JP3031594B2 - Pattern recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a so-called pattern recognition apparatus, and more particularly to a polygon tableware recognition apparatus in a cafeteria settlement system.

[0002] Conventionally, digital image processing technology has been developed with the development of computer devices. One field of image processing is pattern recognition, which processes an image input by a camera or the like by various methods and recognizes it as a specific shape.The problem is how to accurately extract the target, high speed,
Also, was it possible to process a large amount of data?

[0003] The cafeteria settlement system according to the present invention relates to a typical application of this pattern recognition technology. The content of the system is in a popular cafeteria typified by an employee cafeteria, where various types of dishes or single dishes are served. Tableware of various sizes and shapes is placed on a tray, the image is captured by a camera, and image processing is performed.
In the end, the system compares the size and shape of the tableware with each tableware set in advance, and calculates the settlement amount based on the price set for the tableware. In such a case, it is required to accurately and quickly separate and identify the image of the overlapping or protruding tableware.

[0004]

2. Description of the Related Art Conventionally, as a feature extraction technique of an input figure, which is a pattern recognition technique relating to a canteen settlement system, a density gradient for an input image is obtained, and a contour of the figure is extracted from its absolute value. There is a method of recognizing the features of the contour line extracted by the direction vector.

The density gradient is a rate of change in image density at a boundary line such as a line segment. When this change is sharp, a human recognizes the boundary line (edge), and the boundary line is digitally extracted. Needed for This procedure is basically a differential operation. FIGS. 17 and 18 show a conventional example to which this technique is applied.

FIG. 17 shows a technique for detecting the inclination of a polygon based on a density gradient disclosed in Japanese Patent Application Laid-Open No. Sho 59-180649. In the digital image processing, an image divided by a square is usually considered, and so-called binarization in which a pixel corresponding to the square is “1” or “0” is performed. By binarizing, the area can be divided and the outline can be extracted from the background. Whether the areas are connected or separated is determined by whether or not the values of the adjacent pixels are the same. Logical filtering is effective and frequently used as conversion for area division of the binary image. In this method, a comparison operation is performed between a pattern of a small area such as 3 pixels × 3 pixels and a pattern of a filter, and is converted into a predetermined value depending on whether it is true (“1”) or false (“0”). .

First, as the preprocessing, the input graphic E shown in FIG. 17A is subjected to the above-described 3 × 3 spatial filtering to obtain a density gradient vector. When the density gradient vector is obtained, a density boundary can be extracted from its absolute value. Of the outline.

Next, as shown in FIG. 17B, a histogram is obtained in which the angle obtained from the density gradient vector is taken in the horizontal axis direction and the pixel number distribution of each vector is taken in the vertical axis direction. , EL2, and EL3 are in the same direction α1, α2, and α3, respectively, so that a peak appears for each straight line and the features of the outline can be recognized.

Next, a triangular wave from which features are extracted in order to know the inclination of the figure is created in advance as shown in FIG. 17 (c), and the peak position of the extracted feature data is shifted and compared for similarity. Use the detection method. In FIG. 17C, by shifting by the angle θ, it can be determined that the feature data is substantially similar to the feature data shown in FIG. 17B.

FIG. 18 shows another conventional pattern recognition technique using a density gradient, which is a conventional example disclosed in Japanese Patent Application Laid-Open No. 1-134578. According to this technology, an object is to extract data by calculating a density gradient of an input image and extract a unit line segment from the data of the contour to compress data to a minimum feature data for pattern recognition. And Therefore, when tracking a contour point with gradient vector information, the straight line distance from the line segment start point to the search point exceeds a predetermined unit length, and the gradient angle of the search point is set to the line segment start point. When the amount of change exceeds a predetermined amount for each section of the gradient, the contour point is set as a line segment end point. Subsequently, this line segment end point is set as the next line segment start point, and a line segment is determined in the same manner.

FIG. 18 shows a part of the contour points accompanied by the gradient vector information extracted by the density gradient calculation. Ps (1), Pi, Pe (1) indicate sampling points on the contour line, and Gs (1), Gi, Ge (1) indicate gradient angles at the respective sampling points. With the sampling point Ps (1) as the start point of the line segment, the linear distance Li with respect to the sampling point Pi exceeds a predetermined unit length ΔL, and at the same time, the gradient angle Gi with respect to the gradient angle Gs (1) exceeds a predetermined change amount. In this case, the sampling point Pe (1) at that time is set as the line segment end point, and the line segment L (1) is determined. Further, the sampling point Pe (1) is set as the line segment start point, and Ge (1) is set as the next reference gradient angle, and the next line segment L is set.
(2) is determined. Hereinafter, this procedure is repeated to specify the figure.

As described above, even if the curvature of the contour is large, no minute line segment is generated in the calculation, and the feature can be accurately extracted.

[0013]

However, in the above-described feature extraction method based on the density gradient vector, a large number of processing procedures are necessary for performing a differential operation on all image data, and the processing speed is small in a small-scale system. There is a problem that it takes a long time, and an arithmetic unit having a high processing capability is required to improve the processing speed.

Further, in the pattern recognition method using pattern matching used for identifying a figure based on the features of the contour of the extracted figure, it is necessary to register in advance the characteristics and size of the target figure to be recognized. Also,
When comparing the registered information with the actual input figures,
Since a process of correcting distortion in the input stage is required, there is a problem that a device having higher operation processing capability is required.

In view of the above problems, the present invention is applicable to a small-scale image processing system which is capable of reducing distortions at the input stage and variations in the shape of a pattern recognition target object with only minimal data registration. An object of the present invention is to provide a pattern recognition device that can flexibly and quickly identify and measure.

[0016]

In order to achieve the above object, the present invention as defined in claim 1 is characterized in that the image is separated from an image of an original figure.
Polygon discriminating the type of polygon using the straight line
In a pattern recognition device having another means (6),
The polygon discriminating means (6) determines whether an intersection of a plurality of the straight line portions is present.
Of the vertex to be derived and the two
Vertex generator that outputs the angle formed by the line part, and any interior angle
If the input condition is, the arbitrary interior angle is a specific polygon.
Function characteristic that outputs an evaluation value indicating the degree of possible week
And inputting the angle with respect to the vertex to evaluate
A function operation unit that outputs a value, and all functions before opening to the vertex
A set generation unit that generates a set that aggregates the evaluation values,
A plurality of the specific polygons are obtained from the evaluation value of the set.
The representative value that represents the possibility of
A representative value setting unit to be assigned to a specific polygon;
Before the highest value among the representative values calculated by the table value setting unit
Select the specific polygon corresponding to the representative value and select the original figure
An inference discriminator that infers as the type of polygon related to the shape
It is characterized by doing.

The invention according to claim 2 is the invention according to claim 1.
The outline information extracted from the original graphic image
Pertaining to the straight line portion information on the report and the determined original figure
Determine the size and position using the type of polygon,
Vertex coordinates and discrimination determined from the straight line portion information
The center coordinates of the original figure and the type of the polygon
And a figure determining unit for calculating the diameter of a circumcircle of the original figure.
It is characterized in that it is provided with a figure judging means for performing the operation.

The invention according to claim 3 is the invention according to claim 2.
In the above invention, the figure determination means (6) is
The outline figure information related to the polygon is missing and the original figure is
If it is insufficient to make a determination, a plurality of said linear sub-
The length of one side of the original figure is calculated based on the
Using the interior angle value determined from the type and the length of the one side
Having a vertex restoring unit for restoring the dropped vertex coordinates
It is characterized by.

The invention according to claim 4 is the invention according to claim 2.
In the invention, the figure determining means (6) includes:
All the center coordinates and the diameter of the circumscribed circle for the original figure
The contour line information is calculated and compared with each other,
Both the center coordinates and the diameter of the circumscribed circle are below a predetermined tolerance.
If it is below, judge it as the same figure and integrate it,
It has a figure comparison unit that separates and fixes the original figure.
I do.

[0020]

[0021]

According to the first aspect, straight lines can be separated by the following procedure. First, the straight line separating means (6) stores the outline coordinates as shown in FIG. 11 extracted from the recognized image of the dish or the like in the outline coordinate storage memory. On the other hand, the sampling unit performs sampling at an appropriate interval on the contour, and the vector generation unit performs the vector (FIG. 12: V11, V
12, V21, V22), vectors are set for the preceding and following sampling points. Then, the curvature (θ) formed by these vectors (V11 and V12, V21 and V22)
1, θ2) is obtained by the curvature calculation unit, these curvatures are quantified (FIG. 13B), and based on the quantified curvatures, the frequency distribution generation unit generates a frequency distribution as shown in FIG. 13C. Generate. When the straight line separating section separates the sampling points by a certain threshold value, a set of continuous points equal to or smaller than the divided threshold value can be separated as a straight line portion. For example, if straight line separation is performed on a part of the input pattern as shown in FIG. 13A, three straight lines as shown in FIG. 13D can be separated.

According to the second aspect, in the polygon discriminating means (6), the vertex generating section generates a virtual vertex from the straight line separated by the arbitrary straight line separating means. This can be easily obtained from the intersection by extending the line segment of the straight line portion (FIG. 14A). At the same time, the angle corresponding to the interior angle can be calculated from the two line segments related to the virtual vertex. This angle is input to a function operation unit for quantifying the degree of a possible polygon of a virtual vertex based on fuzzy theory, and an evaluation value shown in FIG. 14B is derived. The evaluation values are totaled by the set generation unit, and a relation diagram as shown in FIG. 14B is completed. This shows numerically the degree to which the angle related to each virtual vertex is each polygon in the left column of the figure.

A representative value setting unit sets a representative value for this set of evaluation values, and indicates, by a numeral, which polygon is the figure formed by the outline, as shown in FIG. 14C. Is generated. The type of polygon having the highest numerical value among them can be determined by the inference determining unit as the polygon of the figure (FIG. 14).
In the above, an octagon is determined).

According to the third aspect, the figure determining means (6)
Determines the size, position, and type of the figure on the original image comprehensively based on the known polygon figure. When the entire shape can be assumed geometrically based on the virtual vertex coordinates generated by the polygon discriminating means and the type of polygon, the figure determining unit generates arbitrary diagonal (Ln) information as shown in FIG. The coordinates of the intermediate point are calculated. Thereby, all of the shape, size, and position of the figure are determined.

However, if the original figures overlap as shown in FIG. 16, there is insufficient contour information necessary for determining the size. Therefore, according to the fourth aspect, the vertex generation unit obtains the length of the obtained straight line portion, further generates a theoretical vertex from the determined polygon type, and obtains the interior angle value, By generating a vector obtained from the two, the straight line portion is supplemented as shown in FIG. 16B, and the size can be determined.

However, even if an individual figure is determined,
In some cases, two figures are recognized as two figures by dividing the outline. Thus, according to the fifth aspect, the graphic comparing unit compares the position and size of the center point of the recognized graphic with each other, and when both are approximate values, the two are obtained from the same original graphic. Thus, the figures of all the original images can be specified.

[0027]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of the pattern recognition apparatus according to the present invention will be described with reference to FIGS. In the present embodiment, the pattern recognition device according to the present invention is applied to a cafeteria settlement system. Configuration in the present embodiment First, a typical configuration in the embodiment of the present invention is shown in the block diagram of FIG.

As shown in FIG. 1, the configuration of the present embodiment employs an illumination device 2 in which a shape pattern 1 to be inspected is enhanced with a shadow (contrast) so as to be suitable for image recognition.
An image input device 3 for converting an image of the shape pattern 1 with contrast into a captured video signal, an image processing device 4 for converting the captured video signal from analog to digital, and processing input image information And a CPU (Central processing unit) for executing various processing or recognition processing of the input image information.
nit) 6 and shape pattern 1 extracted from input image information
A contour coordinate storage memory 7 for storing contour line information, and a straight line information storage memory 8 for storing data of a straight line portion among contour line information extracted based on a difference in curvature from the contour line information.
And a fuzzy data storage memory 9 for calculating an interior angle formed by the shape pattern 1 from the straight line information and storing a fuzzy set of each polygon, and a result storage memory 10 for storing a result of recognition of the contour shape. .

As a target shape pattern 1 of the present embodiment, an image of a substantially regular polygonal dish (FIG. 11) is considered.
Specifically, in a cafeteria or the like, the user first selects a plurality of dishes on which his or her favorite items (foods) are served, places them on their trays, and places them in an image reading place for settlement. Here, as an example, it is assumed that three types of dishes A, B, and C are partially overlapped on a tray. An illumination device 2 is provided at the reading place, and a plate on the tray is clearly illuminated, and an image input device 3 (camera or the like) provided on the upper side captures an image of the user's tray, and the pattern recognition device of the present invention. Send to the body.

The problem here is that the dishes usually overlap each other on the tray and are not always input in a completely separated state as shown in FIG. The following is a description including the means for solving this problem. Operation in the present embodiment The operation of the present embodiment will be described with reference to FIG.

In FIG. 2, the input image is first converted into a digital signal by the image processing device 4, and the digital image information is temporarily stored in the input image memory 5 (processing state: step 10). The image information is subjected to contour extraction image processing. Further, in order to separate the overlapping dish images into figures for each dish, contour coordinate data is extracted by a method such as a contour tracing method, and the contour coordinates are extracted. It is stored in the storage memory 7 (step 11). This image data is formed of a set of contour lines, for example, as shown in FIG. In the present embodiment, it is assumed that three dishes (A, B, C) are extracted in a partially overlapping state as shown in the figure.
The contour coordinates extracted in this way are stored in the contour coordinate storage memory 7 and used appropriately in the subsequent processing.
In FIG. 3, the contours thus extracted are i for polygon A, j and k for polygon B, and polygon C
M.

Here, the contour tracing method will be described below. First, this method is based on the premise that the shapes of the target dishes are all regular polygons. The device determines the overlap of A and B or B and C in FIG. 3 and checks which dish the contour is of. That is, in a polygon, the interior angle formed by the two sides is always bent in one direction inward. However, when the polygon overlaps another polygon, the angle between the side on one figure and the side on the other figure at the overlapping point is opposite to the direction of the bending of the interior angle. It will face outward. Therefore, by examining the change in direction along the contour, it can be determined that the figures are overlapping where the contour is bent outward,
At that point, the figure can be cut off and treated as another figure.

Now, the stored partial contours i, j, k,
The contour coordinates are taken at equal intervals with respect to m, sampling points SP1 and SP2 are set therein, and vectors are created therefrom toward contour coordinates which are sampling points before and after the sampling point (V11, V12, V21). , V2
2). Using the angles .theta.1 and .theta.2 formed at each sampling point by the two vectors associated with that point, the curvature at the sampling point such that the features near the vertex (SP1) and the straight line portion (SP2) appear remarkably. (FIG. 2: Step 12). This feature is that, for example, θ1 becomes the smallest in the vicinity of the vertex,
Also, this is a point where θ2 increases in the straight line portion.

Sampling is performed at all points on the contour line, and the obtained curvatures are totaled, as shown in FIG. 4A. The characteristic is that the curvature is large near the vertex and small in the straight line portion. Comes out remarkably. A frequency distribution is created based on the relationship between the curvature data and the frequency of appearance (step 13). This is obtained as shown in FIG. 4 (b), and an area having no frequency occurs in the figure. With this part as a boundary, as shown in the figure, it is divided into the vicinity of the vertex and a straight line part, so that outline information of this straight line part is created and stored in the straight line information storage memory 8 (step 14).

If the above processing is performed on the partial contour i in FIG. 3, the straight lines L1 to L8 are separated. Next, at the intersections of these linear portions or where the graphics overlap and are missing, the intersections generated by the extension of the front and rear linear portions, and a1 to a7 are the vertices of the graphic formed by the partial contour i (hereinafter referred to as virtual vertices). ). Then, an interior angle related to each virtual vertex is obtained (FIG. 2: Step 1).
5).

Next, a fuzzy set serving as a basis for determining a polygon is created using the membership function shown in FIG. 5 (FIG. 2: step 16). This is based on the inference method, which is a technique for fuzzy control that is highly useful as an application of fuzzy theory. For a given input value, the result obtained is represented by a membership function that is a relational expression of “ambiguity”. Here, a triangular function that is easy to calculate is selected, and inference by discrete fuzzy variables, that is, a method of obtaining the most probable polygon for a certain interior angle. This function is used to create a fuzzy set for determining the shape of a figure related to the interior angle with respect to the interior angle, using the interior angle as an input condition.

FIG. 6 shows how the above-mentioned weighting is performed. As shown in FIG. 6A, the membership functions F1, F2, and F3 are functions of the likelihood of being different polygons. Virtual vertices a1, a2,
When the angle of the inner angle of a3 is input, the output data (referred to as grade) of the membership functions F1, F2, and F3 is
Each interior angle is a ratio belonging to each polygon. When this is developed as a relation diagram, it becomes as shown in FIG. 6B, which is a fuzzy set to be obtained.

When such a procedure is performed on the partial contour i as an input condition, a fuzzy set as shown in FIG. 7A is obtained. This is stored in the fuzzy data storage memory 9.

Next, the minimum value of the degree of conformity (grade) of each polygon is extracted as a representative value, and a representative value diagram as shown in FIG. 7B is created. Although there is variation in the angles of the interior angles, it is possible to determine that the polygon to which all interior angles belong is the main pair of contours, and FIG.
7B can be determined to be an "octagon" from FIG. 7B (FIG. 2: Step 17).

The above processing is performed by using other partial contours j,
When k and m are performed, for example, j can be determined as “hexagon”, k can be determined as “hexagon”, and m can be determined as “hexagon”. Next, the size of the figure is determined for the identified partial contour. The size of the figure is calculated by using the virtual vertex coordinates obtained from the straight line portion separation of the outline information and drawing the diagonal line that is most appropriate for obtaining the size and center coordinates from the characteristics according to the type of polygon. . Here, regarding the partial contours i, j, and k in FIG. 3, since all the interior angles are regular polygons with even-numbered angles, only one diagonal line passes through the center coordinates of the figure (FIG. 8). Therefore, if the number of recognized vertices is [(theoretical number of vertices / 2 + 1)], the entire size can be selected. When each partial contour is examined, i becomes 7 virtual vertices. Since the vertices are recognized, the size can be determined, but (6 (square) / 2 + 1 = 4 <
7) The recognized virtual vertices of j and k are each 2
Therefore, the size cannot be determined (FIG. 2: step 18).

Here, when the number of vertices cannot determine the size of the figure, it is necessary to obtain the coordinates of the vertices not obtained as data. As shown in FIG. 9, when the number of vertex coordinates is insufficient, the number of recognized vertices is two.
If it is, the length of the side of the figure to be calculated with the distance between the vertices as one side can be calculated. If the vector is formed using this and the theoretical value of the interior angle obtained from the identified polygon type, the insufficient vertex coordinates can be supplemented (FIG. 2: step 19). That is, in FIG. 9, there are linear portions Lb1, Lb2, and Lb3, and virtual vertices Ab1, Ab
It is assumed that 2 has been calculated. Since the determination result of the type of polygon is a hexagon, the distance between Ab1 and Ab2 is set to the length of one side, and a vector Vb1 having a hexagonal internal angle θ (120 ° in this case) with respect to Lb2 from Ab2 as a viewpoint. The coordinates of the vertices Ab3 and Ab4 are obtained by creating the vector Vb2 in the same manner. Therefore, the center coordinates and the size can be determined from the diagonal lines Ab1 to Ab4.

According to the above procedure, the coordinates of the missing virtual vertex are obtained, the diagonal is drawn based on the coordinates, the center coordinates of the figure are obtained, and the circumscribed circle of the polygon having the coordinates as the center and the length of the diagonal as the diameter is obtained. The size of the figure is set (FIG. 2: Step 20). As described above, a polygon can be obtained for each of the partial contours in FIG.

However, when the dishes are superimposed, there is a disadvantage that the outline of one figure is divided into a plurality of pieces and recognized and discriminated as separate figures. In other words, the outline is recognized as two polygons as a result even though the partial outlines j and k are generated from the same figure.

Therefore, the center coordinates and the size of the recognized polygon are compared, and the center coordinates (x1, y) of a certain figure 1 within a preset range, for example, are recognized.
If 1) is less than or equal to a certain value Dc with respect to the center coordinates (x2, y2) of another figure 2 and if the difference between the diameters of polygons 1 and 2 is less than or equal to α, the figure is the same figure Yes, you can set it.

In this manner, for a group of figures whose sizes and center coordinates have been identified and determined, it is determined in combination whether the size and center coordinates are below an allowable value. Are integrated as one figure, and the final figure can be determined (FIG. 2: step 21).

The type, center coordinates and diameter of the circumscribed flame recognized by the above procedure are stored in the recognition result storage memory 10 (FIG. 2: step 22). The graphic information thus identified and stored is as shown in FIG.

Finally, the settlement device or the like that receives the output of the pattern recognition device of the present embodiment collects a preset price uniquely determined from the size and shape of all the discriminated dishes. Then, the user may be presented with the settlement amount. For example, in FIG. 10, A, B, and C indicate the size and shape (octagon) of the dish of A, 300 yen for “side dish”, 100 yen for “side dish”, and 150 yen for “rice”. If it is a circle,
Since there are no other dishes, the total of 550 yen is settled. Effects of the present embodiment According to the present embodiment, even if the original images of the dishes overlap, the type and size of the dishes can be definitely determined, and the cafeteria settlement system can be constructed by a simple procedure. In particular, the outer shape of the dish does not need to be bent at an accurate angle, and the closest polygon can be specified without fail even if a gentle curve is drawn. Further, the curvature does not need to be common between the plates, and can be determined even with an arbitrary polygonal outer shape. Other Modifications The straight line portion used for the polygon discriminating means does not need to be the above-described straight line separating means, and the straight line portion separated by another method (for example, a polygon inclination detecting method (FIG. 17)). May be used. Similarly, the input of the figure judging means does not need to use the output of the straight line separating means or the polygon judging means, but can be obtained by another method (for example, a contour point tracking method using a density gradient vector (FIG. 18)). The straight line portion and the polygon type may be used as parameters.

[0048]

As described above, according to the first aspect of the present invention, it is not necessary to take a sampling point for each pixel, and a sufficient polygon can be determined by sampling at a large interval. However, the straight portion can be surely separated.

According to the second aspect of the present invention, even if the original image is a figure different from a theoretical polygon such as a rounded shape near a vertex, a reliable figure can be obtained by judging the evaluation value by fuzzy theory. It can be applied to various pattern recognitions by making inferences and changing the membership function that governs the characteristics according to the input object.

According to the third aspect of the present invention, the size and the center position of the figure can be easily specified if the straight line portion data and the type of the figure are given. According to the fourth aspect, even when the original images overlap and only a part of the outline is known, the entire shape can be easily restored.

According to the fifth aspect, even from an original image in which a number of figures overlap, each figure can be distinguished or integrated to determine the figure without erroneous recognition. Applicable to fields requiring high reliability.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a pattern recognition device according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating a flow of a process according to the embodiment of the present invention.

FIG. 3 is a pattern diagram of an input figure in the embodiment of the present invention.

FIG. 4 is an explanatory diagram showing a method of separating a straight line portion from a contour line.

FIG. 5 is an explanatory diagram showing an example of a membership function in the embodiment of the present invention.

FIG. 6 is an explanatory diagram showing a method for creating a fuzzy set.

FIG. 7 is an explanatory diagram showing a method of determining the shape of a contour line.

FIG. 8 is an explanatory diagram showing a diagonal line for determining a size.

FIG. 9 is an explanatory diagram of a method for determining the size of an identified contour line.

FIG. 10 is an explanatory diagram showing an identification result in the example of the present invention.

FIG. 11 is an explanatory diagram showing an example of an input pattern.

FIG. 12 is an explanatory diagram of a method of measuring a curvature of a contour line.

FIG. 13 is an explanatory diagram showing how a straight line is separated from an input pattern.

FIG. 14 is an explanatory diagram of a method of determining a shape by fuzzy sets from straight line information.

FIG. 15 is an explanatory diagram of a method for obtaining the size of an identified shape.

FIG. 16 is an explanatory diagram of a method of supplementing insufficient contour information.

FIG. 17 is a diagram illustrating a conventional technique (1).

FIG. 18 is a diagram illustrating a conventional technique (2).

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Shape pattern 2 ... Lighting 3 ... Image input device 4 ... Image processing device 5 ... Input image memory 6 ... CPU (Central processing unit) 7 ... Contour coordinate storage memory 8 ... Straight line information storage memory 9 ... Fuzzy data storage memory 10 ... Recognition result storage memory A, B, C, E ... input pattern figure i ... contour line related to figure A j, k ... contour line related to figure B m ... contour line related to figure C L1 to L8, Lb1 to Lb6 ... linear part a1 a7, Ab1 to Ab4, N1 to N4, Ann + 1 (1
≤ n ≤ 7) ... virtual vertices SP1, SP2, Ps (1), Pi, Pe (1) ... sampling points V11, V12, V21, V22 ... direction vectors θ, θ1, θ2 ... measured angles F1 to F3 ... Membership function O: found center point L, Ln ... diagonal line EL1 to EL3 ... side α1 to α3 ... density gradient vector Gs (1), Gi, Ge (1) ... gradient angle L for finding size of figure (1), ΔL, Li ... line segment

Continuation of the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06T 7/00 G06T 1/00 G06T 7/60 G06T 5/00

Claims (4)

(57) [Claims] 1. A straight line portion separated from an image of an original figure.
Polygon discriminating means (6) for discriminating the type of polygon using
In the above-described pattern recognition apparatus, the polygon discriminating means (6) includes an intersection of the plurality of linear portions.
The coordinates of the vertex derived from the point and before 2 pertaining to the vertex
Vertex generation unit that outputs the angle formed by the straight line part, and when an arbitrary interior angle is used as an input condition, the arbitrary interior angle is specified
Output an evaluation value that indicates the degree of week that can be a polygon
Inputting the angle with respect to the vertex
A function operation unit that outputs the evaluation value, and a set in which all the evaluation values that open to the vertices are aggregated.
A plurality of specific polygons from the set generation unit to be formed, and the evaluation value related to the set.
The representative value that represents the possibility of
A representative value setting unit to be assigned to the specific polygon, and the most representative value among the representative values calculated by the representative value setting unit.
Select the specific polygon corresponding to the high representative value and select
Inference discriminator that infers as the type of polygon related to the original figure
Pattern recognition apparatus characterized by having, when. 2. The pattern recognition apparatus according to claim 1,
Line information on the contour information extracted from the original figure image.
Using the type of polygon related to the original figure determined
To determine the size and position, and
Vertex coordinates that can be determined more and the determined polygon
Depending on the type, the center coordinates of the original figure and the outside coordinates of the original figure
Figure determining means having a figure determining unit for calculating the diameter of a tangent circle
Pattern recognition apparatus comprising: a. 3. The pattern recognition device according to claim 2,
And the figure determining means (6) relates to the input polygon.
The outline information is missing and it is not sufficient to determine the original figure.
If it is a minute, the original drawing
Calculate the length of one side of the shape and determine from the type of polygon
The vertex locus missing using the angle value and the length of the one side
A pattern recognition device comprising a vertex restoration unit for restoring a target . 4. The pattern recognition apparatus according to claim 2,
And the figure determination means (6) performs a
The center coordinates and the diameter of the circumscribed circle are all the contour information.
Are calculated and compared with each other, the center coordinates and the
If the diameters of the circumscribed circles are both smaller than
Judge as one figure and integrate, and separate and confirm all of the original figures
A pattern recognition apparatus, comprising: