WO2013051619A1 - 類似性検出装置及び指向性近傍検出方法 - Google Patents
類似性検出装置及び指向性近傍検出方法 Download PDFInfo
- Publication number
- WO2013051619A1 WO2013051619A1 PCT/JP2012/075673 JP2012075673W WO2013051619A1 WO 2013051619 A1 WO2013051619 A1 WO 2013051619A1 JP 2012075673 W JP2012075673 W JP 2012075673W WO 2013051619 A1 WO2013051619 A1 WO 2013051619A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- data
- similarity
- search
- parameter
- detection apparatus
- Prior art date
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 114
- 238000000034 method Methods 0.000 title description 31
- 230000004044 response Effects 0.000 claims description 4
- 230000006870 function Effects 0.000 abstract description 88
- 238000004364 calculation method Methods 0.000 abstract description 78
- 238000007726 management method Methods 0.000 description 85
- 239000013598 vector Substances 0.000 description 49
- 238000012545 processing Methods 0.000 description 30
- 238000009826 distribution Methods 0.000 description 25
- 238000004590 computer program Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- IYLGZMTXKJYONK-ACLXAEORSA-N (12s,15r)-15-hydroxy-11,16-dioxo-15,20-dihydrosenecionan-12-yl acetate Chemical compound O1C(=O)[C@](CC)(O)C[C@@H](C)[C@](C)(OC(C)=O)C(=O)OCC2=CCN3[C@H]2[C@H]1CC3 IYLGZMTXKJYONK-ACLXAEORSA-N 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- IYLGZMTXKJYONK-UHFFFAOYSA-N ruwenine Natural products O1C(=O)C(CC)(O)CC(C)C(C)(OC(C)=O)C(=O)OCC2=CCN3C2C1CC3 IYLGZMTXKJYONK-UHFFFAOYSA-N 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000009827 uniform distribution Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2457—Query processing with adaptation to user needs
- G06F16/24575—Query processing with adaptation to user needs using context
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/285—Clustering or classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/766—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using regression, e.g. by projecting features on hyperplanes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
Definitions
- the present invention relates to a feature detection apparatus and a directivity vicinity detection method, and more particularly to a feature detection apparatus and a directivity vicinity detection method for retrieving desired data from a multidimensional continuous data set.
- the neighborhood search technique is an important technique used in various application ranges. For example, in consideration of the similarity of data such as images and web pages, data having similar characteristics is detected from a huge database. Used for applications. As a very simple method, each data feature is expressed as a point in a space with a defined distance (for example, a point in Euclidean space), and the distance to all data from the query point is calculated. Close data can be detected. However, such a method for directly calculating the distance is very costly, and the calculation load increases nonlinearly as the number of data increases, so various methods have been proposed.
- Japanese Patent Laid-Open No. 2004-021430 discloses an image search apparatus that can accurately search for an image including a specific subject as a similar image without complicating the user's operation.
- the image search apparatus is an image search apparatus that extracts an image similar to a reference image as a search key from a search target image group including a plurality of search target images, the reference image and the search target image group Included in the search target image group, means for dividing each image included in the image into a plurality of areas, means for extracting at least one feature amount from each area of the image included in the reference image and the search target image group, and the search target image group
- Patent Document 2 Japanese Patent Application Laid-Open No. 2005-070927 discloses an image feature acquisition method for acquiring an image feature that can detect an image having a similar drawing shape regardless of the degree of coincidence of images. Is disclosed.
- the image feature acquisition method is an image feature acquisition method for acquiring an image feature of a two-dimensional image, and the size of the two-dimensional image is changed to a predetermined size.
- a horizontal component one-dimensional raster image is created by sequentially connecting the starting point of the pixel component in the vertical direction, and the pixel column that continues in the vertical direction from the starting point set at either the top or bottom of the two-dimensional image is defined as the end point of the pixel column. And the starting point of the next pixel row are connected in order in the horizontal direction to create a one-dimensional raster image of the vertical component, and appropriate conversion processing is executed on these one-dimensional raster images.
- Patent Document 3 Japanese Patent Laid-Open No. 10-326286 discloses a similarity search apparatus that can improve the accuracy of similarity search and has a low possibility that important similar data is omitted from the search result.
- the similarity search device creates a vector database for storing a plurality of vector data, each of which is created for each of a plurality of objects and has a plurality of attributes that characterize the object as vector components, and vector data for a specified similar search object Target vector data creation unit, a search condition set generation unit that generates a plurality of search conditions, and a plurality of vector data stored in the vector database for each search condition generated by the search condition set generation unit
- Similar search engine that searches for vector data that satisfies the above search condition and that is similar to the target vector data, and a search result display that displays the results searched by the similar search engine for each search condition It is characterized by having a part.
- LSH Location Sensitive Hashing
- the probability that the data p collides with a certain query q depends only on the distance d (p, q), and therefore on the circumference of one circle centered on the query q. A plurality of data all collide with the query q with equal probability.
- JP 2004-021430 A JP-A-2005-070927 JP-A-10-326286
- the first problem is that the detailed positional relationship represented by the direction is not considered as an index of the closeness of the features of the two data. Specifically, when a distance in one direction is more important than a distance in another direction, it is not possible to perform a neighborhood search while distinguishing the difference in direction. The reason is that the conventional neighborhood search uses a method that depends only on the distance.
- the second problem is that when only the similarity to a certain viewpoint is concerned, the conventional method takes extra calculation time.
- the reason is that the conventional method does not consider the direction, and therefore, after extracting candidates by proximity detection in all directions, a two-step process of extracting only data along the direction of interest is necessary. is there.
- the greater the difference in height of interest depending on the direction when the difference between the weight of the distance along one direction and the weight of the distance along the other direction is very large), the greater the number of candidates detected in the neighborhood, The effect of reducing the calculation time cannot be expected so much.
- the object of the present invention is to specify an arbitrary similarity criterion (arbitrary direction and importance of an arbitrary distance) in neighborhood detection in a space where the difference in characteristics is defined as a distance, It is an object to provide a similarity detection apparatus and a directivity vicinity detection method for detecting a signal at high speed.
- the similarity detection apparatus includes a random number generation unit, an initialization unit, a data registration unit, a search unit, and a data processing unit.
- the random number generator calculates a plurality of pieces of random number information based on a plurality of direction parameters and a plurality of intensity parameters input via the input device.
- the plurality of direction parameters indicate directions in the Euclidean space.
- the initialization unit calculates a plurality of key calculation functions based on the plurality of random number information.
- the data registration unit calculates a plurality of tables corresponding to the plurality of key calculation functions based on the plurality of search target data input through the input device, and records the plurality of tables in the table holding device. .
- Each of the plurality of search target data indicates a point on the Euclidean space.
- the distance between two points respectively indicated by two arbitrary search target data among the plurality of search target data indicates the similarity between the two search target data.
- a table corresponding to an arbitrary key calculation function among the plurality of tables associates a plurality of keys with a plurality of data lists, and data belonging to a data list corresponding to any key of the plurality of data lists The value calculated by being assigned to the arbitrary key calculation function is calculated to be equal to the arbitrary key.
- the search unit refers to the plurality of tables and calculates a candidate data list based on the query indicated by the search condition input via the input device.
- the candidate data list includes a plurality of search data lists corresponding to the plurality of key calculation functions.
- the search data list corresponding to the arbitrary key calculation function of the plurality of search data lists is a query value calculated by substituting the query into the key calculation function of the plurality of data lists.
- a data list corresponding to is shown.
- the data processing unit calculates search result data satisfying the condition indicated by the search condition from a plurality of search data belonging to the candidate data list, and outputs the search result data to the output device.
- a plurality of pieces of random number information are calculated based on a plurality of direction parameters and a plurality of intensity parameters input via the input device.
- a plurality of key calculation functions are calculated based on the plurality of random number information.
- a plurality of tables corresponding to the plurality of key calculation functions are calculated based on a plurality of search target data input via the input device.
- the plurality of tables are recorded in a table holding device.
- the candidate data list is calculated based on the query indicated by the search condition input via the input device. Further, search result data satisfying the condition indicated by the search condition is calculated from a plurality of search data belonging to the candidate data list.
- the search result data is output to an output device.
- the plurality of direction parameters indicate directions in the Euclidean space.
- Each of the plurality of search target data indicates a point on the Euclidean space.
- the distance between two points respectively indicated by two arbitrary search target data among the plurality of search target data indicates the similarity between the two search target data.
- a table corresponding to an arbitrary key calculation function among the plurality of tables associates a plurality of keys with a plurality of data lists, and data belonging to a data list corresponding to any key of the plurality of data lists
- the value calculated by being assigned to the arbitrary key calculation function is calculated to be equal to the arbitrary key.
- the candidate data list includes a plurality of search data lists corresponding to the plurality of key calculation functions.
- the search data list corresponding to the arbitrary key calculation function of the plurality of search data lists is a query value calculated by substituting the query into the key calculation function of the plurality of data lists.
- a data list corresponding to is shown.
- the first effect is that it is possible to detect a neighboring point with a high degree of freedom along a certain direction of interest from a certain query point in an enormous amount of data.
- the reason is that the data management function has a table in which any two points are registered in the same entry with a higher probability as the distance measured by any distance weight (importance) along any direction of interest is shorter. This is because the provision of the proximity proximity detection function is realized.
- the second effect is speeding up of the directional neighborhood search process.
- the reason is that by registering data in a plurality of tables in which desired directivities are registered in advance, only data registered in the same entry as the query point can be handled in the directivity neighborhood search from the query point. This is because the online processing time is greatly shortened because it is not necessary to calculate the distance to all data.
- FIG. 1 is a block diagram showing a similarity detection apparatus according to the present invention.
- FIG. 2 is a block diagram showing the table management apparatus.
- FIG. 3 is a flowchart showing the initialization operation.
- FIG. 4 is a flowchart showing the search operation.
- FIG. 5 is a table showing examples of directivity parameters.
- FIG. 6 is a graph showing collision probability isolines corresponding to directivity parameters.
- FIG. 7 is a table showing examples of directivity parameters.
- FIG. 8 is a table showing an example of a plurality of table sets.
- FIG. 9 is a graph showing a plurality of collision probability isolines corresponding to a plurality of directivity parameters.
- FIG. 10 shows an image.
- FIG. 11 is a graph showing a plurality of search target data.
- FIG. 11 is a graph showing a plurality of search target data.
- FIG. 12 is a table showing an example of a plurality of table sets.
- FIG. 13 is a graph illustrating an example of a search condition query.
- FIG. 14 is a graph showing an example of search result data.
- FIG. 15 is a graph showing a plurality of time-series data.
- FIG. 16 is a graph showing a plurality of search target data among a plurality of time series data.
- FIG. 17 is a graph showing a plurality of collision probability isolines.
- FIG. 18 is a graph showing a plurality of other collision probability isolines.
- FIG. 19 is a graph showing the collision probability distribution.
- FIG. 20 is a graph showing another collision probability distribution.
- FIG. 21 is a graph showing the feature distribution.
- FIG. 22 is a block diagram showing a mobile management system.
- FIG. 23 is a table showing a plurality of table sets.
- the similarity detection apparatus 1 is connected so that a plurality of computers can transmit information to each other in both directions.
- each of the plurality of computers includes a CPU, a storage device, and an interface.
- the CPU controls the storage device and the interface by executing a computer program installed in the computer.
- the storage device records the computer program and temporarily records information created by the CPU.
- the interface outputs information generated by an external device connected to the computer to the CPU, and outputs information generated by the CPU to the external device.
- Examples of the external device include an input device, an output device, a communication device, and a removable memory drive.
- the input device creates information by being operated by the user, and outputs the information to the CPU.
- Examples of the input device include a keyboard, a pointing device, and a touch panel.
- the output device outputs information generated by the CPU so that the user can recognize it.
- Examples of the output device include a display, an acoustic device, and a touch panel.
- the communication device transmits information created by the CPU via a communication network to another computer, and outputs information received from the other computer via the communication network to the CPU.
- the communication device is further used to download a computer program installed in the computer from another computer.
- the removable memory drive is used to read data recorded on the recording medium when the recording medium is inserted.
- the removable memory drive is further used when the computer program is installed in the computer when a recording medium in which the computer program is recorded is inserted.
- Examples of the recording medium include a magnetic disk (flexible disk, hard disk), an optical disk (CD, DVD), a magneto-optical disk, and a flash memory.
- the plurality of computers include a parameter management device 2, a random number generation device 3, a table management device 5, and a data processing device 6.
- the parameter management device 2 includes an input device 7.
- the parameter management device 2 controls the input device 7 so that the parameter list 8 is input to the parameter management device 2 via the input device 7.
- the random number generation device 3 calculates the random number information based on the information output from the table management device 5 when the random number information is requested from the table management device 5.
- the table management device 5 includes a table holding device 10 and an input device 11.
- the table management device 5 controls the input device 11 so that the input data 12 is input to the table management device 5 via the input device 11.
- the table management device 5 calculates a plurality of tables based on the input data 12 and the random number information calculated by the random number generation device 3.
- the table management device 5 controls the table holding device 10 so that the plurality of tables are recorded in the table holding device 10.
- the table management device 5 further calculates a data list based on the information output from the data processing device 6 by referring to the plurality of tables when the data processing device 6 requests the data list.
- the data list is output to the data processing device 6.
- the data processing device 6 includes an input device 14 and an output device 15.
- the data processing device 6 controls the input device 14 so that the search condition 16 is input to the data processing device 6 via the input device 14.
- a search condition 16 indicates a query and a condition.
- the data processing device 6 further requests the data list from the table management device 5 by outputting the query to the table management device 5.
- the data processing device 6 calculates search result data based on the conditions and the data list calculated by the table management device 5.
- the search result data indicates data satisfying the condition among the data indicated by the data list.
- the data processing device 6 further controls the output device 15 so that the search result data calculated by the table management device 5 is expressed so as to be recognized by the user.
- FIG. 2 shows the table management device 5.
- the computer program installed in the table management device 5 is formed of a plurality of computer programs that cause the table management device 5 to realize a plurality of functions.
- the table management device 5 When the table management device 5 realizes the plurality of functions, the table management device 5 includes a random number information acquisition unit 21, an initialization unit 22, a data registration unit 23, and a search unit 24.
- the random number information acquisition unit 21 collects the parameter list 8 input to the parameter management device 2 from the parameter management device 2.
- the parameter list 8 shows table parameters, management parameters, and directional parameters.
- the table parameters indicate the number of table faces P, the window width W, the radix C, and the bit length B.
- the number of faces P indicates a positive integer.
- the window width W indicates a positive real number.
- the cardinal number C represents an integer of 2 or more.
- the bit length B indicates a positive integer.
- the management parameter indicates the number of dimensions D.
- the directivity parameter indicates the set U.
- the set U is formed from the directivity parameter u j and is expressed by the following equation.
- U ⁇ u j
- u j ⁇ R D ⁇ R + ⁇ (j 1,..., D)
- the set RD indicates a D-dimensional number vector space.
- the set R 2 represents a 2-dimensional number vector space.
- the random number information acquisition unit 21 requests the random number generator 3 for random number information by outputting the window width W, the bit length B, the dimension number D, and the set U in the parameter list 8 to the random number generator 3.
- the random number information acquisition unit 21 collects random number information calculated by the random number generation device 3 from the random number generation device 3.
- the random number information acquisition unit 21 repeats the request P times, collects a plurality (P pieces) of random number information from the random number generation device 3, and creates a random number information set Z.
- the set Z p indicates one piece of random number information calculated by the random number generation device 3.
- the set Z p is expressed by the following equation.
- Z p ⁇ Z b
- the random number R (p) b indicates a random number according to the uniform distribution U [0, W].
- the random vector ⁇ (p) b represents a D-dimensional vector.
- ⁇ (p) b V ⁇ ⁇ 1/2 A b
- V (v 1 ,..., V D )
- ⁇ ⁇ 1/2 diag ⁇ 1 / ⁇ 1 ,..., 1 / ⁇ D ⁇
- the initialization unit 22 further calculates a plurality (P) of key calculation functions based on the parameter list 8 collected by the parameter management device 2 and the random number information collected by the random number information acquisition unit 21.
- the key calculation function L (p) (x) corresponding to an arbitrary natural number p (p ⁇ P) among the plurality of key calculation functions is a function of the D-dimensional number vector x.
- L (p) (x) (f (p) 1 (x),..., F (p) B (x))
- the data registration unit 23 controls the input device 11 so that the input data 12 is input to the table management device 5 via the input device 11.
- the input data 12 indicates a data set X.
- the data set X is formed from the data x i and is expressed by the following equation.
- X ⁇ x i
- x i ⁇ R D ⁇ (i 1,..., N)
- the set RD indicates a D-dimensional number vector space.
- the natural number N indicates the total number of elements of the data x i of the data set X, and indicates a natural number greater than 1.
- Data x i indicates the D-dimensional vector number.
- the data registration unit 23 further creates a plurality (P) of tables based on the input data 12 and the plurality of key calculation functions calculated by the initialization unit 22.
- a table corresponding to an arbitrary natural number p among the plurality of tables is formed from a plurality of entries.
- each entry of the plurality of entries indicates a pair (set) of one key and one value, thereby associating the plurality of keys with the plurality of values. That is, an arbitrary key of the plurality of keys corresponds to one value of the plurality of values.
- An arbitrary key of the plurality of keys is converted into a data set X by a key calculation function L (p) (x) corresponding to an arbitrary natural number p of the plurality of key calculation functions calculated by the initialization unit 22.
- the value calculated by substituting any one of the data x is shown.
- the plurality of keys are different from each other and include a plurality of values calculated by substituting all the data x belonging to the data set X into the key calculation function L (p) (x).
- a value Table (p) [L] corresponding to an arbitrary key L among the plurality of values indicates a data list which is a set of predetermined data.
- the value Table (p) [L] has a value calculated by substituting arbitrary data x of the data belonging to the data list into the key calculation function L (p) (x). Is calculated so as to match.
- the data registration unit 23 further controls the table holding device 10 so that the plurality of tables are recorded in the table holding device 10.
- the search unit 24 refers to the plurality of tables created by the data registration unit 23 and calculates a data list based on the query. That is, the search unit 24 calculates a plurality of search keys corresponding to the plurality of tables based on the query q.
- the search key L (p) (q) corresponding to the natural number p among the plurality of search keys is the key calculation function L ( corresponding to the natural number p among the plurality of key calculation functions calculated by the initialization unit 22.
- p A value calculated by substituting the query q into (x) is shown.
- the search unit 24 refers to the plurality of tables created by the data registration unit 23 and calculates the data list List ⁇ x> based on the plurality of search keys.
- the data list List ⁇ x> indicates a set of predetermined data among the data belonging to the data set X.
- the data list List ⁇ x> includes a value Table (p) [L ] So as to include data belonging to the data list indicated by
- the data processing device 6 calculates the search result data x * based on the search condition 16 and the data list List ⁇ x> calculated by the search unit 24.
- the search result data x * indicates a set of data satisfying the condition indicated by the search condition 16 among the data belonging to the data list List ⁇ x>.
- the data processing device 6 further controls the output device 15 so that the search result data x * is displayed.
- the embodiment of the directivity vicinity detection method according to the present invention is executed by the similarity detection apparatus 1 and includes an initialization operation, a data registration operation, and a search operation.
- FIG. 3 shows the initialization operation.
- the user prepares the parameter list 8 and operates the input device 7 to input the parameter list 8 to the parameter management device 2.
- the parameter management device 2 records the parameter list 8 in the storage device.
- the table management device 5 acquires the parameter list 8 input to the parameter management device 2 from the parameter management device 2 (step S1).
- the parameter list 8 shows table parameters, management parameters, and directional parameters.
- the table parameter indicates information necessary for creating the table, that is, indicates the number P of the table, the window width W, the radix C, and the bit length B.
- the number of faces P indicates a positive integer.
- the window width W indicates a positive real number.
- the cardinal number C represents an integer of 2 or more.
- the bit length B indicates a positive integer.
- the management parameter indicates information necessary for other management operations, that is, the dimension number D.
- the directivity parameter includes feature detection direction and distance weight information, expresses directivity in proximity detection, and indicates a set U.
- the set U is formed from the directivity parameter u j and is expressed by the following equation.
- U ⁇ u j
- the direction parameter v j indicates a D-dimensional vector.
- the intensity parameter ⁇ j is a positive real number and indicates the degree of emphasis on the direction parameter v j .
- the table management device 5 substitutes 1 for the table surface number p (step S2).
- the table management device 5 calculates the magnitude relationship between the value indicated by the table surface number p and the number of surfaces P (step S3).
- the table management device 5 sets the window width W, bit length B, dimension number D, and set in the parameter list 8 U is output to the random number generator 3 to request the random number generator 3 for random number information.
- Number generator 3 when requested random number information from the table managing unit 5, on the basis of the set U and the window width W and the bit length B and rank D, and calculates the random number information indicating the set Z p.
- the set Z p is formed from the element Z b and is expressed by the following equation.
- Z p ⁇ Z b
- the random number R (p) b indicates a random number according to the uniform distribution U [0, W), and is generated by B independent trials.
- the random vector ⁇ (p) b represents a D-dimensional vector.
- ⁇ (p) b V ⁇ ⁇ 1/2 A b
- V (v 1 ,..., V D )
- ⁇ ⁇ 1/2 diag ⁇ 1 / ⁇ 1 ,..., 1 / ⁇ D ⁇
- the table management device 5 collects the random number information calculated by the random number generation device 3 from the random number generation device 3 (step S4).
- the table management device 5 increments the table surface number p by 1, that is, the table surface number p is obtained by adding 1 to the value indicated by the table surface number p. (Step S5).
- the table management device 5 calculates the magnitude relationship between the value indicated by the table surface number p and the number of surfaces P (step S3).
- the table management device 5 repeatedly executes steps S4 to S5.
- the table management device 5 calculates a plurality (P) of keys based on the random number information set Z and the parameter list 8. A function is calculated (step S6).
- the random number information set Z indicates a plurality (P pieces) of random number information collected by repeatedly executing step S4.
- the key calculation function L (p) (x) corresponding to the table surface number p among the plurality of key calculation functions is a function of the D-dimensional number vector x.
- the random number information set Z is expressed by the following formula.
- L (p) (x) (f (p) 1 (x),..., F (p) B (x))
- f (p) b (x) is expressed by the above formula (see Equation 1).
- the table management device 5 creates a plurality of (P) tables with empty entries after the plurality of (P) key calculation functions are calculated (step S7).
- the plurality of tables correspond to the plurality of key calculation functions.
- the table management device 5 controls the table holding device 10 to record the plural (P) tables in the table holding device 10.
- the data registration operation is executed after the initialization operation is executed.
- the user prepares input data 12 and operates the input device 11 to input the input data 12 to the table management device 5.
- the input data 12 indicates a data set X.
- the data set X is formed from the data x i and is expressed by the following equation.
- X ⁇ x i
- x i ⁇ R D ⁇ (i 1,..., N)
- the set RD indicates a D-dimensional number vector space.
- the natural number N indicates the total number of data x i that are elements of the data set X, and indicates a natural number greater than one.
- Data x i indicates the D-dimensional vector number.
- the table management device 5 records the input data 12 in the storage device by controlling the storage device.
- the table management device 5 executes a plurality (P) of table creation operations corresponding to the plurality of (P) tables created by the initialization operation. To do.
- the table creation operation corresponding to the table surface number p is formed of a plurality (N) of entry creation operations corresponding to all data belonging to the data set X.
- the table management device 5 uses the plurality of (P) key calculation functions created by the initialization operation.
- the key L (p) (x i ) is calculated by substituting the data x i for the key calculation function L (p) (x) corresponding to the table surface number p.
- the table management device 5 has an entry corresponding to the key L (p) (x i ) in the table corresponding to the table surface number p among the plural (P) tables created by the initialization operation. To determine if
- the table management device 5 When there is an entry corresponding to the key L (p) (x i ) in the table, the table management device 5 stores the data list indicated by the value Table (p) [L (p) (x i )] of the entry. Add data x i to. When there is no entry corresponding to the key L (p) (x i ) in the table, the table management device 5 adds an entry corresponding to the key L (p) (x i ) to the table. At this time, the value Table (p) [L (p) (x i )] of the entry indicates a data list having only the data x i as an element.
- the table management apparatus 5 creates a table corresponding to the table surface number p among the plurality (P) of tables by executing all of the plurality (N) of entry creation operations.
- the table management device 5 creates all of the plural (P) tables by executing all of the plural (P) table creating operations.
- the table management device 5 records the plurality of tables in the table holding device 10 by controlling the table holding device 10.
- FIG. 4 shows the search operation.
- the user prepares the search condition 16 and operates the input device 14 to input the search condition 16 to the data processing device 6.
- a search condition 16 indicates a query q and a condition. Examples of the condition include “obtain one piece of data nearest to the query q” and “obtain K pieces of data from the one close to the query q”.
- the data processing device 6 controls the storage device to record the search condition 16 in the storage device (step S11).
- the data processing device 6 further requests the data list from the table management device 5 by outputting the query q to the table management device 5 (step S12).
- the table management device 5 calculates the magnitude relationship between the value indicated by the table surface number p and the number of surfaces P (step S14).
- the table management device 5 calculates the search key L (p) (q) based on the query q when the value indicated by the table surface number p is not greater than the number of surfaces P (step S14, p ⁇ P) (step S14). S15).
- the query q is assigned to the key calculation function L (p) (x) corresponding to the table surface number p among the plurality of key calculation functions calculated by the initialization operation. The value calculated by this is shown.
- the table management device 5 refers to the table corresponding to the table surface number p among the plurality of tables created by the data registration operation, and the value Table (p) corresponding to the search key L (p) (q ). [L (p) (q)] is acquired (step S16). The table management device 5 adds data belonging to the data list indicated by the value Table (p) [L (p) (q)] to the data list List ⁇ x> (step S17).
- the table management device 5 when there is data belonging to duplicate data list List and ⁇ x> Value Table (p) [L (p ) (q)] and, Value Table (p) [L (P) Only data that is not registered in the data list List ⁇ x> among the data indicated by (q)] is added to the data list List ⁇ x>.
- step S17 the table management apparatus 5 increments the table surface number p by 1, that is, the sum calculated by adding 1 to the value indicated by the table surface number p is added to the table surface number p. Substitute (step S18). After incrementing the table surface number p, the table management device 5 calculates the magnitude relationship between the value indicated by the table surface number p and the number of surfaces P (step S14). When the value indicated by the table surface number p is not larger than the number P of surfaces (step S14, p ⁇ P), the table management device 5 repeatedly executes steps S15 to S18.
- the table management device 5 outputs the data list List ⁇ x> to the data processing device 6 when the value indicated by the table surface number p is larger than the number of surfaces P (step S14, p> P).
- the data list List ⁇ x> thus created is a candidate for data that will be a short distance when the distance is calculated by placing importance on the specific direction of the query q with an appropriate weight.
- the data processing device 6 calculates the search result data x * based on the condition indicated by the search condition 16 and the data list List ⁇ x> (step S19).
- the search result data x * indicates a set of data satisfying the condition among the data belonging to the data list List ⁇ x>. If the condition indicates that “K data is acquired from the one close to the query q” and the data list List ⁇ x> includes only K or less data, the search result data x * All the data of the list List ⁇ x> is shown.
- a normal Euclidean distance may be used, or the distance may be calculated in consideration of the importance specified by the directivity parameter. .
- the data processing device 6 controls the output device 15 to display the search result data x * on the output device 15 (step S20).
- the similarity detection apparatus 1 can perform high-dimensional data by setting a directivity parameter in consideration of interest information indicating the direction of interest and the degree of interest of the user.
- interest information indicating the direction of interest and the degree of interest of the user.
- directional neighborhood detection that preferentially selects data close to the center of interest at high speed in line with the interest.
- the collision probability contour is not an isotropic sphere, and the collision probability contour can be controlled as an ellipsoid having a major axis in the direction of interest. It is possible to detect the directivity neighborhood with weights, and a significant time reduction can be expected.
- the data processing device 6 can be replaced with a plurality of computers in which the random number information acquisition unit 21, the initialization unit 22, the data registration unit 23, and the search unit 24 are realized. . Furthermore, the similarity detection device 1 can be replaced with a single computer that implements the parameter management device 2, the random number generation device 3, the table management device 5, and the data processing device 6. Furthermore, the parameter management device 2, the random number generation device 3, the data processing device 6, the random number information acquisition unit 21, the initialization unit 22, the data registration unit 23, and the search unit 24 are formed from a plurality of computers. You can also. Similarly to the similarity detection device 1, the similarity detection device thus replaced can perform directivity vicinity detection with an arbitrary weight in an arbitrary direction.
- the plurality of directivity parameters 36 are formed from two directivity parameters u 1 and u2 .
- the plurality of direction parameters 37 include a direction parameter v 1 indicating (1 / sqrt (2), 1 / sqrt (2)) and a direction parameter v indicating ( ⁇ 1 / sqrt (2), 1 / sqrt (2)). 2 .
- a plurality of intensity parameter 38 is formed from the intensity parameter sigma 1 showing a 2 and 1 are shown intensity parameter sigma 1 Tokyo.
- the directivity parameter u 1 associates the direction parameter v 1 with the intensity parameter ⁇ 1 .
- the directivity parameter u 2 associates the direction parameter v 2 with the intensity parameter ⁇ 2 .
- U ⁇ (1 / sqrt (2), 1 / sqrt (2)), 2>, ⁇ ( ⁇ 1 / sqrt (2), 1 / sqrt (2)), 1> ⁇
- the direction parameter v 1 indicates a direction inclined 45 degrees counterclockwise from the x axis, as shown in FIG.
- Direction parameter v 2 indicates the direction perpendicular to the direction indicated by the direction parameter v 1.
- Set U since that indicates the twice the value of the value indicated by the intensity parameter sigma 2 is intensity parameter sigma 1, meaning that emphasized twice that the direction of the direction parameter v 1 in the direction of the direction parameter v 2 .
- the input data 12 has an arbitrary distribution of data belonging to the set X.
- the search condition 16 indicates that “the data is the closest to the query q with respect to the input of the query q”.
- the collision probability isolines 31 is formed in an oval with long axis in the direction of the direction parameter v 1. That is, all the data on the collision probability isoline 31 are detected with the same probability.
- Similarity detection device 1 is compared with the direction of the direction parameter v 2, equate twice far away in the direction of the direction parameter v 1, it is possible to search the search result data from the data belonging to the set X. That is, since the similarity detection apparatus 1 can control the detection probability, it becomes possible to perform directivity vicinity detection with the weight of the intensity parameter ⁇ 1 in the direction of the direction parameter v 1 .
- the directivity parameter indicated by the parameter list 8 described in the above-described embodiment is replaced with another directivity parameter.
- u j ⁇ R D ⁇ R + ⁇ (j 1,..., D) At this time, the sets U k are different from each other. That is, either the direction parameter v or the intensity parameter ⁇ is different.
- the set U k is expressed by the following equation.
- U k ⁇ v x, ⁇ x 2 p >, ⁇ v y , ⁇ y 2 p > ⁇
- the direction parameter v x indicates a unit vector in the x-axis direction.
- Direction parameter v y represents a unit vector in the y-axis direction.
- the variable p represents an integer of 0 or more.
- the intensity parameter ⁇ x indicates a basic weight in the x-axis direction.
- the intensity parameter ⁇ y indicates a basic weight in the y-axis direction.
- the similarity detection apparatus creates a plurality of key calculation function sets corresponding to a plurality of sets belonging to the set U indicated by the directivity parameter based on the directivity parameter.
- a key calculation function set corresponding to the set U k out of the plurality of key calculation function sets indicates a plurality of key calculation functions.
- the plurality of key calculation functions are calculated based on the set U k in the same manner as the plurality of key calculation functions calculated based on the set U by the similarity detection apparatus 1 in the above-described embodiment.
- the similarity detection device Based on the input data 12 and the plurality of key calculation function sets, the similarity detection device generates a plurality of table sets 51 corresponding to the plurality of key calculation function sets 52 as shown in FIG. create.
- An arbitrary table set of the plurality of table sets 51 indicates a set formed of a plurality of tables.
- the plurality of tables are created in the same manner as the plurality of tables in the above-described embodiment. That is, a table corresponding to an arbitrary table surface number p among the plurality of tables associates a plurality of keys with a plurality of values.
- An arbitrary key of the plurality of keys is assigned to any data in the data set X by the key calculation function L (p) (x) corresponding to the table surface number p of the plurality of key calculation functions. The value calculated by substituting x is shown.
- a value Table (p) [L] corresponding to an arbitrary key L among the plurality of values indicates a data list which is a set of predetermined data.
- the value Table (p) [L] has a value calculated by substituting arbitrary data x of the data belonging to the data list into the key calculation function L (p) (x). Is calculated so as to match.
- FIG. 9 shows a plurality of collision probability isolines each indicating the directivity specified by the set U k .
- the plurality of collision probability isolines are each formed into an ellipse having a major axis in the direction of the direction parameter, and are different from each other.
- the data on the collision probability isolines corresponding to the set U k among the plurality of collision probability isolines are similar to the collision probability isolines 31 shown in FIG.
- the similarity detection apparatus creates and manages a table in which a table corresponding to each directivity parameter is created for a specified number of surfaces.
- the user inputs another search condition different from the search condition 16 to the similarity detection apparatus.
- the search condition indicates the directivity parameter and the query and condition indicated by the search condition 16.
- the directivity parameter indicates a set of direction and intensity pairs.
- the similarity detection apparatus selects a table set corresponding to the set U k indicating the directivity parameter closest to the directivity parameter from the plurality of table sets 51.
- the similarity detection apparatus calculates a plurality of search data lists using a plurality of tables belonging to the table set, and calculates a data list List ⁇ x>.
- the search data list corresponding to a certain table set in the plurality of search data lists matches the data list indicated by the value corresponding to the query key among the plurality of values in the table set.
- the data list List ⁇ x> includes a plurality of search data lists corresponding to the plurality of table sets.
- Such a similarity detection apparatus is provided with a function of creating a plurality of table sets in advance and selecting a table to be used in accordance with the search, so that the directivity parameter is set in advance. Even if it is unknown, a directivity vicinity search can be performed by inputting the directivity parameter “in this direction at this point with this strength” together during the search operation.
- the searcher inputs the search directivity parameter in the format of the directivity parameter U (a set of direction and intensity pairs). This defines a more intuitive input.
- the data processing apparatus may convert to a directivity parameter format.
- the directivity of the search by specifying a length along the direction of interest that is equated with the unit length in the axial direction of no interest (for example, When the length 1 along the direction inclined 45 ° from the x axis corresponds to the length 3 along the direction inclined ⁇ 45 ° from the x axis, the direction inclined ⁇ 45 ° from the x axis is inclined 45 °. It will be detected with the same probability even if it is three times as far as the actual distance).
- a second embodiment of the directivity vicinity detection method according to the present invention will be described in detail.
- a method for detecting images similar to a certain feature at high speed will be described.
- Image data representing an image can generally be expressed by a D-dimensional feature vector.
- Various methods are conceivable, but here, as an example, as shown in FIG. 10, an image 61 is cut into a plurality of pixels 62-1 to 62-16 with an appropriate mesh, and each pixel 62 Consider image data expressed by defining pixel values with -i.
- a feature vector is created by raster-scanning the pixel value in each pixel and arranging the pixel values.
- values are read in order from left to right starting from the upper left, and when reaching the right end, the same process is repeated.
- the feature vector u of an arbitrary image divided into 4 ⁇ 4 pixels is expressed by a 16-dimensional vector.
- it is expressed by the following equation using 16 unit vectors e 1 to e 16 and 16 pixel values.
- the unit vector e i is a vector in which only the i-th component indicates 1 and the subsequent component indicates 0.
- the user is interested only in a certain feature direction.
- the user may be interested only in a region of interest 63 that is a certain part of the image 61 and ignore other similarities.
- the user attaches importance only to the first component, the second component, the fifth component, and the sixth component of the feature vector u, and does not attach importance to other pixels.
- the directivity vicinity detection method according to the present invention can generally detect a similar image using a combination of pixels at an appropriate ratio as a feature direction. For example, consider a case where importance is attached to the similarity of the ratio of the first, second, fifth, and sixth pixel values.
- v 2 ,..., v 1 6 ⁇ can be configured.
- the color density is expressed as a real value of 0 to 1 as a pixel value, a similar image can be obtained even if the difference in pixel values other than the target portion is somewhat large if the ratio of the darkness of the target portion is close. Will be detected.
- v (ti) represents a multidimensional vector at time ti.
- the time series data include data representing acceleration values in the x-axis direction, the y-axis direction, and the z-axis direction of the acceleration sensor measured at time ti as a three-dimensional vector at a certain point. In that case, a set of points using the time in the three-dimensional vector space as a parameter becomes time series data.
- FIG. 11 shows a plurality of time series data 71-1 to 71-T to be searched in this embodiment.
- the user first inputs a plurality of time-series data 71-1 to 71-T to the similarity detection apparatus in the present embodiment.
- the similarity detection apparatus sequentially registers each point of time-series data in a plurality of tables corresponding to a plurality of directivity parameters in the same manner as the similarity detection apparatus in the second embodiment.
- the plurality of directivity parameters are simply a directivity parameter having a major axis in the x-axis direction and a minor axis in the y-axis direction, and another directivity having directivity in a direction rotated 90 degrees from the major axis.
- a plurality of directional parameter intensities can be specified, but here, for example, a ratio of 3: 1 is set.
- the similarity detection apparatus creates a plurality of table sets 75 corresponding to the plurality of directivity parameters 74, as shown in FIG.
- the number of tables belonging to each table set of the plurality of table sets 75 is K.
- the similarity detection device calculates keys for each point of time-series data and registers them in the value in all tables in the same manner as the similarity detection device in the above-described embodiment.
- the user inputs time-series data 76 and designates a search range parameter 77 indicating a region of interest (search target area) in the time-series data 76 as a query.
- the search range parameter 77 only needs to specify the region of interest of the time series data 76.
- it may be data indicating a range of coordinate values, or data indicating a range of parameters (for example, time t) of each data.
- the search range parameter 77 may be partial time series data itself extracted from the time series data 76 only for data within the region of interest.
- the similarity detection apparatus refers to a plurality of tables belonging to the table set selected by the user from among the plurality of table sets 75 in the same manner as the similarity detection apparatus in the above-described embodiment. Is extracted, and the output device 15 is controlled so that the time series data 78 is expressed so as to be recognized by the user.
- the time series data 78 is formed from data included in a plurality of search data lists corresponding to the plurality of tables.
- a search data list corresponding to a certain table among the plurality of search data lists is a data list indicated by a value corresponding to a query value calculated by assigning the query to a key calculation function corresponding to the table. Show.
- the reference point 85 is a point that best represents the partial time series data 83. For example, the average of all points or the point closest to the average may be used, or the average of the maximum and minimum values (S1 in this example). And the average of S9).
- the similarity detection apparatus has a probability that the reference point 85 and the points of the partial time series data 83 and the partial time series data 84 are registered in the same entry in a plurality of tables belonging to each of the plurality of table sets 75. (Collision probability) is calculated, and a plurality of collision probability distributions corresponding to the plurality of directivity parameters 74 are calculated.
- the plurality of collision probability distributions respectively indicate functions of distances from the reference point 85 to each data belonging to the partial time series data 83 and the partial time series data 84.
- FIG. 17 shows a collision probability isoline corresponding to the first directional parameter among the plurality of directional parameters.
- the collision probability contour line 86 has a major axis in the x-axis direction and a minor axis in the y-axis direction.
- FIG. 18 shows a collision probability isoline corresponding to the second directional parameter among the plurality of directional parameters.
- the collision probability isoline 87 has a major axis in the x-axis direction and a minor axis in the y-axis direction.
- FIG. 19 shows the collision probability distribution.
- the collision probability distribution 88 shows the collision probability that each data belonging to the partial time series data 83 and the partial time series data 84 is registered in the same entry in the table calculated based on the reference point 85 and the first directional parameter. Show.
- FIG. 20 shows another collision probability distribution.
- the collision probability distribution 89 shows the collision probability that each data belonging to the partial time series data 83 and the partial time series data 84 is registered in the same entry in the table calculated based on the reference point 85 and the second directional parameter. Show.
- the number L of actual collisions (registered in the same entry) is measured against the total number K of tables for each directivity parameter, and the collision probability is calculated by L / K. It is approximated by.
- the collision probability distribution 88 and the collision probability distribution 89 indicate that the collision probability between each point of the time series data and the reference point 85 takes different values depending on the positional relationship between the two locations and the direction and intensity indicated by the directivity parameter. ing.
- the collision probability distribution 88 and the collision probability distribution 89 have the same shape as the two partial time series data to be compared are closer to each other. Therefore, by comparing the difference between these shapes, any two partial times can be obtained. It is possible to evaluate the similarity of series data.
- FIG. 21 shows the feature distribution.
- the feature distribution 90 indicates the difference in collision probability of each partial time series data between the collision probability distribution 88 and the collision probability distribution 89, and the shape of the feature distribution 90 of each partial time series data indicates the similarity in the partial time series. It can be evaluated as the similarity of data.
- the method computes the inner product between smooth interpolated function, or a method or the like using the distance evaluation measure such as E a r t h Mover 's Distance is contemplated, limited to this It is not a thing, What is necessary is just to be able to evaluate the similarity of the form of a function (function value when a certain distance is given) quantitatively.
- the time series data designated by the query is limited to the designated search range.
- reference points are calculated from the partial time series data specified by the query, and only data within the search range is extracted from data colliding with reference points in all tables belonging to a plurality of table sets 75 ( In this example, the time range for which time is specified is extracted), the collision probability distribution corresponding to each time series data is created, and the similarity with the collision probability distribution of the partial time series data specified by the query is evaluated. Just do it.
- the search range is specified as one continuous section.
- the time series data to be searched can be registered in advance in a table offline, and what is actually required for online processing is constituted by data colliding with a reference point. This is because there are only the time series data and the corresponding collision probability distribution. In particular, when the number of search target time-series data is enormous, a significant reduction in the number of target data can be expected.
- the time is treated as a parameter.
- the coordinate value can be compared with the value of the time-series data by appropriately multiplying the time itself by a suitable constant or by appropriately converting the time. It is also possible to search by working as one.
- the similarity detection apparatus according to the present invention is applied to a mobile management system 100 as shown in FIG.
- the mobile management system 100 is connected via a network 103 so that a plurality of computers can bidirectionally transmit information to each other.
- Examples of the network 103 include the Internet and a mobile phone network.
- the plurality of computers includes a mobile management apparatus 101 and a plurality of mobile terminals 102-1 to 102-3.
- Each of the plurality of mobile terminals 102-1 to 102-3 is a device that is carried by a user and can acquire current position information. Examples of the device include a smartphone and a GPS logger.
- the mobile management apparatus 101 includes the similarity detection apparatus in the above-described embodiment, manages the movement directions of the plurality of mobile terminals 102-1 to 102-3, and is moving in a similar direction. Provides the function of grouping For example, the mobile management apparatus 101 manages a pair of the current position and a position that is a suitable time ⁇ before each of the plurality of mobile terminals 102-1 to 102-3.
- the space to be handled is a two-dimensional space
- the main movement direction to be detected is the x-axis direction
- the directivity parameter is set so that the intensity is appropriately given in the x-axis direction compared to the y-axis direction.
- the intensity for example, a constant multiple of v ⁇ is set in the x-axis direction, and the y-axis direction is set to, for example, 0.1 times the intensity in the x-axis direction.
- the y-axis direction can define the allowable range of the direction blur.
- a plurality of mobile terminals 102-1 to 102-3 exist at positions 1010, 1020, and 1030, respectively, at a certain time t, and move to positions 1011, 1021, and 1031 respectively at time t + ⁇ .
- the plurality of mobile terminals 102-1 to 102-3 acquire the current position by GPS, and periodically upload the position information to the mobile management apparatus 101 through the network 103.
- the mobile object management apparatus 101 creates a plurality of table sets 112 corresponding to the plurality of directivity parameters 111 as shown in FIG. 23 in the same manner as the similarity detection apparatus in the above-described embodiment. .
- the mobile unit management apparatus 101 manages the positions of a plurality of mobile terminals 102-1 to 102-3 using a plurality of table sets 112, and includes a position 1010, a position 1011, a position 1020, a position 1021, a position 1030, and a position 1031.
- the key of each table is calculated and registered in the value corresponding to the key.
- the mobile management apparatus 101 uses this table set to group mobile terminals that proceed in the x-axis direction from among the plurality of mobile terminals 102-1 to 102-3, and move that includes information on characteristics of the group
- the body group information 105 is output.
- the mobile terminal is moving at an approximate speed in the x-axis direction, the position coordinates at time t and time t + ⁇ are likely to have the same key, but move in the orthogonal y-axis direction. Mobile terminals are likely to have different keys, so they can be distinguished.
- the position information 1010, position 1011, position 1020 and position 1021 which are the position information of the mobile terminal 102-1 and the mobile terminal 102-2 are registered in the same entry, and the position of the mobile terminal 102-3
- the information position 1030 and position 1031 are registered in different entries. Therefore, it is possible to group the mobile terminals 102-1 and 102-2 moving at the same speed in the x-axis direction as the same group.
- a mobile terminal belonging to the same group as the search target mobile terminal can be output as a parallel mobile terminal in the past or present.
- the grouping policy it is possible to modify the grouping according to the policy by setting the grouping policy together if necessary. For example, if the grouping policy is set to “a mobile terminal that has run in parallel with the search target mobile terminal within the past 10 minutes or more within a radius of 100 m” as a grouping policy, It is sufficient to filter and output only those satisfying the conditions.
- the number of mobile terminals registered in each entry in each table is registered in the same entry at both time t and time t + ⁇ .
- a set of mobile terminals registered in the entry is extracted in an order from the largest number.
- the mobile terminals traveling along the same direction are sampled from a group having a large total number among all the mobile terminals.
- the mobile terminal group information such as the size, direction and speed of the mobile terminal group can be created, and the movement state of the mobile terminal can be classified and displayed.
- the mobile management system 100 basically updates the table managed for the location update of the mobile terminal, samples the mobile terminals with high interest from the table by the above method, and performs grouping. There is no need to calculate and group the interrelationships related to the movement of mobile terminals. Therefore, even if the number of mobile terminals increases, the increase in calculation load can be suppressed, and the calculation time can be greatly reduced.
- the key calculation function L (p) (x) and the basic function f (p) in the above-described embodiment are used as the table key calculation method.
- b (x) and the random vector ⁇ (p) b are used, the present invention is not limited to this as long as the asymmetry corresponding to the desired direction and intensity is introduced by the parameter specifying the directivity.
- the data collision in the table is registered in the exact same entry.
- the entry distance of the table is defined and a nearby entry is defined. If the data registered up to (for example, “adjacent entry” that differs by one bit of the key represented by the key calculation function L (p) (x)) collides, the collision range may be expanded.
- a random number generator for calculating a plurality of random number information based on a plurality of direction parameters and a plurality of intensity parameters input via an input device;
- An initialization unit for calculating a plurality of key calculation functions based on the plurality of random number information;
- a data registration unit that calculates a plurality of tables corresponding to the plurality of key calculation functions based on a plurality of search target data input via an input device, and records the plurality of tables in a table holding device;
- a search unit that refers to the plurality of tables and calculates a candidate data list based on a query indicated by a search condition input via an input device;
- the data registration unit associates a plurality of keys with a plurality of data lists when calculating a table corresponding to an arbitrary key calculation function of the plurality of tables,
- the candidate data list includes a plurality of search data lists corresponding to the plurality of key calculation functions
- the search data list corresponding to the arbitrary key calculation function in the plurality of search data lists is a query value calculated by substituting the query into the key calculation function in the plurality of data lists. Similarity detection device showing a data list corresponding to.
- the similarity detection apparatus according to any one of appendices 1 and 2,
- the plurality of direction parameters are set such that two values calculated by substituting two similar state change similar data of the plurality of search target data into the arbitrary key calculation function are equal to each other.
- the similarity detection device is set so that the difference between the two values is smaller than a predetermined value.
- the similarity detection apparatus according to any one of appendices 1 to 3,
- the plurality of search target data is a similarity detection device that indicates an image.
- Appendix 7 The directivity vicinity detection method according to appendix 6, Further calculating a plurality of table sets based on the plurality of search target data; And a step of calculating the candidate data list with reference to the plurality of tables when the table set indicated by the search condition of the plurality of table sets indicates the plurality of tables.
- Appendix 8 The directivity vicinity detection method according to any one of appendices 6 to 7, The plurality of direction parameters are set such that two values calculated by substituting two similar state change similar data of the plurality of search target data into the arbitrary key calculation function are equal to each other. Or the directivity vicinity detection method set so that the difference of the said two values may become smaller than predetermined value.
- Appendix 9 A directivity vicinity detection method according to any one of appendices 6 to 8,
- the plurality of search target data each indicate an image.
- [Appendix 11] It is a similarity detection device that detects similar data at a high speed with an arbitrary similarity criterion for a large amount of search target data.
- the above-described similarity detection device holds a table associated with random number information created by a parameter for setting similarity determination criteria for registering and managing the above-described search target data, In response to a neighborhood search request from any reference point, the above table is used, similarity judgment is performed according to the set similarity criterion, and data that is judged as a neighborhood is output. Similarity detection device.
- the search target data described in the supplementary note 11 or the supplementary note 12 is not only data whose value is statically fixed in advance, but also data in which the value is dynamically updated.
- a similarity detection apparatus characterized by updating a registration state.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Library & Information Science (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Electron Beam Exposure (AREA)
Abstract
Description
U={uj|uj∈RD×R+}(j=1,・・・,D)
ここで、集合RDは、D次元数ベクトル空間を示している。例えば、集合R2は、2次元数ベクトル空間を示している。
uj=<vj,σj>
ここで、方向パラメータvjは、D次元のベクトルを示している。強度パラメータσjは、正の実数であり、方向パラメータvjを重視する程度を示している。
Z={Zp}(p=1,・・・,P)
ここで、集合Zpは、乱数発生装置3により算出された1つの乱数情報を示している。
Zp={Zb|Zb=<Φ(p) b,R(p) b>}(b=1,・・・,B)
ここで、乱数R(p) bは、一様分布U[0,W]に従う乱数を示している。ランダムベクトルΦ(p) bは、D次元のベクトルを示している。
Φ(p) b=VΛ-1/2Ab
V=(v1,・・・,vD)
Λ-1/2=diag{1/σ1,・・・,1/σD}
L(p)(x)=(f(p) 1(x),・・・,f(p) B(x))
X={xi|xi∈RD}(i=1,・・・,N)
ここで、集合RDは、D次元数ベクトル空間を示している。自然数Nは、データ集合Xのデータxiの要素の総数を示し、1より大きい自然数を示している。データxiは、D次元数ベクトルを示している。
U={uj|uj=<vj,σj>}(j=1,・・・,D)
ここで、方向パラメータvjは、D次元のベクトルを示している。強度パラメータσjは、正の実数であり、方向パラメータvjを重視する程度を示している。
Zp={Zb|Zb=<Φ(p) b,R(p) b>}(b=1,・・・,B)
ここで、乱数R(p) bは、一様分布U[0,W)に従う乱数を示し、B回の独立な試行により生成される。ランダムベクトルΦ(p) bは、D次元のベクトルを示している。
Φ(p) b=VΛ-1/2Ab
V=(v1,・・・,vD)
Λ-1/2=diag{1/σ1,・・・,1/σD}
Z={Zp}(p=1,・・・,P)
L(p)(x)=(f(p) 1(x),・・・,f(p) B(x))
ここで、基本関数f(p) b(x)は、前述の式により表現される(数1参照)。
X={xi|xi∈RD}(i=1,・・・,N)
ここで、集合RDは、D次元数ベクトル空間を示している。自然数Nは、データ集合Xの要素であるデータxiの総数を示し、1より大きい自然数を示している。データxiは、D次元数ベクトルを示している。
U={<(1/sqrt(2)、1/sqrt(2))、2>、<(-1/sqrt(2)、1/sqrt(2))、1>}
{Uk}(k=1,・・・,K)
Uk={uj|uj∈RD×R+}(j=1,・・・,D)
このとき、集合Ukは、互いに異なる。すなわち、方向パラメータv又は強度パラメータσのどれかが異なっている。
Uk={<vx,σx2p>、<vy,σy2p>}
ここで、方向パラメータvxは、x軸方向の単位ベクトルを示している。方向パラメータvyは、y軸方向の単位ベクトルを示している。変数pは、0以上の整数を示している。強度パラメータσxは、x軸方向の基本重みを示している。強度パラメータσyは、y軸方向の基本重みを示している。
v1=(e1+e2+e5+e6)/2
以上、本発明の実施の形態及び実施例を詳述してきたが、実際には、上記の実施の形態及び実施例に限られるものではなく、本発明の要旨を逸脱しない範囲の変更があっても本発明に含まれる。
上記の実施の形態及び実施例の一部又は全部は、以下の付記のように記載することも可能である。但し、実際には、以下の記載例に限定されない。
入力装置を介して入力された複数の方向パラメータと複数の強度パラメータとに基づいて複数の乱数情報を算出する乱数発生部と、
前記複数の乱数情報に基づいて複数のキー計算関数を算出する初期化部と、
入力装置を介して入力された複数の検索対象データに基づいて、前記複数のキー計算関数に対応する複数のテーブルを算出し、前記複数のテーブルをテーブル保持装置に記録するデータ登録部と、
前記複数のテーブルを参照して、入力装置を介して入力された検索条件が示すクエリに基づいて候補データリストを算出する検索部と、
前記候補データリストに属する複数の検索データから前記検索条件が示す条件を満足する検索結果データを算出し、前記検索結果データを出力装置に出力するデータ処理部と
を具備し、
前記データ登録部は、前記複数のテーブルのうちの任意のキー計算関数に対応するテーブルを算出する際、複数のキーを複数のデータリストに対応付け、前記複数のデータリストのうちの任意のキーに対応するデータリストに属するデータを前記任意のキー計算関数に代入することにより算出される値が前記任意のキーに等しくなるように該テーブルを算出し、
前記候補データリストは、前記複数のキー計算関数に対応する複数の検索データリストを含み、
前記複数の検索データリストのうちの前記任意のキー計算関数に対応する検索データリストは、前記複数のデータリストのうちの、前記クエリが前記キー計算関数に代入されることにより算出されるクエリ値に対応するデータリストを示す
類似性検出装置。
付記1に記載の類似性検出装置であって、
前記データ登録部は、前記複数の検索対象データに基づいて複数のテーブルセットを更に算出し、
前記検索部は、前記複数のテーブルセットのうちの前記検索条件が示すテーブルセットが前記複数のテーブルを示すときに、前記複数のテーブルを参照して前記候補データリストを算出する
類似性検出装置。
付記1乃至2のいずれかに記載の類似性検出装置であって、
前記複数の方向パラメータは、前記複数の検索対象データのうちの類似する2つの状態変化類似データが前記任意のキー計算関数に代入されることによりそれぞれ算出される2つの値が等しくなるように、又は、前記2つの値の差が所定の値より小さくなるように、設定される
類似性検出装置。
付記1乃至3のいずれかに記載の類似性検出装置であって、
前記複数の検索対象データは、それぞれ画像を示す
類似性検出装置。
付記1乃至4のいずれかに記載の類似性検出装置であって、
前記データ登録部は、前記複数の検索対象データが更新されたときに、前記複数のテーブルを更新する
類似性検出装置。
入力装置を介して入力された複数の方向パラメータと複数の強度パラメータとに基づいて複数の乱数情報を算出するステップと、
前記複数の乱数情報に基づいて複数のキー計算関数を算出するステップと、
入力装置を介して入力された複数の検索対象データに基づいて、前記複数のキー計算関数に対応する複数のテーブルを算出するステップと、
前記複数のテーブルをテーブル保持装置に記録するステップと、
前記複数のテーブルを参照して、入力装置を介して入力された検索条件が示すクエリに基づいて候補データリストを算出するステップと、
前記候補データリストに属する複数の検索データから前記検索条件が示す条件を満足する検索結果データを算出するステップと、
前記検索結果データを出力装置に出力するステップと、
前記複数のテーブルのうちの任意のキー計算関数に対応するテーブルを算出する際、複数のキーを複数のデータリストに対応付け、前記複数のデータリストのうちの任意のキーに対応するデータリストに属するデータを前記任意のキー計算関数に代入することにより算出される値が前記任意のキーに等しくなるように該テーブルを算出するステップと
を具備し、
前記候補データリストは、前記複数のキー計算関数に対応する複数の検索データリストを含み、
前記複数の検索データリストのうちの前記任意のキー計算関数に対応する検索データリストは、前記複数のデータリストのうちの、前記クエリが前記キー計算関数に代入されることにより算出されるクエリ値に対応するデータリストを示す
指向性近傍検出方法。
付記6に記載の指向性近傍検出方法であって、
前記複数の検索対象データに基づいて複数のテーブルセットを更に算出するステップと、
前記複数のテーブルセットのうちの前記検索条件が示すテーブルセットが前記複数のテーブルを示すときに、前記複数のテーブルを参照して前記候補データリストを算出するステップと
を更に具備する
指向性近傍検出方法。
付記6乃至7のいずれかに記載の指向性近傍検出方法であって、
前記複数の方向パラメータは、前記複数の検索対象データのうちの類似する2つの状態変化類似データが前記任意のキー計算関数に代入されることによりそれぞれ算出される2つの値が等しくなるように、又は、前記2つの値の差が所定の値より小さくなるように、設定される
指向性近傍検出方法。
付記6乃至8のいずれかに記載の指向性近傍検出方法であって、
前記複数の検索対象データは、それぞれ画像を示す
指向性近傍検出方法。
付記6乃至9のいずれかに記載の指向性近傍検出方法であって、
前記複数の検索対象データが更新されたときに、前記複数のテーブルを更新するステップ
を更に具備する
指向性近傍検出方法。
大量の検索対象データに対して、任意の類似度判定基準で類似するデータを高速に検出する類似性検出装置であり、
前述の類似性検出装置は、前述の検索対象データを登録・管理するための、類似度判定基準を設定するパラメータによって作成される乱数情報と関連付けられたテーブルを保持し、
任意の参照点からの近傍検索要求に対して、前述のテーブルを利用し、前述の設定された類似度判定基準によって類似性判断を行い、近傍と判断されるデータを出力する
ことを特徴とする類似性検出装置。
付記11に記載の類似性検出装置において、複数の異なる類似度判定基準を有し、それぞれの類似度判定基準と関連付けられた複数のテーブルを有し、検索対象データを全てのテーブルで並列に管理する
ことを特徴とする類似性検出装置。
付記11に記載の類似度判定基準として、興味の方向と、その方向の重要度パラメータの組の集合とする
ことを特徴とする類似性検出装置。
付記11に記載の検索対象データとして、任意の2点の距離がユークリッド空間上の距離として定義されている
ことを特徴とする類似性検出装置。
付記12に記載の類似性検出装置において、検索においてその検索中心とともに検索類似度判定基準を入力し、複数のテーブルにおける近傍検出結果と、各テーブルに指定された類似度判定基準パラメータと検索類似度判定基準との関連性とを用いて出力する近傍を決定する
ことを特徴とする類似性検出装置。
付記11又は付記12に記載の検索対象データとして、予めその値が静的に固定されたデータだけでなく、動的にその値を更新するデータであり、値の更新に合わせて前述のテーブルの登録状態を更新する
ことを特徴とする類似性検出装置。
付記11又は付記12に記載の類似性検出装置として、前述のテーブルを用いた類似性判断結果を利用して、各データの状態変化が類似するものをグルーピングする
ことを特徴とする類似性検出装置。
付記11に記載のテーブルは、キーとバリューのペアで構成され、ある任意のデータの登録に当たって、前記乱数情報を用いてデータを登録すべきキーを計算し、対応するバリューに当たるデータのリストに追加登録することでデータを管理する
ことを特徴とする類似性検出装置。
付記11に記載の類似性判断として、同一もしくは差異の小さいキーを持つデータ同士を類似性が高いと扱う
ことを特徴とする類似性検出装置。
Claims (10)
- 検索対象データに対して、任意の類似度判定基準で類似するデータを検出する類似性検出装置であって、
類似度判定基準を設定するパラメータによって作成される乱数情報と関連付けられたテーブルを保持する手段と、
検索対象データを前記テーブルに登録し管理する手段と、
任意の参照点からの近傍検索要求に対して、前記テーブルを利用し、前記設定された類似度判定基準によって類似性判断を行い、近傍と判断されるデータを出力する手段と
を具備する
類似性検出装置。 - 請求項1に記載の類似性検出装置であって、
複数の異なる類似度判定基準を設定する手段と、
前記複数の異なる類似度判定基準のそれぞれと関連付けられた複数のテーブルを保持する手段と、
前記検索対象データを前記複数のテーブルで並列に管理する手段と
を更に具備する
類似性検出装置。 - 請求項1又は2に記載の類似性検出装置であって、
興味の方向と、該方向の重要度パラメータとの組の集合を、前記類似度判定基準とする手段
を更に具備する
類似性検出装置。 - 請求項1乃至3のいずれか一項に記載の類似性検出装置であって、
任意の2点の距離がユークリッド空間上の距離として定義されているデータを、前記検索対象データとする手段
を更に具備する
類似性検出装置。 - 請求項1乃至4のいずれか一項に記載の類似性検出装置であって、
検索の中心とともに検索類似度判定基準を入力し、前記複数のテーブルにおける近傍検出結果と、前記複数のテーブルの各々に指定された類似度判定基準パラメータと検索類似度判定基準との関連性とを用いて出力する近傍を決定する手段
を更に具備する
類似性検出装置。 - 請求項1乃至5のいずれか一項に記載の類似性検出装置であって、
予め値が静的に固定されたデータ、及び動的に値を更新するデータのいずれも、前記検索対象データとして使用する手段と、
前記検索対象データとして使用されるデータの値の更新に合わせて前記テーブルの登録状態を更新する手段と
を更に具備する
類似性検出装置。 - 請求項1乃至6のいずれか一項に記載の類似性検出装置であって、
前記テーブルを用いた類似性判断結果を利用して、状態変化が類似するデータをグルーピングする手段
を更に具備する
類似性検出装置。 - 請求項1乃至7のいずれか一項に記載の類似性検出装置であって、
前記テーブルをキーとバリューとの組で構成し、ある任意のデータの登録に当たって、前記乱数情報を用いてデータを登録すべきキーを計算し、対応するバリューに当たるデータのリストに追加登録することでデータを管理する手段
を更に具備する
類似性検出装置。 - 請求項1乃至8のいずれか一項に記載の類似性検出装置であって、
同一もしくは差異の小さいキーを持つデータ同士を類似性が高いと判断する手段
を更に具備する
類似性検出装置。 - 類似性検出装置により実施され、検索対象データに対して、任意の類似度判定基準で類似するデータを検出するための指向性近傍検出方法であって、
類似度判定基準を設定するパラメータによって作成される乱数情報と関連付けられたテーブルを保持することと、
検索対象データを前記テーブルに登録し管理することと、
任意の参照点からの近傍検索要求に対して、前記テーブルを利用し、前記設定された類似度判定基準によって類似性判断を行い、近傍と判断されるデータを出力することと
を含む
指向性近傍検出方法。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US14/348,951 US9530081B2 (en) | 2011-10-03 | 2012-10-03 | Similarity detecting apparatus and directional nearest neighbor detecting method |
JP2013537536A JP6070956B2 (ja) | 2011-10-03 | 2012-10-03 | 類似性検出装置及び指向性近傍検出方法 |
SG11201401213UA SG11201401213UA (en) | 2011-10-03 | 2012-10-03 | Similarity detecting apparatus and directional nearest neighbor detecting method |
EP12838222.3A EP2765520B1 (en) | 2011-10-03 | 2012-10-03 | Similarity detection device and directional nearest neighbor method |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2011219547 | 2011-10-03 | ||
JP2011-219547 | 2011-10-03 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2013051619A1 true WO2013051619A1 (ja) | 2013-04-11 |
Family
ID=48043773
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2012/075673 WO2013051619A1 (ja) | 2011-10-03 | 2012-10-03 | 類似性検出装置及び指向性近傍検出方法 |
Country Status (5)
Country | Link |
---|---|
US (1) | US9530081B2 (ja) |
EP (1) | EP2765520B1 (ja) |
JP (1) | JP6070956B2 (ja) |
SG (1) | SG11201401213UA (ja) |
WO (1) | WO2013051619A1 (ja) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2016152527A (ja) * | 2015-02-18 | 2016-08-22 | Kddi株式会社 | 移動分析装置、移動分析方法および移動分析プログラム |
WO2019116515A1 (ja) * | 2017-12-14 | 2019-06-20 | 三菱電機株式会社 | 検索システム及び監視システム |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2013115202A1 (ja) * | 2012-01-30 | 2013-08-08 | 日本電気株式会社 | 情報処理システム、情報処理方法、情報処理装置およびその制御方法と制御プログラム、通信端末およびその制御方法と制御プログラム |
KR102017746B1 (ko) * | 2012-11-14 | 2019-09-04 | 한국전자통신연구원 | 유사도 산출 방법 및 그 장치 |
US9996574B2 (en) | 2015-06-30 | 2018-06-12 | International Business Machines Corporation | Enhancements for optimizing query executions |
US10810458B2 (en) * | 2015-12-03 | 2020-10-20 | Hewlett Packard Enterprise Development Lp | Incremental automatic update of ranked neighbor lists based on k-th nearest neighbors |
WO2017095421A1 (en) * | 2015-12-03 | 2017-06-08 | Hewlett Packard Enterprise Development Lp | Automatic selection of neighbor lists to be incrementally updated |
KR101834504B1 (ko) * | 2016-01-15 | 2018-03-06 | 단국대학교 산학협력단 | 암복호화 장치 및 방법 |
JP6433928B2 (ja) * | 2016-02-15 | 2018-12-05 | 株式会社東芝 | 検索装置、検索方法および検索システム |
KR101834522B1 (ko) | 2016-04-22 | 2018-03-06 | 단국대학교 산학협력단 | 데이터 확인 장치 및 이를 이용하여 데이터를 확인하는 방법 |
US10599664B2 (en) * | 2016-04-22 | 2020-03-24 | Cloudera, Inc. | Interactive identification of similar SQL queries |
US10671615B2 (en) * | 2016-05-27 | 2020-06-02 | Facebook, Inc. | Methods and systems for assigning affinity scores to contacts |
US11055349B2 (en) * | 2018-12-28 | 2021-07-06 | Intel Corporation | Efficient storage and processing of high-dimensional feature vectors |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH10326286A (ja) | 1997-05-27 | 1998-12-08 | Mitsubishi Electric Corp | 類似検索装置及び類似検索プログラムを記録した記録媒体 |
JP2000112973A (ja) * | 1998-10-02 | 2000-04-21 | Ricoh Co Ltd | 空間インデックス方法及び空間インデックス処理プログラムを格納した媒体 |
JP2003157283A (ja) * | 2001-09-04 | 2003-05-30 | Burittsua:Kk | 情報検索プログラム |
JP2004021430A (ja) | 2002-06-13 | 2004-01-22 | Fuji Xerox Co Ltd | 画像検索装置、画像検索方法及び画像検索プログラム |
JP2004199472A (ja) * | 2002-12-19 | 2004-07-15 | Internatl Business Mach Corp <Ibm> | 情報検索のためのデータ構造を生成するコンピュータ・システム、そのための方法、情報検索のためのデータ構造を生成するコンピュータ実行可能なプログラム、情報検索のためのデータ構造を生成するコンピュータ実行可能なプログラムを記憶したコンピュータ可読な記憶媒体、情報検索システム、およびグラフィカル・ユーザ・インタフェイス・システム |
JP2005070927A (ja) | 2003-08-21 | 2005-03-17 | Mitsuru Oba | 画像特徴取得方法、及びそのプログラムとその装置 |
JP2006190191A (ja) * | 2005-01-07 | 2006-07-20 | Sony Corp | 情報処理装置および方法、並びにプログラム |
JP2009116592A (ja) * | 2007-11-06 | 2009-05-28 | Nippon Telegr & Teleph Corp <Ntt> | ベクトル検索装置、ベクトル検索方法、プログラムおよびプログラムを記録した記録媒体 |
JP2011219547A (ja) | 2010-04-06 | 2011-11-04 | Shin-Etsu Chemical Co Ltd | エラストマー球状微粒子をポリオルガノシルセスキオキサンで被覆した複合微粒子およびその製造方法 |
-
2012
- 2012-10-03 EP EP12838222.3A patent/EP2765520B1/en active Active
- 2012-10-03 JP JP2013537536A patent/JP6070956B2/ja not_active Expired - Fee Related
- 2012-10-03 SG SG11201401213UA patent/SG11201401213UA/en unknown
- 2012-10-03 US US14/348,951 patent/US9530081B2/en active Active
- 2012-10-03 WO PCT/JP2012/075673 patent/WO2013051619A1/ja active Application Filing
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH10326286A (ja) | 1997-05-27 | 1998-12-08 | Mitsubishi Electric Corp | 類似検索装置及び類似検索プログラムを記録した記録媒体 |
JP2000112973A (ja) * | 1998-10-02 | 2000-04-21 | Ricoh Co Ltd | 空間インデックス方法及び空間インデックス処理プログラムを格納した媒体 |
JP2003157283A (ja) * | 2001-09-04 | 2003-05-30 | Burittsua:Kk | 情報検索プログラム |
JP2004021430A (ja) | 2002-06-13 | 2004-01-22 | Fuji Xerox Co Ltd | 画像検索装置、画像検索方法及び画像検索プログラム |
JP2004199472A (ja) * | 2002-12-19 | 2004-07-15 | Internatl Business Mach Corp <Ibm> | 情報検索のためのデータ構造を生成するコンピュータ・システム、そのための方法、情報検索のためのデータ構造を生成するコンピュータ実行可能なプログラム、情報検索のためのデータ構造を生成するコンピュータ実行可能なプログラムを記憶したコンピュータ可読な記憶媒体、情報検索システム、およびグラフィカル・ユーザ・インタフェイス・システム |
JP2005070927A (ja) | 2003-08-21 | 2005-03-17 | Mitsuru Oba | 画像特徴取得方法、及びそのプログラムとその装置 |
JP2006190191A (ja) * | 2005-01-07 | 2006-07-20 | Sony Corp | 情報処理装置および方法、並びにプログラム |
JP2009116592A (ja) * | 2007-11-06 | 2009-05-28 | Nippon Telegr & Teleph Corp <Ntt> | ベクトル検索装置、ベクトル検索方法、プログラムおよびプログラムを記録した記録媒体 |
JP2011219547A (ja) | 2010-04-06 | 2011-11-04 | Shin-Etsu Chemical Co Ltd | エラストマー球状微粒子をポリオルガノシルセスキオキサンで被覆した複合微粒子およびその製造方法 |
Non-Patent Citations (2)
Title |
---|
DATAR, M.; IMMORLICA, N.; INDYK, P.; MIRROKNI, V.: "Locality sensitive hashing scheme based on p-stable distributions", 2004, ACM SYMPOSIUM ON COMPUTATIONAL GEOMETRY |
See also references of EP2765520A4 |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2016152527A (ja) * | 2015-02-18 | 2016-08-22 | Kddi株式会社 | 移動分析装置、移動分析方法および移動分析プログラム |
WO2019116515A1 (ja) * | 2017-12-14 | 2019-06-20 | 三菱電機株式会社 | 検索システム及び監視システム |
CN111492371A (zh) * | 2017-12-14 | 2020-08-04 | 三菱电机株式会社 | 检索系统和监视系统 |
JPWO2019116515A1 (ja) * | 2017-12-14 | 2020-10-22 | 三菱電機株式会社 | 検索システム |
US11320798B2 (en) * | 2017-12-14 | 2022-05-03 | Mitsubishi Electric Corporation | Retrieval system and monitoring system |
CN111492371B (zh) * | 2017-12-14 | 2023-05-26 | 三菱电机株式会社 | 检索系统和监视系统 |
Also Published As
Publication number | Publication date |
---|---|
JPWO2013051619A1 (ja) | 2015-03-30 |
US20140324870A1 (en) | 2014-10-30 |
EP2765520A4 (en) | 2016-04-27 |
EP2765520B1 (en) | 2022-09-14 |
JP6070956B2 (ja) | 2017-02-01 |
SG11201401213UA (en) | 2014-08-28 |
EP2765520A1 (en) | 2014-08-13 |
US9530081B2 (en) | 2016-12-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6070956B2 (ja) | 類似性検出装置及び指向性近傍検出方法 | |
KR100353798B1 (ko) | 영상 객체 모양 정보 추출 방법 및 그를 이용한 내용기반 이미지 검색 시스템 및 그 방법 | |
CN108984785B (zh) | 一种基于历史数据和增量的指纹库的更新方法及装置 | |
US8849030B2 (en) | Image retrieval using spatial bag-of-features | |
JP2019045894A (ja) | 検索プログラム、検索方法、及び、検索プログラムが動作する情報処理装置 | |
CN111831660B (zh) | 度量空间划分方式评价方法、装置、计算机设备及存储介质 | |
KR20040005895A (ko) | 거리 측정기를 사용한 이미지 검색법 | |
CN103714148B (zh) | 基于稀疏编码分类的sar图像检索方法 | |
KR20180053731A (ko) | 일정한 처리 시간 내에 k개의 극값을 찾는 방법 | |
Zhang et al. | Dynamic time warping under product quantization, with applications to time-series data similarity search | |
CN113436223B (zh) | 点云数据的分割方法、装置、计算机设备和存储介质 | |
JP5014479B2 (ja) | 画像検索装置、画像検索方法及びプログラム | |
CN116610840A (zh) | 一种相似数据搜索方法、系统及电子设备 | |
CN110083731B (zh) | 图像检索方法、装置、计算机设备及存储介质 | |
Oguri et al. | General and practical tuning method for off-the-shelf graph-based index: Sisap indexing challenge report by team utokyo | |
CN115129915A (zh) | 重复图像检索方法、装置、设备及存储介质 | |
CN113496260A (zh) | 基于改进YOLOv3算法的粮库人员不规范作业检测法 | |
Arun et al. | On integrating re-ranking and rank list fusion techniques for image retrieval | |
JP5616310B2 (ja) | 画像マッチング装置及び画像マッチングプログラム | |
Yang et al. | Adaptive density peak clustering for determinging cluster center | |
JP7121706B2 (ja) | 情報処理装置、情報処理方法、及び情報処理プログラム | |
WO2021115154A1 (zh) | 可移动设备定位数据处理方法、装置、设备及存储介质 | |
Divya Lakshmi et al. | Helly hypergraph based matching framework using deterministic sampling techniques for spatially improved point feature based image matching | |
CN113657179A (zh) | 图像识别及建模方法、装置、电子设备及存储介质 | |
Faustino et al. | k d-SNN: A Metric Data Structure Seconding the Clustering of Spatial Data |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 12838222 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2012838222 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 2013537536 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 14348951 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |