JP3668560B2 - Line figure recognition method and recognition apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a figure registration method for registering a map and a figure included in a drawing read by a photoelectric conversion device such as a scanner or a TV camera in a database, and in particular, it is represented by a set of a plurality of similar lines such as contour lines. The present invention relates to a method for recognizing a line figure which is effective when a figure to be recognized is registered.
[0002]
[Prior art]
Many prior studies on automatic input of contour line information have been reported as conventional line figure recognition methods. For example, in Reference 1: Aoiin et al., `` Study on mountain map information '' (Report of the Institute of Electronics, Information and Communication Engineers, PRL81-37, 1981), after applying thinning processing to the mountain map image to be input, a long vector component A method has been proposed in which contour line candidates are extracted from the determination as to whether or not, and the contour line candidates divided by characters and symbols are connected to each other based on the angle of the vector. However, this method has a problem that many connection errors occur because the determination of the vector angle is local.
[0003]
On the other hand, in Reference 2: Mizutani et al. 3, `` Extraction of contour lines using adjacent relations by Voronoi diagram '' (Electronic Information and Communication Society paper D-II, Vo1.J74-D-II, No. 11, 1991), A method has been proposed in which a Voronoi diagram is created in order to grasp the global adjacency of a line segment, an adjacency graph of the line segment is obtained therefrom, and the divided contour line components are accurately connected. In this method, it is possible to reduce connection errors from the determination of the adjacency graph. However, it is impossible to completely recognize the contour line in the part where the contour line is densely joined in a narrow area and the line is divided by the altitude value, cliff symbol, etc., and many recognition errors occur. Arise. For this reason, when automatic recognition using the batch processing type input method as described above is performed, an intensive error check is often requested from the operator. Since this intensive error check requires enormous effort, there is a problem that the total man-hours from the input to the system to the output are not reduced.
[0004]
Therefore, as described in Reference 3: Yamakawa et al. 1, "Interactive drawing input system" (Nikkei Computer Graphics, April 1987, pp.120-130) In particular, a dialog-guided recognition method that reflects the operator's judgment is attracting attention. In this dialog-guided recognition method, if the vectorization direction is not automatically determined, the vectorization direction is determined by an inquiry from the system to the operator, thus reducing unnecessary work required for error checking. be able to. However, the system configuration in which the search direction, that is, the vectorization direction is inquired every time a search direction is not automatically determined, such as a contour joint or division, is burdened on the operator (user). Is big.
[0005]
On the other hand, apart from such a flow, there are also many reports that recognize the whole figure by describing the target figure structurally by the relationship between the part and the whole, and checking the consistency between the searched parts sequentially. Has been. For example, in Reference 4: Nakamura et al. 1, “Image feature extraction by parallel search” (Journal of the Japanese Society for Artificial Intelligence, Vo1.8, No.1, PP.65-78, 1993), feature extraction based on such structure description It is proposed to realize this by cooperative action of multi-agents. According to this method, even if there is no detailed description about the flow of processing, it is possible to determine processing dynamically at the time of execution, and high-precision image recognition is expected by self-consistent problem solving. . However, this method may require enormous calculation time due to the explosion of the combination. In addition, in a figure having a complicated shape such as a contour line, the structure description itself is not clear.
[0006]
Therefore, in the present invention, as an application of multi-agent theory to line figure recognition, a recognition agent that performs route recognition corresponding to each of a plurality of adjacent lines specified by an operator and a management agent that performs overall management A method for configuring the system is proposed. In this method, highly accurate recognition is possible by describing the autonomous behavior of each agent and the cooperative behavior between agents. However, if the agent assignment is wrong and two agents are set corresponding to one contour line, there is a problem that the recognition result is wrong because the cooperative operation between agents is not performed correctly. In order to realize highly accurate recognition by the coordinated operation of multi-agents, a technique for automatically changing the agent assignment so that the agents are set correctly is important.
[0007]
As an example in which the agent assignment change is realized, there is a document 5: Gyoten et al. 3, “Multi-agent system (...) for character extraction from a document image. The spatial distribution method is applied to the character segmentation problem in the document image, and one agent is set corresponding to one character string in the document image, but the recognition target is different between character segmentation and line figure recognition. For this reason, the agent assignment used for character recognition and the assignment error detection method cannot be directly used for line figures.
[0008]
[Problems to be solved by the invention]
An object of the present invention is to register a part in which a line that is originally continuous becomes discontinuous due to restrictions on the notation of the figure (resolution, annotation, etc.) when a figure on a map or drawing is registered as vectorized data. When a single figure is recognized as one continuous vectorized data by automatically interpolating the part where multiple independent lines are connected (hereinafter referred to as the problem part), the recognition has already been completed. It is an object of the present invention to provide a graphic registration method and apparatus capable of performing line recognition processing without error based on a reliable recognition result and automatically registering the result in a database.
[0009]
Another object of the present invention is to automatically recognize and register a line figure by assigning a recognition agent to a figure before path interpolation and eliminating an erroneous assignment between recognition agents that recognize a target path. The object is to provide a line figure registration method and apparatus.
[0010]
[Means for Solving the Problems]
In order to achieve the above object, a graphic represented by a line on a paper drawing is input to a computer as binary image data, first vectorized data is created by thinning vectorization processing, etc. An agent is assigned to one vectorized data component. When the agent encounters a problem location when recognizing the corresponding vectorized data, the vectorized data of the problem location is interpolated with reference to the adjacent vectorized data, and the series of vectorized data obtained is integrated. Are recognized as two graphic data and registered as second vectorized data. Further, it is determined whether the agent assignment is correct based on the basic characteristics of the line figure such as the adjacency so as to correspond one-to-one with the figure data recognized by the agent assigned to the first vectorized data, If it is wrong, the agent assignment is changed by integrating or dividing the agent so that the contradiction corresponding to the basic characteristics of the line figure is resolved. Further, at the time of initial assignment of agents, line routes of the entire drawing may be recognized based on local connectivity, and agents may be assigned to each recognized line route. The line figure to be recognized may be a contour line extracted from the image.
[0011]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described using an example of recognizing contour lines in a topographic map.
[0012]
(1) Overall system
FIG. 1 is a diagram showing a system configuration for recognizing a line figure according to the present invention. Reference numeral 1 denotes a central processing unit that controls the system such as line recognition and vectorization. Reference numeral 2 denotes an image input device that captures a map / drawing as binary image data. An example is a scanner. Reference numeral 3 denotes a first file device for storing binary image data obtained by a scanner. Reference numeral 4 denotes a second file device for storing vectorized data obtained by pre-processing binary image data stored in the first file device. The preprocessing performed here refers to a process of extracting two or more feature points (bending points and end points) representing binary image data, obtaining a position map of those points, and sequentially vectorizing them. Reference numeral 5 denotes a display device that displays line images, line figures, and the like based on binary image data captured by a scanner, vector data obtained by preprocessing, and vector data after recognition processing. Reference numeral 6 denotes an input device for inputting instructions for operating the system. Reference numeral 7 denotes a memory having a program describing preprocessing, recognition processing, and the like executed by the central processing unit 1, and a storage area used in the processing. It does not matter whether the file devices (1) and (2) and the memory 7 are provided inside the system or provided externally. Further, the file devices (1) and (2) and the memory 7 may be integrated with one large memory and distinguished by dividing the used area.
[0013]
An overview of the processing operation of the entire system in FIG. 1 will be described with reference to FIG. FIG. 2 is a PAD diagram showing an outline of processing of the entire system.
[0014]
First, the topographic map to be processed is read by the scanner 2 and stored in the file device (1) 3 as binary image data (step 401).
[0015]
The original topographic map data stored in the file device (1) 3 is first subjected to vectorization processing (preprocessing program; step 402) as preprocessing, and the input topographic map vector including the recognition target. To obtain data. In this embodiment, contoured vectorized data is obtained. One recognition agent is assigned to one continuous contour line obtained (initial assignment program; step 404), and contour lines are recognized. Until all of the assigned recognition agents have been processed, the conflict resolution processing and cooperative recognition processing between recognition agents are repeated (recognition program; steps 405 to 407). When the contour line recognition processing is completed, the result is stored as a final result in a storage device such as the file device (2) 4 (registration program). If necessary, display on the display device. Next, each process will be described.
[0016]
(2) Vectorization processing (preprocessing program; step 402)
FIG. 3 is a screen display example for explaining the processing of this embodiment. The screen display 300 is a diagram showing the vectorized data after the vectorization processing is performed on the original image data of the topographic map read from the file device (1) 3.
[0017]
In the present invention, graphic processing such as thinning processing by polygonal line approximation and beard removal may be performed after vectorization processing. Further, in order to avoid the deformation of the figure accompanying the vectorization process, image processing such as separation of noise, characters, symbols, etc. may be performed before the vectorization process.
[0018]
(3) Agent assignment to recognition symmetrical figure (initial assignment program; steps 403, 404)
(3-1) Extraction of recognition target area (contour component)
The screen display 300 is a display of binary image data obtained by reading a general topographic map. Such a topographic map includes not only the contour line 301 but also various components such as the house frame 102, the road 103, the altitude value, and the characters 104. In order to extract a line figure as a figure to be processed from the screen display 300, the target area is directly designated by the input device 6 (310). Reference numeral 311 denotes an area designated by the user using the mouse. Further, components other than the line figure such as the house frame 102, the road 103, the altitude value and the characters 104 may be designated and deleted from the pointing device 6, and the remaining display may be designated as the recognition target area (320).
[0019]
Of course, if the read image is composed only of figures to be recognized, or if the image processing that separates the characters / symbols described in (2) above has already been performed, it is simply processed. The target area may be specified.
[0020]
Although the user may designate the recognition target from the outside as in the above method, a method of automatically extracting the recognition target graphic as shown below can be used. This will be described with reference to FIGS.
[0021]
FIG. 4 is a PAD diagram showing a flow of automatic contour line component extraction processing, and FIG. 5 is a diagram showing a table structure used in this method. Reference numeral 500 denotes a vector data table corresponding to the entire drawing, in which information relating to the coordinates of the start point and end point of the vector obtained by the thinning vectorization process (step 402) is stored. First, a list label table corresponding to 500 is created and a label is initialized to 0 (step 400). Reference numeral 510 denotes a list label table, where the list number represents the vector number stored in 500, and the label represents flag information indicating whether or not the recognition of the corresponding vector has been completed. Therefore, for all vectors corresponding to the drawing (step 410), it is determined whether the label of the corresponding list is 0 (step 420). If the label is 0, the line route is recognized using the vector as a starting point, and a label is assigned to the corresponding list (step 430). For example, the number of the recognized line route is given as a label. Reference numeral 520 denotes a recognized line path data table, in which information regarding the number of constituent points and constituent point coordinates of the line path is stored. Thus, the feature amount of the recognized line route is obtained, the line route is classified by clustering in the feature space, and stored in the line route information table (step 440). Reference numeral 530 denotes a line route information table. The label represents the number of the line route stored in 520, and the line route information represents the classification in the feature amount space. For example, 0 is assigned if the line path does not correspond to the recognition target, and 1 is assigned if it corresponds.
[0022]
Next, a line path recognition method in the contour line component extraction method will be described with reference to FIGS. FIG. 6 is a PAD diagram showing the flow of the line path recognition process, and FIG. 7 is an explanatory diagram regarding the search area setting method and vector determination method in this method. First, a circular area is set as a search area (step 611), and a vicinity area of the search base point (the end point of the confirmed route) is searched. Reference numeral 700 is an explanatory diagram regarding the setting of this circular area. Here, a circular area centered on the search base point is set, and a vector having end points in this area is searched. However, vectors that have already been recognized are excluded from search targets. Here, the radius of the circle is set to a value that is sufficiently smaller than the distance between adjacent contour lines. The condition whether or not a vector has an end point in this circle region is expressed by the conditional expression 701. If the vector does not exist in the circular area (step 612), an elliptical area is set as the search area (step 613), and the search area is enlarged and searched again. Reference numeral 710 is an explanatory diagram regarding the method of setting the elliptical area. Here, an ellipse area having two focal points, ie, a search base point (end point of the definite route) and a point on the extension line of the determinate route, is set, and a vector having end points in this region is searched. The condition for whether the vector has endpoints in this region is represented by 711. If the vector does not exist within the elliptical region (step 614), the line path recognition is terminated (step 615). If the vector exists, the vector is determined using the smoothness of the line when the vectors are connected as an evaluation function (step 616). 720 is an explanatory diagram regarding a vector determination method based on the smoothness of a line. Reference numeral 721 represents the definition of the evaluation function and the condition for determining the vector. When two or more vectors exist as candidates, the next vector cannot be determined if both vectors are smoothly connected. Therefore, if the evaluation function is not more than the specified range and the difference between the evaluation functions is not less than the specified range, it is determined as the next vector. If the vector is not determined (step 617), the recognition of the line path is terminated (618). The process from the setting of the search area to the determination of the vector is repeated until the recognition of the line path is completed (step 610). Reference numeral 730 denotes a list of parameters used in line path recognition.
[0023]
Next, a line path classification method in the contour line component extraction method will be described with reference to FIG. 800 is a diagram showing an example of a recognized line route. The recognized line path is composed of a vector portion in which vectorized data exists and a gap portion in which no vectorized data exists. By the way, the contour line figure is displayed by 1) winding and 2) a solid line. 3) It has a feature that it is a certain length or longer. Therefore, if the clustering process in the feature space is used, it is possible to extract a line route corresponding to the contour line figure. Reference numeral 810 represents a feature quantity corresponding to the basic characteristic of winding and a feature quantity condition for satisfying this basic characteristic. The feature quantity of this basic characteristic is represented by the distance / path length between the start and end points. If this feature quantity is below the specified range, it is determined that the winding is winding. Reference numeral 820 represents a feature quantity corresponding to the basic characteristic displayed with a solid line and a feature quantity condition for satisfying the basic characteristic. The feature quantity of this basic characteristic is represented by the sum of vector lengths / path length. If this feature quantity is greater than or equal to the specified range, it is determined that it is displayed as a solid line. Reference numeral 830 represents a feature quantity corresponding to the basic characteristic of a certain length or more and a condition of the feature quantity for satisfying this basic characteristic. The feature quantity of this basic characteristic is represented by the path length.
[0024]
Next, a parameter setting method for extracting contour line components will be described with reference to FIG. When a line route classification method based on feature values is used as a contour line component extraction method, it is necessary to set a feature value range as a parameter. FIG. 9 is a diagram showing an interface for setting such parameters. 900 is a parameter setting screen, 910 is a window for displaying the color path corresponding to the feature amount range designated by the operator by color, and 920 is for setting the feature amount range of the line route by the operator. It is a window. The parameter setting in 920 is performed by extending or contracting a bar representing the feature amount range by operating the mouse, or inputting the feature amount range from the keyboard. If the parameters are set in this way, a line path corresponding to the feature amount range is displayed in the window 910. When the parameter is updated, the corresponding line route is displayed again.
[0025]
(3-2) Agent assignment to recognition target figure (step 404)
Next, an agent is assigned to the recognized symmetrical figure by setting the area.
[0026]
330 in FIG. 3 shows how initial values are set for parallel search. The initial value is specified as follows. The operator uses the input device 6 to draw a search auxiliary line 331 so as to cross a plurality of adjacent contour lines. An intersection of the search auxiliary line 331 and a plurality of contour lines to be recognized is set as a search reference point 332. A plurality of designated search reference points 352 are defined as B1, B2,. The order of the search reference points 332 on the search auxiliary line 331 represents the adjacent relationship of the corresponding contour lines. Then, a contour route search is started in parallel with each search reference point 332 as an initial feature point.
[0027]
A method of automatically assigning agents to each contour line and automatically grasping the adjacency relationship between agents will be described.
[0028]
Next, an initial assigning method of recognition agents in the line figure recognition method according to the present invention will be described with reference to FIG. It is assumed that the contour line component has already been extracted, and that the contour line component has been recognized and registered as a line path according to a local criterion. FIG. 10 is a diagram illustrating initial allocation of recognition agents. 1000 is a diagram showing the correspondence between the line path recognized by the local criterion and the recognition agent. Here, the line path is divided at a cutting point or a branch point. A recognition agent is assigned to each of such line paths one by one. 1010 represents recognition agent data and methods. This recognition agent has line route data (number of points and point coordinates) as data, operations to generate and disappear recognition agents, add point to line route data, and point from line route data As a method, there are an operation for deleting the line route, a request for line route data from another recognition agent, a response to the line route data in response to this request, and the like.
[0029]
Next, a method for the recognition agent to recognize the environment when a problem occurs in the line figure recognition method according to the present invention will be described with reference to FIGS. In order to solve the problem of obtaining the connection relationship between line paths at the cut and joints with the cooperation of multiple recognition agents, it is necessary to grasp the adjacent relationship between the recognition agents as the environment where the recognition agents are placed. . 11 is a PAD diagram showing the flow of processing in the adjacency recognition method, FIG. 12 is a diagram showing a Voronoi diagram used to grasp the adjacency relationship, and FIG. 13 is a setting of a management agent that controls the recognition agent. It is a figure which shows a method. For example, the adjacency relationship between recognition agents is obtained using a Voronoi diagram in the same manner as in Document 2. First, a Voronoi diagram composed of bisectors between line paths is created (step 1100), and an adjacency graph representing an adjacency network is created (step 1110). 1200 represents the Voronoi diagram and 1310 represents the corresponding adjacency graph. Here, a solid line represents a case where the length of a boundary line (Voronoi diagram) between adjacent line paths is longer than a corresponding line path, and a broken line represents a short case. A solid line is called a strong adjacency because it is likely to represent an adjacency between contour lines, and a broken line is called a weak adjacency because the possibility is low. In addition, based on the basic characteristics of contour figures that form a closed loop by themselves or that the end points exist on the page boundary line, if this characteristic is satisfied, recognition ends, and if it is not satisfied, a problem occurs. It is displayed separately. A page is an area serving as a processing unit. For all recognized agents (step 1120), if the recognized agent has a problem (step 1130) and is not registered with any management agent (step 1140), a management agent is generated (step 1150). ), The recognition agent belonging to this and its adjacency are registered (step 1160). 1320 is a diagram showing the relationship between the management agent and the corresponding recognition agent.
[0030]
(4) Contour line recognition processing (recognition program; steps 405 to 407)
Once the adjacency relationship between agents assigned to each contour line is clarified, next, recognition processing as a meaningful contour line is performed using the adjacency relationship as a clue. Even if the user uses the auxiliary line to initially set the adjacency relationship as shown in 330 in FIG. 3, the road cannot be separated by noise removal or by accidentally drawing the auxiliary line overlapping one contour line. In some cases, auxiliary lines are drawn on figures that are not recognized. In addition, when an agent is automatically set, a plurality of agents are assigned to one contour line, or only one agent is assigned to a plurality of contour lines, because of cutting, branching, or joining of contour lines. Sometimes.
[0031]
Contour recognition processing is performed in cooperation with agents assigned to other contour lines that have already been confirmed, as well as conflict resolution processing that integrates and divides agents in order to make one-to-one correspondence between contour lines of a specific height and agents. And cooperative recognition processing for recognizing the contour line.
[0032]
(4-1) Conflict resolution processing
The conflict resolution method between recognition agents in the present invention will be described with reference to FIG. 14, FIG. 15 and FIG. In this embodiment, conflicts between recognition agents are resolved by integrating or dividing recognition agents.
[0033]
FIG. 14 is a diagram showing a method for integrating recognition agents at a problem location such as a cutting section. Reference numeral 1400 represents an initial allocation of recognition agents in the cutting unit. Here, two recognition agents (2 and 4) are assigned to one contour line. Reference numeral 1410 denotes a management agent and a recognition agent corresponding to the disconnection unit. In this case, a weak adjacency exists between the recognition agents (2 and 4). Therefore, if such an adjacency exists, the management agent integrates these two recognition agents to generate a new recognition agent. 1420 represents the integration of the recognition agents by the management agent, and 1430 represents the adjacency relationship of the new agent.
[0034]
FIG. 15 is a diagram showing a method for integrating recognition agents at a problem location such as a joint. 1500 represents the initial assignment of recognition agents at the junction. Here, three recognition agents (2, 5, 6 or 3, 5, 7) are assigned to one contour line. Reference numeral 1510 denotes a management agent and a recognition agent corresponding to the joint. In this case, a weak adjacency exists between the recognition agents (2 and 5, 3 and 5, 5 and 6, and 5 and 7). Therefore, if such an adjacency exists, the management agent integrates these two recognition agents to generate a new recognition agent. 1520 represents the integration of the recognition agents by the management agent, and 1530 represents the adjacency relationship of the new agent.
[0035]
FIG. 16 is a view showing a recognition agent dividing method in a problem part such as a joint. 1600 represents the initial assignment of recognition agents at the junction. Here, one recognition agent (3) is assigned to three contour lines. Reference numeral 1610 denotes a management agent and a recognition agent corresponding to the joint. In this case, there is a contradiction between 1 and 2 and 3 or between 3 and 4 and 5. When such a cooperative network exists, there is a high possibility that the three recognition agents common to both are wrong. Therefore, as shown in 1620, the management agent divides the recognition agent into three. Then, three new recognition agents are generated. 1630 represents the newly generated recognition agent assignment, and 1640 represents the adjacency relationship of the new agent.
[0036]
(4-2) Cooperative recognition processing for path interpolation
FIG. 17 is a diagram showing a message flow between recognition agents according to this method, and FIG. 18 is a diagram showing a route interpolation method based on this method. FIG. 17 shows a message flow between the management agent and the recognition agent corresponding to the disconnection unit in FIG. FIG. 18 shows a schematic diagram of a path interpolation method in such a cutting section. The adjacency relationship between recognition agents is already known. Using this adjacency relationship, the recognition target path is interpolated.
[0037]
In the present invention, when the path interpolation is performed, if a problem location such as the joint portion 102 or the dividing portion 103 is reached, the search direction is determined based on the already determined adjacent contour lines.
[0038]
Hereinafter, as shown by 330 in FIG. 3, the cooperative processing will be described on the assumption that an agent is assigned to a contour line by setting the auxiliary line 331. In the case of 330 in FIG. 3, the search reference point of the agent is set by the position where the auxiliary line is input. On the other hand, as shown in FIG. 10, when agents are automatically assigned to contour line components, each agent may set an arbitrary position as a search reference point.
[0039]
340 is a diagram showing the vectorized data after completing the route search with 330 as an initial value. 341 represents a contour line recognized by the route search, 342 is a rounding point, and 343 is a surrounding point. The round point represents the point where the agent returns to the search reference point when the contour line itself forms a closed loop, and is the end point of the search. On the other hand, the surrounding points are located on the page boundary, and when the search reaches this point, the search returns to the search reference point and the search is performed in the reverse direction. When the search in the reverse direction reaches the surrounding point again, this point is set as the end point of the search. The processing from the setting of the search reference point to the feature point search as described above is repeated for the entire contour line figure, and a wide range of contour line recognition results are obtained.
[0040]
Next, the relationship between the agent performing the recognition process and the data referred to by the agent will be described with reference to FIG. In the present invention, a system is configured by multi-agents, and the processing is formulated by describing the autonomous behavior of each agent and the cooperative operation between agents. In contrast to the conventional procedure-oriented approach, it was necessary to describe all processing flows without omission, whereas in the agent-oriented approach, the cooperative behavior between agents is determined dynamically during program execution. Does not require a detailed description of.
[0041]
Reference numeral 1900 denotes a management agent that manages all agents assigned to the contour lines, 1910 denotes a recognition agent that controls path recognition of each contour line, and 1920 denotes a shared data portion. In the shared data section, input information such as contour image data and vectorized data before recognition processing, index information for performing vector search at high speed, label information attached to the vector after the search, and feature point position coordinates of the contour line The recognition result represented by is stored. 1930 indicates message communication between agents, and 1940 indicates data reference. FIG. 20 shows the function definitions and preconditions for the autonomous behavior of these agents, and FIG. 21 summarizes the specific processing contents for the cooperative operation between agents. 2001 and 2002 represent the function definition and preconditions of the agent that recognizes the line figure (recognition agent), respectively, and 2003 and 2004 represent the function definition and preconditions of the management agent, respectively. Further, 2101 and 2102 in FIG. 21 represent cooperative operations between the management agent and the recognition agent at the joint and the dividing part, respectively.
[0042]
Next, a route search method in the autonomous behavior of the recognition agent will be described with reference to FIG. Suppose now that the i-th contour line is being searched. Reference numeral 2200 denotes vectorized data near the search area. Reference numeral 2210 denotes a confirmed vector for which the search has already been completed, and reference numerals 2220, 2230, and 2240 also denote undefined vectors for which the search has not been completed. The end point 2211 of the vector 2210 represents the current point of the route search and indicates that the search has been completed up to this point. 2221, 2231 and 2241 represent the end points of 2220, 2230 and 2240, respectively. Reference numeral 2250 denotes a search area for searching for the next route candidate. In this figure, 2220 and 2230 are selected as route candidates. If there is a confirmed vector of adjacent contour lines in this search area, a route candidate is selected so as not to cause a contradiction such as an intersection with an adjacent route.
[0043]
When there are a plurality of route candidates, the optimum route candidate is selected based on the adjacent information as shown in FIGS. First, FIG. 23 shows a procedure for configuring a reference line for determining a route candidate based on adjacent information. Reference numeral 2300 denotes that a plurality of vectors as route candidates exist in the search area of the i-th contour line. Here, it is assumed that the route search for the (i + 1) th contour line has already been completed. As indicated by 2310, the shortest distance d from the current point is obtained for the (i + 1) th contour line already determined. Then, as indicated by 2320, the component vector of the (i + 1) th contour line is translated by the distance d, and this is used as the reference line. Next, FIG. 24 shows a method for selecting an optimal route candidate based on this reference line. Reference numeral 2400 denotes the positional relationship between the reference line 2410 and the vector 2430. As a route candidate, a vector whose distance and direction are closest to this reference line should be selected. Therefore, an evaluation function is defined as in 2420, and a vector that minimizes this evaluation function is selected as the next path candidate. Here, d1 and d2 represent the shortest distances from the end points of the vector 2430 to the reference line, respectively. Whether or not to confirm the selected route candidate vector as a route is determined using the determination method shown in FIG.
[0044]
A method for predicting a route based on the route information of the contour lines adjacent to each other at the junction or division will be described with reference to FIG. FIG. 26 shows a prediction method in the divided portion, but the route can be predicted by the same process in the joint portion. Reference numeral 2600 denotes that there is no vector as a route candidate in the search area. Here, it is assumed that the route search for the (i + 1) th contour line has already been completed. As shown in 2610, the shortest distance d from the current point is obtained for the (i + 1) th contour line already determined. Then, as indicated by 2620, the constituent vectors of the (i + 1) th contour line are translated one after another by the distance d, and this is used as the predicted path. If the length of the predicted route is within the specified range, the prediction is continued until a route candidate vector is found.
[0045]
(5) Registration to storage device (registration program)
Data obtained by the above processing is registered in the recording apparatus as continuous vectorized data indicating the drawing input by the scanner 2. The recording device may be the file devices (1) and (2) and the memory 7 shown in the system configuration diagram of FIG. 1, or other external storage medium (not shown).
[0046]
【The invention's effect】
According to the present invention, when registering input graphic data as vectorized data, by using the adjacency relationship of agents assigned to the graphic to be recognized, the competition / cooperation processing between agents is performed. In the problem area where the line that should be continuous is discontinuous or the independent line is joined to another line (the figure is joined to the other figure), the excessive interaction is reduced and uncertain. The automatic processing capacity of the part is improved. As a result, it is possible to greatly reduce the input cost in creating a database of figures represented by lines.
[0047]
Moreover, even if the initial recognition agent assignment is wrong, it is possible to automatically change the agent assignment so that the recognition agent is set correctly, which suppresses unnecessary interaction operations and is used by the user. An easy-to-use system can be provided.
[Brief description of the drawings]
FIG. 1 is a diagram showing a system configuration of the present invention.
FIG. 2 is a PAD diagram showing the overall processing flow of the present invention.
FIG. 3 is a diagram showing a display screen for processing according to the present invention.
FIG. 4 is a PAD showing automatic contour line component extraction processing;
FIG. 5 is a diagram showing the structure of a table used in the automatic contour line component extraction process;
FIG. 6 is a PAD diagram showing line path recognition processing;
FIG. 7 is a diagram for explaining a search area setting method and a vector determination method;
FIG. 8 is a diagram for explaining a line path classification method;
FIG. 9 is a screen display when setting parameters in the automatic contour line component extraction processing;
FIG. 10 is a diagram showing a method for initially assigning recognition agents.
FIG. 11 is a PAD diagram of processing for recognizing adjacency relationships of recognition agents.
FIG. 12 is a Voronoi diagram used to grasp the adjacency relationship.
FIG. 13 is a diagram for explaining a management agent setting method;
FIG. 14 is a diagram for explaining recognition agent integration processing in a cutting unit;
FIG. 15 is a diagram for explaining recognition agent integration processing at a joint;
FIG. 16 is a diagram for explaining recognition agent division processing in a joint portion;
FIG. 17 is a diagram illustrating a message exchange relationship between recognition agents in cooperative recognition.
FIG. 18 is a diagram illustrating a path interpolation process of a cutting unit in cooperative recognition.
FIG. 19 is a diagram illustrating a relationship between a recognition agent, a management agent, and data.
FIG. 20 is a diagram showing a function definition and preconditions regarding an autonomous behavior of an agent.
FIG. 21 is a diagram illustrating specific processing contents regarding cooperative operation between agents;
FIG. 22 is a diagram illustrating route search processing in the autonomous behavior of an agent.
FIG. 23 is a diagram illustrating a procedure for configuring a reference line for determining a route candidate based on adjacent information.
FIG. 24 is a diagram illustrating processing for selecting an optimal route candidate based on a reference line.
FIG. 25 is a diagram illustrating a route determination process for a selected route candidate vector.
FIG. 26 is a diagram for explaining a method of predicting a route based on route information of adjacent contour lines in a dividing part.
[Explanation of symbols]
1 ... b central processing unit 2 ... image input device 3 ... first file device 4 ... second file device 5 ... display device 6 ... input device 7 ... memory

Claims (8)

Capture and display image data with multiple line figures on the display screen,
An auxiliary line that intersects with the plurality of line figures is received via the input means,
Set the intersection of the auxiliary line and each of the plurality of line figures as a search reference point,
Recognizing the adjacency relationship of the line figure set with the search reference point from the order of the plurality of set search reference points,
A line figure recognition method characterized by assigning a recognition agent to each line figure for which the search reference point is set, and searching for a plurality of line figures assigned to the recognition agent in parallel using the adjacency relationship. . 2. The method for recognizing a line figure according to claim 1, wherein the image data is image data of a contour figure, and the line figure is a contour line. 3. The line according to claim 2, wherein the search is performed by predicting a route using the adjacent information and route information of a line figure that has already been established at a joint portion or a dividing portion of the line image. How to recognize shapes. Extract line path from the above image data,
Calculate the value of the ratio between the distance between the start and end points of the line route and the route length, the route length of the line route, the sum of the vector length of the line route and the ratio of the route length,
The respective values for the line path selected is compared with a predetermined value, a method of recognizing a line drawing according to any one of claims 1 to 3, characterized in that the setting of the search reference point. A display device for displaying image data having a plurality of line figures;
An input device for receiving an input of an auxiliary line intersecting with the plurality of displayed line figures;
The intersection of the auxiliary line and each of the plurality of line figures is set as a search reference point, and the adjacency relationship of the line figure set by the search reference point is recognized from the order of the plurality of set search reference points,
A control unit that assigns a recognition agent to each of the line figures for which the search reference point is set, and searches for a plurality of line figures to which the recognition agent is assigned in parallel using the adjacency relationship. Line figure recognition device. 6. The line figure recognition apparatus according to claim 5 , wherein the image data is image data of a contour figure, and the line figure is a contour line. The said control part performs the said search by predicting a path | route using the path | route information of the line figure already determined with the said adjacent relationship in the junction part or division | segmentation part of the said line image, It is characterized by the above-mentioned. The line figure recognition apparatus according to 5 or 6 . The control unit extracts a line path from the image data,
Calculate the value of the ratio between the distance between the start and end points of the line route and the route length, the route length of the line route, the sum of the vector length of the line route and the ratio of the route length,
For comparison to selected the ray path of the respective values to a predetermined value, a line drawing of the recognition device according to any one of claims 5 to 7, characterized in that the setting of the search reference point.

Images