JPH02288978A - Drawing processing method - Google Patents

Abstract

PURPOSE:To exactly segment the part of a complicated symbol at high speed by wholly processing segmentation by a vector by using the formation of a contour to a vector and core line. CONSTITUTION:A loop LA including an unprocessed core line AO (to which a processed flag is not added) is searched. When a length between the both end points of the loop LA is less than a threshold value, the loop of satisfactory straight advance ability in contact with the loop LA is searched concerning right and left sides. The both end points of a found-out whole loop group having the satisfactory straight advance ability are calculated and a perpendicular line falls down from the end points of the respective core line vectors constituting the respective loops to a virtual straight line connecting the both end points and the respective end points and a length to feet are calculated. At such a time, when the whole perpendicular lines are less than the width of the line, it is judged that a core line group corresponding to the loop group is a straight line. Then, the contour vector corresponding to the core line defined as the straight line is removed from the contour vector and the circumscribed rectangle of a group in the remained contour vector not to be removed is calculated. Then, the range of this circumscribed rectangle is defined as a symbol area. Thus, the complicated symbol area can be segmented.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to drawing processing, and particularly to segmentation of handwritten drawings in drawing recognition systems and the like.

[Prior art and problems to be solved by the invention] In conventional drawing recognition systems, the binary image of the drawing is thinned, and the result is used to recognize symbols and vectorize line segments. However, since the same processing is performed on parts that require detailed processing, such as symbols, and parts that require rough processing, such as line segments, processing time is required, and the recognition accuracy of symbols, etc. is not sufficient.

Regarding this type of system, please refer to the paper "Drawing reading device TO8GRARHJ (Yuyo et al.; Information Processing Society of Japan, 33rd Gold Country Conference, pp. 2353-2354)" and the paper [
Mechanical drawing automatic recognition system “ARCADIA-M”
(Nichiban et al.; Information Processing Society of Japan 33rd Gold Country Conference, pp. 1
489-1498) etc.

An object of the present invention is to provide a drawing processing method that can speed up the segmentation of handwritten drawings and the like and improve the accuracy of cutting out regions of fairly complex symbols.

[Means to solve the problem]

The present invention extracts a contour vector by tracing black pixels of a binary image of one drawing, extracts a skeleton using a pair of extracted contour vectors 1 to 1, and extracts a skeleton as a reliable line segment or straight line from the extracted skeleton. The method is characterized in that a region of clusters of contour vectors corresponding to the remaining core lines is cut out as a symbol region.

[For production]

Segmentation of drawings is performed by solid-level processing, which enables high-speed processing, and it is also possible to accurately cut out regions of fairly complex symbols such as arrows, small circles, and double circles.

Note that if you perform more detailed processing on only the symbol area cut out in this way than on parts such as line segments,
It is possible to improve the symbol recognition rate and speed up the entire drawing recognition process.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows an example of a hardware configuration for implementing the present invention. '' is an image memory for storing binary image data such as mechanical drawings, and the binary image data is inputted from an external scanner or the like via the human output interface circuit 2. 3 is a program memory that stores programs such as drawing recognition programs. In this block 11,
Reference numerals 5 and 4 are CPUs that execute programs for various processes such as contour conversion, core line extraction, and segmentation. A data memory 5 temporarily stores data obtained through processing.

The flow of processing directly related to the present invention is shown in FIG. First, the binary image data of the drawing is manually generated and stored in the image memory (step 1). Next, a contour vector is extracted (generated) by tracing the black pixels of the input binary image (step 2).

From the data of this contour vector, the center line of the two contour vectors is extracted as the skeleton line (step
).

Regarding the generation of contour planes and the extraction of core lines, please refer to the paper "Drawing processing using multidimensional data structure" (Osawa et al.: Transactions of the Institute of Electronics and Communication Engineers '85/4 V
ol,, J 68-D Na4. , pp, 84585
2 and the paper “Improvement of core line generation method based on contour lines” (Haka Osawa; Information Processing Society of Japan, 337th MEGANE Kokujinkai p.
p., 'L600-1601). However, other methods may also be used.

In the following explanation of Yagmation, FIG. 33 will be referred to as appropriate. Figure 23 shows an example of the results of converting the axis contour vectors and core lines, where the thin lines are the contour edges, the thick lines are the core lines, and the broken lines are interpolated to connect the contour vectors and the core lines. The vector j~ (corresponds to the gl u e edge 1-le described in the above-mentioned paper ``Improvement of the core line generation method based on contour lines''. For convenience, this is also called the contour line 1~ ).

The example shown in FIG. 3 is the result of contour vectorization and center line extraction processing of a portion (encircled by a broken line in the drawing, for example, as shown in FIG. 5).

Here, the "loop" used in the following explanation will be explained with reference to FIG. In FIG. 7, solid 1 to Δ, B, and C are each solid 1 hell of one core wire. Also, each solid 1 to A, 13. Attached to the end point of C (
As, Bs, and Cs in ) are the starting point coordinates of each solid 1 to A, B, and C, and Ae and Be.

Ce is 1 which is the end point coordinate of each solid line A, B, C, and the loop containing the core line (solid line A) is one segment 1~ achieved by vectors A, B, C. means. Further, the distance between both end points of this loop corresponds to the distance between As and Ce.

Once again, attention will be paid to the example shown in FIG. 3, and the processing contents will be explained. First, a loop LA including an unprocessed (processed flag (=J not added) skeleton AO is searched for (step ■).

Here, loop LA is composed of core wires A, O and A1.

If you find core wire AO and loop LA, loop ■.

Check whether the length between both end points of A is larger than a preset threshold (for example, the size of one character) (step ■)
. If it is too thick, set the core wires AO and Al as reliable line segments in step (2), attach a processed flag to the core wires AO and A1-, and then return to step (2).

Both end points of loop I and A are 1? When the length of JJ is less than the threshold (i/4), search for loops with good straightness that touch A on the right and left side (step ■) 1. Loops that touch both the right and left side When the core wires AO and Ai do not exist, or when the straightness is not good even if they exist,
Add a processed flag to , and return to step ■. In the example of FIG. 3, the loop with good linearity that is in contact with the loop LA is the vector (core line) BO.

This is a loop consisting of B1.

Here, good straightness means that the angle between the loop LA and the loop in contact with it is less than a certain threshold value. Further, the angle between the loops is the angle O shown in FIG. When a loop with good linearity that is in contact with loop LA is found, the procedure of similarly searching for loops with good straightness beyond this loop is repeated.

Find both end points of the entire group of loops with good linearity found in step (2), and draw a perpendicular line from the end point of each core line back 1 to 1 constituting each loop to a virtual straight line connecting the two end points. Then, find each end point and the length to its foot,
Check whether the length exceeds the specified line width (step ■
). FIG. 8 is an explanatory diagram of this determination regarding loops LA and LB. Also, on both sides of the imaginary straight line, add 1~
Another example of where the end points of the file are located is shown in FIG.

If the length of even one perpendicular line exceeds the line width, check the processed flag for each core line corresponding to the loop group and return to step ■; however, if all the perpendicular lines are less than the line width, The corresponding core wire group is determined to be a straight line (step (2)), a processed flag is attached to each core wire, and the process returns to step (2).

In the example shown in FIG. 3, the loop L +3 corresponding to the core wires B, O, and Bl is the loop that is in contact with the right side of the loop LA.
, a loop LC corresponding to the core wire CO, a loop LD corresponding to the core wire DO, and a loop LE corresponding to the core wire EO, among which the loop LB with the best linearity is selected in step (2). Since loops LA and LB satisfy the condition of step (2), they are extracted as straight lines.

When no untreated core wires are found, the process proceeds from step (■) to step [phase]. Here, the contour lines corresponding to the core lines that were determined to be reliable line segments in step ■ and the core lines that were determined to be straight lines in step do.

In the next step 0, a circumscribed rectangle of the remaining contour vector mass is determined, and the range of this circumscribed rectangle is defined as a symbol area.

FIG. 4 shows the segmentation results for the example shown in FIG. 3. In FIG. 4, thick lines are skeleton lines extracted as reliable line segments or straight lines, chain lines are circumscribed rectangles (symbol areas), and thin lines are outlines of symbol parts.

〔Effect of the invention〕

As is clear from the above description, according to the present invention, all segmentation is processed using vectors using contour conversion and core line conversion. It is now possible to accurately cut out parts.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of a hardware configuration for carrying out the present invention, FIG. 2 is an overview flowchart of processing from 1 to 3, and FIG. 3 is a diagram showing an example of the results of contour vectorization and skeletonization. , Fig. 4 is a diagram showing the segmentation result of the example shown in Fig. 3, Fig. 5 is a diagram showing an example of a specific drawing part corresponding to Fig. 3, and Fig. 6 is a diagram showing a loop and a loop. 7 is an explanatory diagram of skeleton lines and loops. Figures 8 and 9 are illustrations of the length of the perpendicular line from the end point of each skeleton vector to the virtual straight line connecting both end points of the loop group. It is an explanatory diagram. 1...Image memory, 3.Program memory, 4...
・CPU, 5...Data memory. . Easy → (

Claims (1)

[Claims] (1) Extract the contour vector by tracing the black pixels of the binary image of the drawing, extract the skeleton line from the extracted pair of contour vectors, extract the skeleton line as a reliable line segment or straight line from the extracted skeleton line, and extract the skeleton line as a line segment or straight line. A drawing processing method characterized by cutting out a region of clusters of contour vectors corresponding to core lines as a symbol region.