JP2946449B2 - Clustering processor - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a clustering processing apparatus for generating a cluster in accordance with the distribution of a data group, and to a clustering processing apparatus for automatically setting an initial cluster spread width when performing clustering.

[0002]

2. Description of the Related Art In a system in which an information processing operation is obtained from obtained cases (data), input data covers a very wide range and the amount is large as compared with the amount of data that can be obtained. Here, if the input data is classified and data is obtained in the classified state, the amount of data to be obtained can be reduced. As described above, there is clustering as a method of classifying input data. Clustering is to merge similar data in input data and combine them into several groups, that is, clusters. The classification by clustering is performed by measuring and quantifying the degree of similarity of each data forming a set, and collecting data having a high degree of similarity into the same group (cluster).

FIG. 6 is a configuration diagram showing a configuration of a conventional clustering processing device which is a device for clustering input data as described above. In the figure, 1 is a data input unit, 2 is a cluster generation determining unit that determines whether a new cluster is to be generated for input data, 3 is a cluster generating unit that newly generates a cluster of input data, 4 is A cluster adjusting unit that adjusts clusters based on input data; 5 is a cluster fusion determining unit that determines whether to fuse a plurality of clusters based on input data; and 6 is a cluster fusion unit that fuses a plurality of clusters. Department. Reference numeral 7 denotes a storage unit for storing parameters and clustering information necessary for newly generating a cluster or for adjusting and merging clusters. The storage unit 7 includes a clustering parameter storage unit 8 and a clustering information storage unit 9. Is done.

FIG. 7 is an explanatory diagram showing a processing situation in such a clustering processing apparatus. For example, when an input data group up to a certain point is in a situation as shown in FIG. Is performed as shown in FIG. Thereafter, as the time elapses, the characteristics of the input data group gradually change to the state shown in FIG. 7B, and accordingly, the adjustment and adjustment of the cluster shown in FIG.
Fusion or generation of a new cluster in the cluster of FIG. 7C, and as a result, clustering is performed as shown in FIG. 7E. That is, the present apparatus generates a cluster corresponding to an input data group that increases with time, and can generate, adjust, and fuse clusters in accordance with sequentially input data.

In the clustering processing device shown in FIG.
First, data (signal) input via the data input unit 1
Is determined by the cluster generation determination unit 2 to generate a new cluster or to belong to an already generated cluster. Generally, as a measure for calculating the degree of belonging of the input data to each cluster,
There is a Euclidean distance shown in equation (1) and a Mahalanobis distance shown in equation (2).

[0006]

(Equation 1)

[0007]

(Equation 2)

Here, in equation (2), σx ² , σy ²
Indicates a variance value, and is represented by equations (3) and (4), respectively. Here, n is the number of data, and μ is an average value.

[0009]

(Equation 3)

[0010]

(Equation 4)

The Mahalanobis distance is used for obtaining a normal distribution used for probability, statistics, and the like. Therefore, the calculation is complicated and the calculation time is long. For this reason,
It is not suitable for online processing that requires quick processing. In this clustering processing device, in order to enable online processing, when calculating the degree of membership of an input signal to a cluster, the calculation time is reduced by using the Euclidean distance of only the four arithmetic operations, and the memory capacity is increased. I try to suppress it.

Here, as a distribution for defining the degree of belonging of the input data to each cluster, a distribution of triangles often used as a membership function of fuzzy control is used. As shown in FIG. 8, the one-dimensional triangle distribution is
The spread α of the distribution is used as a parameter, and the position of the center coordinate μ is defined as the degree of belonging “1”. In the case of multi-dimension, a similar distribution is given independently in each axis direction. Next, an outline of an algorithm of the one-dimensional clustering process will be described with reference to the triangular distribution of FIG.

In the above triangular distribution, a region of a cluster that has already been generated is defined as, and a region around the cluster is defined as. Here, X indicates input data, μ indicates the center position of the cluster, and α indicates the extent from the center of the cluster. Note that K1 is a parameter related to the range in which the cluster is to be enlarged, and K2 is a parameter related to the range in which the cluster is to be reduced, and are stored in the clustering parameter storage unit 8 (FIG. 6). If the data X input in the data input unit 1 falls within the range of the area, a new cluster is generated in the cluster generation unit 3. If the input data X falls within the range of the region, the cluster adjusting unit 4 performs enlargement processing on the spread of the cluster. If the input data X falls within the range of the area, the spread of the cluster is not changed. If the input data X falls within the range of the region (the center of the cluster), the cluster adjusting unit 4 performs a reduction process on the spread of the cluster. Note that if the input data X falls within the range of the area 1 to the center, the center position of the cluster is also changed.

Next, a cluster generation algorithm in the cluster generation section 3 will be described. As described above, when the data X input in the data input unit 1 falls within the range of the area and there is no cluster to which the data X belongs, or when no cluster has been generated yet, the input data is set to the center position μ. A cluster having a membership distribution as shown in FIG. 9 is generated and stored in the clustering information storage unit 9. In this case, the degree of belonging F (x) is calculated by the equation (5).
And the maximum value is “1”. Here, α is a parameter indicating the initial spread of the cluster, and is stored in the clustering parameter storage unit 8.

[0015]

(Equation 5)

Next, an adjustment algorithm in the cluster adjustment section 4 will be described with reference to FIG. FIG.
As shown in FIG. 10A, when the input data X falls within the range of the area, the cluster adjustment unit 4 performs enlargement adjustment on the spread of the cluster as shown in FIG. Will be In this case, the cluster adjusting unit 4 first calculates the cluster expansion αnew based on Expression (6).

[0017]

(Equation 6)

Next, a new center position μnew of the enlarged cluster is calculated based on equations (7) and (8), and the result is stored in the clustering information storage unit 9 as clustering information.

[0019]

(Equation 7)

[0020]

(Equation 8)

Next, FIGS. 10 (c) and 10 (d) are diagrams showing the reduction adjustment of the cluster. As described above, when the input data X falls within the range of the area (the center of the cluster), the cluster adjustment unit 4 performs the cluster reduction adjustment as shown in FIG. First, the center position μnew is calculated based on equation (9). Note that Nold in Expression (9) indicates the number of data input to the cluster up to this point, and n indicates the reciprocal of the number of clusters to which the input data X should belong.

[0022]

(Equation 9)

Next, assuming that a parameter related to reduction stored in the clustering parameter storage unit 8 is K3, the cluster expansion αnew is calculated based on the equation (10).

[0024]

(Equation 10)

When the input data X falls within the range of the region, the spread of the cluster is not changed, and only the center position is changed based on the equation (9). Next, a fusion algorithm in the cluster fusion unit 6 will be described. There are a plurality of clusters to belong to, each cluster is a pair, and the minimum degree of intersection at each axis of the common part of each pair on each axis is the degree of membership of each cluster pair, and the maximum degree of membership is a preset threshold. When the cluster fusion determination unit 5 determines that the time has exceeded TH, the cluster fusion unit 6
Fuses pairs of clusters selected by the following algorithm: That is, in this case, the spread of the new cluster α
Redefine the cluster spread so that new can include all of the original cluster spread αold. However, since the area of the cluster is only expanding in this state, the number of data N (new) is calculated based on the following equation (11), assuming that the end of the cluster is not so important. Further, based on the equations (12) and (13), the center position μ (ne
w), and calculate the cluster spread α (new) independently for each axis.

[0026]

[Equation 11]

[0027]

(Equation 12)

[0028]

(Equation 13)

FIG. 11 is a diagram showing a situation in which two clusters A and B are merged. That is, as shown in FIG. 9A, the input data X is in the range of the common area of the areas of the clusters A and B, and the minimum degree of belonging of the intersection of each axis of this common area exceeds the threshold value TH. If it is determined that there is a cluster, clusters A and B are merged to generate a cluster C as shown in FIG. In this case, assuming that the numbers of data of the clusters A and B are NA and NB, the number of data NC of the merged cluster C is calculated based on Expression (14).

[0030]

[Equation 14]

Further, the center position μ of the fused cluster C
C and the spread αC are calculated based on equations (15) and (16), respectively.

[0032]

(Equation 15)

[0033]

(Equation 16)

Even if the input data X is within the range of the common region of the regions of the clusters A and B, as shown in FIG. Is not exceeded, clusters A and B are not merged.

Here, an example in which the clustering processing device is adapted to a color image processing device will be described below. The color image processing apparatus converts an RGB signal obtained from a color video camera or the like that is capturing an image into an Irg space, and converts the input data of the color image into a vector represented by r and g on an rg variable surface in this space. Is formed as a distribution. Note that the Irg space has an intensity I that is the sum of the signals R, G, and B.
And r obtained by dividing the R (red) signal by the intensity I and g obtained by dividing the G (green) signal by the intensity I. The distribution of the color data of the reference input image formed on the rg plane as shown in FIG.
By forming a reference cluster by performing clustering by the clustering processing device, it is possible to determine whether a certain image is the same as the reference image.

That is, first, the reference image is captured by the camera, and the distribution of the reference image data converted by the color image processing device is clustered by the clustering processing device and set as a reference cluster. Next, when the recognition target image is captured by the camera, the recognition target image matches the reference image depending on how far the distribution state of the recognition image data converted by the color image processing device matches the reference cluster. Degree can be determined.

[0037]

Conventionally, the above-described configuration has the following problems.
First, when clustering the distribution of the data of the reference image obtained by the above-described color image processing apparatus, the distribution of this data is viewed by a human and the initial value α of the spread width of the cluster is used.
Therefore, there is a problem that labor is required for this. And the dimension of the data to be clustered is 4
, It is almost impossible for a human to determine the initial value α of the spread width of the cluster.

The spread of the generated cluster is 1
Since the initial cluster is determined by two parameters (α, α _new ) and the initial cluster is a square, there is a problem that an appropriate initial cluster is not given depending on the distribution of the data group to be clustered. For example, the distribution of the data of the reference image obtained by the color image processing apparatus is shown in FIG.
In the case of the state shown in (a), since all the distribution states are approximately uniformly distributed in the rg plane in the rg plane, there is not much problem even if the initial cluster is a square. However, when the distribution of the data of the reference image obtained by the color image processing apparatus is in a state as shown in FIG. 12B, the state of all distributions of the data of the reference image is changed according to the r-axis direction. If the initial cluster is a square, it takes a long time to generate a cluster suitable for this distribution state.

In the prior art, since the initial cluster is a square, the data of the color image processing apparatus described above has values in the horizontal axis direction and the vertical axis direction of the plane indicating the distribution, that is, each data indicating input data. Dimension values needed to be normalized.

The present invention has been made to solve the above problems, and has as its object to automatically generate an initial cluster in accordance with the distribution of input data.

[0041]

SUMMARY OF THE INVENTION A clustering processing apparatus according to the present invention sequentially and arbitrarily extracts three or more pieces of data from a data group to be clustered, and calculates the absolute difference between the extracted data and the data to be extracted next to this data. Calculate the values, arrange the absolute values of the differences in descending power order, calculate the absolute value of his difference in this order, and use the maximum value of the calculated values to calculate the size of the newly generated cluster. It is characterized by having an initial cluster spread determining means for determining.

[0042]

When a new cluster is generated from the obtained data, the size of the cluster is adapted to the distribution state of the data group to be clustered.

[0043]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a configuration diagram showing the configuration of the clustering processing device of the present invention. Referring to FIG. 1, reference numeral 10 denotes an initial cluster expansion width generated when there is no cluster to which the data input in the data input unit 1 belongs or a cluster has not yet been generated, depending on the state of the input data. Is the initial cluster spread width determination unit, and the other components are the same as those in FIG.

Next, the operation of the initial cluster spread width determination unit 10 of the clustering device will be described. First, 10 data are arbitrarily extracted from the data group to be subjected to the clustering process via the data input unit 1, and the absolute value of the difference is obtained in the order in which the data is extracted. Here, "3", "25",
"5", "27", "7", "1", "4", "3"
The case where the numbers are "2", "34", and "8" will be described.

By taking the difference in the order in which these randomly extracted data are extracted, the absolute value of this difference becomes “22”,
“20”, “21”, “20”, “6”, “3”, “2”
8 "," 3 ", and" 26 ". Next, the absolute values of the obtained differences are rearranged in descending power order, and “28”, “26”,
"22", "21", "20", "20", "6",
"3", "3", and in this order again, the absolute value of the difference between the values is taken, and the maximum value among these is "14".
Is set as the initial cluster spread width. In this case, the initial cluster spread width α is “7”. Then, the initial cluster spread width α thus obtained is stored in the clustering parameter storage unit 8, and the cluster generation unit 3 generates an initial cluster using this value.

When the initial cluster spread width determining unit 10 determines the initial cluster spread width α for clustering data indicating the color distribution of the reference input image formed on the rg plane by the above-described color image processing apparatus, r
And g, the initial cluster spread width is determined as described above. As a result, all distribution states are approximately uniformly distributed in the r direction and the g direction in the rg plane.
In the case of (a), as in the conventional case, the square initial cluster 2
1 is generated, and in the case of FIG. 2B in which the state of all distributions of the data of the reference image is spread in the r-axis direction, unlike the conventional case, a horizontally long rectangular initial cluster 22 is generated. As described above, according to the clustering processing device of the present invention, an initial cluster that is actually adapted to the distribution is generated, so that the clustering processing is performed quickly.

Here, the data group to be clustered is
As shown in FIG. 3, consider a state in which the data is distributed in two regions. Here, the case where there is one dimension will be described. In this case, if data is arbitrarily extracted from the data group as described above, the data in the area 31 and the data in the area 32 should be extracted in a state close to this distribution. If the difference is taken in the order in which the extracted data is extracted, it is considered that some of the absolute values have values close to the distances a, b, c, and d shown in FIG. Therefore, when the differences between the extracted data are arranged in descending order and the absolute values of the differences are obtained in that order, the maximum value D among them is the distance d, distance c, distance a, and distance b arranged in descending order. In this order, the absolute value of the difference is close to the maximum value L.

Here, when the distribution state of the data group to be clustered is, as shown in FIG. 3, the area 31 and the area 32 are sufficiently separated from the size of each area, the difference between the distance c and the distance a is determined. Is the maximum value L. This maximum value L
Is the width α of the initial cluster, even if the initial cluster 33 is generated with the maximum value data X in the area 31, the cluster 33 does not cover the area 32. Accordingly, half of the maximum value D, which is smaller than the maximum value L, is set as the initial cluster width α. Similarly, even if the initial cluster of this width α is generated using the maximum value data of the area 31, this initial cluster becomes It does not cover the area 32.

Embodiment 2 FIG. By the way, in the above embodiment, the case where the data to be subjected to the clustering process does not change with the passage of time has been described, but the application range of the clustering device of the present invention is not limited to this. If the initial cluster spread width determining unit 10 periodically generates the initial cluster spread width α at a predetermined time interval, the data to be subjected to the clustering process is actually distributed to data that changes with time. A suitable initial cluster can be generated.

In a state where data is generated as needed, as shown in FIGS. 4 and 5, the distribution of data changes with time. FIG. 4 is a distribution diagram showing a state in which the actual distribution 41 at a certain time has changed to a more dispersed distribution 42 with time, and FIG. 5 shows a state in which the actual distribution 51 has changed to a more aggregated distribution with time. FIG.

When clustering data whose distribution changes as shown in FIG.
When 3 is fixed, a large number of small clusters are generated for the distribution 42 as shown in FIG. As described above, even if a cluster in a state as shown in FIG. 4B is generated, as described in the section of the related art,
In some cases, clusters repeat fusion, and as a result, a cluster as shown in FIG. 4A may be generated. However, this takes time for the clustering process. If the size of the initial cluster is significantly different from the actual distribution, many small clusters remain. On the other hand, if the initial cluster spread width α is set according to the data distribution state at that time as in the clustering processing device of the second embodiment, an appropriate distribution for the distribution 42 is obtained as shown in FIG. A cluster 44 is generated.

Next, when clustering data in which the initial distribution 51 changes to the distribution 52 as shown in FIG.
If the initial cluster 53 is fixed as in the related art, many large clusters are generated for the distribution 52 as shown in FIG. 5B. As a result, one cluster may be generated even though the distribution 52 is formed from three regions.
On the other hand, if the initial cluster spread width α is set according to the distribution state of the data at that time as in the clustering processing apparatus of the second embodiment, the distribution 52 is reduced as shown in FIG. Thus, an appropriate cluster 54 is generated. As described above, when performing clustering processing on a data group whose distribution changes with the passage of time, a cluster that is not suitable for an actual distribution may be conventionally generated. The apparatus performs an appropriate clustering process that reflects the actual data distribution state.

[0053]

As described above, according to the present invention, the size of a cluster (initial cluster) when a new cluster is generated even if there are many data dimensions of the data group to be subjected to the clustering process. Is determined without relying on human senses. Conventionally, if the dimension of the data is four or more, it is impossible to determine the size of the initial cluster suitable for the distribution of the data group to be clustered. However, the present invention makes it possible. Further, even if there are many dimensions of input data, there is an effect that it is not necessary to normalize them.

There is an effect that an initial cluster having a size suitable for the distribution of the data group to be clustered can be generated. Therefore, the result of the clustering process accurately reflects the state of distribution of the data group.
Even if the distribution of the data to be subjected to the clustering process changes with time, the clustering process that accurately reflects the distribution state can be performed.

[Brief description of the drawings]

FIG. 1 is a configuration diagram illustrating a configuration of a clustering processing apparatus according to an embodiment of the present invention;

2 is a distribution diagram showing a state of distribution of data of a reference image obtained by a color image processing apparatus that performs a clustering process by the clustering processing apparatus of FIG. 1 and a state of an initial cluster to be generated.

FIG. 3 is an explanatory diagram for explaining a basic concept of an operation of an initial cluster spread width determination unit 10;

FIG. 4 is a distribution diagram showing a data group whose distribution state changes with time.

FIG. 5 is a distribution diagram showing a data group whose distribution state changes over time.

FIG. 6 is a configuration diagram illustrating an example of a conventional clustering processing device.

FIG. 7 is a diagram illustrating a state of a clustering process of the clustering device in FIG. 6;

FIG. 8 is a distribution diagram for explaining an outline of a clustering processing algorithm of the clustering device in FIG. 6;

FIG. 9 is an explanatory diagram illustrating a cluster generation algorithm of the clustering device in FIG. 6;

FIG. 10 is an explanatory diagram for explaining an algorithm of cluster enlargement adjustment and cluster adjustment of the clustering device of FIG. 6;

FIG. 11 is an explanatory diagram illustrating a cluster fusion algorithm of the clustering device in FIG. 6;

12 is a distribution diagram illustrating a state of distribution of data of a reference image obtained by a color image processing device that performs a clustering process by the clustering device of FIG. 6;

[Explanation of symbols]

 Reference Signs List 1 data input unit 2 cluster generation determination unit 3 cluster generation unit 4 cluster adjustment unit 5 cluster fusion determination unit 6 cluster fusion unit 7 storage unit 8 clustering parameter storage unit 9 clustering information storage unit 10 initial cluster spread width determination unit

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int. Cl. ⁶ , DB name) G06T 7/00 G10L 3/00

Claims (1)

(57) [Claims] 1. A clustering processing apparatus for generating a new cluster when the obtained data does not fall into an existing cluster and performing a clustering process according to the characteristics of the sequentially obtained data. Is sequentially extracted arbitrarily, the absolute value of the difference between the extracted data and the data to be extracted next to this data is calculated, and the absolute values of the differences are arranged in descending power order. A clustering processing apparatus comprising: an initial cluster spread determining unit that calculates a value and determines a size of the newly generated cluster using a maximum value of the calculated values.