JPH0488481A - Detecting method for symbol candidate area - Google Patents

Abstract

PURPOSE:To increase the detection accuracy by extracting a symbol candidate and a connection line candidate and also extracting their subordinate areas, and deciding the possibilities of the candidates respectively. CONSTITUTION:A graphing information processing part P1 for feature points obtain information wherein the features of a graphic segment are graphed according to feature point information detected by using both the contour vectors and core vectors of a figure. An extraction processing part P2 for the symbol candidate extracts the symbol candidate from the graphing information on the feature points and the rule in a symbol candidate extraction rule storage part M1 is used for the extraction to extract the symbol candidate and its subordinate area from the graphing information on the feature point. An extraction processing part P3 for the connection line candidate uses the graphing information and the rule in a connection line candidate extraction rule storage part M2 to extract the connection line candidate according to whether the possibility is high or low. Further, a detection processing part P4 for symbol and connection line candidate areas detects a symbol candidate area or connection line candidate area for the extraction results of both the candidates. Consequently, the symbol candidate and connection line candidate are securely detected.

Description

[Detailed Description of the Invention] A. Industrial Application Field The present invention relates to a figure recognition system for automatically recognizing figures such as electronic circuit diagrams in which symbols and lines are mixed, and in particular to a figure recognition system for automatically recognizing figures such as electronic circuit diagrams in which symbols and lines are mixed. This invention relates to a method for detecting a raccoon symbol candidate region that is separated into symbols and connecting lines.

B1 Summary of the Invention The present invention extracts feature points of figure line segments through contour vectorization and core line vectorization of a binarized image, graphs the figure line segments based on these feature points, and generates symbols from this graphing information. In a figure recognition system that recognizes symbols and connection lines by separating them into candidate areas and connection line candidate areas, symbol candidates and connection line candidates are extracted from graphing information with priority, and subgroups that can be the same candidate are extracted. By extracting a region from its surroundings, the possibility of it becoming a candidate is determined, and each candidate region is detected based on this determination, thereby ensuring the detection of each candidate.

C1 Conventional technology A graphic recognition system recognizes symbols and line segments in a graphic as separate elements through a processing procedure as shown in FIG. The drawing to be recognized is collected as binarized image information using an image scanner or the like (step Sl),
Contours of the binarized image are extracted (step S2), and then contour vectorization is performed to extract contour vectors (step S3). For example, the binarized image A shown in FIG.
The contour is extracted as a continuous contour image of the boundary points between the black part and the white part, and a contour vectorization process is performed by using the part where the boundary points are continuous as a straight line as one contour vector.

This contour vector is represented as an arrow line segment of figure B in the figure.

Next, in order to trace the line segments of the figure from the contour vector information, thinning processing is performed so that the line width of the figure becomes 1 (unit width), and the thinned points are approximated by vectors to create core line vectors. Perform skeleton vectorization to obtain information (
Step S4).

Next, from the skeleton vector information, the end points and bending points 1 of the figure line segment are
Extract feature points (marked with circles in Figure 4) at branch points (step S5), create graph information of figure line segments based on this feature point information (step S6), and create symbols from this graph information. Detect and separate candidates and connection line candidates (Step S7), perform symbol and line segment recognition processing for each candidate (Step S8), and finally, the data converted into symbol code and line code information by this recognition processing is It is edited and output as figure recognition information.

D3 Problems to be Solved by the Invention In conventional systems, symbol candidates and connection line candidates are detected using relatively long line segments as connection line candidates, and a large number of relatively short line segments existing in an area. Objects are detected as symbol candidates.

However, in dense drawings or drawings with many nearby symbols, many short connecting lines coexist, and detecting candidates based only on the lengths of line segments increases the error rate and makes detection impossible.

Particularly, as a conventional method, there is a method in which candidate detection is performed by preparing rules that allow detection of a specific symbol. This method limits the types of symbols that can be detected, and requires new rules to be added to detect new symbols.

An object of the present invention is to provide a method that ensures the detection of symbol candidates and connection line candidates.

In order to achieve the above object, the present invention extracts a contour vector of a binarized image, extracts a thinned core vector from this contour vector, and extracts a figure from this core vector. Extract the feature points of the line segment, create graph information of the figure line segment based on the feature points, separate the figure into symbol candidates and connection line candidates from this graph information, and In a figure recognition system that performs figure recognition processing, symbol candidates and connection line candidates are extracted with priority according to extraction rules for symbol candidates and connection line candidates using graphing information of each feature point, and surrounding areas of each candidate are extracted. sub-areas that can be the same candidate are extracted from the graphing information of The symbol candidate area and the connection line candidate area are determined and detected by assigning a ranking to the character as a sub-region.

F. Embodiment FIG. 1 is a software configuration diagram showing an embodiment of the present invention. The feature point graphing information processing unit P1 obtains information in which the features of a figure line segment are graphed based on feature point information detected using both the contour vector and core line vector of the figure. FIG. 2 shows an example of graph representation by graphing information processing of feature points, which is based on feature point information extracted from the contour vector and skeleton vector diagram of FIG. 4. Feature point extraction for graphing involves extracting contour vectors corresponding to each skeleton vector, and determining T-shaped branches and cross-crossings based on the number of points of the skeleton vector that has a start point or end point and the shape of the contour vector around the reading point. Furthermore, the presence or absence of filling is determined. In addition, to graph the figure line segment from the feature point information, the shape type of the original image is determined as a section from the state of the vector group between the feature points, and the shape type also includes features such as vector refraction. When detected, it is determined as a subsection, and graphing information is obtained by correlating it with the contour vector.

The graph representation in Figure 2 resulting from these processes is a section consisting of the feature point information whose end point is t1 and the bifurcation point b, and the straight line connecting them.The black circles are vector inflection feature points, and subsections are extracted and features A graph is expressed by connecting points with straight lines.

Returning to FIG. 1, the symbol candidate extraction processing unit P2 extracts symbol candidates from the graphing information of the feature points. For this extraction, the rules in the symbol candidate extraction rule storage M1 are used to extract symbol candidates and their sub-regions from the graphing information of the feature points.

The rule storage unit Ml is composed of a plurality of rules, each rule is prioritized, and symbol candidates and their sub-regions are extracted from graph information consisting of feature points and straight lines connecting them.

For example, the graph information that is connected between the branch feature point S and section SI in Figure 2 is considered to be information that is likely to be a symbol candidate from the loop connection with the inflection point (black circle).
, and a sub-region is searched to see if there is anything that can be a symbol candidate near this graph information.

Through this search, a sub-region consisting of sections Ll, L4, Ll and branch feature points is detected, and when these straight lines are traced based on rule M1, a loop of Ll-L4-Ll-Ll is detected. From the loop nature of this sub-region and the possibility of symbol candidates for section LIO, Ll, L4,
Ll, LIO and their branch feature points are extracted as symbol candidates with high rankings. Therefore, the symbol candidate extraction process does not determine the characteristics of a certain limited symbol, but determines how likely the symbol candidate is, including its sub-region.

Returning to FIG. 1, the connection line candidate extraction processing unit P3 uses the graphing information and the rules in the connection line candidate extraction rule storage unit M2 to determine the probability of connection line candidates, similar to the symbol candidate extraction process. Extract with For example, sections L9 and L8 that connect the endpoint feature point t and the branch feature point S have no loop property, and there are no sub-regions around them that have the possibility of symbol candidates, so there is a possibility that they are connection line candidates. It is determined that there is.

The processing of these extraction processing units P2 and P3 is to determine the possibility of a symbol candidate or a connection line candidate, and some sections may have the possibility of becoming both candidates. Each rule is given a priority based on its priority.

Returning to FIG. 1 again, the symbol/connection line candidate area detection processing unit P4 uses the rules in the candidate area detection rule storage unit M3 to determine the extraction results of both candidates as a symbol candidate area or a connection line candidate area. To detect. The rules for this detection are configured to make decisions based on the priorities of symbol candidates and connection line candidates and their surrounding sub-regions. In the example of FIG. Candidate area (shown as dashed block in Figure 4)
A region having sections L5, L6 and L8, L9, L2, and L3 is detected as a connection line region.

Through the above processing, symbol candidate areas and connection line candidate areas are detected, symbols and connection lines are separated, and the set of contour vectors in the symbol candidate area is compared with the set of contour vectors in the dictionary. This allows symbol recognition.

G1 Effect of the Invention As described above, according to the present invention, in order to separate and extract symbol candidate areas and connection line candidate areas from graphing information, symbol candidates and connection line candidates are extracted and their sub-areas are extracted. Since we have also determined the possibility that each candidate area may become a candidate and used this to detect each candidate area, we have improved the detection accuracy of candidate areas compared to conventional candidate extraction using general properties as rules. enhance In addition, the reliability of candidate areas is expressed in the form of priorities, and detection is performed based on area detection rules, so the detection method is highly versatile and can be applied to drawings of various formats, such as drawings with high density. This is a highly versatile method. Furthermore, changes in symbol types and description methods can be easily handled by simply changing the priority order and modifying the symbol candidate area detection rules.

[Brief explanation of drawings]

FIG. 1 is a software configuration diagram showing an embodiment of the present invention;
FIG. 2 is a diagram showing an example of feature point graph representation, FIG. 3 is a processing flowchart of the figure recognition system, and FIG. 4 is a diagram of contour vectors and skeleton vectors. Pl...Characteristic point graphing information processing unit, P2...Symbol candidate extraction processing unit, P3...Connection line candidate drawing ratio processing unit, P4...Symbol/connection line candidate area detection processing unit , Ml... symbol candidate extraction rule storage unit, M2.
. . . Connection line candidate extraction rule storage unit, M3 . . . Candidate area detection rule storage unit. 1 other person 2nd figure Eh, December 20, 1990

Claims (1)

[Claims] (1) Extract the contour vector of the binarized image, extract the thinned skeleton vector from this contour vector, extract the feature points of the figure line segment from this skeleton vector, and draw the figure line based on the feature points. In a figure recognition system that creates graph information for each feature point, separates the figure into symbol candidates and connection line candidates from this graph information, and performs figure recognition processing on each candidate, the graph information of each feature point is is used to extract symbol candidates and connection line candidates with priority according to the symbol candidate and connection line candidate extraction rules, and to extract sub-regions that can be the same candidate from the graphing information around each candidate, A method for detecting a symbol candidate area, comprising detecting a symbol candidate area from the priority order of candidates and connection line candidates and sub-areas surrounding the candidates.