JP2615247B2 - Line figure feature extraction method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a method for extracting a feature of a line graphic in a character recognition device, a drawing recognition device, and other devices that handle a line graphic.
In particular, the present invention relates to a feature extraction method suitable for obtaining a structural feature such as a skeleton line of a line figure.

[Conventional technology]

As one of the feature extraction methods of a line figure, there is a method of obtaining a structural feature such as a skeleton line of the line figure. The main methods of the prior art include the following methods.

(1) A method of performing thinning processing on the entire image data and tracing black pixels of the obtained thinned image to obtain a skeleton line (Tamura; "Some Considerations on Thinning", IEICE Technical Report, PRL7
5-66 (1975)).

(2) A method of tracing the boundary of a line segment, vectorizing this, and then creating a skeleton line at a portion sandwiched between facing vector pairs (Okada: "Examination of thinning method of binary figure by contour tracing" , IEICE Technical Report, IE78-74 (1978)).

(3) A method of dividing the image data into small meshes, integrating the small meshes to obtain a cell, and recognizing a cell pattern for each cell to obtain a partial skeleton line to obtain an entire skeleton line (Suzuki, Mori; "Method of extracting features of line figures", IEICE Technical Report,
PRU89-105 (1990)).

(4) After finding all the boundary points, an appropriate search range is set while tracing the boundary points, a sequence of boundary points is determined in this range, and a skeleton point is obtained at the center of the paired boundary points. A method for obtaining a skeleton line (JP-A-59-95680).

[Problems to be solved by the invention]

The method (1) has an advantage that a skeletal line can always be obtained regardless of the line width and resolution of a line figure, regardless of the input. However, the method requires a large amount of processing and a large number of lines in the process of thinning. It has the disadvantage that information is lost and undesired phenomena such as distortion and whiskers occur. There is also a disadvantage that the planar area is lost.

The method (2) has an advantage that processing is relatively fast, but it is difficult to automatically set parameters for determining a vector pair in accordance with a line width and a resolution. There is a disadvantage that a skeleton line cannot be obtained near a branch or an intersection, and a desired skeleton line cannot be obtained even when short line segments such as characters and drawing symbols are dense. The planar area is an unknown area.

The method (3) can be applied to most image data because the mesh size is determined according to the line width, and the skeleton line near the branch or intersection can be extracted quite accurately. In the case where is small, the cell pattern becomes complicated, so that the performance is remarkably reduced, and there is a disadvantage that a correct skeleton line cannot be obtained. In this method, a planar area can be detected.

The method (4) is a method mainly for drawing, using a threshold value assuming a line width, and calculating a skeleton line by focusing only on the distance between opposing boundary points. Is not determined according to the image data,
There is a disadvantage that the skeleton line is distorted near the branch or intersection of line segments having different line widths. In addition, the boundary point sequence surrounding the planar area is extracted.

In any case, it has been difficult with the conventional method to obtain a correct skeleton line (particularly near a branch / crossing) by adapting all image data.

In addition, from the viewpoint of matching of the obtained features, when more detailed structural features and statistical features other than the structural features are required (for example, identification of “2” and “Z”,
In some cases, sufficient skeleton lines cannot be obtained due to low quality image data), but it is very difficult to extract these locally and easily and to clarify the correspondence between the features using conventional methods. Was difficult.

Therefore, an object of the present invention is to solve the above-mentioned drawbacks of the conventional method, to accurately extract structural features of arbitrary line graphic image data, and to obtain more detailed structural features and structural features. It is an object of the present invention to provide a feature extraction method for a line figure having a framework that allows local and easy extraction of statistical features whose correspondence is clear.

[Means for solving the problem]

The feature extraction method for a line figure according to the present invention includes the steps of extracting a section that is substantially orthogonal to the direction of the line segment from the line figure image data, obtaining a series of the sections and forming a section series, and erasing areas other than the section series. An area, and a node representing either the cross-sectional series or the unknown area, and an arc creating a cross-sectional series graph representing the connection relationship between them. The cross-sectional series graph is characterized by a line figure. That is the main point.

[Action]

FIG. 3 shows the relationship between the original image data (line graphic image data) and the skeleton lines (core lines) of the sectional series graph. As described above, the line graphic description by the cross-sectional series graph has a hierarchical structure, and the cross-section, which is a structural element, has a property of being substantially orthogonal to the direction of the line segment. Further, the cross-sectional series which is a higher-order structural element is obtained by ordering the cross-sections as if they were time-series (part (a) in FIG. 3). The unknown area is an area not included in this cross-sectional series, and corresponds to a planar area of a line end, a bend, a branch, and an intersection (a part (b) in FIG. 3). In other words, the cross-sectional series graph is a simple one that clearly divides a line figure into simple lines and parts that are not. Based on this, structural features such as skeleton lines and rough structural matching are taken. Becomes easier.

〔Example〕

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic block diagram showing an example of a hardware configuration for realizing the present invention. In FIG. 1, reference numeral 1 denotes a central processing unit (CP) for executing feature extraction processing, recognition processing, and the like.
U) and 2 are program memories for storing programs necessary for the processing of the CPU 1. Here, the CPU 1 executes a process of extracting a sectional sequence graph from the image data of the line figure based on the sectional sequence graph extraction processing program 21 in the program memory 2. Reference numeral 3 denotes an image scanner that scans an original or drawing and inputs image data of a line figure.
Reference numeral 4 denotes a pattern memory for storing the input line graphic image data, and reference numeral 5 denotes a data memory for storing data during the processing in the CPU 1 and result data.

FIG. 2 is a flow chart of the whole process of extracting a sectional series graph from line graphic image data. The target line graphic image data may be binary or multi-valued. However, in the case of multi-value, processing is performed while determining white / black of a pixel with a certain threshold value. Prior to the description of the processing content, terms used herein will be described.

Forward / Backward: The tracking direction when boundary point tracking (see FIG. 3) is performed so as to see a black pixel to the left is referred to as the forward direction of the boundary point, and the opposite direction is referred to as the backward direction.

FIG. 4 shows an example of an extension boundary point. In FIG. 4, a circle is a black pixel at a boundary point, and an arrow therein indicates a tilt direction. An inclination direction ang (p) with respect to the target boundary point p is obtained by a differential operator. From p, the boundary point q that first appears following the black pixel in this direction is referred to as an extension boundary point for p (fourth point).
FIG. (A)). However, as shown in FIG.
When the boundary point tracking is performed by eight connections, a white pixel may be encountered without encountering the boundary point. In this case, one suitable boundary point is selected from the boundary points adjacent to the white pixel. In addition, since the boundary points of a line segment having a width of one pixel are tracked redundantly in the process of obtaining a boundary point sequence, they are treated as different boundary points having the same coordinates, and are mutually extended boundary points. And Hereinafter, the extension boundary point of the boundary point p is referred to as pair
Expressed by (p).

Section; When it can be considered that a white pixel hardly appears in a straight line path connecting any two boundary points q and p, this boundary point pair and its straight path are collectively called a cross section (see FIG. 3), and S Expressed as (q, p). The direction of S matches the direction of p and is opposite to the direction of q. One boundary point can belong to multiple cross sections.

Section series: Intuitively, it can be said that the section is a series in which sections are arranged without any gaps or intersections (for example, the part (a) in FIG. 3). Strictly speaking, a cross-section series is a series of one or more cross-sections, and all adjacent cross-sections included in the series (this is referred to as S (p1, q1), S (p
2, q2)) satisfy the following conditions.

"P1 and p2 are the same or adjacent" and "q1 and q2 are the same or adjacent" and "p1 and p2 are not the same or q1 and q2 are not the same" and "boundary Except for the point, S (p1, q1) and S (p2, q2) do not intersect. "A section without a section in the forward direction is called the head of the section series, and a section without a section in the backward direction. The section series of TA
Call it il. However, in the case of a loop, an appropriate adjacent cross section is selected, and the head and tail are respectively selected. The direction of the section series is
It simply inherits the direction of the cross section that was initially selected, and is not necessary.

The cross-sectional series whose identification number is n is Ln, and the head of Ln is head (L
n), let tail of Ln be tail (Ln).

One section belongs to at most one section series. Similarly,
One boundary point belongs to at most one section series.

Unknown area; boundary points that do not belong to the cross-sectional series and the head or
A region that is surrounded by the tail and does not include a cross-sectional series inside (for example, the portion (b) in FIG. 3).

Hereinafter, the procedure of the sectional sequence graph extraction processing will be described with reference to FIG.

Extraction of All Boundary Points and All Boundary Point Sequences (Process 1) First, with respect to the line figure image data in the pattern memory 4, the boundary points are tracked so that black pixels are seen to the left, and all the boundary point sequences PSi (i = 0... M) and all boundary points pij (j
= 0... Ni), and for each boundary point, its boundary extension point is determined. The extracted all boundary points and all boundary point sequences are stored in the data memory 5.

Initial Cross Section Extraction (Process 2) A cross section orthogonal to the direction of the line segment is obtained as follows, focusing on the relationship between the extracted boundary point and its extended boundary point.

Assuming that the target boundary point among the boundary points pij (i = 0... M, j = 0... Ni) is p, if p satisfies one of the following conditions 1 to 3, the cross section S (p, pair ( p)) is registered. Registration is performed using the data memory 5. Where S (pair (p), p)
If is already registered, do not register.

1.pair (pair (p)) = p 2.ang (p) and ang (pair (p)) almost face each other.

3. The number of boundary points between pair (pair (p)) and p is less than or equal to a certain number.

Through the above processing, the cross section of the portion of the line graphic having a small curvature can be obtained.

Furthermore, in order to obtain a cross section at a portion with a large curvature,
Regarding p that satisfies any of the above 1 to 3, the target boundary point when tracking in the tracking direction (and the reverse direction) from p is p ′
Then, as long as pair (p ') and pair (p) are the same, tracking of p' is continued, and S (p ', pair (p)) is registered. At this time, if S (pair (p), p ') has already been registered, registration is not performed. This process is just a pair
This can be said to be a process of obtaining a cross section as if a circle is drawn with (p ') as the center.

Extraction of Cross Section Series (Process 3) When cross sections are initially determined, a cross section series is obtained by associating them before and after. The processing here is briefly described as follows. "The processing of extracting an unprocessed cross section, selecting an appropriate one from the unprocessed cross sections adjacent thereto, and relating them to the front and rear, and the adjacent cross section does not exist It can be said that all cross-sectional series are extracted by repeating the processing of “Perform until”. Whether a section is untreated
The determination is made based on the label of the boundary point belonging to the cross section (this represents the identification number of the cross section series).

The details of the processing are as follows. The symbols are defined as follows.

LBL (p): label of boundary point p. Indicates the identification number of the section series.

c: a counter indicating a cross-section series identification symbol.

CNG (): Operator whose cross section direction is reversed. That is
CNG (S (q, p)) = S (q, p) SS; a variable used for convenience of explanation;
Represents a cross section.

STEP1: “Initial setting” LBL (pij) ← −1 (i = 0... M, j = 0... Ni) c ← −1 STEP2: Take out one cross section (end if not taken out). Assuming that the extracted cross section is S (q, p), LBL (p) ≧ 0 or
If LBL (q) ≧ 0, it is considered that the search has already been performed, and S
Go to TEP2, otherwise go to STEP3.

STEP3: increment c.

 The section series Lc having the identification number c is registered.

SSpq ← S (p, q), LBL (p) ← c, LBL (q) ← c.

STEP3-1: "Processing in the forward direction" Let p 'be the boundary point next to p and q' be the boundary point before q (see FIG. 5 (a)). LBL (p ') ≧ 0 and LBL
If (q ') ≥ 0, head (Lc) ← SSpq as STEP3
Go to -2. Otherwise, SSp′q ′, SSp′q, SSp
q ′ is obtained as follows.

SSp′q ′ ← S (p ′, q ′): S (q ′, p ′) exists, and LBL (p ′) <0 and LBL (q ′) <0 ← CNG (S (q ′, p ')): S (q', p ') exists, LBL (p') <0 and LBL (q ') <0 ← NULL: In other cases, SSp′q ← S (p ′, q): S (p ', q) exists, LBL (p') <0 ← CNG (S (q, p ')): S (q, p') exists, and LBL (p ') <0 ← NULL : Other than the above: SSpq ′ ← S (p, q ′): S (p, q ′) exists, and LBL (q ′) <0 ← CNG (S (q ′, p): S (q ′, pBL exists, and LBL (q ') <0 ← NULL: In cases other than the above, after SSp'q', SSp'q, and SSpq 'are obtained, the following cases (a) to (e) are followed.

(A) SSp'q '≠ NULL (see FIG. 5 (b)) SSpq and SSp'q' are related to each other.

 Let p ← p ′, q ← q ′, SSpq ← SSp′q ′.

(B) SSp'q '= NULL and SSp'q ≠ NULL and SSp'
q ≠ NULL (see FIG. 5 (b)) S (p ′, q ′) is registered and related to SSpq.

 Let p ← p ′, q ← q ′, SSpq ← S (p ′, q ′).

(C) SSp'q '= NULL and SSpq' ≠ NULL and SSp'q
= NULL (see FIG. 5 (c)) SSpq and SSpq 'are related to each other.

 Let q ← q ′, SSpq ← SSpq ′.

(D) SSp'q '= NULL and SSpq' = NULL and SSp'q
≠ NULL (See FIG. 5 (d)) SSpq and SSpq ′ are related to each other.

 Let p ← p ′ and SSpq ← SSp′q.

(E) SSp'q '= NULL and SSpq' = NULL and SSp'q
= NULL (See FIG. 5 (e)) Head (Lc) ← SSpq and go to STEP3-2.

Return to STEP 3-1 by setting LBL (p) ← c and LBL (q) ← c.

STEP 3-2: “Processing in the backward direction” The description is omitted because it is the same as in STEP 3-1. At the time when the series can no longer be assigned, tail (Lc) ← SSpq,
Return to STEP2 to obtain the next section series.

Through the above processing, all the cross-sectional series are initially obtained for the target line figure. In STEP 2, a slight difference may appear in the processing result depending on which unprocessed cross section is selected first. However, if the conditions 1 to 3 in the processing 2 are sufficiently strict, there is no significant difference.

Deletion of Unnecessary Section Series (Process 4) In the section series obtained in Process 2, there is a possibility that sets intersecting each other exist as shown in FIG. for that reason,
In search of such a pair, one needs to be deleted. By using the fact that the cross-section series is a series of cross-sections without gaps, when determining the intersection, there is no need to check the possibility of intersection of all the cross-sections included in the series, just select one cross-section Good.

It is checked whether there is any crossing between the obtained cross-sectional series. If there is a crossing, the cross-sectional series with the larger width (average of the distance between boundary points belonging to the cross-section) is included in that. After the label of the boundary point to be cleared is cleared (set to -1), the section series is deleted from the registration.

It should be noted that a section series that does not have a sufficient number of sections for the width of the section series may be deleted at this time.

Extension / integration of cross-section series (Process 5) When the influence of noise is large, as shown in FIG. 7 (a), the cross-section series does not reach the end of the line sufficiently, or one line segment has a plurality of cross-sections. May be divided into affiliates. Therefore, as shown in FIG. 7 (b), it is necessary to extend the cross-sectional series or to form a cross-section between the cross-sectional series and then integrate a plurality of cross-sectional series into one cross-sectional series. Become.

First, the integration process will be described. The section series is taken out one by one, and the following processing is performed in each of the front and rear directions. Here, the extracted cross-sectional series is defined as Ln.

(I) Cross-section creation processing between cross-section series (forward) Nothing is performed if head (Ln) has already been processed. If not processed, it is assumed that head (Ln) has been processed.
Then, the boundary point is traced forward from S (p, q) ← head (Ln) pp ← p, the boundary point is traced backward from the first boundary point qq ← q whose label is not −1, and the label is traced. Find the first boundary point that is not -1. Here, if pp = q, a cross-sectional series extension process described later is performed. On the other hand, if LBL (pp) ≠ LBL (qq), there is a branch or an intersection, so nothing is done in such a case.

Next, let LBL (pp) be n '. And head (L
n ′) = S (qq, pp) (or tail (Ln ′) = S (pp, q
q)), it is assumed that this has been processed, and head (L
Since there is a possibility that n) and head (Ln ′) (or tail (Ln ′)) can be integrated, a cross section is created between the cross-sectional series in a manner similar to STEP 3-1.

In STEP3-1, the sequence was determined based on whether or not a cross section exists. However, since it is known that there is no cross section, SSp'q ', SSpq', and SSp'q were temporarily created. Then, a cross section that minimizes a certain evaluation function is adopted.

The evaluation function may be made smaller as it faces the inclination direction of two boundary points included in the cross section, and becomes smaller as its distance becomes shorter. If the smallest evaluation value exceeds a certain threshold, integration is not performed.

(Ii) Integration of cross-section series (forward direction) After the space between cross-section series is filled, if necessary, the direction of one cross-section series is reversed, and then the cross-section series is integrated.

The cross-section creation processing and the integration of the cross-section series between the cross-section series in the rear direction are the same as the processing in the front direction, and a description thereof will be omitted.

Next, the processing for extending the cross-section series may be performed in the same manner as the above-described cross-section creation processing until the cross-section series is sufficiently close to the line end.

Extraction of unknown region and reference to cross-sectional series (process 6) A part that has become a cross-sectional sequence is determined, and a process of setting this as an unknown region is performed.

If there is no cross-sectional series in the boundary point sequence PSi (i = 0... M), this is regarded as one unknown region.

If there is a cross-sectional series, proceed as follows. P, pp, q between the cross-section series that have not been integrated in advance in the section series extension / integration processing (processing 5)
And qq can be referred to each other. These boundary points (that is, boundary points belonging to the section series “hea” or “tail”) detect a part forming a loop and make it an unknown area, and the section series connected to the unknown area is a head section.
It is clarified whether the connection is made by the connection or the tail, and the graph structure consisting of the section series and the unknown region is obtained.

By the above processing 1 to processing 6, a graph structure (cross-section sequence graph) can be extracted from the line graphic image data in which the nodes represent the cross-section series or the unknown area, and the arcs represent the connection relation between them (FIG. 3). reference).

Correction of cross-section series graph (Process 7) The cross-section series obtained by the above-mentioned processes is unnecessary as shown in FIG. 8 (a) because of the cross-section series crossing the inside of the intersection and the influence of small projections and holes. In some cases. In such a case, it is necessary to remove these as shown in FIG. 8 (a) to correct the sectional sequence graph.

First, with respect to all the unknown regions, the representative points thereof (that is, points which can be smoothly connected when skeleton lines are extracted from the connected sectional series) are obtained as follows.

Let K be the number of section series adjacent to the unknown area. K is 1
If it is below, there is no representative point. When K is 2 or more, the sectional series connected to the unknown region of interest is represented by Lk (k
= K ₁ ... k _k ).

An evaluation value F defined by the following equation is obtained for black pixels in the unknown area (including the boundary points).

F = min (cos (θk)) (k = k ₁ ... K _k ) θk; angle Vk between vector Vk and vector Wk; vector representing the traveling direction near the unknown area of interest of cross-sectional series Lk (unknown Wk; head (Lk) (If tail is connected to unknown area, tai
l (Lk)), the vector evaluation value F having the starting point at the center and the ending point at the black pixel of interest is maximum when the black pixel of interest is on an extension of all cross-sectional series, so that F is maximized. A black pixel to be obtained is determined, and this is set as a point representing the unknown area (see FIG. 9).

Next, for all sectional series Li (i = 1... N), Li
The length len (Li) and the width wid (Li) of Li are: len (Li) = (number of cross-sections included in Li) + (distance from the center of head (Li) to the representative point of the unknown area adjacent to head ) + (Distance from the center of tail (Li) to the representative point of the unknown area adjacent to tail) wid (Li) = (average distance between two boundary points of the cross section included in Li) If (Li) is not long enough for wid (Li), it is assumed that Li is not a line segment,
The node representing the section series Li is removed from the section series graph.

The cross-section series graph is completed by the above processes 1 to 7. Here, a process of obtaining a skeleton line graph (core line) based on the sectional series graph will be briefly described.

In the sectional sequence graph, as is apparent from the work process, the nodes representing the sectional sequence are not adjacent to each other, and similarly, the nodes representing the unknown regions are not adjacent to each other (see FIG. 3).

For a series of cross sections, a skeleton line can be easily obtained by selecting some cross sections in addition to the head and tail, and forming and connecting skeleton points having the coordinates of the center.

On the other hand, since the representative point has already been obtained in process 6 for the unknown region, a skeleton line can be obtained by performing the following process according to the number of connections (this is K).

(A) When K = 0 It is considered to represent a surface area. Although it depends on the processing object and purpose, basically, if the area is sufficiently small, one skeleton point is created, otherwise, the surface area is left as it is.

(B) When K = 1 Since nothing can be considered as representing a line end, nothing is done.

(C) When K> 1 It can be considered to represent any one of severely bent, branched, and crossed. A skeleton point having the coordinates of the representative point of the unknown region is created, and connected to the skeleton point at the end (unknown region side) of the adjacent cross-sectional series Lk (k = k ₁ ... K _k ).

In this manner, the skeleton line graph can be obtained from the cross-sectional series graph, but the roughness of the skeleton line graph can be easily changed by changing the interval for selecting the cross section.

In the skeletal line extraction processing described above, the unknown area is much smaller than when the image is divided into rectangles or the like, and the direction of the line segment near the unknown area can be accurately determined from the sectional series. A correct skeleton line can be obtained even in the vicinity of a bend / branch / intersection regardless of the above.

Also, we will not mention specific methods here,
It is also easy to obtain statistical features (statistics of the inclination direction of the boundary point, etc.) locally (between arbitrary cross sections in an arbitrary cross section series).

〔The invention's effect〕

As apparent from the above description, the present invention has the following effects.

(1) By extracting a cross-sectional sequence graph, rough structure matching can be performed with this cross-sectional sequence graph,
Since the roughness of the skeletal line graph can be locally changed and statistical characteristics can be easily obtained, it contributes to the efficiency of the recognition processing and the improvement of the recognition rate. Further, since the cross-sectional series graph has reproducibility, it can be used not only for recognizing line figures but also for applications such as editing, transmission, storage / retrieval.

(2) Since the cross section is formed so as to be substantially perpendicular to the direction of the line segment, it is possible to extract features independent of the line width.

(3) Since the orthogonality of the cross section and the line segment is not so much lost, and the sequences are ordered as such that there is no inconsistency in the context, the skeleton graph can be easily extracted.

(4) By integrating the divided cross-sectional series, unnecessary unknown regions do not appear, which contributes to processing efficiency, and at the same time, reduces the influence of noise, thereby contributing to an improvement in recognition rate.

(5) By extending the cross-sectional series until it approaches the line end, a correct skeleton line graph can be obtained even near the line end, which contributes to an improvement in recognition rate.

(6) Compared to the case where the image is divided into rectangles or the like, the unknown area is extremely small, and the direction of the line segment near the unknown area can be accurately determined from the sectional series. , An accurate skeleton line graph can be obtained, which contributes to an improvement in the recognition rate.

(7) The cross-sectional sequence graph is corrected, so that a short cross-sectional sequence formed inside an intersection and a short cross-sectional sequence surrounding a small hole region are deleted. Therefore, a cross-sectional sequence graph is correctly obtained, and thus a skeleton line graph is also correctly obtained. This contributes to the improvement of the recognition rate.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of a hardware configuration for implementing the present invention, FIG. 2 is a flowchart of a sectional sequence graph extraction process according to the present invention, and FIG. 3 is a relationship between original image data, a sectional sequence graph, and a skeleton line. 4, FIG. 4 is an explanatory view of the extension boundary point, FIG. 5 is an explanatory view of the cross-section series extraction processing, FIG. 6 is a view showing an example of intersection of the cross-section series, and FIG. FIG. 8 is a diagram showing an example of integration, FIG. 8 is a diagram showing an example of correction of a sectional sequence graph, and FIG. 9 is a diagram showing how to find a point representing an unknown area. 1 central processing unit, 2 program memory, 21
Section sequence graph extraction processing program, 3 ... image scanner, 4 ... pattern memory, 5 ... data memory.

Claims (5)

(57) [Claims] 1. A step of extracting a cross section substantially orthogonal to the direction of a line segment from line graphic image data, a step of obtaining a series of the cross sections to form a cross section series, and setting an area other than the cross section series to an unknown area. A node representing either the cross-sectional series or the unknown region, and an arc generating a cross-sectional series graph representing a connection relationship between them, and a method for extracting a line figure feature using the cross-sectional series graph as a feature of the line figure . 2. The method according to claim 1, wherein, when a set of cross sections exists in the cross-sectional series, the cross-sectional series having a larger width is deleted. 3. The method according to claim 1, wherein an evaluation function is determined between the two cross-sectional series and cross-section generation processing is repeated to integrate the two cross-sectional series into one cross-sectional series. Extraction method of line figure. 4. The method according to claim 1, wherein when the end of the series of sections corresponds to a line end, an evaluation function is determined and the processing for generating a cross section is repeated until the end is sufficiently close to the line end to extend the cross section. 1) A method for extracting features of a line figure described. 5. After defining a point representative of all unknown regions in the created cross-sectional series graph,
As the length of each cross-section series, the sum of the number of cross-sections included in the cross-section series and the distance from the end of the cross-section series to the representative point of the adjacent unknown area is obtained. And extracting a node corresponding to the section sequence from the section sequence graph if the size is not sufficiently large. Method.