figure recognition system of the present invention, it is possible to automatically input electrical circuit diagrams and electronic circuit diagrams into a computer, and it is expected to have great effects such as preventing input errors and significantly reducing man-hours. It is something that can be done.

Although the above description has been made regarding the case where electrical circuit diagrams and electronic circuit diagrams are targeted, the application of the line figure recognition system of the present invention is not limited to these cases, but can also be applied to architectural drawings, component design drawings, etc. Needless to say, it is something to be gained.

[Brief explanation of drawings]

Figure 1 is a diagram showing the line segment tracking sensor, Figure 2 is an explanatory diagram showing the region of singularity processing, Figure 3 is an explanatory diagram showing symbol contour tracing, and Figure 4 is an outline information recording method. FIG. 5 is an explanatory diagram showing an example of matching an unknown symbol and a template. FIG. 6 is a block diagram showing the system configuration of an embodiment of the line figure recognition system of the present invention. FIG. 6 is a diagram showing the data input/output relationship between each memory and each processor, FIG. 8 is a block diagram showing an example of a configuration of a line segment determination circuit, and FIG. 9 is a diagram showing an example of a configuration of a feature extraction circuit. Block diagram, 1st
FIG. 0 is a block diagram showing an example of the configuration of a matching circuit. 1... Contour line number, 2... Coordinates of the tracing start point,
3...Outline length, 4...Outline code, 5...Number of intersecting line segments, 6...Coordinates of intersecting line segments, 11
...Unknown symbol, 11 _-1 ...Feature point series, 11
_-2 ...Distance series, 12...Template contents,
12 _-1 ... class, 12 _-2 ... feature point series, 12 _-3 ...
... Distance series, 12 _-4 ... Symbol, 13 ... K-type template, 21 ... Memory control unit, 22 ...
...Judgment control unit, 23... Bus line, 31... Connection information register, 32... Array controller, 33...
Judgment device, 34...Input/output (I/O) buffer, 3
5... Input/output (I/O) controller, 41... Connection information register, 42... Array controller, 43... Peripheral distribution measuring device, 44... Inclination measuring device, 45... Branch measuring device, 46 ... Determiner, 47 ... Contour extractor, 48 ... Contour encoder, 49 ... Input/output (I/O) buffer, 50 ... Input/output (I/O)
Controller, 51...Code length register, 52...Code register, 53...Intersection register, 54...
...Template register, 55...EX-OR
Circuit, 56... Distance measuring device, 57... Comparator, 5
8... Determiner, 59... Input/output (I/O) buffer, 60... Input/output (I/O) controller.

Claims (1)

[Claims] 1. 2 obtained from a line figure via a photoelectric conversion device
Regarding image data consisting of value signals, 3-bit information consisting of the output signal of each detection section in a line segment tracking sensor that has three detection sections for upper, front, and lower in the direction of line segment movement is (010), ( 000), (other), find out whether the line segment is continuous in the traveling direction of the sensor, the line segment has ended, or a singular point on the line segment, and if the line segment is continuous or a straight line extracting means for determining whether or not it is a true end by tracing the line segment in the advancing direction if the line segment ends, and scanning several pixels ahead in the advancing direction when the line segment ends; Regarding the singular point found by this method, the peripheral distribution of the image within a certain area including the singular point, that is, the pixel distribution on the edge when pixels in the direction perpendicular to this edge are gathered together on a specific edge, and the straight line. The characteristics of the geometric shape of the figure are measured from the extending direction and length of the branches and the average slope near the singular point, and the presence or absence of a black circle at the intersection is determined from the area of the intersection. a singular point extracting means for distinguishing between symbols and broadly classifying symbol shapes; information on a series of singular points on a contour line and a series of distances between feature points on a contour line for the symbols found by the singular point extracting means; 1. A line figure recognition system comprising symbol extraction means for identifying and extracting a symbol by comparing the information with a predetermined template.