CN109145139A - A kind of image search method and device, equipment and storage medium - Google Patents
A kind of image search method and device, equipment and storage medium Download PDFInfo
- Publication number
- CN109145139A CN109145139A CN201811116598.5A CN201811116598A CN109145139A CN 109145139 A CN109145139 A CN 109145139A CN 201811116598 A CN201811116598 A CN 201811116598A CN 109145139 A CN109145139 A CN 109145139A
- Authority
- CN
- China
- Prior art keywords
- image set
- image
- retrieved
- feature vector
- eigenvector
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The embodiment of the present application provides a kind of image search method and device, equipment and storage medium, wherein determines the corresponding first eigenvector of image set to be retrieved and the corresponding second feature vector of bottom library image set；Determine the concordance list for carrying the space distribution information of the second feature vector；According to the first eigenvector and the concordance list retrieve in the image set of the bottom library with the image that matches in the image set to be retrieved.
Description
Technical field
The invention relates to the computer vision communications field, a kind of image search method and dress are related to, but are not limited to
It sets, equipment and storage medium.
Background technique
Pedestrian retrieval task shows the inquiry picture of a pedestrian out, needs to search out in a large-scale data set
All pictures of the same person.Solving this problem mainly will extract a feature vector to each pedestrian's picture, utilize
This feature vector distinguishes different pedestrians.
The test of usual pedestrian retrieval carries out as follows: providing image set to be retrieved (probe) and bottom library image set
(gallery) two image sets have recorded the picture and corresponding identity number (Identity, ID) of different people.
Pedestrian retrieval network is all passed through to every picture in probe and gallery in test and extracts feature vector, by probe
In the vector space that constitutes of vector of each vector in gallery in scan for, by search result according to two to
The similarity of amount is ranked up, and ranking is more forward to mean that two vectors are more close, i.e. people on two pictures is the same person
Probability it is bigger.
Traditional test method can sequentially read all images in two image sets, and image is sent into graphics processor
(Graphics Processing Unit, GPU) extracts the feature of image by network, and feature deposit is similar lightening interior
In the database for depositing mapping database (Lightning Memory-Mapped Database, LMMD), scan for so multiple
Miscellaneous degree is higher, influences the speed of search.
Summary of the invention
In view of this, the embodiment of the present application provides a kind of image search method and device, equipment and storage medium.
The technical solution of the embodiment of the present application is achieved in that
The embodiment of the present application provides a kind of image search method, which comprises
Determine the corresponding first eigenvector of image set to be retrieved and the corresponding second feature vector of bottom library image set；
Determine the concordance list for carrying the space distribution information of the second feature vector；
According to the first eigenvector and the concordance list retrieve in the image set of the bottom library with the figure to be retrieved
The image to match in image set.
In the embodiment of the present application, the corresponding first eigenvector of determination image set to be retrieved and bottom library image set pair
The second feature vector answered, comprising:
It is extracted respectively from the image set to be retrieved and bottom library image set using preset M neural network model
The first eigenvector and the second feature vector；Wherein, M is the integer greater than 1；The preset M neural network
Model, which corresponds, to be carried in M video cards of graphics processor (Graphics Processing Unit, GPU).
In the embodiment of the present application, described to use preset M neural network model from the image set to be retrieved and institute
It states and extracts the first eigenvector and the second feature vector in the image set of bottom library respectively, comprising:
The image set to be retrieved is divided by M sub image set to be retrieved using the GPU comprising M video card, and by institute
It states bottom library image set and is divided into M sub- bottom library image sets；
Using the preset M neural network model in the M video cards of the GPU respectively correspondingly from the M
The first eigenvector and the second feature vector are extracted in sub image set to be retrieved and the M sub- bottom library image sets.
In the embodiment of the present application, the concordance list of the determining space distribution information for carrying the second feature vector,
Include:
The second feature vector is inputted into preset search library, obtains the corresponding index column of the second feature vector
Table.
In the embodiment of the present application, bottom library image is retrieved according to the first eigenvector and the index list
The image concentrated and matched in the image set to be retrieved, comprising:
The first eigenvector is inputted into the preset search library, is obtained special with described first in the index list
Levy the first list of vector correlation；
From retrieved in the first list in the image set of the bottom library with the figure that matches in the image set to be retrieved
Picture.
In the embodiment of the present application, it is described from retrieved in the first list in the image set of the bottom library with it is described to be checked
The image to match in rope image set, further includes:
The corresponding additional information of the image set to be retrieved is inputted into the first list, obtain in the first list with
The relevant second list of the additional information；Wherein, the additional information is for characterizing image in the image set to be retrieved
Feature；
From retrieved in the second list in the image set of the bottom library with the figure that matches in the image set to be retrieved
Picture.
In the embodiment of the present application, using preset M neural network model from the image set to be retrieved and described
After the first eigenvector and the second feature vector are extracted in the image set of bottom library respectively, the method also includes:
The first eigenvector and the second feature vector are stored in different npy in the form of binary respectively
In file.
The embodiment of the present application provides a kind of image retrieving apparatus, and described device includes: the first determining module, the second determining mould
Block and the first retrieval module, in which:
First determining module, for determining the corresponding first eigenvector of image set to be retrieved and bottom library image set pair
The second feature vector answered；
Second determining module, for determining the concordance list for carrying the space distribution information of the second feature vector；
First retrieval module, for retrieving bottom library figure according to the first eigenvector and the concordance list
In image set with the image that matches in the image set to be retrieved.
In the embodiment of the present application, first determining module, comprising:
First extracting unit, for using preset M neural network model from the image set to be retrieved and the bottom
The first eigenvector and the second feature vector are extracted in the image set of library respectively；Wherein, M is the integer greater than 1；It is described
Preset M neural network model, which corresponds, to be carried in the M video cards of graphics processor GPU.
In the embodiment of the present application, first extracting unit, comprising:
First segmentation subelement, for the image set to be retrieved to be divided into M son using the GPU comprising M video cards
Image set to be retrieved, and bottom library image set is divided into M sub- bottom library image sets；
First extracts subelement, for the preset M neural network model difference in the M video cards using the GPU
The first eigenvector is extracted from described M sub image set to be retrieved and the M sub- bottom library image sets correspondingly
With the second feature vector.
In the embodiment of the present application, second determining module, comprising:
First input unit obtains the second feature for the second feature vector to be inputted preset search library
The corresponding index list of vector.
In the embodiment of the present application, first retrieval module, comprising:
Second input unit obtains the index for the first eigenvector to be inputted the preset search library
First list relevant to the first eigenvector in list；
First retrieval unit, for from retrieved in the first list in the image set of the bottom library with the figure to be retrieved
The image to match in image set.
In the embodiment of the present application, first retrieval module, further includes:
Third input unit, for obtaining the corresponding additional information input of the image set to be retrieved first list
The second list relevant to the additional information into the first list；Wherein, the additional information for characterize it is described to
Retrieve the feature of image in image set；
Second retrieval unit, for from retrieved in the second list in the image set of the bottom library with the figure to be retrieved
The image to match in image set.
In the embodiment of the present application, described device further include:
First preserving module, for respectively by the first eigenvector and the second feature vector with binary shape
Formula is stored in different npy files.
The present embodiment provides a kind of computer storage medium, it is executable that computer is stored in the computer storage medium
Instruction, after which is performed, can be realized the step in image search method provided by the embodiments of the present application
Suddenly.
The present embodiment provides a kind of computer equipment, the computer equipment includes memory and processor, the storage
Computer executable instructions are stored on device, the processor can be real when running the computer executable instructions on the memory
Step in existing image search method provided by the embodiments of the present application.
The embodiment of the present application provides a kind of image search method and device, equipment and storage medium, wherein firstly, determining
The corresponding first eigenvector of image set to be retrieved and the corresponding second feature vector of bottom library image set；Then, it is determined that carrying institute
State the concordance list of the space distribution information of second feature vector；Finally, being examined according to the first eigenvector and the concordance list
Rope go out in the image set of the bottom library with the image that matches in the image set to be retrieved；In this way, not losing measuring accuracy
In the case of, it accelerates the test speed in common magnitude bottom library and realizes the test in the bottom library of ten million magnitude.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and those figures show meet this public affairs
The embodiment opened, and together with specification it is used to illustrate the technical solution of the disclosure.
Figure 1A is the composed structure schematic diagram of the embodiment of the present application network architecture；
Figure 1B is the implementation process schematic diagram of the embodiment of the present application image search method；
Fig. 2A is the another implementation process schematic diagram of the embodiment of the present application image search method；
Fig. 2 B is the application scenarios schematic diagram of the embodiment of the present application image search method；
Fig. 3 is the composed structure schematic diagram of the embodiment of the present application fish eye images processing unit；
Fig. 4 is the composed structure schematic diagram of the embodiment of the present application computer equipment.
Specific embodiment
To keep the purposes, technical schemes and advantages of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application
In attached drawing, the specific technical solution of invention is described in further detail.Following embodiment does not have to for illustrating the application
To limit scope of the present application.
The present embodiment first provides a kind of network architecture, and Figure 1A is that the composed structure of the embodiment of the present application network architecture is illustrated
Figure, as shown in Figure 1A, which includes two or more computer equipments 11 to 1N and server 31, and wherein computer is set
It is interacted between standby 11 to 1N and server 31 by network 21.Computer equipment can be various types of during realization
The computer equipment with information processing capability of type, for example, the computer equipment may include mobile phone, it is tablet computer, desk-top
Machine, personal digital assistant, navigator, digital telephone, television set etc..
The present embodiment proposes a kind of image search method, can effectively solve the problem that image retrieval computation complexity is high and right
When the bottom library of big data magnitude is retrieved, the problems such as taking a long time, this method is applied to computer equipment, and this method institute is real
Existing function can realize that certain program code can be stored in by the processor caller code in computer equipment
In computer storage medium, it is seen then that the computer equipment includes at least pocessor and storage media.
Figure 1B is the implementation process schematic diagram of the embodiment of the present application image search method, as shown in Figure 1, the method includes
Following steps:
Step S101 determines the corresponding first eigenvector of image set to be retrieved and the corresponding second feature of bottom library image set
Vector.
Here, the image library to be retrieved (probe) can be understood as user and want to carry out in big data image set
The image library of retrieval, for example, the photo that user wants to retrieve oneself from a mobile phone photo album has several, then the photograph of oneself
Piece is image library to be retrieved, and photograph album is bottom library image set (gallery).
Step S102 determines the concordance list for carrying the space distribution information of the second feature vector.
Here, in the embodiment of the present application, determine that the spatial distribution for carrying the second feature vector is believed using the library faiss
The concordance list of breath.It is to be understood that determining that each feature (or each image) is in space in the image set of bottom library using the library faiss
In distribution situation can be quickly from index after inputting the library faiss in order to the subsequent first eigenvector by probe
The first list of image relevant to first eigenvector is inquired in table.
Step S103, according to the first eigenvector and the concordance list retrieve in the image set of the bottom library with it is described
The image to match in image set to be retrieved.
Here, first eigenvector is input in the library faiss, so that the library faiss is being indexed according to first eigenvector
Inquired in table in the image set of the bottom library with the image that matches in the image set to be retrieved.
In a kind of image search method provided by the embodiments of the present application, by first getting the feature vector of image set,
Then the corresponding concordance list of feature vector is obtained, from being determined in the image set of bottom library and in the image set to be retrieved in concordance list
The image to match；In this way, the speed of image characteristics extraction is not only accelerated, but also in the case where not losing measuring accuracy,
It accelerates the test speed in common magnitude bottom library and realizes the test in the bottom library of ten million magnitude.
In other embodiments, the method also includes:
The first eigenvector and the second feature vector are stored in different npy in the form of binary respectively
In file.
Here, first eigenvector and second feature vector are saved in the form of binary, GPU is utilized, floating-point is transported
The speed advantage of calculation can obviously accelerate search speed when scanning on index.
In other embodiments, the step 101 determines the corresponding first eigenvector of image set to be retrieved and bottom library
The corresponding second feature vector of image set, comprising:
It is extracted respectively from the image set to be retrieved and bottom library image set using preset M neural network model
The first eigenvector and the second feature vector.
Here, M is the integer greater than 1；The preset M neural network model one-to-one correspondence is carried on graphics processor
In the M of GPU video cards.
It in the present embodiment, include M neural network model from image set to be retrieved and institute using M video cards in GPU
It states bottom library image set and extracts first eigenvector and second feature vector, in this way, extract feature in such a way that multithreading blocks more
Vector greatly accelerates the speed of image characteristics extraction.
The embodiment of the present application provides a kind of image search method, and Fig. 2A is the another of the embodiment of the present application image search method
Implementation process schematic diagram the described method comprises the following steps as shown in Figure 2 A:
The image set to be retrieved is divided into M son image to be retrieved using the GPU comprising M video cards by step S201
Collection, and bottom library image set is divided into M sub- bottom library image sets.
Here, image set to be retrieved and bottom library image set are divided into and M mind using the GPU comprising M video cards
Through the one-to-one M son of network model (equally corresponding to M video card) image set to be retrieved and M sub- bottom library image sets.Than
As said, it is to be retrieved that image set to be retrieved and bottom library image set are divided by 8 sons using the GPU comprising 8 video cards respectively
Image set and 8 sub- bottom library image sets.
When bottom library image set includes the image of ten million magnitude, for example in security system, there is a suspect at present,
The image of the suspect is image to be searched, and 50,000,000 images in public security office system are bottom library image set, if adopted
With only include a video card GPU, then can not support the search of such magnitude due to the limitation of the factors such as video memory, memory.
In the present embodiment, using the GPU comprising multiple video cards, bottom library image set is divided into multiple sub- bottom library image sets, for example, GPU
In include 8 bracing cable cards, then bottom library image set is divided into 8 sub- bottom library image sets, then, while from this 8 sub- bottom library images
The time scanned for is concentrated, is equivalent in 1/8th scanned for originally using the GPU of single deck tape-recorder single thread, and due to
Speed advantage of the GPU to floating-point operation itself is so scanned in the image set of bottom library using the GPU comprising multiple video cards
When, it can obviously accelerate search speed.
Step S202, it is one-to-one using the preset M neural network model difference in the M video cards of the GPU
The first eigenvector and described second are extracted from described M sub image set to be retrieved and the M sub- bottom library image sets
Feature vector.
Here, in GPU, a video card comprising a neural network model handles a subset, is thus M aobvious
Card is concentrated from M son extract feature vector simultaneously, in this way, using M subset of extraction in the M video cards of the GPU feature to
The time of amount is exactly the time that feature vector is extracted from a subset, greatly plus the block speed of characteristic vector pickup.
The second feature vector is inputted preset search library, it is corresponding to obtain the second feature vector by step S203
Index list.
Here, the corresponding second feature vector of gallery is first inputted into preset search library, obtained each in gallery
The space distribution situation of the feature vector of image, i.e. concordance list.The preset search library can be the library faiss.
The first eigenvector is inputted the preset search library by step S204, obtain in the index list with
The relevant first list of the first eigenvector.
Here, the corresponding first eigenvector of probe is input in the library faiss, then the library faiss is looked into from concordance list
Image relevant to first eigenvector is ask, first list is obtained.
Step S205, from retrieved in the first list in the image set of the bottom library with phase in the image set to be retrieved
Matched image.
In image search method provided by the embodiments of the present application, by the way that image set is divided into M subset, then with multiple
Video card extracts the feature vector that M son is concentrated simultaneously and obtains and phase in image set to be retrieved then in being input to the library faiss
Matched image, as shown in Figure 2 B, it is the image 21 comprising two personages that image set to be retrieved, which is added, first by bottom library image set
22 corresponding feature vectors are input in the library faiss 23, obtain the corresponding space distribution situation of this feature vector；Then by image
21 corresponding feature vectors are input to the library faiss 23, obtain first list relevant to image 21 in library image set 22 on earth；Finally
Additional information relevant to image 21 is input to the library faiss 23, the personage for including with image 21 is filtered out from first list
It is the image of the same person, i.e. image 24；So greatly accelerate the speed of characteristic vector pickup measuring for ten million to be
The bottom library searching of grade does not need to expend longer time.
In other embodiments, after the step S204, the method also includes following steps:
The corresponding additional information of the image set to be retrieved is inputted the first list by step S241, obtains described the
Second list relevant to the additional information in one list.
Here, the additional information is used to characterize the feature of image in the image set to be retrieved.
Step S242, from retrieved in the second list in the image set of the bottom library with phase in the image set to be retrieved
Matched image.
In the present embodiment, after obtaining first list, continue other additional informations of image in image set to be retrieved
It is input to the library faiss, the library faiss is further filtered out from first list and the more matched figure of image to be retrieved
Picture, to keep the result retrieved more accurate.
In the related art extract feature when can only single deck tape-recorder single thread extract, inefficiency, search when can only be in probe
In one by one feature scan for, and the mathematical characteristic of feature itself is not utilized to optimize, is most simple complexity, ten
Divide slow.When gallery picture reaches ten million magnitude, due to the limitation of the factors such as video memory, memory, such magnitude can not be supported
Search.Based on this, the embodiment of the present application proposes a kind of image search method, extracts spy using the mode that multithreading blocks more
Sign, improves efficiency.Aspect indexing is established to bottom library and GPU accelerates, search speed is greatly improved, by 1,000,000 magnitudes
Search on the library of bottom accelerates to 3 minutes or so, and the search for carrying out 1,000,000 magnitudes in the related art may 3 to 5 hours.
In addition, storing algorithm, more card search techniques using Feature Compression in the present embodiment, make the search in 20,000,000 grades of bottom library can be with
It carries out, time-consuming about 1 to 2 hour.For example, in security protection scene, suspect one criminal gang comprising 5 people, then searching
Collect the photo of this five people as the library probe, is then input to the image data base of public security bureau itself (i.e. gallery)
In the library faiss, the concordance list of the space distribution information comprising the corresponding feature vector of image each in gallery is obtained, then
Probe is inputted into the library faiss, first list relevant to 5 suspects are set in the concordance list is obtained, then again by this
In other additional informations input library faiss of 5 people, further screen the first list, obtain in gallery with this 5
All information of image and this 5 people that people matches, to help quickly to chase.
The method can be realized by following steps:
The first step extracts the feature vector in probe and gallery, and saves as binary file.
Here, in the present embodiment, probe and gallery are divided into M son identical with M video card quantity first
Collection；Secondly, extracting probe and gallery feature vector simultaneously with M video cards, and binary file is saved as, extracted
Binary file is converted to again after probe and gallery feature vector (i.e. first eigenvector and second feature vector)
Single npy file is saved.
Second step determines the concordance list of second feature vector using the library faiss.
Here, GPU index is constructed using the library faiss to vector first in test, vector is utilized spatially in index
Distributed intelligence, establish Nearest Neighbor Search, and speed advantage of the GPU to floating-point operation is utilized, searched on index
Search speed can obviously be accelerated when rope.
Here, in order to better understand the present embodiment, the library faiss is explained, the library faiss be for it is dense to
Amount provides the frame of efficient similarity search and cluster.It has characteristics that 1, a variety of search methods is provided；2, speed is fast；3,
It may be present in memory and disk；4, C++ is realized, is provided Python encapsulation and is called；5, most of algorithm supports GPU to realize.
Third step detects the retrieval rate of image search method in the present embodiment using a variety of preset testing algorithms.
Here, a variety of preset testing algorithms can be average exact value (Mean Average Precision,
MAP) algorithm, preceding K algorithm (topK) and accumulation match curve (Cumulative Match Characteristic Curve,
CMC) etc..Using different testing algorithms, tests the image search method that the present embodiment proposes and carry out the speed that image searches element, it is real
Test the speed the result shows that the speed ratio image retrieval in the related technology of image search method provided in this embodiment retrieval image
Fast 10 times of method or so.
4th step retrieves the bottom library of ten million magnitude by the way of compression storage.
Here, when encountering the bottom library of ten million magnitude, the present embodiment takes compression to deposit vector while establishing index
The mode of storage can complete the foundation to the index of 20,000,000 256 dimensional feature vectors on single machine (8 GPU), and complete
Relevant search and test.
In the present embodiment, GPU is utilized and multithreading accelerates the extraction and search of feature, greatly accelerates feature
The time of extraction；Establishing index using the spatial relationship of vector reduces the complexity of search, so that not losing measuring accuracy
In the case of, accelerate the test speed in common magnitude bottom library；And bottom Cooley biggish for data volume is stored with compression algorithm
Vector supports the test in outsole library, to realize the test in the bottom library of ten million magnitude in the case where not losing precision substantially.
The embodiment of the present invention provides a kind of image retrieving apparatus, and Fig. 3 is the composition of the embodiment of the present application image retrieving apparatus
Structural schematic diagram, as shown in figure 3, described image retrieval device 300 includes: the first determining module 301, the second determining module 302
With the first retrieval module 303, in which:
First determining module 301, for determining the corresponding first eigenvector of image set to be retrieved and bottom library image
Collect corresponding second feature vector；
Second determining module 302, for determining the index for carrying the space distribution information of the second feature vector
Table；
First retrieval module 303, for retrieving the bottom according to the first eigenvector and the concordance list
In the image set of library with the image that matches in the image set to be retrieved.
In the embodiment of the present application, first determining module 301, comprising:
First extracting unit, for using preset M neural network model from the image set to be retrieved and the bottom
The first eigenvector and the second feature vector are extracted in the image set of library respectively；Wherein, M is the integer greater than 1；It is described
Preset M neural network model, which corresponds, to be carried in the M video cards of graphics processor GPU.
In the embodiment of the present application, first extracting unit, comprising:
First segmentation subelement, for the image set to be retrieved to be divided into M son using the GPU comprising M video cards
Image set to be retrieved, and bottom library image set is divided into M sub- bottom library image sets；
First extracts subelement, for the preset M neural network model difference in the M video cards using the GPU
The first eigenvector is extracted from described M sub image set to be retrieved and the M sub- bottom library image sets correspondingly
With the second feature vector.
In the embodiment of the present application, second determining module 302, comprising:
First input unit obtains the second feature for the second feature vector to be inputted preset search library
The corresponding index list of vector.
In the embodiment of the present application, first retrieval module 303, comprising:
Second input unit obtains the index list for the first eigenvector to be inputted preset search library
In first list relevant to the first eigenvector；
First retrieval unit, for from retrieved in the first list in the image set of the bottom library with the figure to be retrieved
The image to match in image set.
In the embodiment of the present application, first retrieval module 303, further includes:
Third input unit is used for the corresponding additional information input of the image set to be retrieved preset search
Library obtains second list relevant to the additional information in the first list；Wherein, the additional information is for characterizing institute
State the feature of image in image set to be retrieved；
Second retrieval unit, for from retrieved in the second list in the image set of the bottom library with the figure to be retrieved
The image to match in image set.
In the embodiment of the present application, described device 300 further include:
First preserving module, for respectively by the first eigenvector and the second feature vector with binary shape
Formula is stored in different npy files.
It should be noted that the description of apparatus above embodiment, be with the description of above method embodiment it is similar, have
The similar beneficial effect with embodiment of the method.For undisclosed technical detail in the application Installation practice, this Shen is please referred to
Please embodiment of the method description and understand.
It should be noted that in the embodiment of the present application, if realizing above-mentioned fish-eye image in the form of software function module
As processing method, and when sold or used as an independent product, also can store in a computer-readable storage medium
In.Based on this understanding, the technical solution of the embodiment of the present application substantially the part that contributes to existing technology in other words
It can be embodied in the form of software products, which is stored in a storage medium, including several fingers
It enables and using so that an instant messaging equipment (can be terminal, server etc.) executes each embodiment the method for the application
It is all or part of.And storage medium above-mentioned include: USB flash disk, mobile hard disk, read-only memory (Read Only Memory,
ROM), the various media that can store program code such as magnetic or disk.In this way, the embodiment of the present application is not limited to any spy
Fixed hardware and software combines.
Correspondingly, the embodiment of the present application provides a kind of computer program product again, and the computer program product includes meter
Calculation machine executable instruction after the computer executable instructions are performed, can be realized image retrieval provided by the embodiments of the present application
Step in method.
Correspondingly, the embodiment of the present application provides a kind of computer storage medium again, stores in the computer storage medium
There are computer executable instructions, the described computer executable instructions realize figure provided by the above embodiment when being executed by processor
As the step of search method.
Correspondingly, the embodiment of the present application provides a kind of computer equipment, and Fig. 4 is the group of the embodiment of the present application computer equipment
At structural schematic diagram, as shown in figure 4, the hardware entities of the computer equipment 400 include: processor 401,402 and of communication interface
Memory 403, wherein
The overall operation of the usually control computer equipment 400 of processor 401.
Communication interface 402 can make computer equipment pass through network and other terminals or server communication.
Memory 403 is configured to store the instruction and application that can be performed by processor 401, can also cache device to be processed
401 and computer equipment 400 in each module it is to be processed or processed data (for example, image data, audio data, language
Sound communication data and video communication data), flash memory (FLASH) or random access storage device (Random Access can be passed through
Memory, RAM) it realizes.
The description of above instant computing machine equipment and storage medium embodiment, is similar with the description of above method embodiment
, there is with embodiment of the method similar beneficial effect.For in the application instant messaging equipment and storage medium embodiment not
The technical detail of disclosure please refers to the description of the application embodiment of the method and understands.
It should be understood that " one embodiment " or " embodiment " that specification is mentioned in the whole text mean it is related with embodiment
A particular feature, structure, or characteristic includes at least one embodiment of the application.Therefore, occur everywhere in the whole instruction
" in one embodiment " or " in one embodiment " not necessarily refer to identical embodiment.In addition, these specific features, knot
Structure or characteristic can combine in any suitable manner in one or more embodiments.It should be understood that in the various implementations of the application
In example, magnitude of the sequence numbers of the above procedures are not meant that the order of the execution order, and the execution sequence of each process should be with its function
It can be determined with internal logic, the implementation process without coping with the embodiment of the present application constitutes any restriction.Above-mentioned the embodiment of the present application
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the device that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or device.
In several embodiments provided herein, it should be understood that disclosed device and method can pass through it
Its mode is realized.Apparatus embodiments described above are merely indicative, for example, the division of the unit, only
A kind of logical function partition, there may be another division manner in actual implementation, such as: multiple units or components can combine, or
It is desirably integrated into another system, or some features can be ignored or not executed.In addition, shown or discussed each composition portion
Mutual coupling or direct-coupling or communication connection is divided to can be through some interfaces, the INDIRECT COUPLING of equipment or unit
Or communication connection, it can be electrical, mechanical or other forms.
Above-mentioned unit as illustrated by the separation member, which can be or may not be, to be physically separated, aobvious as unit
The component shown can be or may not be physical unit；Both it can be located in one place, and may be distributed over multiple network lists
In member；Some or all of units can be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
In addition, each functional unit in each embodiment of the application can be fully integrated in one processing unit, it can also
To be each unit individually as a unit, can also be integrated in one unit with two or more units；It is above-mentioned
Integrated unit both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can store in computer-readable storage medium, which exists
When execution, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes: movable storage device, read-only deposits
The various media that can store program code such as reservoir (Read Only Memory, ROM), magnetic or disk.
If alternatively, the above-mentioned integrated unit of the application is realized in the form of software function module and as independent product
When selling or using, it also can store in a computer readable storage medium.Based on this understanding, the application is implemented
Substantially the part that contributes to existing technology can be embodied in the form of software products the technical solution of example in other words,
The computer software product is stored in a storage medium, including some instructions are used so that computer equipment (can be with
Personal computer, server or network equipment etc.) execute each embodiment the method for the application all or part.
And storage medium above-mentioned includes: various Jie that can store program code such as movable storage device, ROM, magnetic or disk
Matter.
The above, the only specific embodiment of the application, but the protection scope of the application is not limited thereto, it is any
Those familiar with the art within the technical scope of the present application, can easily think of the change or the replacement, and should all contain
Lid is within the scope of protection of this application.Therefore, the protection scope of the application should be based on the protection scope of the described claims.
Claims (10)
1. a kind of image search method, which is characterized in that the described method includes:
Determine the corresponding first eigenvector of image set to be retrieved and the corresponding second feature vector of bottom library image set；
Determine the concordance list for carrying the space distribution information of the second feature vector；
According to the first eigenvector and the concordance list retrieve in the image set of the bottom library with the image set to be retrieved
In the image that matches.
2. the method according to claim 1, wherein the corresponding fisrt feature of determination image set to be retrieved to
Measure second feature vector corresponding with bottom library image set, comprising:
Described in being extracted respectively from the image set to be retrieved and bottom library image set using preset M neural network model
First eigenvector and the second feature vector；Wherein, M is the integer greater than 1；The preset M neural network model
It corresponds and is carried in the M video cards of graphics processor GPU.
3. according to the method described in claim 2, it is characterized in that, it is described using preset M neural network model from described
The first eigenvector and the second feature vector are extracted in image set to be retrieved and bottom library image set respectively, is wrapped
It includes:
The image set to be retrieved is divided by M sub image set to be retrieved using the GPU comprising M video card, and by the bottom
Library image set is divided into M sub- bottom library image sets；
Using the preset M neural network model in the M video cards of the GPU respectively correspondingly from described M it is sub to
The first eigenvector and the second feature vector are extracted in retrieval image set and the M sub- bottom library image sets.
4. the method according to claim 1, wherein the determining space for carrying the second feature vector point
The concordance list of cloth information, comprising:
The second feature vector is inputted into preset search library, obtains the corresponding index list of the second feature vector.
5. method according to claim 1 or 4, which is characterized in that according to the first eigenvector and the index column
Table retrieve in the image set of the bottom library with the image that matches in the image set to be retrieved, comprising:
The first eigenvector is inputted into preset search library, obtain in the index list with the first eigenvector phase
The first list of pass；
From retrieved in the first list in the image set of the bottom library with the image that matches in the image set to be retrieved.
6. according to the method described in claim 5, it is characterized in that, described retrieve bottom library figure from the first list
In image set with the image that matches in the image set to be retrieved, further includes:
The corresponding additional information of the image set to be retrieved is inputted into the first list, obtain in the first list with it is described
The relevant second list of additional information；Wherein, the additional information is used to characterize the feature of image in the image set to be retrieved；
From retrieved in the second list in the image set of the bottom library with the image that matches in the image set to be retrieved.
7. according to the method described in claim 2, it is characterized in that, using preset M neural network model from institute described
State extracted respectively in image set to be retrieved and the bottom library image set first eigenvector and the second feature vector it
Afterwards, the method also includes:
The first eigenvector and the second feature vector are stored in different npy files in the form of binary respectively
In.
8. a kind of image retrieving apparatus, which is characterized in that described device includes: the first determining module, the second determining module and
One retrieval module, in which:
First determining module, for determining that the corresponding first eigenvector of image set to be retrieved and bottom library image set are corresponding
Second feature vector；
Second determining module, for determining the concordance list for carrying the space distribution information of the second feature vector；
First retrieval module, for retrieving bottom library image set according to the first eigenvector and the concordance list
In with the image that matches in the image set to be retrieved.
9. a kind of computer storage medium, which is characterized in that be stored with the executable finger of computer in the computer storage medium
It enables, after which is performed, can be realized the described in any item method and steps of claim 1 to 7.
10. a kind of computer equipment, which is characterized in that the computer equipment includes memory and processor, the memory
On be stored with computer executable instructions, can be realized when the processor runs the computer executable instructions on the memory
The described in any item method and steps of claim 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811116598.5A CN109145139B (en) | 2018-09-25 | 2018-09-25 | Image retrieval method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811116598.5A CN109145139B (en) | 2018-09-25 | 2018-09-25 | Image retrieval method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109145139A true CN109145139A (en) | 2019-01-04 |
| CN109145139B CN109145139B (en) | 2021-07-27 |
Family
ID=64823535
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811116598.5A Active CN109145139B (en) | 2018-09-25 | 2018-09-25 | Image retrieval method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109145139B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110659373A (en) * | 2019-08-27 | 2020-01-07 | 浙江大华技术股份有限公司 | Image retrieval method, image retrieval device, computer device and storage medium |
| CN113761241A (en) * | 2021-07-23 | 2021-12-07 | 北京旷视科技有限公司 | Image searching method and device, computer readable storage medium and electronic device |
| CN113780424A (en) * | 2021-09-13 | 2021-12-10 | 四川新网银行股份有限公司 | Real-time online photo clustering method and system based on background similarity |
| CN113868440A (en) * | 2020-06-30 | 2021-12-31 | 华为技术有限公司 | Method, device, equipment and medium for managing feature library |
| WO2023273082A1 (en) * | 2021-07-02 | 2023-01-05 | Zhejiang Dahua Technology Co., Ltd. | Method for constructing feature base library, feature retrieval method, and related device |
| CN116543181A (en) * | 2023-05-10 | 2023-08-04 | 北京微聚智汇科技有限公司 | Anti-partner fraud method and system based on image background feature recognition |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551823A (en) * | 2009-04-20 | 2009-10-07 | 浙江师范大学 | Comprehensive multi-feature image retrieval method |
| CN101576894A (en) * | 2008-05-09 | 2009-11-11 | 中国科学院半导体研究所 | Real-time image content retrieval system and image feature extraction method |
| CN106951484A (en) * | 2017-03-10 | 2017-07-14 | 百度在线网络技术（北京）有限公司 | Picture retrieval method and device, computer equipment and computer-readable medium |
| CN108229358A (en) * | 2017-12-22 | 2018-06-29 | 北京市商汤科技开发有限公司 | Index establishing method and device, electronic equipment, computer storage media, program |
-
2018
- 2018-09-25 CN CN201811116598.5A patent/CN109145139B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101576894A (en) * | 2008-05-09 | 2009-11-11 | 中国科学院半导体研究所 | Real-time image content retrieval system and image feature extraction method |
| CN101551823A (en) * | 2009-04-20 | 2009-10-07 | 浙江师范大学 | Comprehensive multi-feature image retrieval method |
| CN106951484A (en) * | 2017-03-10 | 2017-07-14 | 百度在线网络技术（北京）有限公司 | Picture retrieval method and device, computer equipment and computer-readable medium |
| CN108229358A (en) * | 2017-12-22 | 2018-06-29 | 北京市商汤科技开发有限公司 | Index establishing method and device, electronic equipment, computer storage media, program |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110659373A (en) * | 2019-08-27 | 2020-01-07 | 浙江大华技术股份有限公司 | Image retrieval method, image retrieval device, computer device and storage medium |
| CN113868440A (en) * | 2020-06-30 | 2021-12-31 | 华为技术有限公司 | Method, device, equipment and medium for managing feature library |
| CN113868440B (en) * | 2020-06-30 | 2023-06-27 | 华为技术有限公司 | Feature library management method, device, equipment and medium |
| WO2023273082A1 (en) * | 2021-07-02 | 2023-01-05 | Zhejiang Dahua Technology Co., Ltd. | Method for constructing feature base library, feature retrieval method, and related device |
| CN113761241A (en) * | 2021-07-23 | 2021-12-07 | 北京旷视科技有限公司 | Image searching method and device, computer readable storage medium and electronic device |
| CN113780424A (en) * | 2021-09-13 | 2021-12-10 | 四川新网银行股份有限公司 | Real-time online photo clustering method and system based on background similarity |
| CN116543181A (en) * | 2023-05-10 | 2023-08-04 | 北京微聚智汇科技有限公司 | Anti-partner fraud method and system based on image background feature recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109145139B (en) | 2021-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109145139A (en) | A kind of image search method and device, equipment and storage medium | |
| US9036905B2 (en) | Training classifiers for deblurring images | |
| CN109871490B (en) | Media resource matching method and device, storage medium and computer equipment | |
| CN104094255B (en) | For searching for the method and apparatus of image and computer readable recording medium storing program for performing for performing this method | |
| CN108416028A (en) | A kind of method, apparatus and server of search content resource | |
| US20140133757A1 (en) | Creating social network groups | |
| CN113434716B (en) | Cross-modal information retrieval method and device | |
| CN105956051A (en) | Information finding method, device and system | |
| CN108431800B (en) | Image processing apparatus and display method of image search interface | |
| CN109271545A (en) | A kind of characteristic key method and device, storage medium and computer equipment | |
| CN114741581A (en) | Image classification method and device, computer equipment and medium | |
| CN109241299A (en) | Multimedia resource searching method, device, storage medium and equipment | |
| KR20190124436A (en) | Method for searching building based on image and apparatus for the same | |
| CN115115855A (en) | Training method, device, equipment and medium for image encoder | |
| CN116860916A (en) | Multi-stage text retrieval method, multi-stage text retrieval device, computer equipment and storage medium | |
| CN115909176A (en) | Video semantic segmentation method and device, electronic equipment and storage medium | |
| CN103093213A (en) | Video file classification method and terminal | |
| Thepade et al. | Identification of aerial image land use using fused thepade sbtc and adaptive thresholding with machinelearning ensemble | |
| KR101170676B1 (en) | Face searching system and method based on face recognition | |
| Sudha et al. | Reducing semantic gap in video retrieval with fusion: A survey | |
| CN113221786A (en) | Data classification method and device, electronic equipment and storage medium | |
| CN112100427A (en) | Video processing method and device, electronic equipment and storage medium | |
| CN112333182A (en) | File processing method, device, server and storage medium | |
| CN111553335A (en) | Image generation method and apparatus, and storage medium | |
| CN111797765A (en) | Image processing method, image processing apparatus, server, 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 |