CN107145519B - Image retrieval and annotation method based on hypergraph - Google Patents
Image retrieval and annotation method based on hypergraph Download PDFInfo
- Publication number
- CN107145519B CN107145519B CN201710229126.XA CN201710229126A CN107145519B CN 107145519 B CN107145519 B CN 107145519B CN 201710229126 A CN201710229126 A CN 201710229126A CN 107145519 B CN107145519 B CN 107145519B
- Authority
- CN
- China
- Prior art keywords
- node
- hypergraph
- query
- image
- lower bound
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 230000007704 transition Effects 0.000 claims abstract description 38
- 239000011159 matrix material Substances 0.000 claims abstract description 33
- 238000005516 engineering process Methods 0.000 claims abstract description 15
- 230000000007 visual effect Effects 0.000 claims abstract description 11
- 239000000872 buffer Substances 0.000 claims abstract description 8
- 238000001914 filtration Methods 0.000 claims description 10
- 238000005295 random walk Methods 0.000 claims description 6
- 238000012966 insertion method Methods 0.000 claims description 4
- 238000013016 damping Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 abstract description 2
- 238000012216 screening Methods 0.000 abstract 1
- 230000003139 buffering effect Effects 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000003780 insertion Methods 0.000 description 2
- 230000037431 insertion Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/901—Indexing; Data structures therefor; Storage structures
- G06F16/9024—Graphs; Linked lists
-
- 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
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Library & Information Science (AREA)
- Data Mining & Analysis (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 invention discloses an image retrieval and annotation method based on a hypergraph. According to the method, the comment relation, the annotation relation and the visual similarity relation of the social image are effectively organized by utilizing a hypergraph theory; the computation efficiency of the hypergraph transition probability matrix is improved by using batch, parallel and buffer technologies; by utilizing a node ranking score upper and lower bound estimation and approximation method, the query efficiency is improved; and the query quality is improved by utilizing the user feedback. The method comprises the steps of generating an image t-NN graph according to image visual characteristics, establishing a hypergraph model by combining social association information of the image, calculating a transition probability matrix and storing the transition probability matrix in a B + tree; performing parallel personalized PageRank query, estimating and screening candidate point sets according to upper and lower boundaries, and sorting; and finally, according to the user feedback, inquiring again and optimizing the inquiry result. The invention can effectively organize various multidimensional relations, greatly improves the calculation efficiency and the query efficiency of the transition probability matrix, has better query quality than the prior method and provides the best performance.
Description
Technical Field
The invention relates to a Top-k query technology on a hypergraph, in particular to an image retrieval and annotation method based on the hypergraph.
Background
With the development of social media and mobile internet, social image websites provide a large number of images that are text-tagged by different users. Social images often accompany a variety of information, such as visual features, tags, users, and the like, as well as a variety of behavioral relationships, such as annotations, comments, and the like. The method has wide application in searching and labeling mass social images, and becomes a research hotspot in the fields of databases, data mining and machine learning.
The image retrieval is to search the closest image object according to given information, and has various retrieval types such as similar image retrieval, key image retrieval and the like according to different given information types. The image labeling is to add semantic text information to the appointed image, namely to search the most similar semantic text according to the image. The hypergraph is an extension of a common graph model, wherein the hyperedge can comprise a plurality of nodes, so that the hypergraph can represent a high-dimensional relation and is more suitable for modeling a complex network. The image retrieval and annotation has important application value in the fields of search engines, social media and the like.
Current image retrieval and annotation methods generally manage only a single feature or use a generic graph to model social images. However, a single feature can only represent some relevance and cannot be used for representing real semantic association, and a common graph cannot represent high-dimensional relationships (such as annotation and comment relationships), so that information is lost. Some methods in the field of machine learning use hypergraph models, but solve using complex matrix operations, with huge time and storage overhead and no scalability.
Disclosure of Invention
Aiming at the defects, the invention provides an image retrieval and labeling method based on a hypergraph, which uses the hypergraph to model social images to express a high-dimensional relation, uses batch, parallel and buffering technologies to accelerate hypergraph model construction, uses personalized PageRank to accelerate Top-k query on the hypergraph, and also uses a parallel and approximate method to further accelerate query.
The technical scheme adopted for solving the technical problem comprises the following steps: an image retrieval and annotation method based on hypergraph comprises the following steps:
step (1): establishing a t-NN graph for the image data set by using an image retrieval engine based on content, and establishing a relation between each image and t images with the most similar visual characteristics;
step (2): establishing a hypergraph according to the image t-NN graph and the social association information of the image, calculating a transition probability matrix of the hypergraph, and storing the hypergraph in a B + tree;
and (3): a user submits a query object set and a k value;
and (4): generating a query vector according to a query object set submitted by a user, then carrying out parallel personalized PageRank query on a hypergraph, filtering nodes of the hypergraph by using an upper and lower bound estimation method, and gradually reducing a candidate point set of each iteration in the PageRank query process until k results are obtained;
and (5): and (4) evaluating the k results obtained in the step (4) by the user to generate feedback information, adjusting the query vector in the step (4) according to the feedback information to form a new query vector, repeating the step (4), and finally obtaining a new query result and returning the new query result to the user.
Further, the image retrieval engine in the step (1) extracts visual features of the images according to the MPEG-7 standard based on the metric space SPB tree index structure, establishes the SPB tree, performs t-NN query on each image to obtain t query results most similar to the SPB tree, and further establishes the t-NN graph.
Further, the types of the hypergraph in the step (2) are three:
1) taking each image in the image t-NN image and the t-NN image thereof as a super edge;
2) taking a ternary combination marked by a user on an image as a super edge;
3) and taking a multi-element combination of images evaluated by a user as a super edge.
Further, the specific steps of calculating the hypergraph transition probability matrix in the step (2) are as follows:
a hypergraph is represented by a matrix H of | V | × | E | where the matrix elements H (V, E) are:
wherein v is a node and e is a super edge;
the degree d (v) of node v may be expressed as:
d(v)＝∑e∈E|v∈ew(e)＝∑e∈Ew(e)h(v,e)
wherein w (e) represents the weight of the excess edge e;
the degree δ (e) of the excess edge e can be expressed as:
δ(e)＝|e|＝∑v∈Vh(v,e)
the transition probability matrix of the hypergraph, denoted by P, then the element P [ u, v ] in P represents the transition probability from node u to node v, which can be calculated using the following formula:
further, the transition probability matrix in the step (2) is stored in a B + tree by adopting a batch insertion method, a parallel technology and a buffering technology; when the transition probability matrix is stored in the B + tree, the combination of the out-degree node and the in-degree node of the hypergraph is used as a key, and the transition probability is used as a value.
Further, k in the step (3) is the number of query return nodes specified by the user.
Further, the personalized PageRank in step (4) takes the query vector and the value k as input, and in each iteration, according to the transition probability matrix, the node obtains a ranking score from the in-degree node of the node, where the ranking score is the product of the ranking score and the transition probability of the in-degree node in the previous iteration.
Further, the upper and lower bound estimations in step (4) are specifically as follows:
1) for lower bound estimates, lower bound estimates for the ith rounds i[u]Calculated using the following formula:
wherein c is a damping coefficient, i is a current iteration round,s i-1[u]lower bound estimates for round i-1; lambda [ alpha ]i[u]For the random walk probability value from the query node to the node u in the ith iteration, the specific formula is as follows:
where q represents a query vector, q u]Represents the weight of node u, and ∑u∈Vq[u]1 is ═ 1; in addition, C [ u ] is used]To represent the set of in-degree nodes, λ, of node ui-1[v]The random walk probability value from the query node to the node v in the i-1 th iteration is obtained;
2) for upper bound estimates, the ith round upper bound estimate
Calculated using the following formula:
wherein, Pmax[u]＝max{p[v,u]| V ∈ V } is the maximum value of all the in-degree transition probabilities of the current node u;
further, in the step (4), the nodes of the hypergraph are filtered by using an upper and lower bound estimation method, and the specific filtering steps are as follows:
(4.1) when the size of the candidate point set is larger than k, if the upper bound estimation of a certain node in the candidate point set is larger than the current lower bound estimation of the kth node, the node is a candidate point and may appear in the final k results, otherwise, the node cannot become a result point and is discarded; if the size of the candidate point set is k, all the nodes in the candidate point set are final return results, and the node ranking in the candidate point set is determined through the step (4.2) or the step (4.3);
(4.2) when the size of the candidate node set is k, if the lower bound estimation of a node in the candidate node set is larger than the upper bound estimation of other nodes in the set or the upper bound estimation of the node is smaller than the lower bound estimation of other nodes in the set, determining the sequence of the node, and adding the node into a result node set; stopping iteration when the candidate node sequences are all determined;
and (4.3) when the size of the candidate node set is k, detecting the upper and lower bound estimation difference values of all current candidate nodes, stopping iteration if the upper and lower bound estimation difference values of all the nodes are smaller than a set threshold value, and determining the sequence of the nodes according to the lower bound estimation descending order.
Further, in the step (5), the user evaluates the k results obtained in the step (4), selects a result image which meets the query intention as a feedback node, adjusts the query vector in the step (4) according to the feedback node, inserts the feedback node into the original query vector to form a new query vector, performs weight distribution on the original query node and the newly inserted query node in the new query vector again, performs personalized PageRank query in the step (4) again, and uses an upper and lower bound estimation method to perform filtering, finally obtains a new query result and returns the new query result to the user.
The invention has the beneficial effects that: by adopting the hypergraph theory, the invention enables the multi-dimensional relationships such as comment relationships, annotation relationships, visual similarity relationships and the like of social images to be effectively organized; by adopting a measurement index technology in a spatial database, a batch construction technology, a parallel technology and a buffering technology are fully utilized, so that the calculation efficiency of the hypergraph transition probability matrix is remarkably improved; by adopting an upper and lower bound estimation technology, the technical effect of improving the query efficiency is brought, the iteration times are greatly reduced, and the I/O overhead and the CPU time are obviously reduced; and an optimization mode of user feedback is adopted, so that the query efficiency is obviously improved.
Drawings
FIG. 1 is a flow chart of the steps of carrying out the present invention;
FIG. 2 is a schematic diagram of the working principle of the hypergraph-based image retrieval and annotation system.
Detailed Description
The technical solution of the present invention will be further explained with reference to the accompanying drawings and specific implementation:
as shown in fig. 1 and fig. 2, the specific implementation process and the working principle of the present invention are as follows:
step (1): establishing a t-NN graph for the image data set by using an image retrieval engine based on content, and establishing a relation between each image and t images with the most similar visual characteristics;
step (2): establishing a hypergraph according to the image t-NN graph and the social association information of the image, calculating a transition probability matrix of the hypergraph, and storing the hypergraph in a B + tree;
and (3): a user submits a query object set and a k value;
and (4): generating a query vector according to a query object set submitted by a user, then carrying out parallel personalized PageRank query on a hypergraph, filtering nodes of the hypergraph by using an upper and lower bound estimation method, and gradually reducing a candidate point set of each iteration in the PageRank query process until k results are obtained;
and (5): and (4) evaluating the k results obtained in the step (4) by the user to generate feedback information, adjusting the query vector in the step (4) according to the feedback information to form a new query vector, repeating the step (4), and finally obtaining a new query result and returning the new query result to the user.
In the step (1), the image retrieval engine extracts visual features of the image according to the MPEG-7 standard based on a measurement space SPB tree index structure, wherein the visual features comprise five types of features including measurable color, color distribution, color structure, edge histogram and texture, an SPB tree is established, t-NN query is carried out on each image, t query results which are most similar to the images are obtained, and then a t-NN graph is established.
Establishing a hypergraph model according to the social association information of the image t-NN graph and the image in the step (2), wherein the hypergraph model has three types:
1) taking each image in the image t-NN image and the t-NN image thereof as a super edge;
2) taking a ternary combination marked by a user on an image as a super edge;
3) and taking a multi-element combination of images evaluated by a user as a super edge.
The specific steps of calculating the hypergraph transition probability matrix in the step (2) are as follows:
a hypergraph is represented by a matrix H of | V | × | E | where the matrix elements H (V, E) are:
wherein v is a node and e is a super edge;
the degree d (v) of node v may be expressed as:
d(v)＝∑e∈E|v∈ew(e)＝∑e∈Ew(e)h(v,e)
wherein w (e) represents the weight of the excess edge e;
the degree δ (e) of the excess edge e can be expressed as:
δ(e)＝|e|＝∑v∈Vh(v,e)
the transition probability matrix of the hypergraph, denoted by P, then the element P [ u, v ] in P represents the transition probability from node u to node v, which can be calculated using the following formula:
the transfer probability matrix in the step (2) is stored in a B + tree by adopting a batch insertion method, a parallel technology and a buffer technology; when the transition probability matrix is stored in the B + tree, the combination of the out-degree node and the in-degree node of the hypergraph is used as a key, and the transition probability is used as a value; wherein:
1) the elements are inserted into the B + tree by using a batch insertion method, so that repeated traversal of inserting the elements into the searched insertion positions one by one is reduced, the locality and the space utilization rate of B + tree nodes are improved by inserting the elements in sequence, and the time and space efficiency of B + tree node insertion are improved;
2) according to a formula, the weight of the entry edge of the node is only related to the node and the entry node, the degrees of the node and the hyper-edge are counted by one-time traversal, and then the time efficiency is improved by calculating the transition probability of the hyper-graph by using a parallel technology;
3) calculating the hypergraph transition probability matrix and inserting the elements of the B + tree can be parallel, wherein the data flow between the hypergraph transition probability matrix and the B + tree uses a double buffer, the two buffers are respectively written into the elements of the hypergraph transition probability matrix and read out the elements to be inserted into the B + tree, and if the write buffer is full or the read buffer is empty, the two buffers are exchanged;
the B + tree is stored on a disk and has expandability.
And (4) k is the number of query return nodes specified by the user in the step (3). The query vector comprises a plurality of query nodes, the weight of each query node can be designated by a user so as to emphasize the query nodes, and the sum of the weights of all the query nodes is normalized to be 1.
And (4) taking the query vector and the k value as input, and in each iteration, obtaining a ranking score from the in-degree node of the node according to the transition probability matrix by the node, wherein the ranking score is the product of the ranking score and the transition probability of the in-degree node in the previous iteration.
The upper and lower bound estimations in step (4) are specifically as follows:
1) for lower bound estimates, lower bound estimates for the ith rounds i[u]Calculated using the following formula:
wherein c is a damping coefficient (generally 0.8); and i is the current iteration round,s i-1[u]lower bound estimates for round i-1; lambda [ alpha ]i[u]For the random walk probability value from the query node to the node u in the ith iteration, the specific formula is as follows:
where q represents a query vector, q u]Represents the weight of node u, and ∑u∈Vq[u]1 is ═ 1; in addition, C [ u ] is used]To represent the set of in-degree nodes, λ, of node ui-1[v]The random walk probability value from the query node to the node v in the i-1 th iteration is obtained;
2) for upper bound estimates, the ith round upper bound estimate
Calculated using the following formula:
wherein, Pmax[u]＝max{p[v,u]| V ∈ V } is the maximum value of all the in-degree transition probabilities of the current node u;
and (4) filtering nodes of the hypergraph by using an upper and lower bound estimation method, wherein the specific filtering steps are as follows:
(4.1) when the size of the candidate point set is larger than k, if the upper bound estimation of a certain node in the candidate point set is larger than the current lower bound estimation of the kth node, the node is a candidate point and may appear in the final k results, otherwise, the node cannot become a result point and is discarded; if the size of the candidate point set is k, all the nodes in the candidate point set are final return results, and the node ranking in the candidate point set is determined through the step (4.2) or the step (4.3);
(4.2) when the size of the candidate node set is k, if the lower bound estimation of a node in the candidate node set is larger than the upper bound estimation of other nodes in the set or the upper bound estimation of the node is smaller than the lower bound estimation of other nodes in the set, determining the sequence of the node, and adding the node into a result node set; stopping iteration when the candidate node sequences are all determined;
and (4.3) when the size of the candidate node set is k, detecting the upper and lower bound estimation difference values of all current candidate nodes, stopping iteration if the upper and lower bound estimation difference values of all the nodes are smaller than a set threshold value, and determining the sequence of the nodes according to the lower bound estimation descending order.
Evaluating the k results obtained in the step (4) by the user in the step (5), selecting a result image which meets the query intention as a feedback node, adjusting the query vector in the step (4) according to the feedback node, inserting the feedback node into the original query vector to form a new query vector, and performing weight distribution on the original query node and the newly inserted query node in the new query vector again, wherein the total weight of the original query node is 0.5, the total weight of the newly inserted query node is 0.5, and the newly inserted node equally divides the weight of 0.5; and (5) carrying out personalized PageRank query in the step (4) again, filtering by using an upper and lower bound estimation method, and finally obtaining a new query result and returning the new query result to the user.
Claims (9)
1. An image retrieval and annotation method based on a hypergraph is characterized by comprising the following steps:
step (1): establishing a t-NN graph for the image data set by using an image retrieval engine based on content, and establishing a relation between each image and t images with the most similar visual characteristics;
step (2): establishing a hypergraph according to the image t-NN graph and the social association information of the image, calculating a transition probability matrix of the hypergraph, and storing the hypergraph in a B + tree;
and (3): a user submits a query object set and a k value;
and (4): generating a query vector according to a query object set submitted by a user, then carrying out parallel personalized PageRank query on a hypergraph, filtering nodes of the hypergraph by using an upper and lower bound estimation method, and gradually reducing a candidate point set of each iteration in the PageRank query process until k results are obtained;
and (5): and (4) evaluating the k results obtained in the step (4) by the user to generate feedback information, adjusting the query vector in the step (4) according to the feedback information to form a new query vector, repeating the step (4), and finally obtaining a new query result and returning the new query result to the user.
2. The hypergraph-based image retrieval and annotation method of claim 1, wherein: the image retrieval engine in the step (1) extracts visual features of the images according to the MPEG-7 standard based on the metric space SPB tree index structure, establishes the SPB tree, performs t-NN query on each image to obtain t query results which are most similar to the SPB tree, and further establishes the t-NN graph.
3. The hypergraph-based image retrieval and annotation method of claim 1, wherein: the types of the hypergraph in the step (2) are three types:
1) taking each image in the image t-NN image and the t-NN image thereof as a super edge;
2) taking a ternary combination marked by a user on an image as a super edge;
3) and taking a multi-element combination of images evaluated by a user as a super edge.
4. The hypergraph-based image retrieval and annotation method of claim 1, wherein: the specific steps of calculating the hypergraph transition probability matrix in the step (2) are as follows:
a hypergraph is represented by a matrix H of | V | × | E | where the matrix elements H (V, E) are:
wherein v is a node and e is a super edge;
the degree d (v) of node v may be expressed as:
d(v)＝∑e∈E|v∈ew(e)＝∑e∈Ew(e)h(v,e)
wherein w (e) represents the weight of the excess edge e;
the degree δ (e) of the excess edge e can be expressed as:
δ(e)＝|e|＝∑v∈Vh(v,e)
the transition probability matrix of the hypergraph, denoted by P, then the element P [ u, v ] in P represents the transition probability from node u to node v, which can be calculated using the following formula:
5. the hypergraph-based image retrieval and annotation method of claim 1, wherein: the transition probability matrix in the step (2) is stored in a B + tree by adopting a batch insertion method, a parallel technology and a buffer technology; when the transition probability matrix is stored in the B + tree, the combination of the out-degree node and the in-degree node of the hypergraph is used as a key, and the transition probability is used as a value.
6. The hypergraph-based image retrieval and annotation method of claim 1, wherein: and (4) taking the query vector and the k value as input, and in each iteration, obtaining a ranking score from the in-degree node of the node according to the transition probability matrix by the node, wherein the ranking score is the product of the ranking score and the transition probability of the in-degree node in the previous iteration.
7. The hypergraph-based image retrieval and annotation method of claim 1, wherein: the upper and lower bound estimations in the step (4) are specifically as follows:
1) for lower bound estimates, lower bound estimates for the ith rounds i[u]Calculated using the following formula:
wherein c is a damping coefficient, i is a current iteration round,s i-1[u]lower bound estimates for round i-1; lambda [ alpha ]i[u]For the random walk probability value from the query node to the node u in the ith iteration, the specific formula is as follows:
where q represents a query vector, q u]Represents the weight of node u, and ∑u∈Vq[u]1 is ═ 1; in addition, C [ u ] is used]To represent the set of in-degree nodes, λ, of node ui-1[v]The random walk probability value from the query node to the node v in the i-1 th iteration is obtained;
2) for upper bound estimates, the ith round upper bound estimate
Calculated using the following formula:
wherein, Pmax[u]＝max{p[v,u]| V ∈ V } is the maximum value of all in-degree transition probabilities for the current node u.
8. The hypergraph-based image retrieval and annotation method of claim 1, wherein: in the step (4), nodes of the hypergraph are filtered by using an upper and lower bound estimation method, and the specific filtering steps are as follows:
(4.1) when the size of the candidate point set is larger than k, if the upper bound estimation of a certain node in the candidate point set is larger than the current lower bound estimation of the kth node, the node is a candidate point and may appear in the final k results, otherwise, the node cannot become a result point and is discarded; if the size of the candidate point set is k, all the nodes in the candidate point set are final return results, and the node ranking in the candidate point set is determined through the step (4.2) or the step (4.3);
(4.2) when the size of the candidate node set is k, if the lower bound estimation of a node in the candidate node set is larger than the upper bound estimation of other nodes in the set or the upper bound estimation of the node is smaller than the lower bound estimation of other nodes in the set, determining the sequence of the node, and adding the node into a result node set; stopping iteration when the candidate node sequences are all determined;
and (4.3) when the size of the candidate node set is k, detecting the upper and lower bound estimation difference values of all current candidate nodes, stopping iteration if the upper and lower bound estimation difference values of all the nodes are smaller than a set threshold value, and determining the sequence of the nodes according to the lower bound estimation descending order.
9. The hypergraph-based image retrieval and annotation method of claim 1, wherein: and (5) evaluating the k results obtained in the step (4) by the user, selecting a result image which meets the query intention as a feedback node, adjusting the query vector in the step (4) according to the feedback node, inserting the feedback node into the original query vector to form a new query vector, performing weight distribution on the original query node and the newly inserted query node in the new query vector again, performing personalized PageRank query in the step (4) again, filtering by using an upper and lower bound estimation method, and finally obtaining a new query result and returning the new query result to the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710229126.XA CN107145519B (en) | 2017-04-10 | 2017-04-10 | Image retrieval and annotation method based on hypergraph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710229126.XA CN107145519B (en) | 2017-04-10 | 2017-04-10 | Image retrieval and annotation method based on hypergraph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107145519A CN107145519A (en) | 2017-09-08 |
| CN107145519B true CN107145519B (en) | 2020-07-03 |
Family
ID=59773722
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710229126.XA Active CN107145519B (en) | 2017-04-10 | 2017-04-10 | Image retrieval and annotation method based on hypergraph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107145519B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109271491B (en) * | 2018-11-02 | 2021-09-28 | 合肥工业大学 | Cloud service recommendation method based on unstructured text information |
| CN110413848B (en) * | 2019-07-19 | 2022-04-15 | 上海赜睿信息科技有限公司 | Data retrieval method, electronic equipment and computer-readable storage medium |
| CN112190269B (en) * | 2020-12-04 | 2024-03-12 | 兰州大学 | Depression auxiliary identification model construction method based on multisource brain electric data fusion |
| CN113641857A (en) * | 2021-08-13 | 2021-11-12 | 三星电子（中国）研发中心 | Visual media personalized search method and device |
| US12038968B2 (en) | 2021-08-13 | 2024-07-16 | Samsung Electronics Co., Ltd. | Method and device for personalized search of visual media |
| CN114065758B (en) * | 2021-11-22 | 2024-04-19 | 杭州师范大学 | Document keyword extraction method based on hypergraph random walk |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102024036A (en) * | 2010-11-29 | 2011-04-20 | 清华大学 | Three-dimensional object retrieval method and device based on hypergraphs |
| CN102663010A (en) * | 2012-03-20 | 2012-09-12 | 复旦大学 | Personalized image browsing and recommending method based on labelling semantics and system thereof |
| CN103020116A (en) * | 2012-11-13 | 2013-04-03 | 中国科学院自动化研究所 | Method for automatically screening influential users on social media networks |
| CN103390063A (en) * | 2013-07-31 | 2013-11-13 | 南京大学 | Search method for relevance feedback images based on ant colony algorithm and probability hypergraph |
| CN103955524A (en) * | 2014-05-09 | 2014-07-30 | 合肥工业大学 | Event-related socialized image searching algorithm based on hypergraph model |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6988093B2 (en) * | 2001-10-12 | 2006-01-17 | Commissariat A L'energie Atomique | Process for indexing, storage and comparison of multimedia documents |
-
2017
- 2017-04-10 CN CN201710229126.XA patent/CN107145519B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102024036A (en) * | 2010-11-29 | 2011-04-20 | 清华大学 | Three-dimensional object retrieval method and device based on hypergraphs |
| CN102663010A (en) * | 2012-03-20 | 2012-09-12 | 复旦大学 | Personalized image browsing and recommending method based on labelling semantics and system thereof |
| CN103020116A (en) * | 2012-11-13 | 2013-04-03 | 中国科学院自动化研究所 | Method for automatically screening influential users on social media networks |
| CN103390063A (en) * | 2013-07-31 | 2013-11-13 | 南京大学 | Search method for relevance feedback images based on ant colony algorithm and probability hypergraph |
| CN103955524A (en) * | 2014-05-09 | 2014-07-30 | 合肥工业大学 | Event-related socialized image searching algorithm based on hypergraph model |
Non-Patent Citations (1)
| Title |
|---|
| Hypergraph Spectral Hashing for image retrieval with heterogeneous social contexts;Yang Liu等;《Neurocomputing》;20130120;第49-59页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107145519A (en) | 2017-09-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107145519B (en) | Image retrieval and annotation method based on hypergraph | |
| AU2011210535B2 (en) | Joint embedding for item association | |
| CN109166615B (en) | Medical CT image storage and retrieval method based on random forest hash | |
| CN106503223B (en) | online house source searching method and device combining position and keyword information | |
| WO2013129580A1 (en) | Approximate nearest neighbor search device, approximate nearest neighbor search method, and program | |
| WO2014085776A2 (en) | Web search ranking | |
| CN111310023B (en) | Personalized search method and system based on memory network | |
| CN109376261B (en) | Mode independent retrieval method and system based on intermediate text semantic enhancing space | |
| CN110334290B (en) | MF-Octree-based spatio-temporal data rapid retrieval method | |
| CN110543595A (en) | in-station search system and method | |
| US20100121844A1 (en) | Image relevance by identifying experts | |
| KR101976081B1 (en) | Method, system and computer program for semantic image retrieval based on topic modeling | |
| CN112749330B (en) | Information pushing method, device, computer equipment and storage medium | |
| CN111506760B (en) | Depth integration measurement image retrieval method based on difficult perception | |
| CN115878757A (en) | Concept decomposition-based hybrid hypergraph regularization semi-supervised cross-modal hashing method | |
| CN108647295B (en) | Image labeling method based on depth collaborative hash | |
| CN112836008B (en) | Index establishing method based on decentralized storage data | |
| CN113918807A (en) | Data recommendation method and device, computing equipment and computer-readable storage medium | |
| CN106599305B (en) | Crowdsourcing-based heterogeneous media semantic fusion method | |
| CN116304213B (en) | RDF graph database sub-graph matching query optimization method based on graph neural network | |
| CN103309951A (en) | Method and device for searching multimedia file on the net | |
| CN112650869B (en) | Image retrieval reordering method and device, electronic equipment and storage medium | |
| CN115098728A (en) | Video retrieval method and device | |
| CN105022836B (en) | Compact depth CNN aspect indexing methods based on SIFT insertions | |
| CN114528469A (en) | Recommendation method and device, electronic equipment and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |