CN103500165B - A kind of combination cluster and the high-dimensional vector quantity search method of double key value - Google Patents
A kind of combination cluster and the high-dimensional vector quantity search method of double key value Download PDFInfo
- Publication number
- CN103500165B CN103500165B CN201310365592.2A CN201310365592A CN103500165B CN 103500165 B CN103500165 B CN 103500165B CN 201310365592 A CN201310365592 A CN 201310365592A CN 103500165 B CN103500165 B CN 103500165B
- Authority
- CN
- China
- Prior art keywords
- key
- cluster
- cluster data
- vector
- tree
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
- G06F16/43—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
- G06F16/41—Indexing; Data structures therefor; Storage structures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention is a kind of combination cluster and the high-dimensional vector quantity search method of double key value.In the present invention, it is proposed that double key value index structure CDKB tree of a kind of combination cluster, it carries out a bunch division initially with clustering algorithm to high dimension vector collection, is then that each cluster data builds double key value extension B+Tree, forms CDKB tree.When retrieving, only the cluster data intersected with query context need to be scanned for, realize filtering for the first time by cluster, and by main key and auxiliary key (double key value), realize twice key value to filter, only main key and auxiliary key all need to be carried out similarity mode calculating between the high dimension vector of those in hunting zone and query vector.The index structure that the present invention proposes is compared by cluster and simple double key value size, greatly reduces the operand of similarity mode, greatly accelerates retrieval rate.
Description
Technical field
The invention belongs to the data processing field such as multimedia information retrieval, Intelligent Information Processing, data mining,
Particularly relate to a kind of combination cluster and the high-dimensional vector quantity search method of double key value.
Background technology
Along with computer and the development of information technology, create the multi-medium data of magnanimity, how in magnanimity
Multimedia database in be quickly found out required information be one of current multimedia data storehouse area research
Important Problems.Traditional method is by being manually labeled multi-medium data, then passing through text retrieval
Realize multimedia information retrieval.But artificial mark exists the defect that workload is big and subjectivity is strong, right
For the multi-medium data of explosive growth, the most artificial mark is the most attainable, it is therefore desirable to grind
Study carefully multimedia information retrieval technology based on content.
The technology path realizing multimedia information retrieval based on content is: by eigentransformation, by many matchmakers
The point characteristic vector that volume data is mapped in higher dimensional space, describes multimedia pair with this feature vector
As, obtain feature database；Then the characteristic vector of query object is extracted by same eigentransformation method,
The similar to search of multimedia messages is realized finally by the similarity mode between characteristic vector.The most matchmakers
The similar to search of body information is changed into finds the point set nearest with giving query point in high-dimensional feature space
Process.
In higher dimensional space, just to find and to give the point set that query point is the most close, the method for simple, intuitive
It is sequential scan, the most successively each feature (high dimension vector) in feature database is carried out similar to query point
Degree coupling, returns those feature point sets mated most, obtains retrieving result.Sequential scan is along with feature database
Middle number of features and the increase of characteristic dimension, calculate elapsed time and linearly increase, when the feature in feature database
During large number, sequential scan can not meet real-time demand.In order to accelerate retrieval rate, the most frequently used
Method be just by mean of High-dimensional Index Technology.
In order to realize the management to magnanimity high dimension vector, researchers propose substantial amounts of index structure, its
In the most classical be the R-tree family series index structure with R-tree as representative.R-tree is 20 generation
Discipline is proposed by Guttman the eighties, a kind of index knot designed for managing multidimensional tile data
Structure, it is a kind of height balanced tree utilizing tree construction management data, and each node is with all in this node
The minimum enclosed rectangle (MBR:Minimal Bounding Rectangle) of data represents, real data
Only occur in leaf node.This index structure can also be used for the pipe of higher dimensional space middle data by extension
Reason.In query script, search for downwards to leaf node layer from root node level, inquired about by calculating
Minimum range between each node M BR of vector sum judges whether query context intersects to come real with certain node
Existing beta pruning is filtered, and searches only for comprising the subtree of result, thus accelerates retrieval rate.This index structure
Allow the space overlap between node, have impact on its search efficiency.In order to improve the performance of R-tree, grind
The persons of studying carefully continuously propose R+-tree、R*-tree, SS-tree, SR-tree, X-tree, A-tree etc. index
Structure.But these tree index structures drastically decline, even along with the increase of characteristic dimension, search efficiency
Not as sequential scan, here it is so-called " dimension disaster ".
In addition to tree, there is also the higher-dimension index structure to one-dimensional conversion, such as: pyramid
Technology, NB-tree, iDistance, iMinMax etc..Higher-dimension passes through to the index structure of one-dimensional conversion
Certain rule, is mapped as high dimension vector one-dimensional data (referred to as key value), then uses one-dimensional B+-tree
Managing these key values, key value is at B+The leaf node layer ordered arrangement of-tree.When inquiring about,
First pass through identical higher-dimension and calculate the inquiry key value of query vector to one-dimensional transformation rule, finally according to
Query context, determines key value original position and the end position of search, and it is right to scan these key values successively
The high dimension vector answered, calculates the similitude between query vector and these high dimension vectors, returns those most like
High dimension vector collection, obtain retrieve result.From query script, the index knot of higher-dimension to one-dimensional conversion
Structure under any circumstance performance is superior to or is equivalent to sequential scan, and great many of experiments based on forefathers shows,
This kind of index structure reduces slowly with dimension and the increase of data volume, performance.
Pyramid technology, these higher-dimensions such as NB-tree, iDistance, iMinMax are to one-dimensional conversion index
Structure filters beta pruning by the realization of simply comparing of single key value, although the distance that need not complexity calculates
And have a higher recall precision, but higher-dimension can cause substantial amounts of data message to the process of one-dimensional conversion
Lose, cause different vector to be likely to be of identical one-dimensional k ey value, be only capable of filtering out by single key value
A part of data that ratio is little, the operand causing final similarity mode process is the biggest, looks into
Ask expense the least.
Summary of the invention
It is an object of the invention to propose a kind of combination cluster and the high-dimensional vector quantity search method of double key value,
The method uses clustering algorithm higher dimensional space to carry out a bunch division, then by each higher-dimension in each cluster data
DUAL PROBLEMS OF VECTOR MAPPING is double one-dimensional k ey values.In query script, filter out a part and interrogation zone by cluster
The disjoint cluster data in territory, for each cluster data by increasing by one layer of key value filter course, uses further
Simple key value compares realization and again filters beta pruning, considerably reduces final vector similarity coupling
Operand, dramatically speed up inquiry velocity.
The overall thought of the present invention is as follows: carry out high dimension vector collection bunch dividing initially with clustering algorithm,
Then choose two reference points for each cluster data, and use high dimension vector to this cluster data two each cluster data
High dimension vector in this cluster data is mapped as double one-dimensional k ey value by the distance of individual reference point, and unification chooses this
The key value that in cluster data, a certain employing same reference points obtains is as main key, and another is as auxiliary key,
After, the main key being respectively adopted each cluster data is that each cluster data builds B+-tree, the most each B+-tree leaf
The each main key of node layer binds a pointer pointing to its corresponding auxiliary key, and each auxiliary key binds
One pointer pointing to its corresponding high dimension vector.When retrieving, first only need to intersect with query context
Each cluster data scan for, then in each cluster data intersected with query region, use identical two
Query vector is mapped as inquiring about main key and inquiring about auxiliary key by reference point and mapping method, by inquiring about main key
The main key hunting zone determined in each cluster data with query context, and by inquiring about auxiliary key and inquiry model
Enclose the hunting zone determining auxiliary key, its auxiliary key after main key filters need to be searched at auxiliary key for last
In the range of those high dimension vector and query vectors between carry out similarity mode calculating, return those most like
Vector set, obtain retrieve result.
Concrete innovative point: higher dimensional space is carried out a bunch division, selects for each high dimension vector in each cluster data
Take two reference points and obtain double one-dimensional k ey value, compared, greatly by cluster and twice simple key value
Decrease the final high dimension vector number participating in similarity mode computing, significantly speed up inquiry velocity.
The concrete grammar step of the present invention is: (1) uses clustering algorithm that high dimension vector collection carries out a bunch division,
Obtain cluster centre and the cluster radius of each cluster data；(2) it is that each cluster data builds double key value extension
B+-tree, builds double key value extension B for every cluster data+The process of-tree is: first select for this cluster data
Take two reference points, and use high dimension vector to the distance of the two reference point by the higher-dimension in this cluster data
DUAL PROBLEMS OF VECTOR MAPPING is double one-dimensional k ey values, and unification chooses what a certain employing same reference points in this cluster data obtained
Key value is main key, and another is as auxiliary key, and the main key then using this cluster data is this cluster data structure
Build B+-tree, simultaneously this B+It is auxiliary that each main key of-tree leaf node layer binds its correspondence of sensing
The pointer of key, each auxiliary key binds a pointer pointing to its corresponding high dimension vector, B+-tree leaf
The all main key of node layer forms main key layer, and all auxiliary key form auxiliary key layer；(3) by each number of clusters
According to cluster centre and cluster radius all bind one and point to double key values constructed by its corresponding cluster data and extend
B+The pointer of-tree, forms CDKB-tree；(4), when retrieving, that is filtered out by query context
A little and the disjoint each cluster data of query region, scans for each cluster data intersected with query context,
Searching method in each cluster data is: use identical reference point and mapping method to be mapped by query vector
For inquiring about main key and inquiring about auxiliary key, determine at the main key of this cluster data by inquiring about main key and query context
The original position of layer search and end position, then determined at this bunch by the auxiliary key of inquiry and query context
The auxiliary key hunting zone of data auxiliary key layer, then to main key layer from searching for original position to end position
Between each main key be scanned one by one, it is judged that whether auxiliary key corresponding for this main key searches at auxiliary key
In the range of, if in hunting zone, then carry out between this high dimension vector corresponding for auxiliary key and query vector
Similarity mode calculates, and is returned by the high dimension vector meeting query context, obtains retrieving result.
Further, the clustering algorithm described in step 1 includes that Kmeans clusters.
Further, described in step 2, choose two reference points, including choosing in initial point and cluster
The heart is reference point.
Further, the distance of the two reference point can be used European by the high dimension vector described in step 2
Distance or city block distance.
Further, when the CDKB-tree described in step 3 carries out high dimension vector insertion, first basis
This high dimension vector is to the distance value of each cluster data cluster centre, the number of clusters that this high dimension vector of selected distance is nearest
According to carrying out update, update cluster radius, then according to being inserted into vector to two references of this cluster data
The distance of point obtains being inserted into the main key of vector and auxiliary key value, positions it according to the size of this main key value
This cluster data correspondence B should be inserted into+In a certain leaf node of-tree: if this leaf node less than, then
Directly this main key value being inserted in this leaf node, its auxiliary key is inserted into this position corresponding for main key
Place, is inserted into characteristic vector and is inserted into this position corresponding for auxiliary key, and it is right to make main key produce sensing
Answering the pointer of auxiliary key, its corresponding auxiliary key produces and points to the pointer being inserted into high dimension vector, updates this leaf
The key value that father node is corresponding；If this leaf node is the fullest, the mode of process is as follows:
1) if the left and right brotgher of node of this leaf node exist less than situation, then combine brother around
Node, carries out to be inserted becoming owner of key, auxiliary key and the insertion of high dimension vector, and it is corresponding to update its father node
Key value；
2) if the brotgher of node is the fullest around, then the main key value being inserted into high dimension vector is combined, directly
This leaf node is divided, newly generated leaf node after division is inserted in its father node, with
Time its auxiliary key and high dimension vector be inserted into corresponding storage position, update the key that its father node is corresponding
Value, if father node is the fullest, fission process continues up transmission, and updates the key value of correspondence.
Further, when the carrying out described in step 4 is retrieved, the retrieval mode of employing had both included that scope was looked into
Ask and also include k NN Query.
Further, the query context described in step 4, for range query, is by inquiry half
Footpath determines, is to be determined by the inquiry radius being incremented by by a certain step-length for k NN Query,
Distance value until kth neighbour to query vector is less than inquiry radius.
Accompanying drawing explanation
Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes of the present invention
Point, the schematic embodiment of the present invention and explanation thereof are used for explaining the present invention, are not intended that the present invention
Improper restriction.In the accompanying drawings:
The flow chart of Fig. 1 (a) the method for the invention
The exemplary plot of Fig. 1 (b) index structure of the present invention
The block diagram that Fig. 2 inquires about at the enterprising line range of index structure of the present invention
Fig. 3 carries out the block diagram of k NN Query on index structure of the present invention
Detailed description of the invention
In order to solve the technical problem that needed for making the present invention, technical scheme is clearer, understand, below knot
Close accompanying drawing and the detailed description of the invention of the present invention is described further by embodiment.
A kind of combination cluster that embodiment of the present invention provides and the high-dimensional vector quantity search method of double key values its
Shown in flow chart such as Fig. 1 (a) that index structure builds:
First, use clustering algorithm that high dimension vector collection carries out space bunch and divide, obtain each bunch of high dimensional data；
Secondly calculate cluster centre and the radius of each cluster data, and choose two reference points for every cluster data；Again
Calculate each high dimension vector in each cluster data and the distance between two reference points of this cluster data one by one, obtain
Double one-dimensional k ey values that each high dimension vector is corresponding；Then a certain employing same reference in each cluster data is chosen
The key value that point obtains is as main key, and another is as auxiliary key, and uses the main key of each cluster data for being somebody's turn to do
Cluster data builds B+-tree, inserts auxiliary key corresponding for the main key of each cluster data and high dimension vector data
To corresponding auxiliary key and high dimension vector storage position, the finger of its corresponding auxiliary key is pointed in each main key binding
Pin, each auxiliary key binding is pointed to the pointer of its corresponding high dimension vector, is obtained double key that each cluster data is corresponding
Value extension B+-tree；Finally the cluster centre of each cluster data and cluster radius are bound double key of this cluster data
Value extension B+-tree, forms CDKB-tree index structure.(as shown in Fig. 1 (b), upper strata is clustering information
Layer, centre is each B built by the main key of each cluster data+-tree, bottom for store auxiliary key and higher-dimension to
The auxiliary key layer of amount and characteristic vector layer, each B+The each main key of-tree leaf node layer binds one
Pointing to the pointer of its corresponding auxiliary key, its auxiliary key binds a pointer pointing to corresponding high dimension vector.)
When retrieving, judge whether each cluster data intersects with this inquiry by query vector and query context,
Then the cluster data for intersecting with this inquiry searches further for, and the method for search is: first adopt
With identical reference point and mapping ruler, calculate the distance between query vector and two reference points of this cluster data,
Obtain the main key of the inquiry in this cluster data and inquire about auxiliary key；Then according to inquiring about main key and inquiry model
Enclose, determine at double key values extension B that this cluster data is corresponding+-tree index structure main key layer (i.e. B+-tree
Leaf node layer) main key hunting zone, obtain scanning starting position and the end position of main key layer,
And according to inquiring about auxiliary key and query context, determine at double key values extension B that this cluster data is corresponding+-tree
The auxiliary key hunting zone of index structure auxiliary key layer；Finally, from the scanning starting position of main key layer to knot
Bundle position (main key hunting zone), carries out key scan value, it is judged that the auxiliary key that this main key is corresponding one by one
Whether within auxiliary key hunting zone, if within hunting zone, then calculate this higher-dimension corresponding for auxiliary key
Distance between vector and query vector, returns the high dimension vector of satisfied retrieval result, obtain similar to
Quantity set.
The retrieval mode carrying out retrieving of the present invention includes range query and k NN Query, and scope is looked into
Ask flow chart as in figure 2 it is shown, the flow chart of k NN Query as shown in Figure 3.From the figure 3, it may be seen that k is near
Adjacent inquiry is to be realized by range query.
Above-mentioned high dimension vector can be the characteristic vector of image, video, audio frequency.
It should be appreciated that the above-mentioned description for embodiment is more concrete, can not therefore think
Being the restriction to scope of patent protection of the present invention, the scope of patent protection of the present invention should be with claims
It is as the criterion.
Claims (7)
1. the high-dimensional vector quantity search method combining cluster and double key value, it is characterised in that concrete steps are such as
Under:
1) use clustering algorithm to carry out high dimension vector collection bunch dividing, obtain each cluster data cluster centre and
Cluster radius；
2) it is that each cluster data builds double key value extension B+-tree, builds double key value extension for every cluster data
B+The process of-tree is: first choose two reference points for this cluster data, and use high dimension vector to this two
High dimension vector in this cluster data is mapped as double one-dimensional k ey value by the distance of individual reference point, and unification chooses this
The key value that in cluster data, a certain employing same reference points obtains is main key, and another is as auxiliary key, then
The main key using this cluster data is that this cluster data builds B+-tree, simultaneously this B+-tree leaf node layer
Each main key binds a pointer pointing to its corresponding auxiliary key, and each auxiliary key binds a sensing
The pointer of its corresponding high dimension vector, B+The all main key of-tree leaf node layer forms main key layer, institute
Auxiliary key is had to form auxiliary key layer；
3) cluster centre and the cluster radius of each cluster data are all bound one and pointed to its corresponding cluster data institute structure
Build double key value extension B+The pointer of-tree, forms CDKB-tree；
4), when retrieving, those and the disjoint each cluster data of query region are filtered out by query context,
Scanning for each cluster data intersected with query context, the searching method in crossing each cluster data is:
Identical reference point and mapping method is used to be mapped as query vector inquiring about main key and inquiring about auxiliary key, logical
Cross and inquire about main key and query context determines at the original position of this cluster data main key layer search and stop bits
Put, then determined by the auxiliary key of inquiry and query context and search for model at the auxiliary key of this cluster data auxiliary key layer
Enclose, then main key layer be scanned to each main key end position one by one from search original position,
Judge auxiliary key corresponding for this main key whether in auxiliary key hunting zone, if in hunting zone, the most right
Carry out similarity mode calculating between this high dimension vector corresponding for auxiliary key and query vector, inquiry model will be met
The high dimension vector enclosed returns, and obtains retrieving result.
2. the method for claim 1, it is characterised in that: the clustering algorithm described in step 1 includes
Kmeans clusters.
3. the method for claim 1, it is characterised in that: choose two reference points described in step 2,
Including choosing initial point and cluster centre is reference point.
4. the method for claim 1, it is characterised in that: the high dimension vector described in step 2 to this two
The distance of individual reference point can use Euclidean distance or city block distance.
5. the method for claim 1, it is characterised in that: CDKB-tree described in step 3 is being entered
When row high dimension vector inserts, first according to the distance value of this high dimension vector to each cluster data cluster centre, choosing
Take and carry out update apart from the cluster data that this high dimension vector is nearest, update cluster radius, then according to treating
Insert the vectorial distance to these two reference points of cluster data to obtain being inserted into the main key of vector and auxiliary key value,
Size according to this main key value positions it should be inserted into this cluster data correspondence B+The a certain leaf node of-tree
In: if this leaf node less than, then directly this main key value is inserted in this leaf node, it is auxiliary
Key is inserted into the position that this is corresponding for main key, is inserted into characteristic vector and is inserted into this position corresponding for auxiliary key
Putting place, and make main key produce the pointer pointing to its corresponding auxiliary key, it is to be inserted that its corresponding auxiliary key produces sensing
Enter the pointer of high dimension vector, update the key value that this leaf node father node is corresponding；If this leaf node
The fullest, the mode of process is as follows:
Step one: if the left and right brotgher of node of this leaf node exist less than situation, then combine it left
The right brotgher of node, carries out to be inserted becoming owner of key, auxiliary key and the insertion of high dimension vector, and updates its father node
Corresponding key value；
Step 2: if the brotgher of node is the fullest around, then combine the main key value being inserted into high dimension vector,
Directly this leaf node is divided, newly generated leaf node after division is inserted in its father node,
Its auxiliary key and high dimension vector are inserted into corresponding storage position simultaneously, update the key that its father node is corresponding
Value, if father node is the fullest, fission process continues up transmission, and updates the key value of correspondence.
6. the method for claim 1, it is characterised in that: when the carrying out described in step 4 is retrieved, adopt
Retrieval mode both included that range query also included k NN Query.
7. the method for claim 1, it is characterised in that: the query context described in step 4, for
For range query, inquiry radius determine, be by by a certain step for k NN Query
The long inquiry radius being incremented by determines, the distance value until kth neighbour to query vector is less than inquiry
Till radius.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310365592.2A CN103500165B (en) | 2013-08-21 | 2013-08-21 | A kind of combination cluster and the high-dimensional vector quantity search method of double key value |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310365592.2A CN103500165B (en) | 2013-08-21 | 2013-08-21 | A kind of combination cluster and the high-dimensional vector quantity search method of double key value |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103500165A CN103500165A (en) | 2014-01-08 |
| CN103500165B true CN103500165B (en) | 2016-08-31 |
Family
ID=49865376
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310365592.2A Expired - Fee Related CN103500165B (en) | 2013-08-21 | 2013-08-21 | A kind of combination cluster and the high-dimensional vector quantity search method of double key value |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103500165B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103997783B (en) * | 2014-05-14 | 2017-03-29 | 无锡北邮感知技术产业研究院有限公司 | A kind of outdoor cluster match localization method and device |
| WO2016019484A1 (en) * | 2014-08-08 | 2016-02-11 | Xiaoou Tang | An apparatus and a method for providing super-resolution of a low-resolution image |
| CN104484398B (en) * | 2014-12-12 | 2018-02-02 | 北京国双科技有限公司 | Data aggregation method and device in tables of data |
| CN108846067B (en) * | 2018-06-05 | 2024-10-18 | 洛阳师范学院 | High-dimensional data similarity connection query method and device based on mapping space division |
| CN111309984B (en) * | 2020-03-10 | 2023-09-05 | 支付宝(杭州)信息技术有限公司 | Method and device for retrieving node vector from database by index |
| CN112115281A (en) * | 2020-09-17 | 2020-12-22 | 杭州海康威视数字技术股份有限公司 | Data retrieval method, device and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6289354B1 (en) * | 1998-10-07 | 2001-09-11 | International Business Machines Corporation | System and method for similarity searching in high-dimensional data space |
| CN101299213A (en) * | 2008-06-17 | 2008-11-05 | 中国地质大学（武汉） | N-dimension clustering order recording tree space index method |
| CN102253993A (en) * | 2011-07-08 | 2011-11-23 | 北京航空航天大学 | Vocabulary tree-based audio-clip retrieving algorithm |
-
2013
- 2013-08-21 CN CN201310365592.2A patent/CN103500165B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6289354B1 (en) * | 1998-10-07 | 2001-09-11 | International Business Machines Corporation | System and method for similarity searching in high-dimensional data space |
| CN101299213A (en) * | 2008-06-17 | 2008-11-05 | 中国地质大学（武汉） | N-dimension clustering order recording tree space index method |
| CN102253993A (en) * | 2011-07-08 | 2011-11-23 | 北京航空航天大学 | Vocabulary tree-based audio-clip retrieving algorithm |
Non-Patent Citations (2)
| Title |
|---|
| SS-ClusterTree: a subspace clustering based indexing algorithm over high-dimensional image features;Hongli Xu et al.;《CIVR "08 Proceedings of the 2008 international conference on Content-based image and video retrieval》;20080731;第95-104页 * |
| 基于支持向量机的两阶段模糊聚类在视频检索中的应用;陈颉等;《计算机科学》;20090630;第36卷(第6期);第227-230页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103500165A (en) | 2014-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103500165B (en) | A kind of combination cluster and the high-dimensional vector quantity search method of double key value | |
| Pfoser et al. | Novel approaches to the indexing of moving object trajectories. | |
| Corral et al. | Closest pair queries in spatial databases | |
| Kalashnikov et al. | Main memory evaluation of monitoring queries over moving objects | |
| Rocha-Junior et al. | Top-k spatial keyword queries on road networks | |
| CN109947904A (en) | A kind of preference space S kyline inquiry processing method based on Spark environment | |
| CN103092926B (en) | The three-dimensional space index method of multi-level mixing | |
| CN108052514A (en) | A kind of blending space Indexing Mechanism for handling geographical text Skyline inquiries | |
| CN102609530A (en) | Space database indexing method of regional double-tree structure | |
| CN111639075B (en) | Non-relational database vector data management method based on flattened R tree | |
| CN104346444B (en) | A kind of the best site selection method based on the anti-spatial key inquiry of road network | |
| CN106503223A (en) | A kind of binding site and the online source of houses searching method and device of key word information | |
| CN101692230A (en) | Three-dimensional R tree spacial index method considering levels of detail | |
| CN102306202B (en) | High-dimension vector rapid searching algorithm based on block distance | |
| CN108681577A (en) | A kind of novel library structure data index method | |
| Sreenath et al. | The hcC-tree: an efficient index structure for object oriented databases | |
| CN105138607B (en) | A kind of KNN querying methods based on combination grain distributed memory grid index | |
| CN103514263B (en) | A kind of high-dimensional index structure construction method using double key value and search method | |
| CN105677840A (en) | Data query method based on multi-dimensional increasing data model | |
| JP3938815B2 (en) | Node creation method, image search method, and recording medium | |
| CN108509618A (en) | A kind of big data multidimensional data indexing means based on space filling curve | |
| CN101467149A (en) | Adaptive index with variable compression | |
| KR100612376B1 (en) | A index system and method for xml documents using node-range of integration path | |
| CN107688620B (en) | Top-k query-oriented method for instantly diversifying query results | |
| CN104166672B (en) | A kind of XML keyword query methods |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160831 Termination date: 20180821 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |