CN105740451A - Reference test based multimedia indexing method - Google Patents
Reference test based multimedia indexing method Download PDFInfo
- Publication number
- CN105740451A CN105740451A CN201610076525.2A CN201610076525A CN105740451A CN 105740451 A CN105740451 A CN 105740451A CN 201610076525 A CN201610076525 A CN 201610076525A CN 105740451 A CN105740451 A CN 105740451A
- Authority
- CN
- China
- Prior art keywords
- index
- data
- value
- performance
- 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.)
- 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
- G06F16/40—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)
- Software Systems (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a reference test based multimedia indexing method. The method comprises the following steps of A, selecting test parameters, wherein the test parameters are a k value and a page size; B, calling an indexing dataset and selecting a data volume and a data dimension of the dataset; C, loading an indexing algorithm and executing a retrieval process; and D, collecting performance data returned in the retrieval process to obtain a performance condition of the indexing algorithm, and performing sorting, output and display. According to the method, different indexes can be created according to combinations of a large amount of different parameters and an index with the best performance is finally adopted for quickly querying a multimedia database with large data volume.
Description
Technical field
The present invention relates to a kind of multimedia index method, particularly a kind of multimedia index method based on benchmark test.
Background technology
Benchmark test is a kind of method weighing objective target systematic function to be measured, and the content of benchmark test is to come software or the hardware resource of quantitative evaluation object-computer with procedure set, and it obtains, by running sequence of operations or program, the performance treating survey technology or software.
Along with the development of high-speed transfer and high efficiency storage technologies, bulk information exists with computer-readable form, wherein not only comprises word and sound, more importantly the multimedia video visual information such as figure, image and video.In the face of the increasing multimedia messages that may have access to, in order to not get lost in wherein, people often propose the demand of quick search, are called similarity retrieval, namely find out some the most similar to given query object or an object from data base.
In the last few years, substantial amounts of high dimension data index structure is proposed in succession, is mostly wherein tree index structure, such as R-Tree, R+-Tree, R*-Tree, k-d-b-Tree, SR-Tree, SS-Tree, A-tree, X-Tree etc..Except tree structure, VA-file etc. have employed the method the quantifying compression disk access cost to reduce in query script, but its approximate file is only the simple arrangement of Proximity Vector, its query performance accelerates to be limited in only small scope, is still difficult to meet the needs of practical application.
In existing multimedia index method, when data volume googol is inquired about according to storehouse, its inquiry velocity is slow, it is impossible to meet user's request.
Summary of the invention
It is an object of the invention to, it is provided that a kind of multimedia index method based on benchmark test.The multimedia database that data volume is big can be carried out quick search by the method.
For solving above-mentioned technical problem, technical scheme provided by the invention is as follows: based on the multimedia index method of benchmark test: comprise the following steps:
A, choosing test parameter, described test parameter is k value and page size；
B, call the data set for indexing, choose data volume and the data dimension of data set；
C, loading Index Algorithm, and perform retrieving；
The performance data returned in D, collection retrieving, obtains behavior pattern and the sequence of Index Algorithm output display.
The aforesaid multimedia index method based on benchmark test, in step C, described execution retrieving is, the different combinations of values according to k value, page size and data set, loads Index Algorithm and retrieves.
The aforesaid multimedia index method based on benchmark test, described k value includes 20,50 or 100.
The aforesaid multimedia index method based on benchmark test, described page size includes 32KB, 64KB or 128KB.
The aforesaid multimedia index method based on benchmark test, described data volume includes 200,000,400,000,800,000 or 1,000,000.
The aforesaid multimedia index method based on benchmark test, described data dimension includes 128 peacekeepings 256 to be tieed up.
The aforesaid multimedia index method based on benchmark test, described Index Algorithm includes iDistance, SR-tree, A-tree, pyramid and LSHkit.
The aforesaid multimedia index method based on benchmark test, described performance data includes response time and I/O expense.
Compared with prior art, the application can select the combination of the different values of multiple test parameter (such as, different test data set sizes, different data dimension values, different index pages size values, different k values), create index automatically.Then, for the above-mentioned index automatically created, carry out repeatedly retrieval performance assessment with different inquiries automatically.And according to retrieval performance result, that index that final selected performance is best, as the index that suggestion user uses.Such as, under 256 dimension 1,000,000 data, page-size can set that as 32KB, it is also possible to is set as 64KB.Testing through automatic capability retrieval, it has been found that when parameter is " page-size=64KB ", performance is best, therefore, will allow this index of user's final utilization, and all the other indexes are deleted, or clearly inform the user that, index behavior pattern that all the other automatically derive and sequence.
Five kinds of different types of High-dimensional Index Technology such as iDistance, SR-tree, A-tree, Pyramid, LSHkit are carried out performance test by the present invention by applicant on truthful data collection, and analysis of experimental data is as follows:
Experimental situation
Testing the data adopted herein is Microsoft MSRA-MM2.0 real image data collection.Data volume chooses 200,000,400,000,800,000,1,000,000 four kind, dimension chooses 128D and 256D two kinds, and all data are picture characteristic of correspondence vector.
All experiments herein are all carry out on the server of following configuration, are configured that processor AMD6core3.3GHz, internal memory 8G, hard disk 500G, and operating system is ubuntu12.04-64 bit manipulation system.In this environmental experiment, by relative analysis in different pieces of information amount, different value of K, not under same page size (pagesize) parameter, the knn retrieval performance performance of five kinds of variety classes Index Algorithm of iDistance, SR-tree, A-tree, Pyramid, LSHkit, in order to strengthen the credibility of test result, the Comparative result of response time adds the linear scanning result of three threads.K value refers to, in kNN retrieval mode, it is necessary to finding k the data similar to destination object, k value determines the selection rate in inquiry.
In an experiment, being chosen the reference point of iDistance by K-Means clustering algorithm, quantity is set to 100, to ensure that this index has good performance；LSHkit selects MPLSH index technology therein；For A-tree, 6 bytes are used to carry out approximate representation every dimension, although certain retrieval precision can be lost, but the compression that can be greatly improved data stores degree, thus improving recall precision further.For the data of 128D and 256D, acquiescence uses the page size (pagesize) of 32KB and 64KB.
Testing scheme and interpretation of result
Difference according to test parameter, individually below enumerated data amount size variation, the change of K value, in three kinds of situations of page size variation, the performance of index performance, result is analyzed by the principle that realizes in conjunction with index.
The data volume impact on query performance
Changing in the experiment to performance impact in examination data volume, choosing K value is 100, and page is dimensioned to default size, tests respectively on 128 peacekeeping 256 dimension data collection, and data volume excursion is 20 ten thousand to 100 ten thousand.
On 256 dimension data collection, test result is as Figure 1-Figure 2.
Fig. 1 illustrates under 256 dimension data collection, and each index technology is along with the situation of change of the response time increasing retrieval of data volume, and Fig. 2 illustrates the growth along with data volume, the situation of change of each indexed I/O expense.From figure 1 it appears that the linear scanning being better than three threads retrieval time of each index, but along with the growth of data volume, the response time of each index all increases gradually.Wherein the time efficiency of iDistance is better than other index efficiencies, and A-tree time efficiency in higher-dimension situation is worst.As can be seen from Figure 2 along with the growth of data volume, I/O number of each index is also gradually increased, and wherein, the growth of A-tree is the slowest, and I/O expense is also always up minimum.The I/O expense of iDistance is the highest on the contrary, particularly in data volume more than 400,000 after.
In order to study the situation of change of performance further, comparing index performance situation under 128 dimension data collection, experimental result is as Figure 3-Figure 4.
Compared with the test result under 256 dimension data collection, each index performance situation is substantially the same, and in response time, each index is superior to three thread linear scannings, and the response time gap of each index is little.SR-tree, relative to other index technology, has achieved better performance.In I/O expense, A-tree still achieves best performance, although still the performance of iDistance is worst, but the ratio differed with other index technologies reduces a lot.
Fig. 5-Fig. 6 be each index technology 1,000,000 data set, under the test environment of K value 100, the contrast situations of 128 dimensions and 256 dimension response times, the block diagram on the left side represents 128 dimensions, and the right is 256 dimensions.
By the observation of the performance test results to index each under two kinds of dimension data collection, the experimental principle in conjunction with each index is analyzed, existing sum up following some:
1. although under all test conditions, respectively index on response time, be superior to linear scanning, this is because in experiment, k value is 100 to the maximum, compared to 20～data of 1,000,000, selection rate is very low.Therefore, " dimension disaster " remains existence.And in higher-dimension situation, the response time of various indexes and I/O number are obviously higher than low-dimensional situation.
2. under 256 dimension data collection, iDistance is better than other indexes on response time, and this elects the center of clustering of data as reference point in advance mainly due to iDistance, and in the process of inquiry, seeking scope is more accurate and effective.
3., under 128 dimension data collection, the response time of SR-tree is optimum.This is owing to SR-tree adopts the minimum boundary rectangle (MBR) common portion with minimum border circular (MBS) as node region, in low-dimensional situation, data area overlap is less, decreases multichannel inquiry, therefore, achieve and inquire about performance preferably.
4. in I/O expense, A-tree is always up minimum, and this have employed the technology of compressive features vector mainly due to it.A-tree belongs to based on approximate High-dimensional Index Technology, uses bit code to carry out approximate representation in each dimension, and after such approximate compression, the data of each page of storage are just more, have bigger fan-out.But meanwhile, due to approximate compression, when fetching data, A-tree has a decoding process, adds computing cost, and therefore on response time, A-tree does not preponderate.
5. contrary with A-tree, iDistance I/O number in several indexes is the highest.Particularly under High Dimensional Data Set, along with the increase of data volume, I/O time number sharply increases.This is mainly due to being progressively expand search radius to obtain retrieval result in the KNN process of iDistance, and in high dimensional data, more big radius can cause more overlapping region, can access more data, therefore causes that I/O expense is maximum.
The k value impact on query performance
Changing in the experiment to performance impact at examination k value, choosing data volume is 1,000,000, and page is dimensioned to default size, tests respectively on 128 peacekeeping 256 dimension data collection, and the value change of K value is 20,50,100.
On 256 dimension data collection, test result is as shown in Figs. 7-8.
On 128 dimension data collection, test result is such as shown in Fig. 9-Figure 10.
From the test result of this part it can be seen that no matter under 256 dimensions or 128 dimension data environment, along with the increase of K value, response time and the I/O expense of each index all have small size change, but rate of change is all relatively low.
Wherein, the I/O expense of SR-tree is slightly larger relative to the rate of change of other index technologies.
By the analysis to this partial test result, following 2 points of existing summary:
1.K value is maximum elects 100 as, the data volume relative to 1,000,000, and selection rate is still very low.Therefore, property indices all changes not quite.
The rate of change of 2.SR-tree is slightly larger to be likely due to when the change of K value is big, its mindist and minmaxdist numerical value can become big, the node got rid of when carrying out beta pruning is less, it is necessary to carries out the inquiry of more multichannel, thus causes that response time and I/O expense are more.
Page size (pagesize) impact on query performance
In the examination page size variation experiment to performance impact, choosing data volume is 1,000,000, and K value is 100, tests on 128 dimension data collection, and the value change of page sizes values is 32KB, 64KB, 128KB.Its test result is such as shown in Figure 11-Figure 12.
In figure 12 it can be seen that along with the increase of pagesize, I/O number of various Index Algorithm is obviously reduced.This is due to bigger pagesize, can make to store in each page more subdata.Generally similar data can be stored in same one page or contiguous page, and therefore, when searching the data of as much, the page of access will reduce thus reducing I/O expense.
But in the test result of Figure 11, along with the increase of pagesize, the response time change of each index is different.The response time change of SR-tree, LSHkit is little, and the response time of A-tree then becomes longer, iDistance, pyramidal slightly reduces on the contrary.The reason causing this phenomenon is likely to have two:
1. the utilization rate that bigger pagesize stores in identical internal memory may decline, and so causes that in internal memory, the data of storage reduce, and increase response time.
2., in retrieving, computing cost is generally much less than I/O expense, but when computing cost rises to certain degree, computing cost also cannot be ignored.Along with the increase of pagesize, I/O time number reduces, but computing cost is likely to higher or reduces, and causes the change of response time.As, in the retrieving of A-tree, it is desirable to have the process of a decoding, cause the prolongation of response time because of the increase of computing cost.
By MIB-RI, multimedia index technology five kinds conventional is carried out performance test at Microsoft MSRA-MM2.0 real image data collection, be based on the index SR-tree of division respectively, based on the approximate index LSHkit indexing A-tree, the index iDistance based on distance, the index pyramid based on dimensionality reduction and dimension transformation and position-based sensitive hash of vector.
By experiment, it has been found that under different experimental conditions, five kinds of indexes performance in several two performance indications of response time and I/O time.Experiment have studied data volume, k value and the impact on index technology performance of the tri-kinds of parameters of pagesize, observe various index performance change under different dimensions simultaneously.In three kinds of comparisons indexing response time, add linear scanning and contrast.Finally, the principle performance to various indexes that realizes in conjunction with index carries out interpretation of result, selects the Index Algorithm of optimum.
Accompanying drawing explanation
Fig. 1 is under 256 dimension data collection, and each index technology is along with the variation diagram of the response time increasing retrieval of data volume；Wherein, abscissa is data volume, and vertical coordinate is response time；
Fig. 2 is under 256 dimension data collection, and each index technology is along with the variation diagram of the I/O expense increasing retrieval of data volume；Wherein, abscissa is data volume, and vertical coordinate is I/O expense；
Fig. 3 is under 128 dimension data collection, and each index technology is along with the variation diagram of the response time increasing retrieval of data volume；Wherein, abscissa is data volume, and vertical coordinate is response time；
Fig. 4 is under 128 dimension data collection, and each index technology is along with the variation diagram of the I/O expense increasing retrieval of data volume；Wherein, abscissa is data volume, and vertical coordinate is I/O expense；
Fig. 5 be each index technology 1,000,000 data set, under the test environment of K value 100, the response times contrast of 128 dimensions and 256 dimensions；Wherein, abscissa is data volume, and vertical coordinate is response time；
Fig. 6 be each index technology 1,000,000 data set, under the test environment of K value 100, the I/O expenses contrast of 128 dimensions and 256 dimensions；Wherein, abscissa is data set amount, and vertical coordinate is I/O expense；
Fig. 7 is under 256 dimension data collection, and each index technology is along with the variation diagram of the response time increasing retrieval of k value；Wherein, abscissa is k value, and vertical coordinate is response time；
Fig. 8 is under 256 dimension data collection, and each index technology is along with the variation diagram of the I/O expense increasing retrieval of k value；Wherein, abscissa is k value, and vertical coordinate is I/O expense；
Fig. 9 is under 128 dimension data collection, and each index technology is along with the variation diagram of the response time increasing retrieval of k value；Wherein, abscissa is k value, and vertical coordinate is response time；
Figure 10 is under 128 dimension data collection, and each index technology is along with the variation diagram of the I/O expense increasing retrieval of k value；Wherein, abscissa is k value, and vertical coordinate is I/O expense；
Figure 11 is under 128 dimension data collection, and each index technology is along with the variation diagram of the response time increasing retrieval of page size；Wherein, abscissa is page sizes values, and vertical coordinate is response time；
Figure 12 is under 128 dimension data collection, and each index technology is along with the variation diagram of the I/O expense increasing retrieval of page size；Wherein, abscissa is k value, and vertical coordinate is I/O expense.
Detailed description of the invention
Embodiment.Multimedia index method based on benchmark test: comprise the following steps:
A, choosing test parameter, described test parameter is k value and page size；
B, call the data set for indexing, choose data volume and the data dimension of data set；
C, loading Index Algorithm, and perform retrieving；Described Index Algorithm includes iDistance, SR-tree, A-tree, pyramid and LSHkit.
The performance data returned in D, collection retrieving, obtains behavior pattern and the sequence of Index Algorithm output display.Described performance data includes response time and I/O expense.By the contrast of performance data, the Index Algorithm of optimum can be selected.
In step C, described execution retrieving is, the different combinations of values according to k value, page size and data set, loads Index Algorithm and retrieves.
Described k value includes 20,50 or 100, it is also possible to be set to other numerical value.
Page size includes 32KB, 64KB or 128KB, it is also possible to be set to other numerical value.
Described data volume includes 200,000,400,000,800,000 or 1,000,000, it is also possible to be set to other numerical value.
Described data dimension includes 128 peacekeepings 256 to be tieed up.
When performing retrieval, it is possible to be combined according to the parameter of test parameter and data set, to assess the performance of Index Algorithm.Such as: k value can be 20,50,100 etc.；Page size can value be 32KB, 64KB, 128KB；Data set size can value be 200,000,400,000,800,000,1,000,000 etc..So, the combination of the retrieval for carrying out Performance Evaluation can be:
Page-size=32KB, when data set is sized to 200,000, performs the retrieval that k value is the values such as 20,50,100；
Page-size=64KB, when data set is sized to 400,000, performs the retrieval that k value is the values such as 20,50,100；
………
Test parameter be disposed to test indices performance under various circumstances, by these performance, come Analytical Index technology can obtain under which kind of parameter optimum search efficiency.In the design of MIB, choose data volume, k value and page size etc. as test parameter, next these several test parameters are carried out a detailed description.
1. data volume
A major challenge of High-dimensional Index Technology research is exactly growing data volume, study the treatment technology that the most popular problem is exactly big data at present, therefore, judge whether certain index technology is applicable to the trend of future development, need examination index when data volume changes, if to ensure that search efficiency is still maintained at higher state.
Therefore, in the design of MIB, the change of data volume is a very important test environment factor.
2.k value
In kNN retrieval mode, k value represents and requires to look up how many data similar to destination object.In other words, k value determines when the selection rate in time inquiry.The size of selection rate determines in retrieving, and index needs the size of the data space accessed.For some index technology, kNN searching algorithm is that the radius by being stepped up search is until finding enough destination objects.Such as in iDistance index technology, if not finding enough data points in initial search radius, then search radius increases the △ r more data point of searching successively.In this process, the overlap of different reference point search radius can be caused, and then increase more calculating and I/O number, eventually affect the efficiency of whole inquiry.
Because above-mentioned reason, in the design of MIB, using k value as a test parameter with the test indices performance change when lookup rate changes.
3. page size
Page size (Pagesize) belongs to index inner parameter, and it is how many that page size determines the data that can store in every one page.More big page can store more data, reduces the number of times accessing disk, and when I/O expense reduces, usual effectiveness of retrieval can increase.But in actual retrieving, being not that the page is the bigger the better, page size also can affect the utilization rate of internal memory, because the bigger page easily produces more page breakage.
Accordingly, it would be desirable to observe in testing under certain test environment, arranging great page-size, index can reach good recall precision.
From the foregoing, it will be observed that existing index technology, for same index technology, such as iDistance, SR-tree, A-tree, pyramid and LSHkit etc., when creating these indexes, it is necessary to technical staff is based on the experience of oneself, to (different data volume under various application scenarios, different data dimensions, different selection rates or k value), artificial specifies various parameters (such as, page-size required when creating index, compression ratio, page idleness etc.).If selected suitable of parameter when creating index, then index performance can relatively good, otherwise, and inappropriate indexing parameter or cause indexing poor-performing.Technical scheme, it it is the combination according to substantial amounts of different parameters, create different indexes, and the final index that index finally adopting performance best uses as user, thus avoiding in existing index technology, the Performance comparision of the index that user creates depends on the subjective situation about selecting of technical staff.
Claims (8)
1. based on the multimedia index method of benchmark test, it is characterised in that: comprise the following steps:
A, choosing test parameter, described test parameter is k value and page size；
B, call the data set for indexing, choose data volume and the data dimension of data set；
C, loading Index Algorithm, and perform retrieving；
The performance data returned in D, collection retrieving, obtains behavior pattern and the sequence of Index Algorithm output display.
2. the multimedia index method based on benchmark test according to claim 1, it is characterised in that: in step C, described execution retrieving is, the different combinations of values according to k value, page size and data set, loads Index Algorithm and retrieves.
3. the multimedia index method based on benchmark test according to claim 1, it is characterised in that: described k value includes 20,50 or 100.
4. the multimedia index method based on benchmark test according to claim 1, it is characterised in that: described page size includes 32KB, 64KB or 128KB.
5. the multimedia index method based on benchmark test according to claim 1, it is characterised in that: described data volume includes 200,000,400,000,800,000 or 1,000,000.
6. the multimedia index method based on benchmark test according to claim 1, it is characterised in that: described data dimension includes 128 peacekeepings 256 to be tieed up.
7. the multimedia index method based on benchmark test according to claim 1, it is characterised in that: described Index Algorithm includes iDistance, SR-tree, A-tree, pyramid and LSHkit.
8. the multimedia index method based on benchmark test according to claim 1, it is characterised in that: described performance data includes response time and I/O expense.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610076525.2A CN105740451A (en) | 2016-02-03 | 2016-02-03 | Reference test based multimedia indexing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610076525.2A CN105740451A (en) | 2016-02-03 | 2016-02-03 | Reference test based multimedia indexing method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105740451A true CN105740451A (en) | 2016-07-06 |
Family
ID=56245825
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610076525.2A Pending CN105740451A (en) | 2016-02-03 | 2016-02-03 | Reference test based multimedia indexing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105740451A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112835961A (en) * | 2021-03-01 | 2021-05-25 | 国家机床质量监督检验中心 | Method and system for quickly aligning periodically acquired data |
| CN117093611A (en) * | 2023-10-16 | 2023-11-21 | 北京人大金仓信息技术股份有限公司 | Database combined index suggestion processing method, storage medium and computer device |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559134A (en) * | 2013-11-11 | 2014-02-05 | 曙光信息产业(北京)有限公司 | Configuration method and device for benchmark test component |
-
2016
- 2016-02-03 CN CN201610076525.2A patent/CN105740451A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559134A (en) * | 2013-11-11 | 2014-02-05 | 曙光信息产业(北京)有限公司 | Configuration method and device for benchmark test component |
Non-Patent Citations (2)
| Title |
|---|
| JICHAO LIU，MEI CHEN等: "Benchmarking high-dimensional index structures for multimedia data retrieval", 《WIT TRANSACTIONS ON ENGINEERING SCIENCES- 2013 INTERNATIONAL CONFERENCE ON FUTURE COMPUTER AND INFORMATION TECHNOLOGY》 * |
| NG COLOSSI等: "Benchmarking access structures for high-dimensional multimedia data", 《PROC. OF THE IEEE INTL. CONF. ON MULTIMEDIA AND EXPO(ICME2000)》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112835961A (en) * | 2021-03-01 | 2021-05-25 | 国家机床质量监督检验中心 | Method and system for quickly aligning periodically acquired data |
| CN117093611A (en) * | 2023-10-16 | 2023-11-21 | 北京人大金仓信息技术股份有限公司 | Database combined index suggestion processing method, storage medium and computer device |
| CN117093611B (en) * | 2023-10-16 | 2024-03-19 | 北京人大金仓信息技术股份有限公司 | Database combined index suggestion processing method, storage medium and computer device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Keyword search in spatial databases: Towards searching by document | |
| US8625907B2 (en) | Image clustering | |
| US9514187B2 (en) | Techniques for using zone map information for post index access pruning | |
| Wang et al. | A data-adaptive and dynamic segmentation index for whole matching on time series | |
| US20040148293A1 (en) | Method, system, and program for managing database operations with respect to a database table | |
| US9141666B2 (en) | Incremental maintenance of range-partitioned statistics for query optimization | |
| Bartolini et al. | Query processing issues in region-based image databases | |
| US9104726B2 (en) | Columnar databases | |
| Guo et al. | A new ETL approach based on data virtualization | |
| CN105740451A (en) | Reference test based multimedia indexing method | |
| Günnemann et al. | Subspace clustering for indexing high dimensional data: a main memory index based on local reductions and individual multi-representations | |
| JP2010277329A (en) | Neighborhood retrieval device | |
| US10394870B2 (en) | Search method | |
| Bertella et al. | A systematic review of spatial approximations in spatial database systems | |
| Bertella et al. | The application of spatial approximations to spatial query processing: A systematic review of literature | |
| CN115168390A (en) | Multi-dimensional asset retrieval analysis method, system, terminal and storage medium | |
| Terry et al. | Indexing method for multidimensional vector data | |
| Li et al. | A Survey of Multi-Dimensional Indexes: Past and Future Trends | |
| Moënne-Loccoz | High-dimensional access methods for efficient similarity queries | |
| Li et al. | A locality-aware similar information searching scheme | |
| Georgiadis et al. | Skyline-based dissimilarity of images | |
| Aung et al. | Index structure for nearest neighbors search with required keywords on spatial database | |
| Feng et al. | Effective optimizations of cluster-based nearest neighbor search in high-dimensional space | |
| Winarko et al. | A signature-based indexing method for efficient content-based retrieval of relative temporal patterns | |
| US11874868B2 (en) | Generating and presenting multi-dimensional representations for complex entities |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160706 |