KR20220109731A - Method, apparatus and computer program for clustering using mean-feature - Google Patents

Abstract

The present invention relates to a clustering method using a mean-feature, and a device and a computer program thereof, and more particularly, to a method for performing the clustering for a plurality of object samples in a clustering device, which comprises: a cluster generating step of generating a plurality of clusters by classifying a plurality of object samples; a mean-feature calculation step of calculating a mean-feature including each mean value of a plurality of features used to classify the object samples, with respect to each of the clusters; and an additional sample clustering step of performing clustering on one or more newly added second object samples using the mean-feature of the clusters. Therefore, a computing resource can be dramatically reduced, and quickly and effectively processed.

Description

Clustering method, apparatus and computer program using mean-feature {Method, apparatus and computer program for clustering using mean-feature}

The present invention relates to a clustering method, apparatus, and computer program using average-features, and more particularly, in clustering an object such as a face included in a photo or video, the average value of features for a plurality of object samples By performing clustering using the average-feature corresponding to .

Recently, with the widespread use of digital devices such as smartphones, users create content such as photos and videos, or search for content that other users have created and uploaded to servers, etc., and select and use the content they want. This is done on a daily basis.

However, it can be a very cumbersome task for a user to find content that includes an object such as a person or thing desired by the user from a large amount of content. That is, the user can open the large amount of content one by one and check whether a desired person or object is included. In this case, the user spends a lot of time and effort to find the desired content from the large amount of content There are difficulties involved

In contrast, by recognizing or classifying objects included in the photo or video in advance using machine learning, etc. on a server, etc., and providing the object in response to a search word input by a user, a photo or video including an object desired by the user, etc. A service to make it easier to use is being tried. For example, in Korean Patent Application Laid-Open No. 10-2012-0064581, a face region of a person included in a given image is detected, a facial feature descriptor is extracted from the detected face region, and the person included in the image is recognized. Disclosed is a method and apparatus for classifying an image to provide a user with or by classifying it.

However, there is a problem that a lot of computing resources are required to accurately recognize and classify objects by analyzing a large amount of photos or videos using machine learning or the like.

In particular, it is a very difficult task to accurately classify a person's face included in a photo or video. In order to store and process facial features, a huge amount of storage and computation is required.

For this reason, from the service provider's point of view, in order to cluster objects that are difficult to classify, such as a human face, it may be difficult to construct and operate a system, and it may be difficult to analyze a large amount of data and There are also problems that take a considerable amount of time to process.

Republic of Korea Patent Publication No. 10-2012-0064581 (published on June 19, 2012)

The present invention was devised to solve the problems of the prior art as described above, and it dramatically reduces computing resources such as storage space and amount of computation required for clustering objects such as human faces. An object of the present invention is to provide a clustering method, apparatus, and computer program that can reduce

In particular, in the present invention, it is possible to solve the problem that it takes a lot of time to construct and operate a system for clustering objects that are difficult to classify, such as a human face, and it can improve the problem that it takes a lot of time to process a large number of objects. An object of the present invention is to provide a clustering method, an apparatus, and a computer program.

In a clustering method according to an aspect of the present invention for solving the above problems, in a method of performing clustering on a plurality of object samples in a clustering device, cluster generation for generating a plurality of clusters by classifying the plurality of object samples step; an average-feature calculating step of calculating, for each of the plurality of clusters, an average-feature including each average value for a plurality of features used to classify the object sample; and an additional sample clustering step of performing clustering on one or more newly added second object samples using the average-features for the plurality of clusters. It is characterized in that it includes.

A computer program according to another aspect of the present invention is characterized in that it is a computer program stored in a computer-readable medium for executing each step of the above-described clustering method in a computer.

According to another aspect of the present invention, there is provided a clustering apparatus for clustering a plurality of object samples, comprising: a cluster generator configured to generate a plurality of clusters by classifying the plurality of object samples; an average-feature calculation unit for calculating, for each of the plurality of clusters, an average-feature including an average value of each of the plurality of features used to classify the object sample; and an additional sample clustering performing unit configured to perform clustering on one or more newly added second object samples by using the average-features of the plurality of clusters.

According to an embodiment of the present invention, it is possible to dramatically reduce computing resources such as storage space and computational amount required for clustering an object such as a human face and process it quickly and effectively. do.

In addition, according to an embodiment of the present invention, in particular, it solves the problem of huge cost in building and operating a system for clustering objects that are difficult to classify, such as a human face, and requires a considerable amount of time to process a large number of objects. problems can be improved.

BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are included as a part of the detailed description to help the understanding of the present invention, provide embodiments of the present invention, and together with the detailed description, explain the technical spirit of the present invention.
1 is a block diagram of a system for providing a clustering service according to an embodiment of the present invention.
2 is a detailed example of providing a clustering service according to an embodiment of the present invention.
3 is a flowchart of a clustering method according to an embodiment of the present invention.
4 to 5 are diagrams illustrating an object feature and cluster generation using the same according to an embodiment of the present invention.
6 is a diagram illustrating average-feature calculation for a cluster according to an embodiment of the present invention.
7 and 8 are diagrams illustrating cluster merging according to an embodiment of the present invention.
9 is a diagram for explaining a similarity determination using a hashing function according to an embodiment of the present invention.
10 to 11 are diagrams illustrating specific examples of a clustering method according to an embodiment of the present invention.
12 is a diagram illustrating a specific embodiment of a system for providing a clustering service according to an embodiment of the present invention.
13 to 14 are diagrams illustrating pseudo code for a clustering method according to an embodiment of the present invention.
15 to 17 are diagrams for explaining an effect of a clustering method according to an embodiment of the present invention.
18 is a block diagram of a clustering apparatus according to an embodiment of the present invention.

It should be noted that the technical terms used in the present invention are only used to describe specific embodiments, and are not intended to limit the scope of the present invention. In addition, the technical terms used in the present invention should be interpreted as meanings generally understood by those of ordinary skill in the art to which the present invention belongs, unless otherwise defined in the present invention, and excessively comprehensive It should not be construed as a human meaning or in an excessively reduced meaning. In addition, when the technical term used in the present invention is an erroneous technical term that does not accurately express the spirit of the present invention, it should be understood by being replaced with a technical term that can be correctly understood by those skilled in the art. In addition, general terms used in the present invention should be interpreted as defined in advance or according to the context before and after, and should not be interpreted in an excessively reduced meaning.

Also, the singular expression used in the present invention includes the plural expression unless the context clearly dictates otherwise. In the present invention, terms such as "consisting of" or "comprising" should not be construed as necessarily including all of the various elements or several steps described in the invention, some of which elements or some steps are included. It should be construed that it may not, or may further include additional components or steps.

In addition, terms including ordinal numbers such as first, second, etc. used in the present invention may be used to describe the components, but the components are not limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings, but the same or similar components are assigned the same reference numerals regardless of reference numerals, and overlapping descriptions thereof will be omitted.

In addition, in the description of the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, it should be noted that the accompanying drawings are only for easy understanding of the technical spirit of the present invention, and should not be construed as limiting the technical spirit of the present invention by the accompanying drawings.

Hereinafter, exemplary embodiments of a clustering method, apparatus and computer program according to the present invention will be described in detail with reference to the accompanying drawings.

First, FIG. 1 shows a configuration diagram of a system 100 for providing a clustering service according to an embodiment of the present invention. As can be seen in FIG. 1 , the clustering service providing system 100 according to an embodiment of the present invention includes a clustering device 120 and a user for performing clustering on objects such as people or things included in a picture or a video. It may be configured to include the terminal 110 .

In this case, as the user terminal 110 , a mobile terminal such as a smart phone, a tablet PC, a PDA, or a mobile phone may be used, and in addition, various types of terminals such as a personal computer (PC) and a notebook PC may be employed.

In addition, in the present invention, clustering can be performed by classifying objects such as faces of people included in photos or moving pictures, but the present invention is not necessarily limited thereto and can be used for clustering various objects. may be

Accordingly, in the clustering device 120 according to an embodiment of the present invention, as can be seen in FIG. 2 , an object (FIG. 2(a)) such as a human face included in a photo or a video is classified and clustered. (cluster) is created (FIG. 2 (b)), and using this, content such as a photo or video including a specific person is provided to the user terminal 110 according to the user's request, etc. It may also be possible (FIG. 2(c)).

In this case, the clustering device 120 may be implemented using one or two or more servers, but the present invention is not necessarily limited thereto. It may be implemented using various devices capable of processing information, such as a system.

In addition, the network 130 connecting the user terminal 110 and the clustering device 120 may include a wired network and a wireless network, and specifically, a local area network (LAN), a metropolitan area network (MAN) : It may include various communication networks such as Metropolitan Area Network) and Wide Area Network (WAN). In addition, the network 130 may include a well-known World Wide Web (WWW). However, the communication network 130 according to the present invention is not limited to the above-listed networks, and may include at least a part of a well-known wireless data network, a well-known telephone network, or a well-known wired/wireless television network.

3 is a flowchart of a clustering method according to an embodiment of the present invention. As can be seen in FIG. 3 , in the clustering method according to an embodiment of the present invention, in the clustering apparatus 120 performing clustering on a plurality of object samples, the plurality of object samples are classified A cluster generation step (S110) of generating a cluster of S120) and an additional sample clustering step (S130) of performing clustering on one or more newly added second object samples using the average-features of the plurality of clusters.

Hereinafter, the clustering method according to an embodiment of the present invention will be reviewed in detail with reference to FIG. 3 by dividing each step.

First, in the cluster generation step ( S110 ), a plurality of clusters are generated by classifying the plurality of object samples.

In this case, in the cluster generation step 110 as an embodiment of the present invention, as shown in FIG. 4 , clustering may be performed on the plurality of object samples using a conventional clustering algorithm such as DBSCAN. However, the present invention is not necessarily limited thereto, and various algorithms capable of appropriately clustering objects such as a human face may be used.

More specifically, as can be seen in FIG. 5 , in the cluster creation step ( S110 ), an object ( 502 , 503 in FIG. 5 ) such as a human face included in a photo 501 or a video may be identified first. .

Subsequently, features are calculated for the identified objects ( 502 and 503 in FIG. 5 ). In particular, in the case of an object that is difficult to classify, such as a human face, it is necessary to use a plurality of features to secure accuracy.

In contrast, the feature of the object may be implemented as a multi-dimensional feature vector ( 504 and 505 in FIG. 5 ).

As a more specific example, in the clustering method according to an embodiment of the present invention, a 512-dimensional feature vector may be used for accurate classification of a human face. In this case, a 512-dimensional feature vector for each object sample is obtained. A storage space for storing and an amount of computation for an operation may be greatly increased.

In contrast, in the clustering method according to an embodiment of the present invention, by using a mean-feature for each cluster including a plurality of object samples instead of the feature vector for each object sample, the object In addition to dramatically reducing computational resources such as storage space and computational amount required for performing clustering, it can be processed quickly and effectively, and in particular, it clusters objects that are difficult to classify, such as human faces. It is possible to solve the problem of costly construction and operation of the system for this purpose, and to improve the problem of taking a considerable amount of time to process a large number of objects.

Accordingly, in the average-feature calculation step S120, for each of the plurality of clusters, an average-feature including each average value of the plurality of features used to classify the object sample is calculated.

In this case, in the average-feature calculation step S120, with respect to the N-dimensional feature vector used to classify the object sample, the average value of a plurality of object samples for each dimension is calculated to calculate the N-dimensional average-feature can do.

More specifically, the average-feature may be calculated using Equation 1 below.

[Equation 1]

Equation 1 exemplifies the case of calculating the average-feature by calculating the average value for each dimension with respect to n feature vectors having 512 dimensions.

However, the present invention is not limited thereto, and dimensions and the like may vary depending on the characteristics of an object used, and furthermore, it is possible to configure the average-feature in a format other than a feature vector.

More specifically, as can be seen in FIG. 6 , when a plurality of features (F1-F5 in FIG. 6(a)) for five object samples exist, in the cluster generation step S110, an algorithm such as DBSCAN A cluster, which is a cluster of similar objects, can be generated by comparing the similarity between features ((a)-(d) of FIG. 6) using

Next, in the average-feature calculation step S120, average values for a plurality of features included in the cluster (F1-F5 in FIG. 6(e)) are calculated so that the average-features for the cluster can be calculated. (M1 in Fig. 6(f)).

Furthermore, in the clustering method according to an embodiment of the present invention, in the mean-feature calculation step S120, when the distance between the mean-feature of the first cluster and the mean-feature of the second cluster is within a predetermined reference value , a third cluster may be created by merging the first cluster and the second cluster.

More specifically, as can be seen in FIG. 7 , clustering is performed on a plurality of features to form a first cluster (F2-F5 in FIG. 7A ) and a second cluster (F3-F4 in FIG. 7A ). ) is calculated, and the distance between the average feature for the first cluster (M1 in Fig. 7(b)) and the average-feature (M2 in Fig. 7(b)) of the second cluster is within a predetermined reference value, In the present invention, a third cluster is created by merging the first cluster and the second cluster, and an average feature (M3 in FIG. 7(c)) is calculated and used for clustering, through which the average-number of features It is possible to more effectively suppress the increase in the storage space and the amount of computation by preventing the increase in .

Furthermore, in the clustering method according to an embodiment of the present invention, the average-feature for the third cluster considers the number of object samples included in the first cluster and the number of object samples included in the second cluster. Thus, the weighted average of the average-features of the first cluster and the average-features of the second cluster may be calculated.

As a more specific example, as shown in FIG. 8 , when a plurality of clusters are generated using an algorithm such as DBSCAN and an average-feature for each cluster is generated (M1-M4 in FIG. 8(a)) ), when the average-feature distance of some clusters among the plurality of clusters is within a predetermined reference value, a new cluster may be created by merging the some clusters (M1-M3 in FIG. 8(b)), in which case the The mean-feature (C1 in FIG. 8(b)) for the new cluster is the number of object samples included in the first cluster (3 corresponding to M1 in FIG. 8(b)) and included in the second cluster. In consideration of the number of object samples (four corresponding to M2 in FIG. 8(b)) and the third cluster (four corresponding to M3 in FIG. 8(b)), the average-features of each cluster are weighted. It can be calculated by averaging (C1 = 3/11 * M1 + 4/11 * M2 + 4/11 * M3 in FIG. 8(b)).

Furthermore, in the clustering method according to an embodiment of the present invention, when the average-feature distance of the cluster generated by merging the average-feature of the existing cluster is within a predetermined reference value, the existing cluster and the newly generated The process of creating a new cluster by merging the old clusters again can be repeated.

As a more specific example, in FIG. 8(c) , the average-feature of a fifth cluster (C1 in FIG. 8(c)) generated by merging the first to third clusters and the existing fourth cluster When the distance between the mean-features (M4 in FIG. 8(c)) is within a predetermined reference value, a sixth cluster may be created by merging the clusters (M1-M4 in FIG. 8(c)), in which case the first The average-feature for 6 clusters (C2 in FIG. 8(c)) is the number of object samples included in the fifth cluster (11 in FIG. 8(c)) and the number of object samples included in the fourth cluster. Considering the number (four corresponding to M4 in FIG. 8(c) ), the average-features of each cluster may be weighted and calculated (C2 = 11/15 * C1 + 4 in FIG. 8(c) ) /15 * M4).

Accordingly, in the clustering apparatus 120 and method according to an embodiment of the present invention, in order to perform clustering, the average-feature for each cluster is stored without storing the feature for each object sample belonging to a plurality of clusters. Therefore, it is possible to dramatically reduce computational resources such as storage space and computational amount required to perform clustering on objects, and perform fast and effective clustering.

Next, in the additional sample clustering step ( S130 ), clustering is performed on one or more newly added second object samples using the average-features of the plurality of clusters.

In this case, in the additional sample clustering step ( S130 ), when one or more new clusters are generated by the second object sample, an average-feature for the new cluster may be calculated and used for clustering.

In addition, in the additional sample clustering step ( S130 ), when the number of object samples included in one or more clusters is changed by the second object sample, the average-features for the one or more clusters may be updated and used for clustering. have.

Accordingly, in the clustering method according to an embodiment of the present invention, it is possible to quickly process the clustering task by effectively reducing the amount of computation required to perform clustering on an object such as a human face.

In addition, in the clustering method according to an embodiment of the present invention, in the additional sample clustering step ( S130 ), the average-feature for the existing first cluster and the second cluster added or changed by the second object sample When the average-feature distance is within a predetermined reference value, a third cluster may be generated by merging the first cluster and the second cluster.

Furthermore, in the additional sample clustering step (S130), the average-feature for the third cluster generated by the merging is the number of object samples included in the first cluster and the number of object samples included in the second cluster. In consideration of the number, the weighted average may be calculated by averaging the average-feature of the first cluster and the average-feature of the second cluster.

In addition, in the additional sample clustering step ( S130 ), when the distance between the mean-feature for the existing fourth cluster and the mean-feature of the merged third cluster is within a predetermined reference value, the fourth cluster and the The process of creating a new cluster by merging the third cluster again may be repeated.

In addition, in the clustering method according to an embodiment of the present invention, in the additional sample clustering step ( S130 ), the similarity is determined by limiting the object sample or cluster corresponding to the hashing function value for the second object sample. to perform clustering.

As a more specific example, as can be seen in FIG. 9 , the hashing function value corresponding to the first cluster ( 910 in FIG. 9 ) corresponds to the first value (Bucket 1), and the newly added second object sample ( When the hashing function value of 940 of FIG. 9 also corresponds to the first value (Bucket 1), in the additional sample clustering step (S130), it is limited to an object sample or cluster corresponding to the first value (Bucket 1). By determining the similarity and performing clustering, it is possible to effectively reduce the amount of computation.

Furthermore, in FIG. 9, cases corresponding to a plurality of different hashing function values (Bucket 3 and Bucket 4) as in the second cluster (920 in FIG. 9) in FIG. 9 and cases corresponding to the plurality of hashing function values are considered. It is also possible to determine the degree of similarity with respect to the added object sample.

10 and 11 show specific examples of a clustering method and apparatus according to an embodiment of the present invention.

First, as can be seen in FIG. 10 , when the clustering process is started, the LSH (Local Sensitive Hashing) block 1010 corresponds to the hashing function value for the object sample to be processed as described in FIG. 9 above. Comparison candidates are extracted so that similarity can be compared by limiting to object samples or clusters.

Also, in the DBSCAN block 1020 , the clustering operation is performed by comparing the similarity with the average-feature of the object sample or cluster extracted as the comparison candidate using the DBSCAN algorithm. In this case, as described above, various clustering algorithms may be used in addition to the DBSCAN algorithm.

Subsequently, in the mean-feature block 1030 , a mean-feature for the generated cluster is calculated or updated.

Also, in the Merge Mean-feature block 1040 , similar mean-features located within a predetermined distance may be merged.

Furthermore, as these processes are repeatedly performed, clusters and mean-features are additionally merged, thereby further reducing storage space and computational amount required for clustering.

In addition, as can be seen in FIG. 11, in the prior art (FIG. 11(a)), for clustering one object sample (A1 in FIG. 11(a)), a plurality of object samples (FIG. 11(a)) F2-F8) had to be calculated, but in the clustering method according to an embodiment of the present invention, a cluster is created for a plurality of object samples ((b3 in FIG. 11)) through the DBSCAN block 1020, and , after calculating the average-feature through the mean-feature block 1030 ((b2) of FIG. 11), the object sample or cluster extracted as a comparison candidate through the LSH block 1010 By limiting and comparing similarities, it is possible to dramatically reduce the amount of storage space and computation required for clustering.

12 shows a specific embodiment of the clustering service providing system 100 according to an embodiment of the present invention.

12, in the clustering service providing system 100 according to an embodiment of the present invention, the dataset 1210 may be composed of image data 1211 and label information 1212, and the like. It may be stored in a specific storage 1220 .

In contrast, the image data extraction server 1230 has a function 1231 for extracting various types of metadata and feature information from the image of the dataset 1210 , and the extracted metadata and feature information are stored in the database 1240 . can be stored in

Next, the clustering simulator 1250 performs a clustering operation by applying the clustering method according to the present invention as described with reference to FIGS. 10 to 11 .

At this time, the user can execute the simulator through the web UI 1260 ( 1261 ), and perform visualization ( 1262 ) and quality evaluation ( 1263 ) of the calculated result.

13 to 14 illustrate pseudo code for a clustering method according to an embodiment of the present invention.

First, FIG. 13 exemplifies a pseudo code for calculating an average-feature for merging a plurality of clusters in a clustering method according to an embodiment of the present invention.

13, in calculating the new average-feature (new_feature in FIG. 13) for the merging of the plurality of clusters, the number of object samples included in each cluster is considered (weight:= cluster. elements), which can be calculated by averaging the average-features of each cluster.

Also, FIG. 14 exemplifies pseudo code for classifying a new object sample (element) using an average-feature in the clustering method according to an embodiment of the present invention.

As can be seen in FIG. 14 , for each of the plurality of clusters, the new object sample to which the new object sample belongs by considering the distance between the mean-feature of each cluster and the new object sample (getDistance(cluster, element)) A target cluster (target_cluster) may be calculated.

Furthermore, FIGS. 15 to 17 show the degree of reduction in the storage space and the amount of computation and the clustering performance in the clustering method according to an embodiment of the present invention, respectively.

At this time, in FIGS. 15 to 17, S1 is the result of the conventional DBSCAN algorithm, S2 is the result of applying the mean-feature according to the present invention to the DBSCAN algorithm, and S3 is the DBSCAN algorithm according to the present invention. S4 is the result of applying mean-feature and mean-feature merged (merged mean-feature), and S4 is the mean-feature, mean-feature merged (merged mean-) according to the present invention to the DBSCAN algorithm. feature) and LSH (Local Sensitive Hashing) are applied.

15 to 17, PRESET1 for 12 people, PRESET2 for 30 people, PRESET3 for 60 people, and PRESET4 for 90 people. Or 20, it refers to the case where the test data set is configured and simulation is performed.

First, as can be seen in FIG. 15 , when comparing the required storage space, when the storage space required in the conventional DBSCAN algorithm (S1) is 100 and compared, in S2-S4 according to the present invention, each of 22.83 to 31.64 is As it has a range, it can be confirmed that the required storage space can be remarkably reduced. Furthermore, in terms of storage space, the S3 showed relatively better results.

In addition, as can be seen in FIG. 16, when comparing the amount of computation required, when comparing the amount of computation required in the conventional DBSCAN algorithm (S1) as 100, in S2-S4 according to the present invention, the range of 14.26 to 48.23 is respectively As a result, it can be confirmed that the amount of computation required can also be remarkably reduced. Furthermore, in terms of computational amount, S4 showed relatively better results.

In addition, as can be seen in FIG. 17, when comparing the clustering performance, when comparing the clustering performance (based on the Jaccard Index) in the conventional DBSCAN algorithm (S1) as 100, in S2-S4 according to the present invention, each of 103.23 to 103.23 to Since it has a range of 115.44, it can be confirmed that the degradation of clustering performance is not too large, which is an acceptable level.

In addition, the computer program according to another aspect of the present invention is characterized in that it is a computer program stored in a computer-readable medium in order to execute each step of the above salpin clustering method in a computer. The computer program may be a computer program including a machine language code generated by a compiler, as well as a computer program including a high-level language code that can be executed in a computer using an interpreter or the like. In this case, the computer is not limited to a personal computer (PC) or notebook computer, and includes a central processing unit (CPU) such as a server, smart phone, tablet PC, PDA, mobile phone, etc. to process any information that can execute a computer program. includes the device.

In addition, the medium may continuously store a program executable by a computer, or may be temporarily stored for execution or download. In addition, the medium may be various recording means or storage means in the form of a single or several hardware combined, it is not limited to a medium directly connected to any computer system, and may exist distributed on a network. Examples of the medium include a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk, and those configured to store program instructions, including ROM, RAM, flash memory, and the like. In addition, examples of other media may include recording media or storage media managed by an app store that distributes applications, sites that supply or distribute various other software, or servers.

In addition, FIG. 18 exemplifies the configuration of the clustering device 120 according to an embodiment of the present invention.

As can be seen in FIG. 18 , the clustering device 120 according to an embodiment of the present invention includes a cluster generator 121 , an average-feature calculator 122 , and an additional sample clustering performer 123 . can be

Next, below, the clustering device 120 according to an embodiment of the present invention will be divided for each component. A more detailed description of the clustering device 120 according to an embodiment of the present invention can be inferred from the description of the clustering method according to an embodiment of the present invention described above, and a more detailed description will be omitted below.

First, the cluster generating unit 121 generates a plurality of clusters by classifying the plurality of object samples.

In addition, the average-feature calculating unit 122 calculates an average-feature including each average value of a plurality of features used to classify the object sample with respect to each of the plurality of clusters.

Finally, the additional sample clustering performing unit 123 performs clustering on one or more newly added second object samples using the average-features of the plurality of clusters.

Furthermore, in the clustering apparatus 120 according to an embodiment of the present invention, the average-feature calculating unit 122 includes a plurality of N-dimensional feature vectors for each dimension with respect to the N-dimensional feature vector used to classify the object sample. By calculating the average value of the object sample, the N-dimensional average-feature may be calculated.

In addition, the average-feature calculating unit 122 merges the first cluster and the second cluster when the distance between the average-feature of the first cluster and the average-feature of the second cluster is within a predetermined reference value. to create a third cluster.

Furthermore, when the distance between the mean-feature for the existing fourth cluster and the mean-feature of the merged third cluster is within a predetermined reference value, the fourth cluster and the third cluster are merged again to form a new cluster. The creation process may be repeated.

Also, the additional sample clustering performing unit 123 may perform clustering by determining the similarity by limiting the object sample or cluster corresponding to the hashing function value of the second object sample.

Accordingly, in the clustering method, apparatus, and computer program according to an embodiment of the present invention, computing resources such as storage space and amount of computation required for clustering an object such as a human face can be drastically reduced and processed quickly and effectively.

In addition, in the clustering method, apparatus, and computer program according to an embodiment of the present invention, in particular, the problem of building and operating a system for clustering objects that are difficult to classify, such as a human face, is solved, and a large amount of money is required. It is possible to improve the problem of taking a significant amount of time to process the object of

The above embodiments and drawings described herein are illustrative only, and do not limit the scope of the present invention in any way. In addition, the connection or connection members of the lines between the components shown in the drawings illustratively represent functional connections and/or physical or circuit connections, and in an actual device, various functional connections, physical connections that are replaceable or additional may be referred to as connections, or circuit connections. In addition, unless there is a specific reference such as "essential", "importantly", etc., it may not be a necessary component for the application of the present invention.

In the specification of the present invention (especially in the claims), the use of the term "above" and similar referential terms may be used in both the singular and the plural. In addition, when a range is described in the present invention, each individual value constituting the range is described in the detailed description of the invention as including the invention to which individual values belonging to the range are applied (unless there is a description to the contrary). same as In addition, the steps presented in the method invention among the present invention are not necessarily intended to limit the order of their precedence and after, and the order may be appropriately changed as necessary unless a certain step must be preceded according to the nature of each process. have. The use of all examples or exemplary terms (eg, etc.) in the present invention is merely for the purpose of describing the present invention in detail, and unless defined by the claims, the scope of the present invention is limited by the examples or exemplary terminology. it's not going to be In addition, those skilled in the art can understand that various modifications, combinations, and changes can be made according to design conditions and elements within the scope of the appended claims or their equivalents.

100: clustering service providing system
110, 110a, 110b: user terminal
120: clustering device
121: cluster creation unit
122: average-feature calculation unit
123: Additional sample clustering unit
130: network

Claims (16)

A method for clustering a plurality of object samples in a clustering device, the method comprising:
a cluster generation step of classifying the plurality of object samples to generate a plurality of clusters;
an average-feature calculating step of calculating, for each of the plurality of clusters, an average-feature including each average value for a plurality of features used to classify the object sample; and
an additional sample clustering step of performing clustering on one or more newly added second object samples using the average-features for the plurality of clusters;
A clustering method comprising a. According to claim 1,
In the average-feature calculation step,
For an N-dimensional feature vector used to classify the object sample,
A clustering method, characterized in that by calculating an average value of a plurality of object samples for each dimension, an average of N-dimensional features is calculated. According to claim 1,
In the additional sample clustering step,
and calculating a mean-feature for the new cluster when one or more new clusters are generated by the second object sample. According to claim 1,
In the additional sample clustering step,
and updating the average-feature for the one or more clusters when the number of object samples included in the one or more clusters is changed by the second object sample. According to claim 1,
In the average-feature calculation step,
When the distance between the mean-feature of the first cluster and the mean-feature of the second cluster is within a predetermined reference value,
The clustering method of claim 1, wherein the first cluster and the second cluster are merged to generate a third cluster. 6. The method of claim 5,
The mean-feature for the third cluster is
Considering the number of object samples included in the first cluster and the number of object samples included in the second cluster,
Clustering method, characterized in that calculated by weighted average of the mean-feature of the first cluster and the mean-feature of the second cluster. 6. The method of claim 5,
When the distance between the mean-feature for the existing fourth cluster and the mean-feature of the merged third cluster is within a predetermined reference value,
and repeating the process of creating a new cluster by merging the fourth cluster and the third cluster again. According to claim 1,
In the additional sample clustering step,
When the distance between the mean-feature of the existing first cluster and the mean-feature of the second cluster added or changed by the second object sample is within a predetermined reference value,
The clustering method of claim 1, wherein the first cluster and the second cluster are merged to generate a third cluster. According to claim 1,
In the additional sample clustering step,
A clustering method, characterized in that clustering is performed by determining similarity by limiting to an object sample or cluster corresponding to a hashing function value for the second object sample. According to claim 1,
In the clustering device,
A clustering method, characterized in that in order to perform clustering, the average-feature for each cluster is stored and used instead of storing the feature for each object sample belonging to a plurality of clusters. A computer program stored in a computer-readable medium for executing each step of the clustering method according to any one of claims 1 to 10 on a computer. A clustering device for clustering a plurality of object samples, the clustering device comprising:
a cluster generator to classify the plurality of object samples to generate a plurality of clusters;
an average-feature calculating unit for calculating, for each of the plurality of clusters, an average-feature including an average value for a plurality of features used to classify the object sample; and
an additional sample clustering performing unit configured to cluster one or more newly added second object samples by using the average-features of the plurality of clusters;
A clustering device comprising a. 13. The method of claim 12,
In the average-feature calculation unit,
For an N-dimensional feature vector used to classify the object sample,
A clustering device, characterized in that by calculating an average value of a plurality of object samples for each dimension, an average of N-dimensional features is calculated. 13. The method of claim 12,
In the average-feature calculation unit,
When the distance between the mean-feature of the first cluster and the mean-feature of the second cluster is within a predetermined reference value,
and generating a third cluster by merging the first cluster and the second cluster. 15. The method of claim 14,
When the distance between the mean-feature for the existing fourth cluster and the mean-feature of the merged third cluster is within a predetermined reference value,
and repeating the process of creating a new cluster by merging the fourth cluster and the third cluster again. 15. The method of claim 14,
In the additional sample clustering performing unit,
The clustering apparatus according to claim 1, wherein the clustering is performed by determining the similarity by limiting the object sample or cluster corresponding to the hashing function value of the second object sample.

Images