CN111159443A - Image characteristic value searching method and device and electronic equipment - Google Patents
Image characteristic value searching method and device and electronic equipment Download PDFInfo
- Publication number
- CN111159443A CN111159443A CN201911402366.0A CN201911402366A CN111159443A CN 111159443 A CN111159443 A CN 111159443A CN 201911402366 A CN201911402366 A CN 201911402366A CN 111159443 A CN111159443 A CN 111159443A
- Authority
- CN
- China
- Prior art keywords
- search
- vector
- target
- loading
- node
- 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.)
- Granted
Links
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
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/53—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (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 relates to the technical field of image processing, in particular to a method, a device and electronic equipment for searching image characteristic values, wherein the method comprises the following steps: acquiring a search request, wherein the search request comprises a vector to be searched corresponding to a characteristic value of an image to be searched; according to the classification attribute of the vector to be searched, inquiring a target database node with the classification attribute in each database node of each storage main body; loading the search vector of the original image characteristic value corresponding to the classification attribute in the target database node to a target computing node corresponding to the target database node; comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector; and acquiring an original image characteristic value corresponding to the target search vector in the target database node. The embodiment of the invention can accelerate the response speed and improve the search efficiency.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to a method and an apparatus for searching an image feature value, and an electronic device.
Background
At present, with the rapid development of the internet and the widespread use of computers, image retrieval technology becomes a method for assisting users to efficiently acquire required images. In the existing retrieval system, for the application of uniform storage and search of images, the original characteristic value is mainly searched directly by a single machine at present. The single machine has limited storage space, is not easy to dynamically and transversely expand, and has slow response speed when searching the original characteristic value of large data volume, thereby influencing the searching efficiency of the system. Therefore, the problems of low response speed and low search efficiency exist in the current image information search.
Disclosure of Invention
The embodiment of the invention provides a method for searching an image characteristic value, which can accelerate the response speed and improve the searching efficiency.
In a first aspect, an embodiment of the present invention provides a method for searching an image feature value, where the method is applied to a search system, where the search system includes a plurality of storage bodies, each storage body includes at least one database node and at least one computing node corresponding to the at least one database node, and the method for searching an image feature value includes the following steps:
acquiring a search request, wherein the search request comprises a vector to be searched corresponding to a characteristic value of an image to be searched;
according to the classification attribute of the vector to be searched, inquiring a target database node with the classification attribute in each database node of each storage main body;
loading a search vector of an original image characteristic value corresponding to the classification attribute in the target database node to a target computing node corresponding to the target database node, wherein the target computing node is a computing node corresponding to the target database node in a plurality of computing nodes;
comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector;
and acquiring an original image characteristic value corresponding to the target search vector in the target database node.
In a second aspect, an embodiment of the present invention provides an apparatus for searching an image feature value, which is applied to a search system, where the search system includes a plurality of storage bodies, each storage body includes at least one database node and at least one computing node corresponding to the at least one database node, and the apparatus includes:
the device comprises a first acquisition module, a second acquisition module and a search module, wherein the first acquisition module is used for acquiring a search request which comprises a vector to be searched corresponding to a characteristic value of an image to be searched;
the searching module is used for inquiring a target database node with the classification attribute in each database node of each storage main body according to the classification attribute of the vector to be searched;
a loading module, configured to load a search vector of an original image feature value corresponding to the classification attribute in the target database node to a target computing node corresponding to the target database node, where the target computing node is a computing node corresponding to the target database node in multiple computing nodes;
the comparison module is used for comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector;
and the second acquisition module is used for acquiring the original image characteristic value corresponding to the target search vector in the target database node.
In a third aspect, an embodiment of the present invention provides an electronic device, including: the image feature value searching method comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize steps in the image feature value searching method provided by the embodiment of the invention.
In a fourth aspect, the embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps in the image feature value search method provided by the embodiment of the present invention.
In the embodiment of the invention, because a plurality of storage main bodies are configured, and each storage main body comprises at least one database node and at least one computing node corresponding to the database node, the transverse expansion limitation caused by single-machine search is avoided, and the ductility of a cluster is favorably realized; searching target database nodes with the same classification attribute in a plurality of storage main bodies according to the classification attribute of the vector to be searched, then loading an original characteristic value corresponding to the classification attribute in the target database to a target computing node, and comparing the search vector with the same classification attribute with the vector to be searched, wherein the searching of the target database nodes through the classification attribute can accelerate the searching speed without traversing all database nodes; the data volume of the vector is small, and the response time of comparison can be shortened by comparing the vector to be searched with the search vector, so that the searching efficiency is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a system architecture diagram of a search system provided by an embodiment of the present invention;
fig. 2 is a flowchart of a method for searching an image feature value according to an embodiment of the present invention;
FIG. 3 is a flowchart of another image feature value searching method according to an embodiment of the present invention;
FIG. 4 is a flowchart of another image feature value searching method according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an apparatus for searching an image feature value according to an embodiment of the present invention;
FIG. 6 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
FIG. 8 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
FIG. 9 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
FIG. 10 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a system architecture diagram of a search system to which the image feature value search method according to the embodiment of the present invention is applied. The system architecture 100 includes a plurality of storage agents 1001, each storage agent 1001 includes at least one database node 1001a and at least one target computing node 1001b corresponding to the at least one database node, and the outside of the system 1001 is connected to the mobile device terminal 1002 through a network.
The database node 1001a and the target computing node 1001b are connected to each other via a network to realize data transmission, and are arranged in parallel in the same storage body 1001. The storage body 1001 may also be provided in plurality, such as the second storage body, … …, nth storage body in fig. 1. Each storage agent may have a corresponding database node and a target computing node corresponding to the database node, so as to facilitate the lateral expansion of the database node 1001a and the target computing node 1001 b.
The database node 1001a is used to store the original image feature values, and after the target computing node 1001b searches for the target search vector, the corresponding original image feature values may be found in the database node 1001a based on the target search vector.
The target calculation node 1001b is configured to load a search vector of an original image feature value in the database node 1001a, and compare the search vector with a vector to be searched in the target calculation node 1001b to obtain a target search vector. The search vector may also be deleted in the target computing node 1001 b. The target computing node 1001b may include an image processor (GPU) or a memory, and the target computing node 1001b may further increase the number of GPUs or the size of the memory.
The user can send a search request through the mobile terminal 1002 and receive a target image feature value searched by the system. The mobile terminal 1002 may be an electronic device having a display screen and capable of reading and receiving data transmitted by communication, and the mobile terminal 1002 includes, but is not limited to, a smart phone, a tablet computer, a laptop portable computer, a desktop computer, and the like, and may also be called a terminal, a user terminal, a client, a smart terminal, and the like.
Specifically, when the system receives a search request sent by a user through the mobile terminal 1002, the characteristic value of the image to be searched included in the search request is obtained first, and then the vector to be searched in the characteristic value of the image to be searched is extracted and the corresponding classification attribute is obtained. Then, the system loads the search vector of the original image feature value in the database node 1001a with the same classification attribute to the target computing node 1001b, compares the vector to be searched with the search vector on the target computing node 1001b to obtain a target search vector, searches out the corresponding original image feature value in the database node 1001a according to the target search vector, and returns the original image feature value to the mobile terminal 1002 sending the search request by the user. The database node 1001a and the target computing node 1001b are disposed in the storage master 1001 having the same classification attribute as the database node 1001 a.
It should be understood that the numbers of mobile terminals, database nodes, target computing nodes, storage agents and networks described above are merely illustrative and may be specifically adjusted according to implementation needs.
As shown in fig. 2, fig. 2 is a flowchart of a method for searching an image feature value according to an embodiment of the present invention, including the following steps:
201. and acquiring a search request, wherein the search request comprises a vector to be searched corresponding to the characteristic value of the image to be searched.
In this embodiment, the electronic device on which the image feature value search method is executed may obtain the search request and the like through a wired connection manner or a wireless connection manner. It should be noted that the wireless connection mode may include various connection modes, for example: wired, wireless connection, or fiber optic cable. It should be noted that the Wireless connection manner may include, but is not limited to, a 3G/4G connection, a WiFi (Wireless-Fidelity) connection, a bluetooth connection, a wimax (worldwide interoperability for Microwave access) connection, a Zigbee (low power local area network protocol), a uwb (ultra wideband) connection, and other Wireless connection manners known now or developed in the future.
The search request may be a request sent by a user based on a mobile terminal, the feature value of the image to be searched may be an original image feature value corresponding to the search request, or may also be a long feature value, and the original image feature value may be binary data obtained by converting image information acquired by an image acquisition device. The vector to be searched may be a sorting code generated by a vector sorting algorithm and corresponding to an original image feature value, and may also be referred to as a short feature value to be searched. The sorting code may be generated according to a vector sorting algorithm, which is a preconfigured algorithm. The vector to be searched has the characteristics of short length, small operand and small occupied space relative to the characteristic value of the image to be searched, and the size of the occupied space is smaller than that of the characteristic value of the original image, for example: and generating 500 corresponding sorting codes by using the 5000 acquired characteristic values of the image to be searched through a vector sorting algorithm. The image capturing device may be a device with an image capturing function, including but not limited to a camera, and the like.
202. And inquiring a target database node with the classification attribute in each database node of each storage main body according to the classification attribute of the vector to be searched.
The classification attribute may be an attribute condition for distinguishing different types of vectors to be searched, and may include one or more of identity attributes (gender, name, age, identification number, etc.), appearance attributes (height, circumference, worn hat, glasses, worn clothes, trousers, shoes, and hat, clothes, trousers, style of shoes, color information, etc.), behavior attributes (communication, walking, running, driving, etc.), and the like, for example: the classification attribute is female wearing a skirt, and the classification attribute is male wearing slippers or male wearing short sleeves. Referring to fig. 1, a plurality of storage bodies may be configured as required, at least one database node is constructed in each storage body, and when a database node needs to be added, the database node may be expanded laterally, so as to enhance the extensibility of the search system application.
The storage agent may be a machine including at least one database node and at least one target computing node corresponding to the at least one database node, for example: for example: servers, processors, controllers, workstations, and the like. The database nodes and the calculation contacts are deployed in the same storage main body, so that query of extracting the sequencing codes and the image characteristic values and transmission of data can be facilitated. Different memory principals may have different classification attributes. The database nodes can store original image characteristic values with different classification attributes, and when the data quantity of the original image characteristic values is increased, the storage of more original image characteristic values can be realized by increasing the number of the database nodes.
The classification attribute of the storage subject and the database node deployed on the storage subject may have the same classification attribute, and of course, when there are multiple database nodes in the same storage subject, each database node may store an attribute that is more refined under the classification attribute of the storage subject, for example: the classification attribute of the storage main body is wearing, and the classification attributes of the plurality of database nodes can be respectively corresponding to clothes, trousers, shoes and the like. Therefore, the storage subject with the same classification attribute can be found through the classification attribute of the vector to be searched, and then the target database node with the same classification attribute can be found in the storage subject, for example: and if the classification attribute of the vector to be searched is trousers, the classification attribute of the storage main body is found as wearing, and then the target database node with the classification attribute of trousers is found in the storage main body.
A database interface may be provided in the database node, and when data is recorded into the database node, data may be added through the database interface based on a partition field used by a data table created in the database node, and the data may be distributed to storage areas of different database nodes for storage.
203. And loading the search vector of the original image characteristic value corresponding to the classification attribute in the target database node to a target computing node corresponding to the target database node, wherein the target computing node is a computing node corresponding to the target database node in the plurality of computing nodes.
The target computing node may be a node for storing the search vector, and performing computing processing on the vector to be searched and the search vector, and the target computing node and the target database node have the same classification attribute, and each database node may correspond to one target computing node. And finding out the target computing node corresponding to the target database node according to the classification attribute. The sequencing code of the characteristic value of the original image can be calculated in the target calculation node, and meanwhile, the calculated result can be stored in a GPU or a memory of the corresponding target calculation node, so that the searching operation of a user is facilitated. After the sequencing code is generated, the sequencing code can be updated to the target database node recorded with the original image characteristic value corresponding to the sequencing code, and when the target computing node is restarted, only the sequencing code generated in the target database node corresponding to the target computing node needs to be loaded into a GPU or a memory, and recalculation is not needed.
As a feasible embodiment, when a large number of new search vectors are written into a database node, because the storage space of one database node is limited, the database node can be expanded horizontally, meanwhile, the computing nodes (including the target computing node) corresponding to the database node also need to be increased correspondingly according to the increased number of the database node, and the increased computing nodes also need to create a new service to bind with the corresponding database node. When there is insufficient GPU or memory in the target compute node, a GPU card may be added or memory space may be increased, for example: and setting 8 GPUs to simultaneously store the search vectors, or increasing the memory space from 1G to 4G. In addition, a larger memory space can be obtained by reducing the loading time threshold of the search vector, for example: the standard set loading time threshold of 3 months was reduced to 1 month.
204. And comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector.
The search vectors in the target computing node may include a plurality of vectors, comparing the to-be-searched vectors having the same classification attribute with the search vectors may be comparing the to-be-searched vectors with each search vector in the target computing node, calculating a similarity between the to-be-searched vectors and each search vector in the target computing node, and the process of calculating the similarity may be completed in the target computing node, and after completion, the searched target search vectors may be returned to an interface forwarding layer called by a user through a forwarding service corresponding to the interface forwarding layer, where the interface forwarding layer may be a program specially used for forwarding information, and the forwarding service may be a program used for implementing a forwarding operation. Because the search vectors proposed by using the vector sorting algorithm in the target computing node are all short eigenvalues, the calculation amount is small, and the similarity between the vector to be searched and each search vector in the target computing node can be quickly calculated, so that topN search vectors with the highest similarity can be obtained and serve as the target search vectors.
205. And acquiring an original image characteristic value corresponding to the target search vector in the target database node.
After receiving the target search vector, the interface forwarding layer may forward the target search vector to a target database node, and call an original image feature value corresponding to the target search vector from the target database node, where the obtained original image feature value is a search result corresponding to the search request, and the original image feature value corresponding to the target search vector includes image information corresponding to the search request, for example: a traffic image captured at a certain place at a certain time, and the like.
In the embodiment of the invention, because a plurality of storage main bodies are configured, and each storage main body comprises at least one database node and at least one target computing node corresponding to the database node, the transverse expansion limitation caused by single-machine search is avoided, and the ductility of a cluster is favorably realized; searching target database nodes with the same classification attribute in a plurality of storage main bodies according to the classification attribute of the vector to be searched, then loading an original characteristic value corresponding to the classification attribute in the target database to a target computing node, and comparing the search vector with the same classification attribute with the vector to be searched, wherein the searching of the target database nodes through the classification attribute can accelerate the searching speed without traversing all database nodes; the data volume of the vector is small, and the response time of comparison can be shortened by comparing the vector to be searched with the search vector, so that the searching efficiency is improved.
As shown in fig. 3, fig. 3 is a flowchart of another image feature value searching method provided in the embodiment of the present invention, including the following steps:
301. and acquiring a search request, wherein the search request comprises a vector to be searched corresponding to the characteristic value of the image to be searched.
302. And inquiring a target database node with the classification attribute in each database node of each storage main body according to the classification attribute of the vector to be searched.
303. The target computing node comprises a plurality of loading areas, and the residual storage space corresponding to the loading areas, with the same classification attributes, of the original image characteristic values of the search vectors in the target database node is obtained.
304. And judging whether the residual storage space meets the required loading space of the search vector.
The target computing node can upload the current residual storage space of each loading area, the classification attribute of the loading area and the space size occupied by each classification attribute to an interface transfer layer, the interface transfer layer can load a search vector from a target database node and judge whether the search vector of the classification attribute exists in the loading area, and when the search vector exists in the loading area, the size relation between the residual storage space of the loading area and the loading space required by the search vector can be judged.
305. And if the residual storage space meets the loading space required by the search vector, loading the search vector to a loading area corresponding to the residual storage space.
If the remaining storage space satisfies the loading space required by the search vector, it indicates that the remaining storage space of the corresponding loading area is greater than or equal to the loading space required by the search vector, for example: the method comprises an A loading area and a search vector, wherein the A loading area can store the classification attribute of the search vector, and if the residual storage space of the A loading area is 2G and the storage space occupied by the a search vector is 1G, the a search vector can be directly stored in the A loading area.
306. And if the residual storage space does not meet the loading space required by the search vector, inquiring whether the residual storage space of the rest classification attributes in the target computing node meets the loading space required by the search vector, and if so, loading the search vector.
If the remaining storage space does not satisfy the load space required by the search vector, it indicates that the remaining storage space of the corresponding load area is smaller than the load space required by the search vector, for example: a, B, C, there are two search vectors d and e, and the three loading areas and the two search vectors have the same classification attribute, when a part of d is stored in the loading area A, and the loading area A is full and has no residual storage space, the loading area B also stores the search vector d, and the residual storage space is 2G, and at this time, the residual storage space of the loading area C is 5G, and the storage space occupied by the search vector e is 1G, then the loading area C with the largest residual storage space will store e.
When there is a large number of search vectors to be loaded to a target computing node and the remaining storage space of a loading area in the target computing node having the same classification attribute as the search vector is insufficient, it may be considered that the search vector is stored in a loading area with a different classification attribute, and the remaining storage space of the loading area with a different classification attribute is sufficient for the required storage space of the search vector, for example: the classification attribute of the search vector is a male wearing a hat, and the loading area of the target computing node with the same classification attribute is full, so that the loading area meeting the loading space required by the search vector can be inquired again, the searched loading area can be a male wearing sunglasses or an empty loading area, but the residual storage space of the inquired loading area is at least the storage space required by the search vector which needs to be loaded.
307. And comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector.
After the target database nodes with the same classification attributes as the vectors to be searched are found, the search vectors in the target database nodes can be loaded to the target computing nodes corresponding to the target database nodes, and similarity calculation is carried out on the vectors to be searched and the loaded search vectors on the target computing nodes so as to find topN search vectors with the highest similarity to the vectors to be searched.
308. And acquiring an original image characteristic value corresponding to the target search vector in the target database node.
Optionally, step 307 includes:
and traversing the search vectors with the same classification attributes as the vectors to be searched in the target calculation node, and calculating the similarity between the vectors to be searched and the search vectors.
By calculating the similarity between the vector to be searched and the search vector and the sum of the vector to be searched and the search vector, a more accurate target search vector can be obtained. The user can search according to the attribute information, for example: women who wear hats in the last 3 months were searched. The mobile terminal can directly call the query interface of the database node to search data and return searched result data. The user may also search in the form of a vector, for example: the most similar portrait in the last 3 months needs to be calculated in the target computing node according to the searching vector which is most similar to the vector to be searched in the last 3 months, and the most similar can be limited by a similarity threshold. When the query is carried out in this way, one end of a user sending out a search request can call an interface of an interface forwarding layer, all search vectors are searched for at a target computing node, then topN search vectors with the highest similarity in the target computing node are returned to the interface forwarding layer through a forwarding service, and the interface forwarding layer can forward the topN search vectors to corresponding target database nodes.
As a possible embodiment, the user may also search in combination with the vector and the classification attribute, for example: the most similar men wearing glasses last three months. The search request reaches an interface forwarding layer, the interface forwarding layer firstly queries the most similar topN search vectors in the target computing node according to a vector search mode and returns the most similar topN search vectors to the target database node, then the interface forwarding layer performs attribute filtering (screens out males meeting wearing glasses) in the target database node according to the conditions (vectors and classification attributes) of the most similar topN search vectors, eyes and males returned by the target computing node, and then the filtered search vectors are obtained.
And screening the search vectors with the similarity reaching the similarity threshold value to obtain the target search vectors.
After the similarity between the vector to be searched and the search vector with the same classification attribute is obtained, all the similarities may be compared with a preset similarity threshold, the search vector meeting the similarity threshold is screened out, and the screened search vector will be used as a target search vector, for example: the similarity between the vector A to be searched and the search vector B, C, D, E, F, G is 60, 75, 80, 85, 90 and 92 in sequence, and the similarity threshold is 80, so that the target search vector is D, E, F, G.
Optionally, the method may further include:
and detecting whether the search vectors with the same classification attribute exist in the target computing node or not and storing the search vectors in different loading areas.
Specifically, the loading area with the same classification attribute may indicate that when the history search vector is stored, the remaining storage space of the same classification attribute is insufficient, and therefore the history search vector is transferred to the remaining storage areas with different classification attributes. After the search vectors are updated in the target computing node, the remaining storage space of the partial loading area is increased, and at this time, the search vectors with the same classification attribute can be converged and concentrated into the same loading area with the same classification attribute as the search vectors.
If the search vectors with the same classification attributes are stored in different loading areas, the search vectors with the same classification attributes in the different loading areas are extracted, and the search vectors with the same classification attributes are merged into the same loading area.
If the search vectors with the same classification attribute are stored in different loading areas, the search vectors with the same classification attribute in multiple loading areas may be merged into one loading area, for example: the loading area A stores men wearing hats, the loading area C stores men wearing hats and women wearing hats, and the classification attribute of the loading area C is that the women wearing hats, at the moment, the men wearing hats in the loading area C can be transferred to the loading area A, and the men wearing hats in the loading area B are deleted, so that the problem that the same search vector repeatedly exists in a plurality of loading areas, resource waste and space utilization rate reduction are caused is solved. The search vectors with the same classification attributes are merged into the same loading area, so that a dynamic scheduling of a storage space can be realized, the search vectors with the same classification attributes are merged into one loading area as much as possible, and the loading area transferred out of the search vectors can be used for storing the search vectors of other classification attributes.
As a possible embodiment, the step of merging the search vectors with the same classification attribute into the same loading area includes:
and searching a target loading area of which the residual storage space meets the requirement of occupying the storage space.
The occupied storage space is the total storage space required by the search vectors with the same classification attributes in different loading areas. When merging the loading areas corresponding to the search vectors with the same classification attributes, the remaining storage space of the loading area where each search vector to be merged is located needs to be judged, and the storage space which is occupied is judged according to the size of the remaining storage space of the loading area where each search vector to be merged is located, for example: the search vector with the classification attribute 1 in the A loading area occupies 300MB, the residual storage space is 500MB, the search vector with the classification attribute 1 in the B loading area occupies 200MB, the residual storage space is 800MB, the search vector with the classification attribute 1 in the C loading area occupies 200MB, the residual storage space is 1G, and only if the classification attribute of the C loading area is the same as the classification attribute of the search vector and the residual storage space is 1G which is more than 300MB +200MB, the search vector with the same classification attribute 1 in the A loading area and the B loading area can be transferred to the C storage area.
And migrating the search vectors with the same classification attributes in different loading areas to a target loading area for storage.
After the migration and storage of the search vector are completed, the search vector transferred in the loading area from which the search vector is transferred can be deleted, and a new search vector having the same classification attribute as the storage area is written only after the deletion is completed.
In the embodiment of the invention, after the search request is obtained, the target database node with the same classification attribute as the vector to be searched is searched in the plurality of database nodes in the plurality of storage main bodies, the search vector with the same classification attribute as the vector to be searched in the target database node is extracted and loaded into the target computing node, the search speed can be accelerated by searching the target database node through the classification attribute, traversal of all database nodes is not needed, and the search time is saved; the data volume of the search vector is small, and the response time of comparison can be shortened by comparing the vector to be searched with the search vector, so that the searching efficiency is improved; and before the search vector is stored in the target computing node, the residual storage space of the loading area in the target computing node is judged, and the search vector is respectively stored in different loading areas according to the size of the residual storage space and the classification attribute of the loading area. If the search vectors with the same classification attribute exist in the plurality of loading areas, the search vectors with the same classification attribute can be merged and migrated to the same loading area for storage, and dynamic scheduling of the search vectors in the loading areas is achieved.
As shown in fig. 4, fig. 4 is a flowchart of another image feature value searching method provided in the embodiment of the present invention, including the following steps:
401. and acquiring a search request, wherein the search request comprises a vector to be searched corresponding to the characteristic value of the image to be searched.
402. And inquiring a target database node with the classification attribute in each database node of each storage main body according to the classification attribute of the vector to be searched.
403. And loading the search vector of the original image characteristic value corresponding to the classification attribute in the target database node to a target computing node corresponding to the target database node, wherein the target computing node is a computing node corresponding to the target database node in the plurality of computing nodes.
404. And acquiring corresponding original image characteristic values from the database nodes based on the target search vector.
405. And deleting the search vector in the target computing node, and judging whether the search vector has an original image characteristic value corresponding to the search vector in the target database node.
Specifically, since new search vectors are continuously written into the target computing node, the storage space of the GPU or the memory in the target computing node is limited, and the search vectors in the storage space need to be updated. Deletion may be performed by setting a load time threshold, and deleting search vectors that exceed the load time threshold, for example: and setting the loading time threshold to be 3 months, and deleting the search vectors more than 3 months ago. The deleting operation may be performed by setting a timer in the target computing node, and when the timer detects a search vector exceeding a loading time threshold, the timer automatically deletes the search vector in the GPU or the memory, so that the effective use of the GPU or the memory can be improved.
As a possible embodiment, when a user needs to search for an original image feature value exceeding a loading time threshold, a query may be performed based on a UDF function customized to a database node in advance, for example: the loading time threshold is 3 months, and the user wants to query the original image characteristic value for 8 months. The UDF function is an extensible interface of the database nodes, and can be integrated into each database node in a mode of the UDF function by using the Euclidean theorem, wherein the Euclidean theorem is a mode of calculating similarity between search vectors.
406. And if the original image characteristic value corresponding to the search vector does not exist, directly deleting the search vector in the target computing node.
When the search vector on the target computing node does not have the original image feature value in the database node, the search vector on the target computing node can be directly deleted by calling an interface of the database node, for example: when the image is collected through the camera, only one license plate number, namely one character string, is collected, and then the license plate number is directly deleted in the target computing node.
407. And if the original image characteristic value corresponding to the search vector exists, deleting the corresponding original image characteristic value in the database node and the search vector in the target calculation node in sequence.
When the search vector on the target computing node has an original image characteristic value in the database node, an interface of an interface forwarding layer may be called to delete the original image characteristic value corresponding to the search vector to be deleted in the database node, and then the corresponding search vector is deleted in the GPU or the memory in the target computing node according to the deleted original image characteristic value.
In this embodiment, a search vector of an original image feature value in a database node is loaded to a target computing node, which is actually an action executed many times, and as long as a search request is obtained each time, a corresponding search vector is loaded to the target computing node, but a storage space of the target computing node is limited, so that the search vector stored in a GPU or a memory of the target computing node and exceeding a loading time threshold needs to be deleted, in this embodiment, the search vector may be deleted according to whether the original image feature value of the search vector exists in the database node, and deletion of the search vector may allow a storage space of the target computing node to implement dynamic scheduling, and may not reduce a search speed due to an excessive number of stored search vectors, and simultaneously may ensure that a new search vector may be continuously entered into the storage space of the target computing node, and realizing data updating.
As shown in fig. 5, fig. 5 is a schematic structural diagram of an apparatus for searching an image feature value according to an embodiment of the present invention, where the apparatus includes:
a first obtaining module 501, configured to obtain a search request, where the search request includes a vector to be searched corresponding to a feature value of an image to be searched;
a searching module 502, configured to query, according to the classification attribute of the vector to be searched, a target database node having the classification attribute in each database node of each storage main body;
a loading module 503, configured to load a search vector of an original image feature value corresponding to a classification attribute in a target database node to a target computing node corresponding to the target database node, where the target computing node is a computing node corresponding to the target database node in the multiple computing nodes;
a comparison module 504, configured to compare the vector to be searched with the search vector in the target computing node, to obtain a target search vector;
and a second obtaining module 505, configured to obtain an original image feature value corresponding to the target search vector in the target database node.
Optionally, the target computing node includes a plurality of loading areas, as shown in fig. 6, fig. 6 is a schematic structural diagram of another image feature value searching apparatus provided in an embodiment of the present invention, where the loading module 503 includes:
an obtaining unit 5031, configured to obtain a remaining storage space corresponding to a loading area where a search vector of an original image feature value corresponding to a classification attribute has the same classification attribute in a target database node, and determine whether the remaining storage space meets a required loading space of the search vector;
a loading unit 5032, configured to load the search vector into a loading area corresponding to the remaining storage space if the remaining storage space meets the loading space required by the search vector;
a first querying unit 5033, configured to query whether the remaining storage spaces of the remaining classification attributes in the target computing node satisfy the required loading space of the search vector if the remaining storage spaces do not satisfy the required loading space of the search vector, and if so, load the search vector.
Optionally, as shown in fig. 7, fig. 7 is a schematic structural diagram of another image feature value searching apparatus provided in an embodiment of the present invention, and the apparatus 500 further includes:
a detecting module 506, configured to detect whether search vectors with the same classification attribute exist in target computing nodes and are stored in different loading areas;
the merging module 507 is configured to, if the search vectors with the same classification attribute are stored in different loading areas, extract the search vectors with the same classification attribute in the different loading areas, and merge the search vectors with the same classification attribute into the same loading area.
Optionally, as shown in fig. 8, fig. 8 is a schematic structural diagram of another image feature value search apparatus provided in an embodiment of the present invention, and the merging module 507 includes:
the second query unit 5071 is configured to search for a target loading area whose remaining storage space satisfies an occupied storage space, where the occupied storage space is a total storage space required by search vectors with the same classification attribute in different loading areas;
a migration unit 5072, configured to migrate search vectors in different load areas with the same classification attribute to a target load area for storage.
Optionally, as shown in fig. 9, fig. 9 is a schematic structural diagram of another image feature value searching apparatus provided in the embodiment of the present invention, and the comparison module 504 includes:
the calculating unit 5041 is configured to traverse search vectors in the target computing node, which have the same classification attribute as the vector to be searched, and calculate a similarity between the vector to be searched and the search vectors;
the screening unit 5042 is configured to screen a search vector with a similarity reaching a similarity threshold to obtain a target search vector.
Optionally, as shown in fig. 10, fig. 10 is a schematic structural diagram of another image feature value searching apparatus provided in the embodiment of the present invention, and the apparatus 500 further includes:
a determining module 508, configured to delete a search vector in a target computing node, and determine whether an original image feature value corresponding to the search vector exists in a target database node of the search vector;
a deleting module 509, configured to directly delete the search vector in the target computing node if there is no original image feature value corresponding to the search vector;
the deleting module 509 is further configured to delete the corresponding original image feature values in the database nodes and the search vector in the target computing node in sequence if the original image feature values corresponding to the search vector exist.
The image characteristic value searching device provided by the embodiment of the invention can realize each process realized by the image characteristic value searching method in the method embodiment and can achieve the same beneficial effect, and in order to avoid repetition, the repeated description is omitted.
Referring to fig. 11, fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, and as shown in fig. 11, the electronic device 1100 includes: a memory 1102, a processor 1101, and a computer program stored on the memory 1102 and executable on the processor 1101, wherein:
the processor 1101 is configured to call the computer program stored in the memory 1102, and perform the following steps:
acquiring a search request, wherein the search request comprises a vector to be searched corresponding to the characteristic value of the image to be searched;
according to the classification attribute of the vector to be searched, inquiring a target database node with the classification attribute in each database node of each storage main body;
loading a search vector of an original image characteristic value corresponding to the classification attribute in a target database node to a target computing node corresponding to the target database node, wherein the target computing node is a computing node corresponding to the target database node in a plurality of computing nodes;
comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector;
and acquiring an original image characteristic value corresponding to the target search vector in the target database node.
Optionally, the step executed by the processor 1101 of querying, in each database node of each storage subject, a target database node having a classification attribute according to the classification attribute of the vector to be searched includes:
inquiring a storage main body with the same classification attribute as that of the vector to be searched;
and searching a target database node with the same classification attribute as the vector to be searched in the storage main body with the same classification attribute, and inquiring the search vector with the same classification attribute as the vector to be searched in the target database node.
Optionally, the target computing node includes a plurality of loading areas, and the step executed by the processor 1101 of loading the search vector of the original image feature value corresponding to the classification attribute in the target database node to the target computing node corresponding to the target database node includes:
acquiring a residual storage space corresponding to a loading area, in a target database node, of which the search vectors of the original image characteristic values corresponding to the classification attributes have the same classification attributes, and judging whether the residual storage space meets the required loading space of the search vectors or not;
if the residual storage space meets the loading space required by the search vector, loading the search vector to a loading area corresponding to the residual storage space;
and if the residual storage space does not meet the loading space required by the search vector, inquiring whether the residual storage space of the rest classification attributes in the target computing node meets the loading space required by the search vector, and if so, loading the search vector.
Optionally, the processor 1101 is further configured to perform the following steps:
detecting whether search vectors with the same classification attribute exist in a target computing node or not and storing the search vectors in different loading areas;
if the search vectors with the same classification attributes are stored in different loading areas, the search vectors with the same classification attributes in the different loading areas are extracted, and the search vectors with the same classification attributes are merged into the same loading area.
Optionally, the step of comparing the vector to be searched with the search vector in the target computing node, executed by the processor 1101, to obtain the target search vector includes:
traversing the search vectors with the same classification attributes as the vectors to be searched in the target computing nodes, and computing the similarity between the vectors to be searched and the search vectors;
and screening the search vectors with the similarity reaching the similarity threshold value to obtain the target search vectors.
Optionally, the step performed by the processor 1101 of merging the search vectors with the same classification attribute into the same loading area includes:
searching a target loading area with the residual storage space meeting the storage space occupation requirement, wherein the storage space occupation requirement is the total storage space required by the search vectors with the same classification attributes in different loading areas;
and migrating the search vectors with the same classification attributes in different loading areas to a target loading area for storage.
Optionally, the processor 1101 is further configured to delete a search vector in the target computing node, and determine whether an original image feature value corresponding to the search vector exists in the target database node of the search vector;
if the original image characteristic value corresponding to the search vector does not exist, directly deleting the search vector in the target computing node;
and if the original image characteristic value corresponding to the search vector exists, deleting the corresponding original image characteristic value in the database node and the search vector in the target calculation node in sequence.
The electronic device 1100 provided by the embodiment of the present invention can implement each implementation manner in the embodiment of the method for searching for image feature values, and has corresponding beneficial effects, and for avoiding repetition, details are not repeated here.
The embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor 1101, the computer program implements each process of the image feature value searching method provided in the embodiment of the present invention, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware related to instructions of a computer program, and the program can be stored in a computer readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.
It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and may be performed in other orders unless explicitly stated herein. Moreover, at least a portion of the steps in the flow chart of the figure may include multiple sub-steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed alternately or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
The terms "including" and "having," and any variations thereof, in the description and claims of this application and the description of the above figures are intended to cover non-exclusive inclusions. The terms "first," "second," and the like in the description and claims of this application or in the above-described drawings are used for distinguishing between different objects and not for describing a particular order. Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
The above disclosure is only for the purpose of illustrating the preferred embodiments of the present invention, and it is therefore to be understood that the invention is not limited by the scope of the appended claims.
Claims (10)
1. A method for searching image characteristic values is applied to a search system, and is characterized in that the search system comprises a plurality of storage bodies, each storage body comprises at least one database node and at least one computing node corresponding to the at least one database node, and the method for searching the image characteristic values comprises the following steps:
acquiring a search request, wherein the search request comprises a vector to be searched corresponding to a characteristic value of an image to be searched;
according to the classification attribute of the vector to be searched, inquiring a target database node with the classification attribute in each database node of each storage main body;
loading a search vector of an original image characteristic value corresponding to the classification attribute in the target database node to a target computing node corresponding to the target database node, wherein the target computing node is a computing node corresponding to the target database node in a plurality of computing nodes;
comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector;
and acquiring an original image characteristic value corresponding to the target search vector in the target database node.
2. The method of claim 1, wherein the target computing node includes a plurality of loading regions, and wherein loading the search vector of raw image feature values corresponding to the classification attributes in the target database node into the target computing node corresponding to the target database node comprises:
acquiring a residual storage space corresponding to a loading area, in the target database node, of which the search vectors of the original image characteristic values corresponding to the classification attributes have the same classification attributes, and judging whether the residual storage space meets the required loading space of the search vectors or not;
if the residual storage space meets the loading space required by the search vector, loading the search vector to a loading area corresponding to the residual storage space;
and if the residual storage space does not meet the required loading space of the search vector, inquiring whether the residual storage space of the rest classification attributes in the target computing node meets the required loading space of the search vector, and if so, loading the search vector.
3. The method of claim 2, wherein the method further comprises:
detecting whether search vectors with the same classification attribute exist in the target computing node and are stored in different loading areas;
if the search vectors with the same classification attributes are stored in different loading areas, extracting the search vectors with the same classification attributes in the different loading areas, and merging the search vectors with the same classification attributes into the same loading area.
4. The method of claim 3, wherein the step of merging the search vectors having the same classification attribute into the same load region comprises:
searching a target loading area of which the residual storage space meets the requirement of occupying a storage space, wherein the occupied storage space is the total storage space required by the search vectors with the same classification attributes in the different loading areas;
and migrating the search vectors with the same classification attributes in the different loading areas to the target loading area for storage.
5. The method of claim 1, wherein the step of comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector comprises:
traversing the search vectors with the same classification attributes as the vectors to be searched in the target computing nodes, and calculating the similarity between the vectors to be searched and the search vectors;
and screening the search vector with the similarity reaching a similarity threshold value to obtain the target search vector.
6. The method of claim 1, wherein the method further comprises:
deleting the search vector in the target computing node, and judging whether the search vector has an original image characteristic value corresponding to the search vector in the target database node;
if the original image characteristic value corresponding to the search vector does not exist, directly deleting the search vector in the target computing node;
and if the original image characteristic value corresponding to the search vector exists, deleting the corresponding original image characteristic value in the database node and the search vector in the target calculation node in sequence.
7. An apparatus for searching image feature values, applied to a search system, wherein the search system includes a plurality of storage bodies, each storage body includes at least one database node and at least one computing node corresponding to the at least one database node, the apparatus includes:
the device comprises a first acquisition module, a second acquisition module and a search module, wherein the first acquisition module is used for acquiring a search request which comprises a vector to be searched corresponding to a characteristic value of an image to be searched;
the searching module is used for inquiring a target database node with the classification attribute in each database node of each storage main body according to the classification attribute of the vector to be searched;
a loading module, configured to load a search vector of an original image feature value corresponding to the classification attribute in the target database node to a target computing node corresponding to the target database node, where the target computing node is a computing node corresponding to the target database node in multiple computing nodes;
the comparison module is used for comparing the vector to be searched with the search vector in the target computing node to obtain a target search vector;
and the second acquisition module is used for acquiring the original image characteristic value corresponding to the target search vector in the target database node.
8. The apparatus for searching for image feature values according to claim 7, wherein the target computing node includes a plurality of loading areas, the loading module including:
an obtaining unit, configured to obtain a remaining storage space corresponding to a loading area in the target database node, where a search vector of an original image feature value corresponding to the classification attribute has the same classification attribute, and determine whether the remaining storage space satisfies a required loading space of the search vector;
a loading unit, configured to load the search vector to a loading area corresponding to the remaining storage space if the remaining storage space satisfies a loading space required by the search vector;
and the first query unit is used for querying whether the residual storage spaces of the rest classification attributes in the target computing node meet the required loading space of the search vector if the residual storage spaces do not meet the required loading space of the search vector, and loading the search vector if the residual storage spaces meet the required loading space of the search vector.
9. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, the processor implementing the steps in the image feature value search method according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, characterized in that a computer program is stored thereon, which computer program, when being executed by a processor, carries out the steps in the image feature value search method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911402366.0A CN111159443B (en) | 2019-12-31 | 2019-12-31 | Image characteristic value searching method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911402366.0A CN111159443B (en) | 2019-12-31 | 2019-12-31 | Image characteristic value searching method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111159443A true CN111159443A (en) | 2020-05-15 |
| CN111159443B CN111159443B (en) | 2022-03-25 |
Family
ID=70559455
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911402366.0A Active CN111159443B (en) | 2019-12-31 | 2019-12-31 | Image characteristic value searching method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111159443B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112650775A (en) * | 2020-12-30 | 2021-04-13 | 深圳云天励飞技术股份有限公司 | Data searching method and device, electronic equipment and storage medium |
| CN115510089A (en) * | 2022-11-15 | 2022-12-23 | 以萨技术股份有限公司 | Vector feature comparison method, electronic equipment and storage medium |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050050028A1 (en) * | 2003-06-13 | 2005-03-03 | Anthony Rose | Methods and systems for searching content in distributed computing networks |
| US20060020597A1 (en) * | 2003-11-26 | 2006-01-26 | Yesvideo, Inc. | Use of image similarity in summarizing a collection of visual images |
| CN103020138A (en) * | 2012-11-22 | 2013-04-03 | 江苏乐买到网络科技有限公司 | Method and device for video retrieval |
| WO2017020741A1 (en) * | 2015-08-06 | 2017-02-09 | 阿里巴巴集团控股有限公司 | Method, apparatus and system for retrieving image, acquiring image information and identifying image |
| CN107665180A (en) * | 2016-07-27 | 2018-02-06 | 阿里巴巴集团控股有限公司 | Decomposing system and the method for handling request |
| CN108304431A (en) * | 2017-06-14 | 2018-07-20 | 腾讯科技（深圳）有限公司 | A kind of image search method and device, equipment, storage medium |
| WO2018134964A1 (en) * | 2017-01-20 | 2018-07-26 | 楽天株式会社 | Image search system, image search method, and program |
| CN108520222A (en) * | 2018-04-02 | 2018-09-11 | 惠州学院 | Identification identification method, device, storage medium and electronic equipment based on face's intelligent scanning |
| CN108664514A (en) * | 2017-03-31 | 2018-10-16 | 阿里巴巴集团控股有限公司 | A kind of image search method, server and storage medium |
| CN110019876A (en) * | 2017-12-25 | 2019-07-16 | 深圳云天励飞技术有限公司 | Data query method, electronic equipment and storage medium |
| CN110119405A (en) * | 2019-03-28 | 2019-08-13 | 江苏瑞中数据股份有限公司 | Distributed parallel database method for managing resource |
| CN110321448A (en) * | 2019-06-27 | 2019-10-11 | 腾讯科技（深圳）有限公司 | A kind of image search method, device and storage medium |
| CN110489585A (en) * | 2019-07-08 | 2019-11-22 | 南京邮电大学 | Distributed image searching method based on supervised learning |
-
2019
- 2019-12-31 CN CN201911402366.0A patent/CN111159443B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050050028A1 (en) * | 2003-06-13 | 2005-03-03 | Anthony Rose | Methods and systems for searching content in distributed computing networks |
| US20060020597A1 (en) * | 2003-11-26 | 2006-01-26 | Yesvideo, Inc. | Use of image similarity in summarizing a collection of visual images |
| CN103020138A (en) * | 2012-11-22 | 2013-04-03 | 江苏乐买到网络科技有限公司 | Method and device for video retrieval |
| WO2017020741A1 (en) * | 2015-08-06 | 2017-02-09 | 阿里巴巴集团控股有限公司 | Method, apparatus and system for retrieving image, acquiring image information and identifying image |
| CN107665180A (en) * | 2016-07-27 | 2018-02-06 | 阿里巴巴集团控股有限公司 | Decomposing system and the method for handling request |
| WO2018134964A1 (en) * | 2017-01-20 | 2018-07-26 | 楽天株式会社 | Image search system, image search method, and program |
| US20190205331A1 (en) * | 2017-01-20 | 2019-07-04 | Rakuten, Inc. | Image search system, image search method, and program |
| CN108664514A (en) * | 2017-03-31 | 2018-10-16 | 阿里巴巴集团控股有限公司 | A kind of image search method, server and storage medium |
| CN108304431A (en) * | 2017-06-14 | 2018-07-20 | 腾讯科技（深圳）有限公司 | A kind of image search method and device, equipment, storage medium |
| CN110019876A (en) * | 2017-12-25 | 2019-07-16 | 深圳云天励飞技术有限公司 | Data query method, electronic equipment and storage medium |
| CN108520222A (en) * | 2018-04-02 | 2018-09-11 | 惠州学院 | Identification identification method, device, storage medium and electronic equipment based on face's intelligent scanning |
| CN110119405A (en) * | 2019-03-28 | 2019-08-13 | 江苏瑞中数据股份有限公司 | Distributed parallel database method for managing resource |
| CN110321448A (en) * | 2019-06-27 | 2019-10-11 | 腾讯科技（深圳）有限公司 | A kind of image search method, device and storage medium |
| CN110489585A (en) * | 2019-07-08 | 2019-11-22 | 南京邮电大学 | Distributed image searching method based on supervised learning |
Non-Patent Citations (3)
| Title |
|---|
| ALBERT GORDO 等: ""Deep Image Retrieval: Learning Global Representations for Image Search"", 《HTTPS://LINKSPRINGER.FENSHISHANG.COM/CHAPTER/10.1007/978-3-319-46466-4_15》 * |
| 张运超等: "一种融合重力信息的快速海量图像检索方法", 《自动化学报》 * |
| 李高平: "自适应搜索的快速分形图像编码算法", 《小型微型计算机系统》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112650775A (en) * | 2020-12-30 | 2021-04-13 | 深圳云天励飞技术股份有限公司 | Data searching method and device, electronic equipment and storage medium |
| CN112650775B (en) * | 2020-12-30 | 2024-01-05 | 深圳云天励飞技术股份有限公司 | Data searching method and device, electronic equipment and storage medium |
| CN115510089A (en) * | 2022-11-15 | 2022-12-23 | 以萨技术股份有限公司 | Vector feature comparison method, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111159443B (en) | 2022-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109189991B (en) | Duplicate video identification method, device, terminal and computer readable storage medium | |
| CN110321448B (en) | Image retrieval method, device and storage medium | |
| CN108334632B (en) | Entity recommendation method and device, computer equipment and computer-readable storage medium | |
| CN108288208B (en) | Display object determination method, device, medium and equipment based on image content | |
| CN111159443B (en) | Image characteristic value searching method and device and electronic equipment | |
| CN110147455A (en) | A kind of face matching retrieval device and method | |
| CN108376064B (en) | Rule engine system and related method of rule engine | |
| CN112416960A (en) | Data processing method, device and equipment under multiple scenes and storage medium | |
| CN110580508A (en) | video classification method and device, storage medium and mobile terminal | |
| CN113326810A (en) | Face recognition method, system, device, electronic equipment and storage medium | |
| CN111949354B (en) | Page content updating method and device | |
| US11429660B2 (en) | Photo processing method, device and computer equipment | |
| US11544276B2 (en) | Search device, method and program recording medium | |
| US20190266191A1 (en) | Graph partitioning method and apparatus | |
| CN108170664B (en) | Key word expansion method and device based on key words | |
| CN112364199B (en) | Picture searching system | |
| US11106737B2 (en) | Method and apparatus for providing search recommendation information | |
| CN112433779A (en) | Application site preloading method, device and storage medium based on ERP system | |
| CN111860655A (en) | User processing method, device and equipment | |
| CN113094530B (en) | Image data retrieval method and device, electronic equipment and storage medium | |
| JP2015097036A (en) | Recommended image presentation apparatus and program | |
| CN112214627A (en) | Search method, readable storage medium and electronic device | |
| JP7180769B2 (en) | Data management device, control method, and storage medium | |
| CN111090370A (en) | Picture management method and device and computer readable storage medium | |
| CN110750661B (en) | Method, device, computer equipment and storage medium for searching image |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 518000 1st floor, building 17, Shenzhen Dayun software Town, 8288 Longgang Avenue, Yuanshan street, Longgang District, Shenzhen City, Guangdong Province Applicant after: Shenzhen Yuntian lifeI Technology Co., Ltd Address before: 518000 1st floor, building 17, Shenzhen Dayun software Town, 8288 Longgang Avenue, Yuanshan street, Longgang District, Shenzhen City, Guangdong Province Applicant before: SHENZHEN INTELLIFUSION TECHNOLOGIES Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |