JP2930903B2 - Region extraction method and region extraction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an area extracting method and an area extracting apparatus. More specifically, for example, an environment monitoring system using image data, a detection system that detects a moving area based on a local flow vector obtained by optical flow calculation, and the like, analyze a pattern to be detected from an image, The present invention relates to an area extracting method and an area extracting apparatus for extracting a corresponding partial area.

[0002]

2. Description of the Related Art Conventionally, as an image recognition technique, a template matching method, a recognition method by associating feature points between different spaces (hereinafter referred to as a "feature point associating method") and the like have been known.

[0003] The template matching technique is one of techniques for detecting a geometrically deformed pattern of a specific pattern (second image layer) from one image (first image layer). As shown in (1), a method is provided in which a plurality of geometrically deformed patterns of a certain pattern are prepared in advance as templates, the correlation is obtained while moving the template on the target image, and a portion having a large correlation is detected.

However, in this template matching method, when the number of geometric deformation patterns such as enlargement / reduction, rotation, distortion and the like and the amount of parallel movement on an image become large, the calculation time required for extracting the corresponding area becomes enormous. This makes it difficult to perform partial matching with the template.

On the other hand, the feature point associating method is, as shown in FIG. 14, for each of two spaces of a target image (first space) and a pattern to be detected (second space).
A feature amount based on image information is obtained for each of a plurality of feature points arranged in a space, and a feature amount of a feature point in one space and a feature amount of a feature point in another space are matched. This is a method of correlating feature points between two spaces by calculating for each point. That is, in this method, a local correspondence between two spaces is obtained by comparing local feature amounts.

As an algorithm for determining corresponding points, for example, an algorithm disclosed in "Image Analysis Handbook" (supervised by Mikio Takagi and Hirohisa Shimoda, published by The University of Tokyo, pp. 707-746) is known.

However, when there are a plurality of geometrically deformed patterns of the pattern to be detected in the target image, the degree of the local geometric deformation from the target image is similar due to the correspondence of the local feature points. However, there is a problem that it is difficult to take out a partial area that is present. Such a problem will be described with reference to FIG.

That is, as shown in FIG. 15, an area D1 surrounded by a left frame on the first space is defined by a first relation between corresponding points in the second space of feature points included therein. Is substantially equivalent to the relationship between feature points in the space. On the other hand, as for the relationship between the corresponding points in the second space of the right region D2, it is assumed that the feature points in the original first space are far from the structure formed. In such a case, it is necessary to extract only the portion of the left area D1, and even if there are a plurality of such portions in the first space, they must be able to be extracted. Regardless, such extraction was difficult.

[0009]

As described above, in the feature point associating method, when there are a plurality of geometrically deformed patterns of the pattern to be detected in the target image, the target is determined by local feature point correspondence. There is a problem that it is difficult to extract a partial region having a similar degree of local geometric deformation from an image.

Therefore, the present invention relates to a method based on the similarity of local feature amounts such as a feature point associating method and the like, wherein each of a plurality of feature points arranged in a first space is a In a state where a local point locally corresponds to a point in the second space, a partial region that matches by translation, scaling, rotation, and distortion of the pattern in the second space can be extracted from the first space. The purpose is to: Another object of the present invention is to be able to simultaneously extract a plurality of partial regions even when a plurality of partial regions exist in the first space.

[0011]

SUMMARY OF THE INVENTION The present invention is conceptually
For each feature point arranged in the first space (entire space), an adjacent feature point is connected to form an edge vector, and the second feature point of the two feature points which are the end points of each edge vector is set.
The corresponding points in the space (subspace) are connected to form an edge vector in the second space.

Next, the relationship between the edge vector in the first space and the corresponding edge vector in the second space is represented by a point on the coordinate of the edge transformation parameter characterized by the magnification and the rotation angle of the vector. Is expressed as an edge transformation parameter.

Then, the activity of each edge is determined based on the distribution state of the edge conversion parameters obtained for all the edges, and the edges having an edge activity not less than a certain threshold value are connected to each other. A predetermined area specified by an edge is extracted.

That is, according to the first aspect of the present invention, an entire space in which a plurality of feature points are arranged and a partial space in which corresponding points to the feature points are arranged are provided. A region extraction method for extracting a predetermined region composed of a specific feature point group from the entire space based on the corresponding points obtained, wherein a first edge and a first edge are extracted based on position information of the plurality of feature points. A first edge vector is generated, a second edge and a second edge vector are generated based on the position information of the corresponding point, and an edge is generated based on the first edge vector and the second edge vector. Calculating a conversion parameter, calculating an edge activity based on the edge conversion parameter, comparing the edge activity with a predetermined threshold value, and when the comparison result satisfies a predetermined condition, Based on Tsu di activity, by selecting the corresponding edge to the particular edge vector from the first edge vector, a region extraction method characterized by extracting a predetermined region.

According to the second aspect of the present invention, an entire space in which a plurality of feature points are arranged and a partial space in which corresponding points to the feature points are arranged are provided, and are arranged on the partial space. A region extraction method for extracting a predetermined region composed of a specific feature point group from the entire space based on the corresponding points obtained, wherein a first edge and a first edge are extracted based on position information of the plurality of feature points. Generating a first edge vector and generating a second edge and a second edge vector based on the position information of the corresponding point; and generating a second edge vector and a second edge vector based on the first edge vector and the second edge vector. Calculating edge conversion parameters;
Calculating an edge activity based on the edge conversion parameter; comparing the edge activity with a predetermined threshold; and one comparison result obtained by the comparing step satisfies a predetermined condition. Until the calculating step and the comparing step are repeated, and if one comparison result obtained by the comparing step satisfies a predetermined condition, the first edge vector is calculated based on the edge activity. And extracting a predetermined area by selecting an edge corresponding to a specific edge vector from.

Further, in the invention according to claim 3, when calculating the edge conversion parameter, an enlargement ratio between corresponding edge vectors, a rotation angle or a parameter uniquely obtained from them is used as an edge conversion parameter. It is characterized by.

According to the fourth aspect of the present invention, an entire space in which a plurality of feature points are arranged and a partial space in which corresponding points to the feature points are arranged are provided, and are arranged on the partial space. In the region extraction method for extracting a predetermined region formed by a specific feature point group from the entire space based on the corresponding points thus obtained, the first edge and the second edge are extracted based on the position information of the plurality of feature points. Generating a first edge vector obtained from each of the first edge, and generating a second edge and a second edge vector respectively obtained from the second edge based on the position information of the corresponding point; Calculating an edge conversion parameter based on the first edge vector and the second edge vector, and setting a predetermined value to one edge activity. Calculating an edge activity update amount based on the edge conversion parameter; calculating a new edge activity based on the one edge activity and the edge activity update amount; Resetting the edge activity as one edge activity, determining whether the edge activity satisfies a predetermined convergence condition, and determining whether the edge activity satisfies a predetermined convergence condition. When it is determined, the calculating step or the determining step is repeated. When the determining step determines that the predetermined convergence condition is satisfied, the first edge vector of the first edge vector is determined based on the edge activity. A region extraction method characterized by extracting a predetermined region by selecting an edge corresponding to a specific edge vector from among them. It is.

Here, as described in claim 5, the step of calculating the edge activity update amount according to the present invention comprises:
Performing a non-linear conversion on the one edge activity, and using a predetermined weighting coefficient for the one edge activity based on the one edge activity and the edge conversion parameter obtained by the non-linear conversion. And performing a weighting calculation.

According to a sixth aspect of the present invention, the step of calculating the edge activity update amount according to the present invention includes the step of performing the non-linear conversion on the result obtained by the non-linear conversion in accordance with the one edge activity. The method includes a step of obtaining an average value within a specified predetermined range, and a step of performing a smoothing operation based on a difference between the average value and a result obtained by performing the nonlinear conversion.

Further, as set forth in claim 7, the step of calculating the edge activity update amount according to the present invention comprises the following steps:
The update amount of the edge activity is changed based on the edge conversion parameter.

According to an eighth aspect of the present invention, the step of performing the weighting calculation according to the present invention is characterized in that the first edge based on the feature point on the whole space and the first edge based on the corresponding point on the subspace are used. Calculating the degree of dependency between the second edge and the second edge; and determining a weighting coefficient based on the calculated degree of dependency.

Further, in the present invention, an area extraction method including the following items specifying the invention may be used. That is, first, the present invention also provides a region extraction method characterized in that when calculating the edge activity update amount, the edge activity and the edge activity update amount are added together with a coefficient. Good.

Secondly, according to the present invention, when extracting a predetermined area from the selected edge, an edge whose edge activity exceeds a threshold value is extracted, and A region extraction method may be characterized in that a connected edge sharing one of the end points of the edge vector is extracted from the inside.

In addition, a region extraction method may be characterized in that a connected edge is extracted and a region corresponding thereto is output. Further, the present invention provides a method for extracting a predetermined area from the selected edge, by performing a weighted average operation of edge conversion parameters of edges included in the extracted corresponding area, thereby performing global edge conversion within the corresponding area. An area extraction method characterized by obtaining parameters may be used.

Here, a partial area to be finally extracted is determined by selecting a range of the edge conversion parameter after obtaining a global edge conversion parameter in the corresponding area. May be used.

On the other hand, when the present invention is grasped as an object method, it is as follows. That is, the present invention described in claim 9 is provided with an entire space in which a plurality of feature points are arranged and a partial space in which corresponding points to the feature points are arranged,
An area extracting apparatus for extracting a predetermined area constituted by a specific feature point group from the entire space based on corresponding points arranged on the partial space, wherein the area extracting apparatus extracts location information of the plurality of feature points. Means for generating a first edge and a first edge vector, and generating a second edge and a second edge vector based on the position information of the corresponding point; and the first edge vector And an edge conversion parameter calculating means for calculating an edge conversion parameter based on the second edge vector, and calculating an edge activity based on the edge conversion parameter, and comparing the edge activity with a predetermined threshold. Comparing means for determining a specific edge from among the first edge vectors based on the edge activity when the comparison result satisfies a predetermined condition; Selection means for selecting the corresponding edge vector, based on the edge selected by the edge selection means, a region extraction apparatus characterized by comprising extracting means for extracting a predetermined region.

According to the tenth aspect of the present invention, an entire space in which a plurality of feature points are arranged, and a partial space in which corresponding points to the feature points are arranged, are provided, and are arranged on the partial space. An area extracting apparatus for extracting a predetermined area formed by a specific feature point group from the entire space based on the corresponding points, wherein a first edge and a first edge are extracted based on position information of the plurality of feature points. First edge vector generation means for generating a first edge vector; second edge vector generation means for generating a second edge and a second edge vector based on the position information of the corresponding point; 1
Edge conversion parameter calculating means for calculating an edge conversion parameter based on the edge vector and the second edge vector; edge activity calculating means for calculating an edge activity based on the edge conversion parameter; Comparing means with a predetermined threshold value;
When one comparison result obtained by the comparison means does not satisfy a predetermined condition, the iterative execution means for executing the edge activity calculation means and the comparison means, and one comparison result obtained by the comparison means An edge selection unit that selects an edge corresponding to a specific edge vector from the first edge vectors based on the edge activity when a predetermined condition is satisfied; and an edge selected by the edge selection unit. And a region extracting means for extracting a predetermined region based on the region.

Here, as set forth in claim 11, the region extracting means according to the present invention comprises: means for comparing with another threshold value based on the edge selected by the edge selecting means; When the comparison result obtained by the comparing means is satisfied, the apparatus further comprises means for extracting an edge connected to the selected edge.

[0029]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A feature point extracting method and a feature point extracting apparatus according to the present invention will be described below with reference to the drawings. In the following, for simplicity, the space between the feature points and the corresponding points is both two-dimensional, but the same embodiment can be applied to a general K-dimensional space (K is an integer of 1 or more).

FIG. 1 is a block diagram showing a configuration of a feature point extracting apparatus according to the present invention. The block diagram shown in FIG.
The feature point extracting apparatus according to the present invention is configured from the viewpoint of functional elements, and each functional element is realized by hardware, software, or a combination thereof.

The feature point position information input section 11 is for sending position information of N (N is an arbitrary natural number) feature points arranged in the first space to the edge calculating section 12. On the other hand, the corresponding point position information input unit 13 is for sending the position information of the corresponding point in the second space for each feature point in the first space to the edge calculating unit 12. The position information in the first or second space is, for example,
It is represented by a set of position vectors.

The edge calculation unit 12 calculates an edge vector for an edge formed based on the position information input from the feature point position information input unit 11 and the corresponding point position information input unit 13. Here, an edge refers to a line segment connecting two adjacent points in a certain space. Therefore, in the first space, when one of the two feature points is set as a start point and the other is set as an end point, a line segment connecting these is an edge vector. Similarly, line segments connecting points (corresponding points) in the second space corresponding to the start point and the end point of the edge vector in the first space are edge vectors. Note that the number of edge vectors is the same in the two spaces.

The edge conversion parameter calculator 14 calculates an edge conversion parameter based on the edge vector in each space calculated by the edge calculator 12. Here, the edge conversion parameter refers to a parameter indicating what type of conversion has been performed on each set of edge vectors.

The edge activity calculator 15 calculates the edge activity of each edge in the first space based on the edge conversion parameters, the edge vector, and the edge position information (edge position). The edge activity is a value that depends on the distribution of each edge in the edge conversion parameter space and the distance between edge positions. The edge activity calculating unit 15 sends the obtained edge activity to the output unit 16.

The output section 16 is for extracting a corresponding area based on the edge activity obtained for each edge by the edge activity calculating section 15 and displaying the corresponding area. That is, when the edge activity is equal to or higher than a certain threshold,
Further, a connected edge sharing one of the end points of the edge vector is selected, a region corresponding to the selected edge is extracted, and the corresponding region is displayed if necessary.

Next, details of each functional element shown in FIG. 1 will be described. First, the calculation of the edge vector by the edge calculation unit 12 will be described. As described above, an edge refers to a line segment connecting two adjacent points in a certain space, and a line segment connecting two points in a space is an edge vector.

FIG. 2 is a diagram for explaining edge vectors in the first and second spaces. As shown in FIG. 7A, the position vector of the start point of one edge vector in the first space is represented by r ⁺ , and the end point is represented by r ⁻ . Also, the edge vector in the second space corresponding to it s ^+, s ^- represents a. At this time, the edge vector p in the first ^{space, p = (r + -r -} ) and represents the edge vector q on the second space corresponding to this, q = ^- and (s ⁺ -s) Shall be represented.

Then, the edge calculator 12 sends the obtained edge vector to the edge conversion parameter calculator 14 and the edge activity calculator 15. Next, the edge conversion parameter calculator 14 will be described in detail.

As described above, the edge conversion parameter calculation unit 14 calculates an edge conversion parameter based on the edge vector in each space calculated by the edge calculation unit 12. The edge conversion parameter is a parameter for indicating what kind of conversion has been performed on each set of edge vectors. That is, the i-th edge vector is defined as p
_i , and the corresponding edge vector is qi,

[0040]

(Equation 1) The parameter that can uniquely identify the transformation matrix A _i is the edge transformation parameter. Note that the superscript t
Represents transposition.

More specifically, the edge conversion parameter in the embodiment of the present invention refers to, for example, an enlargement ratio between two corresponding edge vectors, a rotation angle, or a parameter uniquely obtained from them. . That is, assuming an edge vector as shown in FIG. 3, from the rate of change η _i of the magnitude of the edge vector and the rotation angle θ _i from the edge vector p _i to q _i , the transformation matrix A _i is as follows: Express.

[0042]

(Equation 2)

[0043] _{Furthermore, (v ix, v iy)} = (η i cosθ, η i sinθ i) when placed as a parameter edge conversion parameter represented by a set of (v _ix, v _iy). At this time, the system as shown in FIG. 4 is the edge conversion parameter space. Then, the edge vector p _i (x, y) components, respectively (p _ix, q _iy) and, in the edge vector q _i (x, y)
If the components are (p _ix , q _iy )

[0044]

(Equation 3) It is. Here, it is assumed that the absolute value | p | of the edge vector p always takes a positive value.

Next, the calculation of the edge activity by the edge activity calculator 15 will be described in detail. As described above, the edge activity calculator 15 calculates the edge conversion parameter, the edge vector, and the edge position information (edge position).
, The edge activity of each edge in the first space is obtained, and the edge activity is sent to the output unit 16.

FIG. 5 is a block diagram showing the configuration of the edge activity calculating section 15. As shown in FIG. As shown in the figure, the edge activity calculating unit 15 includes an edge activity initial setting unit 51, an edge activity update amount calculating unit 52, an edge activity updating unit 53, and a convergence determining unit 54.

The edge activity initial value setting section 51 gives an initial value to the edge activity for each edge. In the embodiment of the present invention, h _i (0) = 0 is given, but the present invention is not particularly limited to this.

The edge activity update amount calculation section 52 calculates the edge activity based on the edge activity at a certain point in time for each edge.
Also, the update amount of the edge activity is calculated by using the edge conversion parameter and the edge position of the edge as inputs.

Here, the details of the edge activity update amount calculation section 52 will be described. FIG. 6 is a block diagram illustrating a configuration of the edge activity update amount calculation unit 52. The non-linear converter 61 performs a function conversion of the edge activity according to a predetermined function given in advance. FIG. 7 is a diagram illustrating an example of a conversion function of the nonlinear converter 61. In the embodiment of the present invention, a sigmoid function having λ as a constant is used. That is,

[0050]

(Equation 4) It is expressed as In the present invention, the conversion function is
Although not limited to this sigmoid function, a monotonically increasing function is generally used.

The weighted adder 62 shown in FIG. 6 performs a weighted addition operation in a region near an edge in the edge conversion parameter space based on the edge conversion parameter, the edge position, and the output value from the nonlinear converter 61. . FIG.
FIG. 9 is a diagram showing a manner in which weighted addition is performed on a non-linear conversion result of the activity of an edge near a certain edge in an edge conversion parameter space.

The smoothing operation unit 63 shown in FIG. 6 calculates the average of H _i (n) in the area near the edge position of each edge,
A value proportional to the difference between the calculated value and H _i (n) is represented by i
It is added to the i-th edge activity update amount i (n).

The thus configured edge activity update amount calculating section 52 operates as follows. Note that the edge activity of the i-th edge of the n-th iteration is denoted by _hi (n), and the output value from the nonlinear converter 61 is denoted by _Hi (n).

That is, the weighted adder 62 calculates the edge conversion parameter, the edge position, and H _i (n) (i = 1
ＮN _e ) as inputs, a weighted addition operation of H _i (n) is performed in a region near the i-th edge in the edge conversion parameter space, and this is temporarily used as the i-th edge activity update amount △ i (N). Further, the smoothing operation unit 63
Then, the average of H _i (n) near the edge position of each edge is obtained, and a value proportional to the difference from H _i (n) is added to the i-th edge activity update amount △ i (n). Here, an example of △ i will be described. Δi (n) can be represented as follows.

[0055]

(Equation 5) here,

[0056]

(Equation 6) _{However, γ, s a, s b} , s c is a constant.

The first item of Equation 5 corresponds to the weighted adder 62. At this time, d (r, i, j) (r is an upper suffix and i, j is a lower suffix) is a distance between edges i and j in the first space, d (s, i, j) ) (S is the upper suffix, i, j is the lower suffix) is the distance between the edges i and j in the second space, d (v, i, j) (v is the upper suffix, i, j j is a subscript)
Represents a distance in the edge transformation parameter space. Specifically,

[0058]

(Equation 7) It can be expressed as.

The second item of Expression 5 is a part corresponding to the smoothing operation unit 63. U _i represents the adjacent edge of the first space of edge i, N _u denotes the number of adjacent edges. The third item in Equation 5 corresponds to the edge conversion parameter constraint calculation. FIG. 9 is a diagram illustrating an example of an added value to the edge activity update amount based on the value of the edge conversion parameter. That is, as shown in the figure, a function form that reduces the edge activity can be obtained for a region where the value of the edge enlargement ratio j is close to 0 and a region that is considerably larger than 1.

The edge activity update amount calculating section 52
In the above, it is also possible to introduce a process of increasing or decreasing the edge activity update amount i (n) according to the value of the edge conversion parameter. This makes it possible to increase or suppress the edge activity in a certain region of the edge conversion parameter space.

FIG. 10 is a block diagram showing the configuration of the edge activity updating unit 53 shown in FIG. 5. As shown in FIG. 10, the edge activity updating unit 53 includes an adder 71 with a coefficient.
Accordingly, the edge activity h _i (n) and the edge activity update amount i (n) are added, and the new edge activity h _i (n +
Find 1). That is, α (> 0) is used as a coefficient,

[0062]

(Equation 8) To obtain _hi (n + 1).

In the edge activity update amount calculating section 21, the weight coefficient for the weighted adder 62 may include a value depending on the reliability of the correspondence between edges (degree of dependency). This makes it possible to reflect the reliability of the correspondence between edges in the weighting coefficient, so that more accurate region extraction can be performed.

FIG. 11 is a block diagram showing a configuration of convergence determining section 54 shown in FIG. As shown in the figure, the convergence determining unit 54 includes an internal memory 111, an edge activity change calculating unit 112, a threshold comparing unit 113, and a threshold holding unit 114.

The internal memory 111 stores the edge activity because the past edge activity is used.
The edge activity change calculator 112 calculates the amount of change between the current edge activity and the past edge activity stored in the internal memory 111 for all edges.
The past edge activity at this time is not limited to the immediately preceding edge activity. Therefore, for example, past edge activity may be accumulated as a history, and a predetermined amount of past history may be used.

The threshold value comparing section 113 determines whether or not to terminate the current operation according to the threshold value stored in the threshold value storing section 114 in advance. That is, if the output of the edge activity change calculator 112 is equal to or greater than the predetermined threshold, the edge activity is input again to the edge activity update amount calculator 51 to continue the iterative calculation. On the other hand, if it is smaller than the threshold value, the edge activity calculation is terminated and the edge activity is sent to the output unit 16.

FIG. 12 is a block diagram showing a configuration of the output section 16 shown in FIG. The output unit 16 includes an edge extraction unit 121, a connected edge region extraction unit 122, and an extraction region display unit 123.

The edge extracting section 121 extracts only edges having an edge activity equal to or higher than the threshold value. The connected edge area extracting unit 122 further extracts only connected edges that share an end point between the edges, and assigns a unique number to one connected edge area.

The extraction area display section 133 displays a partial area in the first space using the position information of the edge included in each connected edge area. The output unit 16 is for extracting a corresponding region based on the edge activity obtained for each edge by the edge activity calculating unit 15 and displaying the corresponding region. That is, the edge activity is
A connected edge that is equal to or larger than a certain threshold value and shares one of the end points of the edge vector is selected, an area corresponding to the selected edge is extracted, and the corresponding area is displayed if necessary.

[0070]

As described above, according to the present invention, each of the plurality of feature points arranged on the first layer is converted into the second feature point by the method based on the similarity of the local feature amounts. It is possible to extract, from a first layer on which feature points are arranged, a partial area that matches by translation, scaling, rotation, or distortion of a second layer in a state where the partial point locally corresponds to one point on the layer. it can. Further, even when a plurality of partial regions as described above exist on the first layer, those regions can be simultaneously extracted.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a feature point extraction device according to the present invention.

FIG. 2 is a diagram for explaining edge vectors on first and second spaces.

FIG. 3 is a diagram showing an example of a state of conversion of an edge vector from a first space to a second space.

FIG. 4 is a diagram showing an example of an edge conversion parameter space.

FIG. 5 is a block diagram showing a configuration of an edge activity calculating unit 15;

FIG. 6 is a block diagram showing a configuration of an edge activity update amount calculation unit 52;

FIG. 7 is a diagram illustrating an example of a conversion function of a nonlinear converter 61.

FIG. 8 is a diagram showing a state in which a weighted addition is performed on a non-linear conversion result of the activity of an edge near a certain edge in an edge conversion parameter space.

FIG. 9 is a diagram illustrating an example of an added value to an edge activity update amount based on a value of an edge conversion parameter.

FIG. 10 is a block diagram showing a configuration of an edge activity updating unit 53 shown in FIG. 5;

FIG. 11 is a block diagram showing a configuration of a convergence determining unit 54 shown in FIG.

FIG. 12 is a block diagram showing a configuration of an output unit 16 shown in FIG.

FIG. 13 is a diagram showing an example of extraction of a corresponding partial area in an image recognition problem.

FIG. 14 is a diagram showing an example of correspondence between points between two layers and an edge vector.

FIG. 15 is a diagram showing an example of correspondence between regions between two layers.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Feature point position input part 12 ... Edge calculation part 13 ... Corresponding point position input part 14 ... Edge conversion parameter calculation part 15 ... Edge activity calculation part 16 ... Output part 51 ... Edge activity initial setting part 52 ... Edge activity Update amount calculation unit 53 Edge activity update unit 54 Convergence determination unit 61 Nonlinear converter 62 Weighted adder 63 Smoothing operation unit 101 Adder 111 Internal memory 1 112 Edge activity change calculation unit 113: threshold value comparison unit 114: threshold value holding unit 121: edge extraction unit 122: connected edge region extraction unit 123: extraction region display unit

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06T 7/00 G06T 9/20

Claims (11)

(57) [Claims] An overall space in which a plurality of feature points are arranged and a subspace in which corresponding points for the feature points are arranged are provided, and based on the corresponding points arranged in the subspace, the entire space is provided. A region extraction method for extracting a predetermined region formed by a specific feature point group from a space, wherein a first edge and a first edge vector are generated based on position information of the plurality of feature points, Generating a second edge and a second edge vector based on the position information of the corresponding point; calculating an edge conversion parameter based on the first edge vector and the second edge vector; Is calculated based on the edge activity, and the edge activity is compared with a predetermined threshold. When the comparison result satisfies a predetermined condition, the first activity is calculated based on the edge activity. By selecting the corresponding edge to the particular edge vector from the edge vector, region extraction method characterized by extracting a predetermined region. 2. An overall space in which a plurality of feature points are arranged, and a subspace in which corresponding points for the feature points are arranged, and based on the corresponding points arranged in the subspace, the entire space is provided. A region extraction method for extracting a predetermined region formed by a specific feature point group from a space, wherein a first edge and a first edge vector are generated based on position information of the plurality of feature points, Generating a second edge and a second edge vector based on the position information of the corresponding point; calculating an edge conversion parameter based on the first edge vector and the second edge vector; Calculating an edge activity based on the edge conversion parameter; comparing the edge activity with a predetermined threshold value; The calculating step and the comparing step are repeated until one comparison result obtained by the comparison satisfies a predetermined condition. When the one comparison result obtained by the comparing step satisfies a predetermined condition, the edge Extracting a predetermined area by selecting an edge corresponding to a specific edge vector from the first edge vectors based on the degree of activity. Upon calculating the wherein the edge transformation parameters corresponding magnification between the edges vector, according to claim 1 or 2, characterized in that the rotational angle or edge transformation parameters that uniquely derived parameters from their Region extraction method. 4. An overall space in which a plurality of feature points are arranged and a partial space in which corresponding points for the feature points are arranged, and the entire space is provided based on the corresponding points arranged on the partial section. In an area extraction method for extracting a predetermined area constituted by a specific feature point group from a space, a first edge obtained based on the position information of the plurality of feature points and a first edge obtained from the first edge, respectively. Generating an edge vector and generating a second edge and a second edge vector respectively obtained from the second edge based on the position information of the corresponding point; and the first edge vector and the second edge vector. Calculating an edge conversion parameter based on the edge vector of the following; setting a predetermined value to one edge activity; Calculating an edge activity update amount based on a meter; calculating a new edge activity based on the one edge activity and the edge activity update amount; Resetting as the edge activity, and determining whether or not the edge activity satisfies a predetermined convergence condition.If the determination step determines that the predetermined convergence condition is not satisfied, The calculating step and the determining step are repeated. When it is determined that the predetermined convergence condition is satisfied by the determining step, a specific edge vector is selected from the first edge vectors based on the edge activity. And extracting a predetermined area by selecting an edge corresponding to the area. 5. The method according to claim 1, wherein the step of calculating the edge activity update amount comprises: performing a non-linear conversion on the one edge activity; and calculating the non-linearly converted one edge activity and the edge conversion parameter. 5. The method according to claim 4, further comprising: performing a weighting calculation on the one edge activity based on a predetermined weighting coefficient. 6. The method according to claim 1, wherein the step of calculating the edge activity update amount comprises the step of obtaining an average value within a predetermined range specified according to the one edge activity, for a result obtained in the step of performing the nonlinear conversion. The method according to claim 5, further comprising: performing a smoothing operation based on a difference between the average value and a result obtained by performing the nonlinear conversion. 7. A step of calculating the edge activity update amount based on said edges transformation parameters claims 4 to 6 region extraction method wherein changing the edge activity update amount. 8. The step of performing the weighting calculation includes calculating a degree of dependency between a first edge based on a feature point in the overall space and a second edge based on a corresponding point in the subspace. The method according to claim 5, further comprising: determining a weighting coefficient based on the calculated degree of dependency. 9. An entire space in which a plurality of feature points are arranged and a subspace in which corresponding points for the feature points are arranged, and the entire space is provided based on the corresponding points arranged in the subspace. What is claimed is: 1. An area extracting apparatus for extracting a predetermined area formed by a specific feature point group from a space, wherein a first edge and a first edge vector are generated based on position information of the plurality of feature points, Edge vector generating means for generating a second edge and a second edge vector based on the position information of the corresponding point; and calculating an edge conversion parameter based on the first edge vector and the second edge vector. Edge conversion parameter calculating means; calculating means for calculating an edge activity based on the edge conversion parameter; comparing the edge activity with a predetermined threshold value; A selection unit that selects an edge corresponding to a specific edge vector from the first edge vectors based on the edge activity when the comparison result satisfies a predetermined condition; Extracting means for extracting a predetermined area based on the edge. 10. An overall space in which a plurality of feature points are arranged;
Given a subspace in which corresponding points for the feature points are arranged, a predetermined area constituted by a specific feature point group is extracted from the entire space based on the corresponding points arranged in the subspace. An area extraction device, comprising: a first edge vector generation unit configured to generate a first edge and a first edge vector based on the position information of the plurality of feature points; and a second edge vector generation unit configured to generate a first edge vector based on the position information of the corresponding point. 2nd edge and 2nd
A second edge vector generating means for generating an edge vector, an edge conversion parameter calculating means for calculating an edge conversion parameter based on the first edge vector and the second edge vector, Edge activity calculating means for calculating the edge activity, and comparing means for comparing the edge activity with a predetermined threshold value, wherein one comparison result obtained by the comparing means does not satisfy a predetermined condition An iterative execution unit that executes the edge activity calculating unit and the comparing unit; and if the one comparison result obtained by the comparing unit satisfies a predetermined condition, the first based on the edge activity, Edge selecting means for selecting an edge corresponding to a specific edge vector from the edge vectors of Based on the selected edge, region extraction apparatus characterized by comprising a region extracting means for extracting a predetermined region. 11. The area extracting means, based on an edge selected by the edge selecting means,
10. A means for comparing with another threshold value, and means for extracting an edge connected to the selected edge when the comparison result obtained by the comparing means is satisfied. 11. The region extracting apparatus according to claim 10.