JPH02193263A - System for automatically recognizing drawing - Google Patents

Abstract

PURPOSE:To automatically recognize a drawing with a high precision by adding a straight line detection correcting step and correcting or restoring the collapse of a right-angled part or the deviation of an intersection part generated in image processing. CONSTITUTION:A drawing automatic recognizing device consists of an image reader 31 as the drawing input device, an image processor 32 as the private processor for image picture data, a host computer 38, and a plotter 43 which prints out data. The deviation from an actual drawing, namely, the deviation of intersection of straight lines and the collapse state of a right-angled part which are generated in the image processing of line narrowing or the like are corrected by the straight line detection correcting step to restore lines approximating those described on the input drawing, and thereby, data closer to an original image is generated as input data of the succeeding analytic recognition processing. Thus, the drawing is automatically recognized with a high precision.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to an automatic recognition method for recognizing drawings with handwritten characters, etc. The present invention relates to an automatic drawing recognition system that can correct and restore line distortion caused by processing straight line data.

[Conventional technology]

FIG. 3 is a flowchart showing a conventional drawing automatic recognition method presented, for example, at the Information Processing Society of Japan's 33rd (late 1986) National Conference 2N-2 "Study of Element Separation Method in Systematic Drawing Reading"; In STI, STI is the first step of inputting data such as characters, figures, symbols, etc. on the drawing as image data, and Sr2 is the second step of converting the input data into binary information in pixel units. Binarization processing step as the step of Sr3
Sr4 uses image processing methods such as dilation and contraction for the binarized data, and the point on the drawing (area extraction processing step as the third step to extract the outline of the filled-in area; Sr4 is the binarization process) For each piece of line information, a fourth step is a thinning process to extract the skeleton of the central part.Sr1 performs code conversion on the data obtained as a result of the thinning process so that the image information can be handled by a computer. , a point sequence processing step as the fifth step in which pixels are represented by coordinate values, Sr6 further vectorizes each line represented by the point sequence, and converts one line into 1
The vectorization processing step Sr1 is an analysis recognition processing step that recognizes symbols, characters, lines, etc. on the drawing using the data obtained as a result of the processing up to step Sr6.

Next, the operation will be explained with reference to the image data processing diagrams shown in FIGS. 4 and 5. First, step STI
In step ST2, the line diagram 10.20 drawn on the drawing is input to the image reader, and the line information written on the drawing is binarized in step ST2, and the binarized image image information in pixel units is obtained. Create image data as. Although the pixel size is determined by the resolution of the image reader, the image data 11 and 21 are usually created on the order of several to ten-odd pixels per mm. Next, using the image data 11 and 21, line information and area (filling)
In order to extract the outline of the region so that information can be separated, region extraction processing is performed pixel by pixel in step ST3 using a logical filter, and then on the image data that has been binarized by the thinning processing in step ST4. , extract the center pixel of the line with the thickness. Next, using the thinned data in the point sequence processing of step ST5, code conversion processing is performed to make the image data usable by a computer. That is, the intersection points and end points of the lines are extracted, these are defined as feature points, the line between the feature points is defined as one line 12, 22, and the line is represented by an arrangement of a plurality of pixels. Next, step ST
Using the result of the point sequence processing performed in step 6, each line is expressed as a point sequence vector 13.23. As a result, one line 14.24 is represented by one or more vectors. After performing the above processing, analysis and recognition processing of characters, symbols, lines, etc. is performed on a computer using the code data obtained in step ST8.

[Problem to be solved by the invention]

Conventional automatic drawing recognition methods consist of the steps described above, so they can be used to recognize deviations in actual drawings caused by image processing such as line thinning, such as deviations in the intersections of straight lines and collapse of right angles. The next stage of analysis and recognition processing must be performed, and in this case, in the analysis and recognition processing step, if there are large bends in the line, it may not be possible to accurately extract a straight line, or a right angle corner may not be extracted accurately. There were some problems, such as the fact that the parts were crushed, making it difficult to accurately recognize and judge the image data.

This invention was made in order to solve the above-mentioned problems, and after code conversion processing is performed to remove line distortions that occur as a result of image processing, it automatically performs straight line correction. The purpose of this invention is to obtain an automatic drawing recognition method that can restore lines to a form close to the original state, and that can reproduce symbols, lines, etc. with higher accuracy in subsequent analysis recognition processing.

[Means to solve the problem]

The automatic drawing recognition method according to the present invention corrects distortions in straight lines that occur during image processing such as line thinning processing in the straight line detection correction step, and restores the lines to a shape close to that described in the input drawing. In this way, data that is closer to the original image is created as input data for the subsequent analysis and recognition process.

[For production]

In the straight line detection and correction step in this invention, image data obtained as a result of line thinning processing and vectorization processing is used as input data, and vectors close to horizontal or vertical are tracked and extracted, and short lines formed by collapsing at right angles are By further extracting vectors and deviations in intersections due to short vectors that occur at intersections, several divided vectors are finally combined into a single vector and the linearity of the original original image is restored. .

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to the drawings. 1st
In the figure, STI is the drawing input step as the first step in which data on paper is actually input as image data, and Sr1 is the binarization step in which the input data is converted into binary information in pixel units. processing step,
Sr3 is a region extraction processing step as a third step of extracting the ring part of the area that is painted black on the drawing using image processing methods such as dilation and contraction for the binarized data; A thinning processing step as the fourth step for extracting the skeleton of the central part of each binarized line information, Sr5 uses a computer to handle the image information of the data obtained as a result of the thinning processing. In the fifth step, Sr6, converts each line into a vector and transforms each line into one or more vectors. A vectorization processing step as a sixth step of performing the processing represented by
Sr7 is a straight line detection correction processing step as a seventh step that uses the data obtained up to step 8T6 to correct and restore straight line collapse, distortion, etc. that occurred in image processing, and Sr8 is the result of the processing up to step ST7. Using the obtained data, recognize symbols, characters, lines, etc. on the drawing. This is an analysis recognition processing step as the eighth step in the computer.

Next, an automatic drawing recognition method will be explained using the above flowchart with reference to FIGS. 4 and 5.

First, the drawing written on the drawing is inputted to the image reader 31 in step STI, and the line information written on the drawing is binarized in step ST2, and the line information written on the drawing is binarized in pixel units.
Create image data as digitized image information. The size of a pixel is determined by the resolution of the image reader 310, and is usually on the order of several pixels to more than 10 pixels per 1 mm. Next, the image data 11.2
1, and in order to extract the outline of the area so that line information and area (fill) information can be separated, area extraction processing is performed pixel by pixel using a logical filter. Thereafter, a pixel coordinate string at the center of a thick line is extracted from the image data binarized by the thinning process in step ST4.

Next, using the thinned data in the point sequence processing of step ST5, code conversion processing is performed to make the image data usable by a computer.

In other words, extract the intersection points and end points of the lines, use them as feature points,
A line is drawn between the feature points, and the line 12°22 is shown as an arrangement of a plurality of pixels. Next, by vectorization processing in step ST6, the point sequence of each line is expressed as vectors 13 and 23 using the result of point sequence formation. In FIGS. 4 and 5, line segments indicated by Ln (n=1, 2, . . . ) are vectors forming each line. When vectorized in this way, deviations at right angles and deviations at intersections occur as shown in the figure. To correct this, horizontal vectors such as Ll and L2 and vertical vectors such as L, , and L5 are first extracted as reliable, and particularly short vectors such as 5 that are connected to them are extracted due to distortion. It is determined that this has occurred, and finally the coordinate values are taken into consideration and re-vectorized like a straight line 14 to restore the shape to a shape close to the original diagram 10. Similarly, in the case of an intersection, an extremely short line such as the vector Lg may occur at the intersection, causing a deviation, but in this case as well, first extract the reliable horizontal and vertical vectors in the surrounding area, and then use the above method. Similarly, the deviation caused by the short vector is corrected and finally the vector is re-vectorized into the shape of the straight line 24, thereby restoring the shape close to the original diagram 20.

Then, after correcting the planned image data of the straight line portions through the above processing, these data are used to perform analysis recognition processing of characters, symbols, lines, etc. on a computer in step ST8.

FIG. 2 is a block connection diagram showing an automatic drawing recognition device that implements the method of the present invention. In the figure, numeral 31 is an image reader which is a drawing input device, and numeral 32 is an image processor, which is exclusively used for processing image data. It is a processing device. 33 is an image reader interface that interfaces the input device 31 with this data; 34 is a central processing unit of the image processor 32; 35 is a system memory of the image processor 32; 36 is a host interface that interfaces data with a host computer to be described later; Reference numeral 37 is an image memory that stores the image data of the input image as it is.

On the other hand, 38 is a host computer, 39 is an image processor interface, 40 is a central processing unit of the host computer i38, 41 is an input/output interface, 42 is a storage device, 43
is a plotter that prints out data.

Next, the operation will be explained.

First, the image reader 31 reads characters written on paper with a resolution of several pixels to more than ten pixels per millimeter, and transmits the pixel data to the image data system 32 side. The image reader interface 33 in the image processor 32 receives this data, and the received pixel data is stored in the image memory 37 as image data.

Using the image data stored in the image memory 37, the central processing unit 34 of the image processor 32 performs image data conversion processing, code data conversion processing, etc.

These processes include processing such as area extraction processing and thinning processing. In the area extraction process, outline pixels are extracted from areas where black pixels are gathered, that is, areas that are painted black in the input drawing. Furthermore, in the line thinning process, for a line with a width of several pixels on the image data (a line on the input drawing), processing is performed to extract both series that form the skeleton of the central part of the pixel. In these processes, a 3×3 logical filter is used,
For each pixel, the central processing unit performs filtering processing (checks the pattern of the central pixel of the 3×3 filter and its surrounding pixels, and converts image information pixel by pixel). For example, if all surrounding pixels are black, processing such as converting the central pixel into a black pixel is performed.

On the other hand, the host computer 38 stores the processed image data from the image processor 32 in the storage device 42, processes it in the central processing unit 40 as necessary, outputs it through the input/output interface 41, and inputs it to the plotter 43. and print it out.

〔Effect of the invention〕

As described above, according to the present invention, a straight line detection correction step is added to the automatic drawing recognition procedure, and the configuration is configured to correct and restore the collapse of the right angle portion and the deviation of the intersection point that occur during image processing. Data with higher reliability can be obtained in the recognition processing step, and therefore, there is an effect that highly accurate drawing automatic recognition processing can be performed.

[Brief explanation of the drawing]

Fig. 1 is a flowchart showing an automatic drawing recognition system according to an embodiment of the present invention, Fig. 2 is a block connection diagram showing a drawing recognition processing device that executes drawing recognition processing, and Fig. 3 is a flowchart showing a drawing automatic recognition system according to an embodiment of the present invention. The flowcharts shown in FIGS. 4 and 5 are image data processing diagrams. STI is a drawing input step, Sr1 is a binarization processing step, Sr1 is a region extraction step, Sr1 is a thinning processing step, Sr1 is a point sequence processing step, Sr1 is a vectorization processing step, Sr1 is a straight line detection correction step, S
r1 is an analysis recognition process 31ij step. In addition, the same symbols in the figures indicate the same or equivalent parts. Patent applicant: Mitsubishi Electric Corporation Figure 4↓

Claims (1)

[Claims] A drawing input step of inputting characters, etc. on the drawing using an image reader, a binarization processing step of binarizing the input image data, and expansion, contraction, etc. of the binarized data to create the characters, etc. A region extraction step extracts the outline of the blacked-out area, a thinning process step thins and extracts the skeleton of the central part of the extracted outline line information, and a thinning process step extracts the skeleton of the central part of the extracted outline. The image data is converted into a code that can be handled by a computer, and the process step is to convert the image data into a code that can be handled by a computer, and to convert the pixels into a sequence of points using coordinate values. In an automatic drawing recognition method comprising a vectorization processing step for representing characters, and an analysis recognition processing step for analyzing and recognizing the characters, etc. from this vectorized data, a horizontal or nearly vertical vector obtained in the vectorization processing step is used. track and extract,
An automatic drawing recognition method comprising a straight line detection and correction step for restoring the linearity of vectors by extracting short vectors that are crushed at right angles or deviations of short vectors that occur at intersections.