JPH01129368A - Picture recognition device - Google Patents

Abstract

PURPOSE:To efficiently extract a symbol by deciding a next segment to be traced based on a feature point detected during a tracing and extracting a closed loop formed by the plural segments traced until a start point is detected again as the symbol. CONSTITUTION:The feature point is detected from a picture and this detected feature point is stored. Then, the start point tracing the segment is selected from the stored feature points and the segment to be traced and a tracing direction are decided from this start point. The, based on the feature point detected during the tracing, the segment to be traced and the tracing direction are decided and the closed loop formed by the plural segments traced until the start point is detected is extracted as the symbol. Thereby, even in the picture in which the symbol is overlapped on a connecting line, for instance, only the symbol can be effectively and efficiently extracted.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image recognition device, and more particularly to an image recognition device that distinguishes and recognizes symbols and connection lines from images such as block diagrams and circuit diagrams.

[Prior Art] Conventionally, an image recognition device that reads an image from a drawing such as a block diagram or a circuit diagram and recognizes the drawing employs, for example, the following recognition method.

First, the symbols read on the drawing and the connecting lines connecting the symbols are each stored as segmented line images. Line segments are then traced based on this stored line image, and when a closed loop is extracted, this portion is determined to be a symbol.

In addition, when a feature point such as a branch point or an intersection is reached while tracing a line segment, the next line segment is detected by searching the line image clockwise or counterclockwise around each feature point. are doing.

[Problem to be solved by the invention] However, when a connecting line and a symbol overlap, a closed loop is extracted while tracing some line segments of the symbol, and this closed loop is It is judged as.

Therefore, it often happens that symbols cannot be reliably extracted using the above-described symbol determination method.

An example of the conventional symbol extraction method for extracting symbols from drawings in which these connecting lines and symbols overlap is shown in Figure 6.
This will be explained based on FIG.

FIG. 6 shows the tracing of the line segment from the connector 50 to the connector 51. In this case, the tracing of the line segment is from the branch point A to the T block →
The process is performed in the order of ding-te ■ → ding-T.

Furthermore, FIG. 7 shows the tracing of the line segment from the connector 52 to the connector 53, and in this case, the tracing of the line segment is from the branch point E to EF-F cho → cho "→" cho-cho cho 1-cho. - hKL - LE is performed in this order.

In this way, in FIGS. 6 and 7, closed loops are extracted while tracing some line segments of the symbols, so originally symbols A'B' CD and 7 in FIG.
It is difficult to accurately extract the symbols E'J' KL in the figure.

As described above, it is not possible to reliably distinguish between symbols and connection lines and extract only the symbols, leading to problems such as the possibility of erroneously recognizing drawings.

The present invention has been made in view of the above-mentioned problems, and its purpose is to provide image recognition that can reliably and efficiently extract only symbols, even if the symbols and connecting lines are similar to each other. The point is to provide equipment.

[Means for Solving the Problems] In order to solve the above-mentioned problems and achieve the purpose, an image recognition device according to the present invention has symbols having regular shapes and connection lines connecting the symbols. An image recognition device that recognizes an image composed of a plurality of line segments, comprising a detection means for detecting feature points from the image, a storage means for storing the feature points detected by the detection means, and a storage means for storing the feature points detected by the detection means. a selection means for selecting a starting point for tracing a line segment from among feature points; a first determining means for determining a line segment to be traced and a tracing direction of the line segment from the starting point selected by the selection means; a tracing means for tracing a line segment in the tracing direction determined by the first determining means; a line segment for detecting a feature point during the tracing by the tracing means; and a line segment to be traced next based on the feature point; and a symbol extracting means for extracting, as a symbol, a closed loop formed by a plurality of line segments traced until the second determining means detects the starting point. .

[Operation] According to the above configuration, a line segment is traced from the starting point, and the next line segment to be traced is determined based on the feature points detected during the tracing. Then, a closed loop formed by a plurality of line segments traced until the starting point is detected again is extracted as a symbol.

In this way, efficient symbol extraction becomes possible.

[Embodiments] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram showing the configuration of an image recognition device according to the present invention.

1 is a CPU that controls the entire device. 2 is an RO that stores a control program, an error processing program, a program for processing according to the flowchart in FIG. 4, etc.
Reference numeral M3 denotes a RAM that functions as a working area when various programs are executed and a temporary save area during error processing. Reference numeral 4 denotes an image input unit that scans a drawing in which block diagrams, circuit diagrams, etc. are recorded and inputs an image.

5 binarizes and thins the video signal input from the image input unit 4 to obtain a line image, integrates the branch points of each line image, and uses a 3×3 dot masking matrix to identify end points, branch points, and intersections. This is a structure analysis unit that performs extraction processing, etc. 6 is a polygonal line approximation unit that approximates the crooked point with a polygonal line and converts it into a vector in order to find the crookedness point from the thinned line image; 7 is a polygonal line approximation unit that approximates the polygonal line and converts it into a vector; This is a structural graph creation unit that creates a structural graph in which all feature points (all collectively referred to as feature points) are treated as knots, and line segments connecting these feature points are line segment edges. Reference numeral 8 denotes a structure graph storage unit that stores the structure graph created by the structure graph creation unit 7.

Further, 9 is a symbol candidate extraction unit that extracts symbol candidates based on the structure graph stored in the structure graph storage unit 8, and 10 is a symbol candidate extraction unit that classifies the symbols extracted from the symbol candidate extraction unit 9, and classifies the types of each classified symbol. A symbol classification unit 11 is a symbol storage unit that stores the symbols classified by the symbol classification unit 10 for each classification. Reference numeral 12 denotes a connection line extraction section that extracts connection lines based on the structure graph stored in the structure graph storage section 8, and reference numeral 13 denotes a connection line storage section that stores the connection lines extracted by the connection line extraction section 12.

Reference numeral 14 denotes a connection relationship analysis unit that analyzes connection relationships based on symbols stored in the symbol storage unit 11 and connection lines stored in the connection line storage unit 13; and 15, connection between symbols and connections from the connection relationship analysis unit 14. This is a recognition result storage unit that stores recognition results of connection relationships with lines. This connection relationship includes the type of symbol, the positional relationship between the symbol and the connection line, etc. Further, 16 is a display unit that displays an input image based on the recognition result stored in the recognition result storage unit 15, and 17 is a keyboard provided with control keys and the like. Furthermore, the display section 16
Instead, a printer section may be provided, or both the display section 16 and the printer section may be provided.

The image recognition device configured as described above is particularly shown in FIG. 2(a).
The target is a case where symbols shaped like those shown in (e) are connected by a connecting line. In addition, Fig. 2 (a) ~
The symbols shown in (e) are symbols used in block diagrams, logic circuit diagrams, etc., and words to explain these symbols will be omitted.

Furthermore, FIG. 3 shows a structural graph composed of feature points and line segment edges obtained by polygonal line approximation. In the figure, the feature point "a" is an end point where only - tree line segments appear, and the feature point "b" is a branch point where three tree line segments appear. Further, the feature point "c" is an intersection where four tree line segments appear, and the feature point "d" is a bending point obtained by polygonal line approximation.

Next, a method for extracting symbols by the symbol candidate extracting section 9 of this embodiment will be explained below according to the flowchart of FIG. Note that this is an explanation when reading a drawing having a plurality of symbols.

First, after storing a structural graph as shown in FIG. 3 in the structural graph storage section 8, the symbol candidate extraction section 9 selects one inflection point to be the starting point of extraction in order to extract symbols (step S1). , the coordinates of this bending point are stored (step S2). This bending point is an element constituting a symbol, and a plurality of bending points are included for one symbol.

Therefore, the coordinates of the bending points are stored in groups for each symbol. RA each group as a feature point storage group.
Store in M3.

Now, since two line segments extend in different directions from the bending point in a negative rotation of 360 degrees, the directions of these line segments are next determined (step S3). Then, two angles formed by the two tree line segments are detected with the bending point as the center, and if each angle (only one angle is sufficient) is approximately equal to 180°, the process returns to step S1 again and another bending point is detected. The same process as described above is performed based on the selection. Also detected 2
If the two angles are clearly different, the two angles are compared (step S5). Focus on the smaller angle of the two angles, and among the two tree line segments that form this smaller angle, trace the line segment that extends to the left of the bending point (hereinafter referred to as the tracing line segment). ). The method for determining this tracing line segment will be explained with reference to FIG.

In FIG. 5, U is the selected inflection point, the line segment extending to the left of this inflection point U is the line segment T, and the line segment extending to the right is the line segment T. There are two angles formed by these line segments: angle α and angle β. Here, there is a relationship between angle α<angle β, and if we pay attention to the smaller angle α side, the lower line segment T extending to the left of the bending point U corresponds to the tracing line segment. If the angle β side has a smaller angle, the line segment UW corresponds to the tracing line segment. Once the tracing line segment is determined in this way, the tracing line segment is traced until a feature point is detected from the branch point (step S7, step S8).

Next, when a feature point is detected, the following processing is performed depending on the type of feature point.

First, if it is determined to be a "starting point", it means that the inflection point selected in step S1 has been returned to, so this closed loop is determined to be a symbol, and the feature point storage group of this symbol is sent to the symbol classification unit 10. Send (step S
9, step 510). Then, a search is made to see if there are any other bending points, and when a new bending point is detected, the process returns to step S1 (step S11, step 512).

Further, if it is determined to be an "end point", it means the end point of the tracing line segment, and further tracing is impossible, so the process returns to step S1 again (step 513).

Alternatively, if it is determined to be an "inflection point," the coordinates of this inflection point are stored in the feature point storage group of the symbol being extracted (step S14, step 515).Then, the line segment that continues beyond this inflection point is added.゛ Step S7 as trace line segment
Let's try tracking again.

Furthermore, if it is determined that it is a “branching point”, Rule 1 described below
Then, the next tracing line segment is detected, the coordinates of this branching point are stored in the feature point storage group of the symbol being extracted, and the process returns to step S7.Steps S16, S17,
step 818). According to Rule 1, ``At a branch point, rotate the immediately preceding line segment counterclockwise around the branch point, and determine the first line segment that overlaps as the tracing line segment.''

If it is determined to be an "intersection", the next tracing line segment is detected according to Rule 2, which will be described later, the coordinates of this intersection are stored in the feature point storage group of the symbol being extracted, and the process returns to step S7 (step S19, step S20, step 521). According to Rule 2, ``At an intersection, the line segment that exits in the direction closest to the tracing direction of the immediately preceding line segment is determined to be the tracing line segment.'' Note that if it cannot be determined that the intersection is an intersection during the process of step S19, it is determined that the intersection cannot be recognized and the process is stopped, or an error process is performed.

An overview of a series of processing steps from symbol candidate extraction processing to output of recognition results will be described below.

As in the above-described symbol candidate extraction processing, symbols are extracted one by one, and at the same time, the symbol classification unit 10 classifies each symbol having a similar shape. Incidentally, symbols having a predetermined shape are registered in advance in the symbol classification section 10, and the extracted symbols are classified into corresponding symbols among the registered symbols. Then, for each symbol whose type has been determined, a type determination flag is set and stored in the symbol storage unit 11. In this way, symbols that were still candidate symbols at the time of extraction are determined to be symbols. Furthermore, if the extracted symbol does not correspond to the registered symbol, the extracted symbol is excluded from the candidates. In this way, the processing up to symbol extraction and storage is repeated continuously until no symbols can be extracted anymore. In particular, for the line segments that make up the symbol, for the two line segments that come out from the bending point, set flag 1 for the corresponding vector (line segments are stored as vectors) on the structure graph. , or delete the line segment.In this way, it is possible to prevent the inflection point that becomes the starting point from being duplicated.Furthermore, a flag is set for each vector on the structure graph that does not correspond to the symbol. 0" is set.

Next, a method for extracting connection lines will be explained.

First, the connection line extraction unit 12 checks the flags set for vectors in the structure graph in the structure graph storage unit 8, and determines vectors that are not set to flag 1'' as connection lines. While extracting the symbol, the connecting line is stored in the connecting line storage section 13. Similar to the above-described symbol extracting process, the connecting line is stored in the connecting line storage section 13. Similar to the above-described symbol extracting process, this connecting line is continued until it can no longer be extracted.

When all the symbols and connection lines are extracted in this way, the connection relation analysis unit 14 analyzes how the connection lines are connected to the symbols. This analysis includes analysis of the matching relationship between the coordinates of feature points constituting the symbol and the coordinates of both connectors of the connecting line. The analysis result of the connection line analysis section 14 is then stored in the recognition result storage section 15 as an image recognition result.

As explained above, in this embodiment, even in a block diagram or a logic circuit diagram in which symbols and connection lines overlap, it is possible to accurately grasp the tracing direction of the tracing line segment. Closed loops can be extracted. This reliable closed-loop extraction allows symbols to be extracted efficiently and accurately.

Furthermore, in this embodiment, when tracing a line segment, it is also possible to modify the method of determining the next line segment to be traced at a feature point. This first modification will be explained below based on FIG. 7.

In the first modification, contrary to the method of determining the tracing line segment explained in FIG. 5, when focusing on the smaller angle formed by the two line segments, The line segment that is located is determined as the tracing line segment. In this case, tracking is performed in the opposite direction to the closed-loop tracking direction according to the embodiment described above. Also, rule 1 has rule 1°, and rule 2 has rule 2°.
correspond to each other. That is, according to Rule 1', it is defined as ``At a branch point, rotate the previous line segment clockwise around the branch point, and determine the first line segment that overlaps as the tracing line segment.'' According to ``, at an intersection, the line segment that exits in the direction closest to the tracing direction of the immediately preceding line segment is determined to be the tracing line segment.''

In this embodiment and the modified examples, the same operation and effect can be obtained by providing regularity regardless of the closed loop tracking direction.

In addition, as a second modification example, step S in FIG. 4(a)
There is a method of calculating the direction of the line segment found in step 3 in advance.

As a method for this, the direction of each line segment, that is, the direction of the vector may be calculated at the stage of the polygonal line approximation unit 6.

[Effects of the Invention] As described above, the present invention can reliably and efficiently extract symbols even from images in which symbols and connection lines overlap.

[Brief Description of the Drawings] Figure 1 is a block configuration diagram showing the configuration of an image recognition device according to the present invention, Figures 2 (a) to (e) are diagrams showing the shapes of various symbols, and Figure 3 is a block diagram showing the configuration of an image recognition device according to the present invention. An explanatory diagram for explaining feature points, FIGS. 4(a) and (b) are flowcharts for explaining the symbol extraction method of this embodiment, and FIG. 5 is for determining a tracing line segment from a branch point of this embodiment. 6 and 7 are explanatory diagrams illustrating the conventional symbol extraction method. In the figure, 1...CPU, 2...ROM, 3...RA
M. 4... Image human power section, 5... Structure analysis section, 6... Polygonal line approximation section, 7... Structure graph creation section, 8... Structure graph storage section, 9... Symbol candidate extraction section , 10...
・Symbol division neck, 11...Symbol storage section, 12・
... Connection line extraction section, 13... Connection line storage section, 14...
・Connection relationship analysis section, 15... Recognition result storage section, 16.
...Display section, 17...Keyboard section, 50-53...
・It is a connector. (a) (b) Figure 3 (c) (d) (e) Figure 2 Figure 5

Claims (5)

[Claims] (1) In an image recognition device that recognizes an image in which regularly shaped symbols and connection lines connecting the symbols are formed by a plurality of line segments, a detection means for detecting feature points from the image; a storage means for storing feature points detected by the detection means; a selection means for selecting a starting point for tracing a line segment from among the feature points stored in the storage means; a first determining means for determining a line segment and a tracing direction of the line segment; a tracing means for tracing the line segment in the tracing direction determined by the first determining means; , and determines the next line segment to be tracked and the tracking direction of the line segment based on the feature points.
and symbol extracting means for extracting, as a symbol, a closed loop formed by a plurality of line segments traced until the second determining means detects the starting point. Device. (2) The image recognition device according to claim 1, wherein the starting point is a bending point. (3) A patent claim characterized in that the first determining means includes a first selecting means for regularly selecting one of two line segments opening around a starting point as a tracing line segment. The image recognition device according to item 2 of the range. (4) At least when the feature point is a branch point or an intersection, the second determining means selects the next tracing line segment according to regularity based on the line segment immediately before reaching the feature point. 2. The image recognition device according to claim 1, further comprising two selection means. (5) The image recognition according to claim 1, wherein the second determining means includes a tracking stop means for stopping the next tracking from the feature point when the feature point is an end point. Device.