CN107391601A - A kind of construction method of the high dimensional indexing of face feature vector - Google Patents
A kind of construction method of the high dimensional indexing of face feature vector Download PDFInfo
- Publication number
- CN107391601A CN107391601A CN201710525541.XA CN201710525541A CN107391601A CN 107391601 A CN107391601 A CN 107391601A CN 201710525541 A CN201710525541 A CN 201710525541A CN 107391601 A CN107391601 A CN 107391601A
- Authority
- CN
- China
- Prior art keywords
- node
- child node
- child
- objects
- enclosing circle
- 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.)
- Pending
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
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Library & Information Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention belongs to facial image searching field, in particular relates to a kind of construction method of the high dimensional indexing of face feature vector, comprises the following steps：A face characteristic is chosen to be stored in root node；The object newly inserted is stored in root node, if root node is discontented with, is inserted directly into, if number of objects has reached maximum M in root node, carries out one dividing into three operation；The object newly inserted is stored in child node, if child node is discontented with, is inserted directly into, if the number of objects in child node has reached maximum M, carries out weight insertion operation；The object newly inserted continues to be stored in child node, until the number of objects in child node reaches maximum M again, carries out splitting operation；The storage of the object newly inserted, all objects are deposited terminate after complete the foundation of index.The present invention improves recall precision, meets the needs of others' face characteristic retrieval of millions.
Description
Technical field
The invention belongs to facial image searching field, more particularly to a kind of structure side of the high dimensional indexing of face feature vector
Method.
Background technology
In recent years, smart city in all parts of the country was in development like a raging fire, and Video Surveillance Industry is fast-developing, entirely
Security protection market scale is grown rapidly, and the importance of recognition of face is more and more stronger.
How to find out that to be provided simultaneously with a suspects of multiple identity be long-standing problem in the other citizenship storehouse of millions
The problem of public safety related work.Existing working method is mainly by the modes such as artificial judgment, random inspection, Wu Fazhun
Really, rapidly find out and eliminate safe hidden trouble.Meanwhile current index creates a mechanism, tieed up with the characteristic vector of human face data
Several increases, recall precision drastically decline, and can not meet the needs of others' face characteristic retrieval of millions.
Therefore, there is an urgent need to a kind of Indexing Mechanism that can be directed to face feature vector retrieval, recall precision is improved, with full
The needs of others' face characteristic retrieval of sufficient millions.
The content of the invention
According to problems of the prior art, the invention provides a kind of structure of the high dimensional indexing of face feature vector
Method, recall precision is improved, meet the needs of others' face characteristic retrieval of millions.
The present invention uses following technical scheme：
A kind of construction method of the high dimensional indexing of face feature vector, comprises the following steps：
S1, choose a face characteristic and be stored in as an object in root node, the maximum capacity of root node is
M；
S2, a face characteristic newly the inserting object new as one is stored in root node, if root node
It is discontented, then it is inserted directly into；If number of objects has reached maximum M in root node, one dividing into three operation is carried out to root node,
Root node is split into three child nodes, i.e. child node A, child node B and child node C, by M object in root node and newly
One object of insertion is stored in three child nodes respectively accordingly, and the storage number of objects in three child nodes is many
In m, the maximum capacity of three child nodes is M；
S3, a face characteristic newly the inserting object new as one are stored in a child node at random, if
The fixed child node is A, if child node A is discontented with, is inserted directly into；If the number of objects in child node A has reached maximum
M, then weight insertion operation is carried out to M object in child node A and the object newly inserted so that the object in child node A
Quantity is less than M；
S4, a face characteristic newly the inserting object new as one continues to be stored in child node A, until son
Number of objects in node reaches maximum M again, and now M object in child node and the object newly inserted are carried out
Splitting operation, obtain three child node of next layer after child node A divisions；Similarly, until child node B and child node C are completed
Splitting operation, i.e., obtain three child node of next layer after child node B and child node C divisions；
S5, repeat step S3~S4, storage of the face characteristic data newly inserted as object, all face characteristics
Data are deposited terminate after complete the foundation of index.
Preferably, the heavy insertion operation described in step S3 comprises the following steps：
S11, using smallest enclosing circle algorithm and Euclidean distance algorithm, calculate M object in child node A and newly insert
Distance D of one object to child node A smallest enclosing circle center；
S12, M object in child node A and the object newly inserted are sorted from big to small by D values, (M+ before taking-up
1)/2 or M/2 object is as unit is inserted again, is inserted into same layer other child nodes.
Preferably, the splitting operation described in step S4 comprises the following steps：
S21, three objects are arbitrarily selected from M object in child node A and the object newly inserted, respectively as
The center of three initial clusterings；
S22, the average V of the object in each initial clustering, the i.e. center to each initial clustering are calculated, calculate and remove in this
All objects and the average at this center, average V calculation formula outside the heart are as follows:
Wherein, SiRepresent the set of the object of all initial cluster centers, uiRepresent the center of ith cluster, xjRepresent son
In M object in node A and the object newly inserted, except this center uiOutside remaining all object；
S23, according to average V, the distance of each object and three initial cluster centers is calculated by Euclidean distance algorithm, will
Cluster centre of that the minimum initial cluster center of the distance of object and three initial cluster centers as this object, obtains weight
Three clusters newly divided；
S24, repeat step S22~S23, until poor between the average of three initial cluster centers and each a upper average
The absolute value of value is less than specified threshold K；
S25, three clusters finally given, as three child node of next layer after division.
Preferably, after the completion of the index construct, object therein can be retrieved；Retrieving includes following step
Suddenly：
S31, a query object P is inputted, to each not visited n omicronn-leaf child node, is calculated using smallest enclosing circle
Method and Euclidean distance algorithm calculate the ranking value for the smallest enclosing circle that each father node is formed with its child node, and by they according to
Ascending order or descending sort, it is stored in movable branch list；
S32, according to prune rule, movable branch list is trimmed, deletes unnecessary branch；
S33, to each smallest enclosing circle repeat step S32 in movable branch list, until movable branch list is sky；
S34, for leafy node, wherein each object is calculated to query object P distance using Euclidean distance algorithm,
The object minimum to query object P distance is to retrieve object.
It is further preferred that can delete the object obtained by retrieval, specific deletion process comprises the following steps：
S41, by step S31~S34, determine the leafy node where object to be deleted；
S42, the object in leafy node is deleted；
S43, if number of objects is less than m in leafy node, this leafy node is deleted, all objects in node are walked
Heavy insertion operation described in rapid S11~S12；Conversely, any operation is not done.
Preferably, the M is set according to the number of object in index, and the number of object is more, and M setting values are bigger, right
The number of elephant is fewer, and for M setting values with regard to smaller, M is integer；The m is sized according to M, m≤(M+1)/3, and m is integer；
The K is sized according to the needs of clustering precision, and clustering required precision is higher, then the value of K settings is just smaller, gathers
Class dividing precision requires lower, then the value of K settings is bigger.
It is further preferred that the prune rule is specific as follows：
MINDIST refers to that object refers to object to smallest enclosing circle to the minimum range of smallest enclosing circle, MINMAXDIST
Ultimate range, if the MINDIST of node R smallest enclosing circle is than the MINMAXDIST of the smallest enclosing circle of other nodes
Greatly, then node R is wiped out；Conversely, then node R retains, and to node R smallest enclosing circle, query object P to an object O
Distance be more than query object P to node R ultimate range MINMAXDIST (P, R), then object O wiped out；Conversely, then object
O retains, and to node R smallest enclosing circle, if query object P to node R minimum range MINDIST (P, R) is more than inquiry
Object P to object O distance, then node R wiped out.
The advantages of the present invention are：
1) present invention is constantly inserted into the child node of tree using face characteristic data as object to establish index, for son
When number of objects reaches maximum M first in node, weight insertion operation is carried out to this child node, makes number of objects in this child node
Less than maximum M, with the storage of new insertion object, when number of objects reaches maximum again in child node, to this child node
Splitting operation is carried out, the object in child node is stored in its next straton node, so circulation, can deposit thousand a large amount of faces
The object of characteristic, meet the needs of others' face characteristic retrieval of millions.
2) by the construction method of high dimensional indexing of the present invention and the index established, when being retrieved to retrieval object, profit
With prune rule, leafy node existing for retrieving object is found, without considering intermediate node, improves recall precision.
3) when deleting object, first retrieved and the index established by the construction method of high dimensional indexing of the present invention
The leafy node gone out where object, after deleting object, if number of objects is less than m in leafy node, this leafy node is deleted, will
All objects carry out weight insertion operation in node；Conversely, any operation is not done；So that the number of objects in leafy node is located always
More than the m values of setting, reduce the quantity of leafy node, be advantageous to improve recall precision.
Brief description of the drawings
Fig. 1 is the flow chart of the construction method of the high dimensional indexing of the face feature vector of the present invention.
Fig. 2 is the flow chart of the heavy insertion operation in the present invention.
Fig. 3 is the flow chart of the splitting operation in the present invention.
Fig. 4 is the process flow diagram flow chart after the completion of index construct, retrieved in the present invention to object therein.
Fig. 5 is the flow chart deleted in the present invention the object obtained by retrieval.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other under the premise of creative work is not made
Embodiment, belong to the scope of protection of the invention.
As shown in figure 1, a kind of construction method of the high dimensional indexing of face feature vector, comprises the following steps：
S1, choose a face characteristic and be stored in as an object in root node, the maximum capacity of root node is
M；
S2, a face characteristic newly the inserting object new as one is stored in root node, if root node
It is discontented, then it is inserted directly into；If number of objects has reached maximum M in root node, one dividing into three operation is carried out to root node,
Root node is split into three child nodes, i.e. child node A, child node B and child node C, by M object in root node and newly
One object of insertion is stored in three child nodes respectively accordingly, and the storage number of objects in three child nodes is many
In m, the maximum capacity of three child nodes is M；
S3, a face characteristic newly the inserting object new as one are stored in a child node at random, if
The fixed child node is A, if child node A is discontented with, is inserted directly into；If the number of objects in child node A has reached maximum
M, then weight insertion operation is carried out to M object in child node A and the object newly inserted so that the object in child node A
Quantity is less than M；
Specifically, as shown in Fig. 2 described heavy insertion operation comprises the following steps：
1) smallest enclosing circle algorithm and Euclidean distance algorithm are utilized, calculates M object in child node A and newly insert one
Distance D of the individual object to child node A smallest enclosing circle center；
Wherein, smallest enclosing circle algorithm and Euclidean distance algorithm are existing algorithm.
2) M object in child node A and the object newly inserted are sorted from big to small by D values, (M+ before taking-up
1)/2 or M/2 object is as unit is inserted again, is inserted into same layer other child nodes；
When M is set as odd number, then (M+1)/2 object is as unit is inserted again before taking out, when M is set as even number, then
M/2 object is as slotting unit again before taking-up.
S4, a face characteristic newly the inserting object new as one continues to be stored in child node A, until son
Number of objects in node reaches maximum M again, and now M object in child node and the object newly inserted are carried out
Splitting operation, obtain three child node of next layer after child node A divisions；Similarly, until child node B and child node C are completed
Splitting operation, i.e., obtain three child node of next layer after child node B and child node C divisions；
Specifically, as shown in figure 3, described splitting operation comprises the following steps：
1) three objects are arbitrarily selected from M object in child node A and the object newly inserted, respectively as three
The center of individual initial clustering；
2) the average V of the object in each initial clustering, the i.e. center to each initial clustering are calculated, calculating removes this center
Outer all objects and the average at this center, average V calculation formula are as follows:
Wherein, SiRepresent the set of the object of all initial cluster centers, uiRepresent the center of ith cluster, xjRepresent son
In M object in node A and the object newly inserted, except this center uiOutside remaining all object；
3) according to average V, the distance of each object and three initial cluster centers is calculated by Euclidean distance algorithm, will be right
As cluster centre of that the minimum initial cluster center of the distance with three initial cluster centers as this object, obtain again
Three clusters of division；
4) repeat step 2)~step 3), until poor between the average of three initial cluster centers and each a upper average
The absolute value of value is less than specified threshold K；
5) three clusters finally given, as three child node of next layer after division；
S5, repeat step S3~S4, storage of the face characteristic data newly inserted as object, all face characteristics
Data are deposited terminate after complete the foundation of index.
As shown in figure 4, after the completion of the index construct, object therein can be retrieved, retrieving includes as follows
Step：
1) a query object P is inputted, to each not visited n omicronn-leaf child node, utilizes smallest enclosing circle algorithm
The ranking value for the smallest enclosing circle that each father node is formed with its child node is calculated with Euclidean distance algorithm, and by them according to liter
Sequence or descending sort, it is stored in movable branch list；
2) according to prune rule, movable branch list is trimmed, deletes unnecessary branch；
The prune rule is specific as follows：
MINDIST refers to that object refers to object to smallest enclosing circle to the minimum range of smallest enclosing circle, MINMAXDIST
Ultimate range, if the MINDIST of node R smallest enclosing circle is than the MINMAXDIST of the smallest enclosing circle of other nodes
Greatly, then node R is wiped out；Conversely, then node R retains, and to node R smallest enclosing circle, query object P to an object O
Distance be more than query object P to node R ultimate range MINMAXDIST (P, R), then object O wiped out；Conversely, then object
O retains, and to node R smallest enclosing circle, if query object P to node R minimum range MINDIST (P, R) is more than inquiry
Object P to object O distance, then node R wiped out.
3) to the above-mentioned steps 2 of each smallest enclosing circle repeated retrieval process in movable branch list), until activity point
Branch list is sky；
4) for leafy node, wherein each object is calculated to query object P distance using Euclidean distance algorithm, is arrived
The minimum object of query object P distance is to retrieve object.
As shown in figure 5, deleting the retrieval object obtained by retrieval, specific deletion process comprises the following steps：
1) by above-mentioned retrieving, the leafy node where object to be deleted is determined；
2) object in leafy node is deleted；
3) if number of objects is less than m in leafy node, this leafy node is deleted, all objects in node are carried out above-mentioned
Heavy insertion operation；Conversely, any operation is not done.
It is pointed out that the M is set according to the number of object in index, the number of object is more, and M setting values are got over
Greatly, the number of object is fewer, and for M setting values with regard to smaller, M is integer；The m is sized according to M, m≤(M+1)/3, and m is
Integer；The K is sized according to the needs of clustering precision, and clustering required precision is higher, then the value of K settings is got over
Small, clustering required precision is lower, then the value of K settings is bigger.
In summary, the present invention is constantly inserted into the child node of tree to establish rope using face characteristic data as object
Draw, when reaching maximum M first for number of objects in child node, weight insertion operation is carried out to this child node, makes this child node
Middle number of objects is less than maximum M, with the storage of new insertion object, when number of objects reaches maximum again in child node,
To this child node progress splitting operation, the object in child node is stored in its next straton node, so circulation, all objects
It is located only within leafy node, the object of others' face characteristic of millions can be deposited, while cause in leafy node
Number of objects is constantly in more than the m values of setting, is reduced the quantity of leafy node, is improved recall precision.
Claims (7)
1. a kind of construction method of the high dimensional indexing of face feature vector, it is characterised in that comprise the following steps：
S1, choose a face characteristic and be stored in as an object in root node, the maximum capacity of root node is M；
S2, a face characteristic newly the inserting object new as one is stored in root node, if root node is discontented with,
Then it is inserted directly into；If number of objects has reached maximum M in root node, one dividing into three operation is carried out to root node, will
Root node is split into three child nodes, i.e. child node A, child node B and child node C, by M object in root node and new insertion
An object be stored in respectively accordingly in three child nodes, and the storage number of objects in three child nodes is no less than m,
The maximum capacity of three child nodes is M；
S3, a face characteristic newly the inserting object new as one are stored in a child node at random, and setting should
Child node is A, if child node A is discontented with, is inserted directly into；If the number of objects in child node A has reached maximum M,
Weight insertion operation is carried out to M object in child node A and the object newly inserted so that the number of objects in child node A
Less than M；
S4, a face characteristic newly the inserting object new as one continues to be stored in child node A, until child node
In number of objects reach maximum M again, line splitting now is entered to M object in child node and the object newly inserted
Operation, obtain three child node of next layer after child node A divisions；Similarly, until child node B and child node C complete to divide
Operation, i.e., obtain three child node of next layer after child node B and child node C divisions；
S5, repeat step S3~S4, storage of the face characteristic data newly inserted as object, all face characteristics
Storage completes the foundation of index after terminating.
A kind of 2. construction method of the high dimensional indexing of face feature vector according to claim 1, it is characterised in that step
Heavy insertion operation described in S3 comprises the following steps：
S11, using smallest enclosing circle algorithm and Euclidean distance algorithm, calculate M object in child node A and one newly inserted
Distance D of the object to child node A smallest enclosing circle center；
S12, M object in child node A and the object newly inserted are sorted from big to small by D values, (M+1)/2 before taking-up
Or M/2 object be as unit is inserted again, it is inserted into same layer other child nodes.
A kind of 3. construction method of the high dimensional indexing of face feature vector according to claim 2, it is characterised in that step
Splitting operation described in S4 comprises the following steps：
S21, three objects are arbitrarily selected from M object in child node A and the object newly inserted, respectively as three
The center of initial clustering；
S22, the average V of the object in each initial clustering, the i.e. center to each initial clustering are calculated, is calculated in addition to this center
All objects and this center average, average V calculation formula is as follows:
<mrow>
<mi>V</mi>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mn>3</mn>
</munderover>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>x</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msub>
<mi>S</mi>
<mi>i</mi>
</msub>
</mrow>
</munder>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>j</mi>
</msub>
<mo>-</mo>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
Wherein, SiRepresent the set of the object of all initial cluster centers, uiRepresent the center of ith cluster, xjRepresent child node
In M object in A and the object newly inserted, except this center uiOutside remaining all object；
S23, according to average V, by the distance of the Euclidean distance algorithm each object of calculating and three initial cluster centers, by object
Cluster centre of that the minimum initial cluster center as this object, is drawn again with the distance of three initial cluster centers
Three clusters divided；
S24, repeat step S22~S23, until difference between the average of three initial cluster centers and each a upper average
Absolute value is less than specified threshold K；
S25, three clusters finally given, as three child node of next layer after division.
4. the construction method of the high dimensional indexing of a kind of face feature vector according to claim 3, it is characterised in that described
After the completion of index construct, object therein can be retrieved；Retrieving comprises the following steps：
S31, input a query object P, to each not visited n omicronn-leaf child node, using smallest enclosing circle algorithm and
Euclidean distance algorithm calculates the ranking value for the smallest enclosing circle that each father node is formed with its child node, and by them according to ascending order
Or descending sort, it is stored in movable branch list；
S32, according to prune rule, movable branch list is trimmed, deletes unnecessary branch；
S33, to each smallest enclosing circle repeat step S32 in movable branch list, until movable branch list is sky；
S34, for leafy node, wherein each object is calculated to query object P distance using Euclidean distance algorithm, to looking into
Ask the minimum object as retrieval object of object P distance.
5. the construction method of the high dimensional indexing of a kind of face feature vector according to claim 4, it is characterised in that can be right
The object obtained by retrieval is deleted, and specific deletion process comprises the following steps：
S41, by step S31~S34, determine the leafy node where object to be deleted；
S42, the object in leafy node is deleted；
S43, if number of objects is less than m in leafy node, this leafy node is deleted, all objects in node are subjected to step
Heavy insertion operation described in S11~S12；Conversely, any operation is not done.
6. a kind of construction method of the high dimensional indexing of face feature vector according to Claims 2 or 3 or 5, its feature exist
In：The M is set according to the number of object in index, and the number of object is more, and M setting values are bigger, and the number of object is fewer,
For M setting values with regard to smaller, M is integer；The m is sized according to M, m≤(M+1)/3, and m is integer；The K is according to cluster
The needs of dividing precision are sized, and clustering required precision is higher, then for the value of K settings with regard to smaller, clustering precision will
Ask lower, then the value of K settings is bigger.
7. the construction method of the high dimensional indexing of a kind of face feature vector according to claim 4, it is characterised in that described
Prune rule is specific as follows：
MINDIST refers to that object refers to object to smallest enclosing circle most to the minimum range of smallest enclosing circle, MINMAXDIST
Big distance, if the MINDIST of node R smallest enclosing circle is bigger than the MINMAXDIST of the smallest enclosing circle of other nodes,
Then node R is wiped out；Conversely, then node R retains, and to node R smallest enclosing circle, query object P to object O away from
From the ultimate range MINMAXDIST (P, R) more than query object P to node R, then object O is wiped out；Conversely, then object O is protected
Stay, and to node R smallest enclosing circle, if query object P to node R minimum range MINDIST (P, R) is more than query object
P to object O distance, then node R wiped out.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710525541.XA CN107391601A (en) | 2017-06-30 | 2017-06-30 | A kind of construction method of the high dimensional indexing of face feature vector |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710525541.XA CN107391601A (en) | 2017-06-30 | 2017-06-30 | A kind of construction method of the high dimensional indexing of face feature vector |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107391601A true CN107391601A (en) | 2017-11-24 |
Family
ID=60334922
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710525541.XA Pending CN107391601A (en) | 2017-06-30 | 2017-06-30 | A kind of construction method of the high dimensional indexing of face feature vector |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107391601A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689964A (en) * | 2019-09-12 | 2020-01-14 | 银江股份有限公司 | Health data sample searching method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6389424B1 (en) * | 1998-10-28 | 2002-05-14 | Electronics And Telecommunications Research Institute | Insertion method in a high-dimensional index structure for content-based image retrieval |
| CN102831241A (en) * | 2012-09-11 | 2012-12-19 | 山东理工大学 | Dynamic index multi-target self-adaptive construction method for product reverse engineering data |
| CN105868355A (en) * | 2016-03-29 | 2016-08-17 | 贵州大学 | Large-scale multimedia data spatial index method |
-
2017
- 2017-06-30 CN CN201710525541.XA patent/CN107391601A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6389424B1 (en) * | 1998-10-28 | 2002-05-14 | Electronics And Telecommunications Research Institute | Insertion method in a high-dimensional index structure for content-based image retrieval |
| CN102831241A (en) * | 2012-09-11 | 2012-12-19 | 山东理工大学 | Dynamic index multi-target self-adaptive construction method for product reverse engineering data |
| CN105868355A (en) * | 2016-03-29 | 2016-08-17 | 贵州大学 | Large-scale multimedia data spatial index method |
Non-Patent Citations (1)
| Title |
|---|
| 韩冬柏: "基于R_树的最近邻查询研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689964A (en) * | 2019-09-12 | 2020-01-14 | 银江股份有限公司 | Health data sample searching method and system |
| CN110689964B (en) * | 2019-09-12 | 2022-08-26 | 银江技术股份有限公司 | Health data sample searching method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111639237B (en) | Electric power communication network risk assessment system based on clustering and association rule mining | |
| CN108920720B (en) | Large-scale image retrieval method based on depth hash and GPU acceleration | |
| CN102662974B (en) | A network graph index method based on adjacent node trees | |
| CN108052514B (en) | Mixed space indexing method for processing geographic text Skyline query | |
| CN104268280B (en) | A kind of Hierarchical storage and querying method based on key value database | |
| CN110334391A (en) | A kind of various dimensions constraint wind power plant collection electric line automatic planning | |
| CN102306176B (en) | On-line analytical processing (OLAP) keyword query method based on intrinsic characteristic of data warehouse | |
| CN102110171B (en) | Method for inquiring and updating Bloom filter based on tree structure | |
| CN110070121A (en) | A kind of quick approximate k nearest neighbor method based on tree strategy with balance K mean cluster | |
| CN103020321B (en) | Neighbor search method and system | |
| CN106503223A (en) | A kind of binding site and the online source of houses searching method and device of key word information | |
| CN104346444B (en) | A kind of the best site selection method based on the anti-spatial key inquiry of road network | |
| CN108446357A (en) | A kind of mass data spatial dimension querying method based on two-dimentional geographical location | |
| CN104615734B (en) | A kind of community management service big data processing system and its processing method | |
| CN108009265B (en) | Spatial data indexing method in cloud computing environment | |
| CN107273471A (en) | A kind of binary electric power time series data index structuring method based on Geohash | |
| CN105550368A (en) | Approximate nearest neighbor searching method and system of high dimensional data | |
| CN104504251B (en) | A kind of community division method based on PageRank algorithms | |
| CN108241713A (en) | A kind of inverted index search method based on polynary cutting | |
| CN103377237A (en) | High dimensional data neighbor search method and fast approximate image search method | |
| CN112214485A (en) | Power grid resource data organization planning method based on global subdivision grid | |
| CN107391601A (en) | A kind of construction method of the high dimensional indexing of face feature vector | |
| CN105025013B (en) | The method for building up of dynamic IP Matching Model based on priority Trie trees | |
| CN103164487B (en) | A kind of data clustering method based on density and geological information | |
| CN106951519A (en) | Quick track index update method based on mesh generation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171124 |