JPH04238585A - Drawing interpretation processing system - Google Patents

Abstract

PURPOSE:To perform consistent automation from the identification of a graphic to the interpretation of the whole picture by performing collation with a scene dictionary in which relation between the graphic and a scene is stored after separating the superposition of the graphic in an image. CONSTITUTION:A graphic separation part 12 which separates the superposition when plural graphics are superimposed in a fetched image in a direction to connect them at an intersection smoothly by tracing a line segment is provided. Also, a graphic identification part 13 which identifies separated graphics, and a drawing interpretation part 14 which interprets a content represented in a drawing by collating it with the scene dictionary in which the relation between the graphic and the scene is stored based on the identified result of the graphic are provided. By employing such constitution, a binary image can be obtained from a drawing input part 11, and it is separated at the graphic separation part 12 when those binary images are superimposed. The graphic is identified at the graphic identification part 13, and the symbol of each graphic can be decided. The drawing interpretation part 14 interprets the whole drawing based on an obtained symbol.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to automation of drawing processing, and relates to a drawing interpretation processing method that automates not only automatic identification of drawn figures but also all processing up to interpretation of the content of drawings. There are various types of drawings, such as maps, blueprints, and graphic drawings, and the need to interpret them arises in many situations, such as sorting, diagnosis, and evaluation. Currently, most of this work is done manually. The present invention can be expected to save labor.

0002

[Prior Art] Conventionally, elements in drawings (single figures and characters)
Various methods have been proposed for automatic identification of drawings, but no method has been proposed for interpreting the contents of the entire drawing from these elements. Furthermore, the methods that have been proposed as scene interpretation processing methods (Japanese Patent Application No. 1-130185, Japanese Patent Application No. 1-191035) can be applied to the final stage of drawing interpretation; Automation was not considered.

0003

In the prior art, only the automatic identification of figures in drawings was performed, and the final interpretation of the drawings was done manually. In order to achieve further labor savings, consistent automation, including interpretation, is necessary.

[0004] An object of the present invention is to achieve consistent automation ranging from automatic identification of figures in drawings to interpretation of the contents of the entire drawing.

[0005]

[Means for Solving the Problems] FIG. 1 shows a diagram of the basic configuration of the present invention. 11 is a drawing input unit that imports the target drawing as a binary image, and 12 is a drawing input unit that allows humans to perceive figures in the imported image by tracing line segments in directions that connect smoothly at intersections. A figure separator 13 separates the figures in the same way, and a figure identifier 13 identifies the separated figures invariably against changes in rotation and size by a combination of complex logarithmic transformation and frequency plane conversion. Reference numeral 14 denotes a drawing interpretation unit that interprets the contents of the drawing by checking it with a scene dictionary that stores the relationship between the drawing and the scene based on the identification result of the drawing. This configuration makes it possible to fully automate drawing interpretation, which is the problem mentioned above.

[0006]

[Operation] A binary image is obtained from the drawing input section 11, and the
If the value images overlap, they are separated in the graphic separation unit 12. Then, the figure identification section 13 identifies each of the separated figures and determines the symbol of each figure. The drawing interpretation unit 14 interprets the entire drawing based on the obtained symbols.

[0007]

Embodiment FIGS. 2 to 5 show examples of processing results in each part of FIG. 1. This shows the effect of the present invention. FIG. 2 shows the processing results of the drawing input section 11. Although the input drawing is generally a multivalued image, it is converted into a binary image through processing by the drawing input unit 11. FIG. 3 shows the processing results of the figure separation unit 12. The overlapping figures in FIG. 2 are separated by the process of the figure separation unit 12. FIG. 4 shows the processing results of the figure identification section 13. Each figure in Figure 3 is
Through the process in step 3, it is converted into a symbol. FIG. 5 shows the processing results of the drawing interpretation section 14. Each symbol in FIG. 4 is activated by the process of the drawing interpretation unit 14, and the interpretation "drawing of the study room" is output, which has the maximum sum of the activation values.

The drawing input section 11 shown in FIG. 1 can be easily realized using a commercially available image scanner and appropriate binarization processing.

The figure separator 12 can be realized by the method shown in FIG. 6 when each figure is a simple closed loop (single closed curve with no intersections). That is, first, in line thinning 31, 8-connected line thinning is performed on the input line image. Next, in labeling 32, labeling is performed to separate connected figures. Next, in intersection detection 33, intersections are extracted for each connected figure. An intersection can be detected as a point where the number of 8 connections is "4". If there is no intersection, it is assumed that a simple closed loop has already been established, and processing moves on to the next connected figure. If there is an intersection, it is determined that it is a complex loop, and processing 34 is performed. In simple closed loop extraction 34, simple closed loops are extracted by tracing line segments in directions that connect smoothly at intersections. When processes 33 and 34 are completed for all connected figures, the figures are completely separated by relabeling process 35. Note that the ``law of smooth continuity'' described in Gestalt psychology is followed here by tracing line segments in directions that connect them smoothly at intersections. In other words, figure separation is performed in the same way as humans perceive it. Further, a more detailed method for figure separation similar to that perceived by humans is proposed in Reference (1), and can be used.

[0010] The figure identification section 13 in FIG.
This can be realized by applying the method shown in No. 1232 (Pattern Identification Method). FIG. 7 shows the flow of the process. First, in figure centroid calculation 41, the centroid of the target figure is determined. Next, in complex logarithm mapping 42, the coordinate system expressing the target figure is converted from an orthogonal system to a logarithmic-polar coordinate system centered on the center of gravity. Through this processing, the rotation or size change of a figure is converted into a change in position. Next, by R transformation 43, the signal is converted into a frequency plane so as to be invariant to changes in position. Here, the R transform is used because it is easier to process than the Fourier transform. The details are shown in document (2). Finally, in the comparison 44 with the figure dictionary, identification of the target figure is completed by comparing it with the figure dictionary. The figure dictionary is a collection of all possible figures that have been subjected to centroid extraction, complex logarithm mapping around the centroid, and R conversion.

The drawing interpretation section 14 in FIG.
This can be realized by applying the method shown in No. 0185 (Scene Interpretation Processing Method). Figure 8 shows the process flow.
Shown below. First, in scene dictionary creation 51, a scene dictionary is created. The scene dictionary stores numerical values (characteristic values) representing the strength of the relationship between an assumed scene (drawing content) and each figure in the figure dictionary. Once this is created, there is no need to create it later. However, patent application No. 1-1910
It is also possible to sequentially update (learn) the dictionary by applying the method shown in No. 35 (Scene Interpretation Processing Method).

Next, in addition 52 of characteristic values, characteristic values in the scene dictionary corresponding to the figure name which is the output result of the figure identification section 13 are added for each scene. Finally, scene judgment 5
3, the scene showing the maximum value among the added values in the characteristic value addition step 52 is determined to be the content of the input drawing and is output.

References (1) Akira Shimatani, Isamu Yorizawa: Quantitative Analysis of Figure Segmentation Characteristics, Television Science and Technology Report, VAI90-18 (1990) (2) Re
itboeck H. J. and Altmann
J. :A model for size and
rotation-invariant patter
n processing in the visual
l system, Biol. Cyber. ,Vo
l. 51, pp. 113-121 (1984)

[0014]

[Effects of the Invention] According to the present invention, it is possible to automate the entire process from drawing input to drawing interpretation. By applying the present invention, tasks such as sorting, diagnosing, and evaluating a large number of drawings can be easily performed, resulting in significant labor savings.

Note that the present invention is applicable not only to drawings originally created as binary images, but also to natural images with shading, as long as a highly accurate contour image can be obtained. Its effectiveness is enormous.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the principle configuration of the present invention.

FIG. 2 shows processing results of a drawing input unit.

FIG. 3 shows processing results of a figure separation unit.

FIG. 4 shows processing results of a figure identification unit.

FIG. 5 shows processing results of the drawing interpretation section.

FIG. 6 shows the flow of processing of a figure separation unit.

FIG. 7 shows the flow of processing of a figure identification unit.

FIG. 8 shows the flow of processing of the drawing interpretation section.

[Explanation of symbols]

11 Drawing input section 12 Figure separation section 13 Shape identification section 14 Drawing interpretation department

Claims (1)

[Claims] Claim 1: In a drawing interpretation processing method that identifies figures drawn in a drawing and interprets the contents of the drawing based on the identification result, a drawing input unit and a plurality of figures in the captured image are used. If there is an overlap, the overlap is traced in the direction of smooth connection at the intersection, and the shape separation part separates the shape, the shape identification part identifies the separated shapes, and the shape identification result is calculated. 1. A drawing interpretation processing method, comprising: a drawing interpretation unit that interprets the contents of a drawing by comparing it with a scene dictionary that stores relationships between figures and scenes based on the drawings.