JPS62114079A - Closed loop graphic extracting device - Google Patents

Abstract

PURPOSE:To extract a closed loop graphic from in a graphic having a deformation and a noise, by extracting successively a point for satisfying condition which has been prescribed in a closed loop graphic extracting dictionary, from in fundamental feature points which have been extracted and registered from an input drawing. CONSTITUTION:With regard to all picture elements of a binary image data from an image input device 1, a fundamental feature extracting device 2 encodes a pattern shape of its minute peripheral area, and outputs a point for satisfying a condition which has been designated in a fundamental feature extracting dictionary, as a fundamental feature point candidate, to a fundamental feature table 3. On the other hand, in a closed loop graphic extracting dictionary 5, a P feature code for using one of two adjacent apexes of a closed loop graphic of a recognition object, and detecting the other, an N feature code which must not exist in a detection range, and a parameter for prescribing a detecting direction and a detecting range are stored successively, and a closed for satisfying a condition which has been prescribed in the dictionary 5, from in a fundamental feature table 3.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a closed-loop figure extraction device in a figure recognition device.

(Prior art) Conventionally, this divine device thins the figure to be processed, traces the line segments of the thinned figure, and also traces the outline of the figure even on a platform that does not thin the figure. Then, a closed-loop figure is extracted by examining the connectivity or change in direction of adjacent points, extracting the feature points necessary for extracting the figure, and detecting the connection relationship between the feature points.

(Invention is the solution + Js shi J, uto Jru problem) But 'r
However, when thinning the line, there is a problem of generation of noise such as whiskers, and when line thinning is not carried out,
The method of tracing the outline of a figure has the problem that it is not always possible to extract a stable closed-loop figure, despite various deformations such as broken lines, protrusion, unevenness, etc., and noise.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above-mentioned problems and provide an apparatus capable of extracting a desired closed-loop figure with high accuracy and high speed from among figures having various shapes and noises.

(Means for Solving the Problems) In order to solve the above problems, the present invention includes an image input means for inputting an Ifli image to obtain a binary image, and scanning the input image to obtain at least two refraction points. branch point.

A basic feature extraction means that extracts intersections and end points as basic features of a graphic pattern in advance, and a basic feature table that registers feature code names and position information assigned to each basic feature, and recognition that uses multiple bending points as vertices. For each shape of the target closed-loop figure, a closed-loop figure extraction dictionary that specifies the feature code name, detection direction, and detection range of arbitrary adjacent vertices starting from one of the other vertices, and the basic features extracted and registered from the input drawing. Which of the points satisfies the regulations stipulated in the dictionary? A closed-loop figure extraction means is provided for sequentially extracting the desired figure.

(effect) According to the present invention, the inflection point is determined from the binarized figure.

Extract the basic features such as branch points, intersections, and end points, and calculate the direction, range, or feature code name, etc. of other bending points from the bending point, etc. It is compared with a dictionary defined in advance for a specific closed-loop figure, and only when all the conditions defined in the dictionary are satisfied is extracted as a predetermined closed-loop figure.

(Embodiment) FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is an image input device for inputting an original image to obtain a binary image, and 2 is a basic feature extraction @ , 3 is a basic feature table, 4 is a closed-loop figure extraction device which is the central part of the present invention,
5 is a closed-loop figure extraction dictionary, and 6 is an extraction result memory for outputting the extracted results.

The basic feature extraction device 2 also includes a pixel feature extraction device 7, a pixel feature code memory 8, a basic feature extractor 9, a basic feature extraction dictionary 10, and a representative point determiner 11, as shown in FIG. Consisting of

The pixel feature extraction device 7 is a device that codes the pattern shape of a minute peripheral area for all pixels of the input image and outputs the code to the pixel feature code memory 8. Further, the basic feature extractor 9 performs a rask scan on the pixel feature code memory 8 and outputs points that satisfy the conditions specified in the basic feature extraction dictionary 10 to the basic feature table 3 as basic feature point candidates. It is a device. Further, the representative point determiner 11 is a device that determines a representative point from among the basic feature point candidates.

While the pixel features are micro area features, the basic features are more micro area features that are extracted by comprehensively determining the appearance state of pixel features around the point of interest. The procedure for performing such a feature extraction method is shown in Figure 5 (described later).
As can be seen in example a), the same feature as the −4 one pixel feature at the upper left inflection point of the frame-like closed loop figure also exists in a part of the white circle figure on the lower side, and the structure of the figure cannot be determined by pixel features alone. The first reason is the inability to use personal characteristics, and other important reasons include:
This is because by extracting a double structure as in this method, stable feature extraction is possible even when partial patterns are cut or deformed.

The details of the basic feature extraction device 2 will be explained below.

The pixel feature extraction device 7 includes a binary image buffer memory 12 that temporarily stores binary image data from the image input device 1, a pixel feature extractor 13, and a threshold value setter 14.

FIG. 3 is an explanatory diagram of the pixel feature extractor 13.

FIG. 5(a) shows a mask 14 conceptually representing the pixel feature extractor 13, and shows "individual pixels" to be recognized (
Hereinafter, this will be referred to as a pixel of interest. ) 15, area of interest 1jA (S
) 16, a dead zone region 17 that is located around the region of interest 16 and is not involved in pixel feature extraction, and a plurality of sensing regions 1Illi18a, 18b, 1 set around the dead zone region 17.
8c, IF'l (1°18e, 18f, 18o, 18h
The region of interest 16 and the sensing region II!1i188 to 18F are each composed of 9 pixels (3 pixels x 3 pixels), and the dead zone region 1[l
i! The mask 17 has a width of 3 pixels, and the mask 14 as a whole consists of 225 pixels (15 pixels×15 pixels).

The pixel feature extractor 13 scans the binary image buffer memory 12 and extracts the region of interest 16 and the sensing regions 188 to 1.
The number of black pixels included in each area of 8h is calculated using the front threshold 111'i
Compared with the LC candy set in the setting device 10, (number of black pixels) ≧
If (threshold value), it is determined to be 1゛1'', and if (number of black pixels) < (threshold value), it is determined to be rOJ. The determination results for each area are shown in Figure 3 (b). 18a, 18b, 1fl
c, 18d, 16, 18e.

18f, 180.18h are outputted to the pixel time ff1l]-to memory 8 as inner search feature codes 19 of 9 pixels 1 to 18h.

The reason why the dead zone area 17 is set in the pixel feature extractor 13, ie, the mask 14, is to absorb variations in line width in a figure and to express features with a small number of bits. Here, the resolution of the image input device 1 is set to 8 lines/
tnm %, that is, the area per pixel is 0.125 x
If the threshold value is set to 0.125 mm and the threshold value is set to 3, it becomes 0.125 mm.
A line segment having a line width of 125 to 1.125 mm can be recognized as one line segment.

FIG. 4 shows an example of pixel feature extraction by the pixel feature extraction device 7, in which the mask 14 scans leftward over a binary image 20 in the shape of a square bracket having a line width of two pixels. As it progresses, the inner search feature code that is 1q at this time is shown. In FIG. 4(a), the binary image 20 is in the 6th eye area 16. Sensing area 18d.

180.18 block is shown, and in this case, [00
0111100] is obtained. In addition, Fig. 4(b) shows the scanning position as 'C' compared to Fig. 4(a).
The case where the movement is 7 pixels is shown, and the binary image 20 is the area of interest 1.
416. intersects the sensing regions 18e and 18q, and at this time r
A pixel feature code 22 of oooool 1010J is obtained.

FIG. 5 shows that the basic feature in the basic feature extractor 9 is 7.
- Demonstrates the principle of extraction.

As shown in FIG. 5(a), whether the upper left corner of the frame-shaped figure 23 is scanned or the upper left line of the circular figure 24 is scanned, the pixel feature extraction device 7 receives the same information. A pixel feature code indicating the refraction point is obtained. Therefore, if it is attempted to determine the features of a figure using only pixel feature codes that can express only the features in minute parts of the figure, problems will arise.

Therefore, in the basic feature extractor 9, the pixel feature code (hereinafter referred to as the focused Starting from the extraction position of the pixel feature code (referred to as pixel feature code), one or more other extraction positions of the pixel feature (referred to as pixel feature code) are set in the surrounding area, and features are extracted from these.

In FIG. 5(a), the same target pixel feature code (2
In the value display rl 11100100J or roooo
Areas 25a to 25d and 26 in the same positional relationship from the extraction points 23a and 24a of 1010J]-do)
Comparing the pixel feature codes obtained from a to 26d, areas 25a, 26a, 125C, 26C1, and 25d
and 26d can be determined to be figures having special characteristics. That is, the area to the right of the pixel feature code of interest as a starting point is a pixel feature code indicating a horizontal line segment, the areas above and to the left are pixel features indicating blank spaces, and the area below is indicating a vertical line segment. Defining it as a feature figure that has the basic feature of left square bracket only when the pixel feature code exists,
In the case of FIG. 5(a), only the extraction point 23a can be extracted as a point having the target feature figure 27.

In addition, even if the target pixel feature code, here the code in the form of a left square bracket, cannot be obtained from the extraction point 28 due to a break in the line segment as shown in FIG. 5(b), the pixel feature code of interest and each region 1ifR29a, 29b, 29C, 29d
pixel features obtained from] - It becomes possible to extract figures 27 having the same basic features.

The basic feature extractor 9 scans the pixel feature code memory 8 based on this principle, determines the conditions specified in the basic feature extraction dictionary 10, and determines whether there are points that satisfy the conditions. For example, the coordinates of the extraction point are output as feature point candidates to a predetermined location in the basic feature table 3 corresponding to the respective basic feature code.

FIG. 6 shows the configuration of a basic feature extraction dictionary (hereinafter simply referred to as dictionary 1o) 10 referred to by the basic feature extractor 9, in which one page of the dictionary is one basic feature extraction dictionary. . Below, using the first page of the dictionary as an example, its contents and the dictionary ■
The extraction operation according to No. 10 will be explained.

The basic feature code name C1 is stored at the beginning of the dictionary 10,
Next, the key (KFY) feature mask and ]-code are stored. The key feature is a pixel feature that is the starting point for basic feature extraction, and code (COI) F) 1 is its code name. The mask specifies the validity/invalidity of each bit of the pixel feature]-do, and by using the mask and the pixel feature]-do, it is determined whether the current scanning point in the pixel feature]-to memory 8 is a key feature point. This is b, which makes the determination. That is, for the pixel feature of the current scanning point]-doPF, (PF△HASKI)■(C0DE i△HASK
1) = 0...If (1) is satisfied, the detection areas 1 to N shown in the subsequent dictionary
The pixel characteristics appearing in 1 are checked, and if the expression (1) is not satisfied, the process proceeds to the next page.

The number of detection regions N1 is the number of determination regions of pixel features that appear around the key feature, and for the rises from 1 to N1, the distance between the key feature and the detection region d x 1i,
d yli, detection area size Δx1i.

Δyli, the mask and code of the positive (PO8ITVE) feature, and the mask and code of the negative (NEG^TTVE) feature are stored in order.

FIG. 7 shows the positional relationship between the key feature and the surrounding detection area. In the figure, 30 is the key feature, 31a, 31
b, 31c, and 31d are detection areas. Note that in the figure, the subscript j corresponds to the page of the dictionary. Here, the positive feature refers to a pixel feature that must exist within the detected head area, and the negative feature refers to a pixel feature that must not exist within the same area. The meaning of mask and ]-do is the same as the meaning of mask and ]-do in the key feature. That is, the detection area i around the current scanning point
The pixel feature at one point (x, y) in P F ixy
How, (P Fixv A Pmaskil)■(Pcod
eN△P maskil) −〇 −・・−(2)(
p Fixy A N maskN)■(N cod
ei1△Nmaskil)-1...(3) is all the points (x,
Only when y) is satisfied, it is determined that the scan is a basic feature point candidate for code C1, and if either formula (2) or (3) is satisfied and the number h is reached, the next page is Proceed to dictionary processing.

In this way, the basic feature extractor 9 sequentially performs the above-mentioned judgments from the first page of the dictionary 10 while scanning the pixel feature memory 8, and if there is a point that satisfies all the conditions of the page, , the position coordinates are output to a predetermined location in the basic feature table 3 corresponding to the basic feature code name of the dictionary page, and are used as basic feature point candidates.

By the way, as seen in the corner 33 of the left-angled bracket-shaped figure 320 shown in FIG. 8, the basic feature candidate points have a property of appearing adjacent to a range corresponding to the figure line width.

The representative point determiner 11 searches the basic feature table 3 in which the coordinate points are stored for each point, and determines the center point of the rectangular frame of the coordinate points in an adjacent relationship as shown in FIG. The coordinate point set thus far is replaced with the representative point, and the result is again output to the basic feature table 3 in the format shown in FIG. FIG. 9(a) is a diagram explaining the shape of the basic feature in the table and the name of ]-do, and FIG. 9(b) shows the detailed content of the basic feature of ]-do 1 as an example. So, (xl, yl), (x2, y2
)...(xn, yn) is a representative point of the basic feature points of the code name (hereinafter referred to as basic feature points).

Figure 10 (a), (b), (c) C. Showing a logic circuit diagram for this embodiment - closed loop figure 34
35.36, and the shape of the basic feature necessary for extracting the same figure is shown by a thick line.

FIG. 11 is a diagram illustrating the contents of the extraction dictionary 5 in this embodiment for extracting the target figure.

At the beginning of the extraction dictionary, the feature code name Cs of the extraction start point for sequentially extracting 4 IQ points is stored, and the information necessary for sequentially extracting 10 adjacent points is stored at the beginning of the dictionary.
The configuration is such that one extraction process is stored as one page.

Figure (a) shows the processing content for each page. In the example of a one-page dictionary, the starting point is 50 feature codes ([),
This indicates that the point (L) of feature code 6 is to be detected in the Y-axis stop direction, and from now on, starting from the detection point on the previous page, the next vertex will be detected and pushed out.

Specifically, the detection target code (
Hereinafter, it will be referred to as the P feature code. )Features that must not exist within the detection range]-do (hereinafter referred to as F and N feature codes),
Parameters ΔX and ΔY defining the detection direction and range are stored in this order. The N feature code is, for example, as shown in Figure 10 (
In the case of figure 34 in a), 1 - vertex 3 of side = 1a to 34b
In between, feature codes 1, 3, and 6 of basic feature point team 1 and 3
．． 8, etc., which shows a feature inconsistent with the shape of the closed loop figure 34.

When the detection direction is the X-axis direction, the parameters ΔX and ΔY are
Δ×−±XM (The positive and negative signs correspond to the positive and negative directions of the X-axis, respectively.), ΔY=O1 If the detection direction is the Y-axis direction, Δ
X-0. ΔY−±YM (the positive and negative signs correspond to the positive and negative directions of the Y axis, respectively), and the values of XM and YM indicate the relative distance from the starting point considering the maximum dimension required to extract the target figure. ing.

FIG. 12 shows a closed loop figure, for example, one of the frame-shaped figures 37.
(The detection range of one point 378 to 37d is shown, and in the figure, DX, I)V is a parameter described later.

In this way, the closed-loop figure extraction dictionary 5 sequentially stores, starting from one of two adjacent vertices of the closed-loop figure to be recognized, the P feature code, N feature code, detection direction, and detection range for detecting the other. The closed-loop figure extraction device 4 of the present invention uses this extraction dictionary 5 to sequentially extract fish that satisfy the conditions specified in the dictionary 5 from the basic feature table 3 shown in FIG. Then, the target figure is extracted.

Hereinafter, the closed-loop figure extraction process will be explained along the flowchart of FIG.

First, in step SP1, the number of basic feature points having the extraction start point code name Cs of the figure to be extracted stored at the beginning of the dictionary is determined from the basic feature table 3 and stored in a constant N. Next, in step SP2, a counter n is used to count this number.
The initial value of is set to 1. In step SP3, the coordinates of the n-th feature point (xn) whose feature code name is C8 are determined corresponding to the counter n.

yn) at the starting point coordinates (xo, yO). In step SP4, the initial value of the variable S indicating the page counter of the extraction dictionary 5 is set to 1, and in steps SP5 to 5P11, the extraction process of points satisfying the rule f1 specified in the dictionary from pages 1 to 4 is performed in the basic feature table. Perform the search in step 3.

That is, in step 5P5r, the detection area Rs (xO, yO) is calculated from the starting point (xo, yO) and ΔX, ΔY of the dictionary as shown in equations (4) and (5) below.

When the detection direction is the Y-axis direction (ΔX=O), R8(XO,
VO)=((X,V”)1xo −Dx/2≦×≦xo
+Dx/2゜111in (yo, yo1ΔY
)≦y≦max (yo, yo ΔY))
....-(4) When the detection direction is the X-axis direction (ΔY
−0), Rs (xo, yO)= ((x, V')1
111in (XO, xO+ΔX)≦X≦max
(xo, xo+ΔX).

VO-DV/2≦y≦yo+I)y/2)・・・・・・
(5) (However, l) ) Next, in step SP6, among the P feature code points specified on the eighth page of the dictionary, those existing in the sieved region bltRs are extracted from the basic feature table 3.

If there is no point within area R3, step 5
If there is one or more points in step SP7, the point closest to the start point (x0°yO) is stored in (xps, yps), and it is used as a candidate for the P feature code point. and 16° Next, in step SP8, the N feature code points specified in the dictionary are checked. That is, when the detection direction is the Y-axis direction, ΔY in equation (4) is replaced with (yps-yO), and when the detection direction is the X-axis direction, Δ× in equation (5) is replaced with (xps-xO). 'C-J to replace
:S, R3' (XO, yO) is calculated, and the N feature -
” - point is 1 in the region Rs' (xo, yO)
If white also exists, the process advances to step 5P14, and if no point exists, the coordinates (xps, yps) become the P feature code point.

In step SP9, it is determined whether the page of the dictionary is the final page, and if it is not the final page, the process starts from the extracted P feature code point (xps, yps) and performs step 5 PIO).
, advances the dictionary by one page (step 5P11), returns to step SP5, and repeats steps SP5 to SP11 until the processing of the final page is completed.

If it is determined in step SP9 that the processing of the last page of the dictionary has been completed, a closed loop condition determination is performed in step 5P12.If the condition f1 is not adjusted, the process proceeds to step 5P14; > (However, s=1 to 4
) are output to the extraction result memory 6 as the four vertices of the closed loop figure (step 5P13), and the process proceeds to step 5P14.

Step 5P14.5P15 is the step SPI~5
The process up to P13 is performed six times for all n, and when n=N, the extraction of closed loop figures included in the input screen is completed.

(Effects of the Invention) As explained above, according to the present invention, since external thinning is not performed, basic feature points can be extracted stably and accurately, and geometric line segments can be traced easily. Because the basic feature points of the target figure are extracted only by searching the basic feature table without searching, high-speed external pressing is possible. , it is possible to extract the target figure at high speed and with good viscosity without accidentally extracting figures with partially similar shapes, and further, by changing the contents of the dictionary, it is possible to extract all kinds of drawings, e.g. It has the advantage that it can also be used to extract polygonal buildings from city maps.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the closed loop figure extraction device of the present invention, FIG. 2 is a block diagram showing the configuration of a basic feature extraction device, and FIGS. 3(a) and 3(b) are pixel feature extraction Figure 4 (a) and (b) are diagrams showing an example of pixel feature extraction, Figures 5 (a) and (b) are diagrams explaining the principle of basic feature extraction, and Figure 6 is the basic FIG. 7 is a diagram showing the configuration of the feature table, FIG. 7 is a diagram showing the positional relationship between key features and detection areas, FIG. 8 is a diagram showing basic feature candidate points, and FIG. 9 (
a) and (b) are diagrams showing the structure of the basic feature table, 10th
Figures (a), (b), and (C) are dimensional drawings of examples of closed loop shapes.
11(a) and 11(b) are diagrams showing the structure of a closed-loop figure extraction dictionary, FIG. 12 is a diagram showing the detection range of the vertices of a closed-loop figure, and FIG. 13 is a diagram showing the flow of the closed-loop figure extraction process. be. 1... Image input device, 2... Basic feature extraction device, 3
...Basic feature table, 4.. Closed loop figure extraction device, 5.. Closed loop figure extraction dictionary, 6.. Extraction result memory. Patent Applicant Oki Electric Industry Co., Ltd. Agent Ben Sokujukichi 1) Seiko 18a-18h: Sensing area (b) Pixel characteristics Explanatory diagram of fine earthenware Figure 3 (MmI picture @ r Tohe (a) (b) Pixel Figure 4 shows an example of feature extraction Figure 4 (a) (b) Explanation of basic features Figure 6 Diagram of the positional relationship between key features and detection areas Figure 7・ ・ ・ ・ ・"""
・ ・ ・ ・ ・ ・ ・ ...Figure 8 x axis (8) (b) (C) Diagram showing an example of a closed-loop figure Figure 10 (a) (b) Diagram showing the structure of a closed-loop figure extraction dictionary Figure 11 □ X axis Detection range Diagram showing the inspection range of the vertex of the closed loop figure Figure 12

Claims (2)

[Claims] (1) Image input means for inputting an original image to obtain a binary image, and basic feature extraction for scanning the input image and extracting in advance at least refraction points, branch points, intersections, and end points as basic features of a graphic pattern. means, a basic feature table that registers feature code names and position information assigned to each basic feature, and one of arbitrary adjacent vertices for each shape of a closed-loop figure to be recognized whose vertices are multiple inflection points. A closed-loop figure extraction dictionary that specifies the feature code name, detection direction, and detection range of the other vertex with the starting point as the starting point, and points that satisfy the conditions specified in the dictionary from among the basic feature points extracted and registered from the input drawing. 1. A closed-loop figure extracting device comprising: closed-loop figure extracting means for extracting a desired figure by sequentially extracting the figures. (2) The closed-loop figure extraction device according to claim 1, wherein the closed-loop figure extraction dictionary includes a basic feature that must not exist between adjacent vertices of a closed-loop figure to be recognized.