CN109918532A - Image search method, device, equipment and computer readable storage medium - Google Patents
Image search method, device, equipment and computer readable storage medium Download PDFInfo
- Publication number
- CN109918532A CN109918532A CN201910175051.0A CN201910175051A CN109918532A CN 109918532 A CN109918532 A CN 109918532A CN 201910175051 A CN201910175051 A CN 201910175051A CN 109918532 A CN109918532 A CN 109918532A
- Authority
- CN
- China
- Prior art keywords
- image
- hash
- depth
- coding
- retrieved
- 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
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000013507 mapping Methods 0.000 claims abstract description 61
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000005457 optimization Methods 0.000 claims abstract description 12
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 11
- 238000010801 machine learning Methods 0.000 claims abstract description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 34
- 239000011159 matrix material Substances 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 18
- 238000013135 deep learning Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 238000007689 inspection Methods 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 14
- 230000006870 function Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 3
- 229910002056 binary alloy Inorganic materials 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000001737 promoting effect Effects 0.000 description 2
- 101100289792 Squirrel monkey polyomavirus large T gene Proteins 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000006386 neutralization reaction Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- 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
-
- 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
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Library & Information Science (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of image search method, device, equipment and computer readable storage mediums.Wherein, method includes with the image in image data base to for input, the characteristic similarity of distance, label classification and image pair between the Hash coding pair penetrated using image map uses machine learning optimization algorithm optimization penalty values to obtain depth Hash mapping model with training as penalty values;Image to be retrieved is mapped as Hash to be retrieved using depth Hash mapping model to encode;The target image for meeting preset condition with the Hamming distance difference of Hash to be retrieved coding is searched in the Hash code database constructed in advance, search result as image to be retrieved in image data base is exported, Hash code database be by every image in image data base through depth Hash mapping model mapping after gained.The application efficiently solves the problems, such as that same category image Hash coding is excessively consistent in the related technology, to realize the accurate retrieval of same category image.
Description
Technical field
The present embodiments relate to technical field of image processing, more particularly to a kind of image search method, device, equipment
And computer readable storage medium.
Background technique
In recent years, with internet further popularize and big data technology deep application, can all have daily number with
The image of hundred million meters generates.The increase of the collection neutralization scale of image data resource is so that the prior art is increasingly difficult to meet user's figure
As the demand of retrieval.Therefore, the characteristic information for how effectively describing image, using which kind of data structure carry out efficient index and
The problems such as Fast Similarity Retrieval, becomes the research hotspot in this direction.
In face of large-scale image data, there is the property for being easy to compare with storage in view of binary coding, it can be very big
Promotion similarity retrieval speed and save more computer resources, therefore generally image is mapped when carrying out image retrieval
At binary coding.
The appearance of deep learning has pushed the development of computer vision, also provides more effectively for study Hash mapping method
Tool.The relevant technologies neural network model maps Hash coding, and so the training deep learning model in the way of image, leads to
Loss function is crossed to constrain the model parameter, achieves more good result.
But there are such problems for the obtained binary coding of the training: the Hash coding of similar image is excessively consistent,
Thus similar image almost has no discrimination, can not identify and retrieve the more similar image of image, that is to say, that related skill
Art can not retrieve same category image.
Summary of the invention
The embodiment of the present disclosure provides a kind of image search method, device, equipment and computer readable storage medium, realizes
The accurate retrieval of same category image.
In order to solve the above technical problems, the embodiment of the present invention the following technical schemes are provided:
On the one hand the embodiment of the present invention provides a kind of image search method, comprising:
Image to be retrieved Hash to be retrieved is mapped as using the depth Hash mapping model constructed in advance to encode;
It searches in the Hash code database constructed in advance and meets in advance with the Hamming distance difference of the Hash coding to be retrieved
If the target image of condition, using as the image to be retrieved image data base search result;
Wherein, the depth Hash mapping model is with the image in described image database to for input, with the figure
The characteristic similarity of distance, label classification and described image pair between the Hash coding pair that picture obtains mapping is penalty values, is adopted
Optimize the penalty values training gained with machine learning optimization algorithm;The label classification is opened for identifying described image centering two
The similitude of image;The Hash code database is by every image in described image database through the depth Hash mapping model
Gained after mapping.
Optionally, the training process of the depth Hash mapping model includes:
Using the convolutional neural networks model that full articulamentum is Hash coding layer as the net of the depth Hash mapping model
Network structure;
Using the image in described image database to the input as the network structure, using following formula as the net
The penalty values of network structure optimize the penalty values training network structure using stochastic gradient descent:
In formula, Loss is the penalty values, and i-th image and jth image form described image pair, and n is described image
Image total number in database, Si,jFor the depth characteristic similarity of i-th image and jth image, oiFor i-th image
Hash coding, ojHash for jth image encodes, | | oi-oj||2For Hash coding pair between distance, α is hyper parameter,
yi,jFor label classification, yi,j=1 indicates that i-th image and jth image are same category image, yi,j=0 indicates i-th figure
As being different classes of image with jth image.
Optionally, the depth characteristic similarity value of i-th image and jth image can be from the similarity precalculated
Matrix obtains, the similarity matrix calculating process are as follows:
It is raw using the depth characteristic of each image in the convolutional neural networks model extraction described image database constructed in advance
At depth characteristic vector set;
The Euclidean distance between vector two-by-two is calculated separately in the depth characteristic vector set, described image data are generated
The similarity matrix in library;
Wherein, the convolutional neural networks are to utilize image training gained in described image database based on deep learning.
Optionally, the calculating process of the depth characteristic similarity value of i-th image and jth image are as follows:
The depth characteristic similarity value of i-th image and jth image is calculated using following formula:
In formula, fiFor the depth characteristic vector of i-th image, fjFor the depth characteristic vector of jth image, yi,j=1 table
Show that i-th image and jth image are same category image, yi,j=0 i-th image of expression is different classes of with jth image
Image, α, β are hyper parameter.
Optionally, the convolutional neural networks model is VGG-16 network model.
Optionally, the generating process of the Hash code database are as follows:
Every image in described image database is input to the depth Hash mapping model, by the way that threshold value is arranged,
The output of the Hash coding layer of the depth Hash mapping model is mapped as Hash coding;
The Hash code database is generated according to the Hash of every image coding;
Wherein, m codings of i-th image of described image databaseSuch as following formula:
In formula,For the m output in Hash coding layer of i-th image, θ is the threshold value.
Optionally, the Hamming distance searched in the Hash code database constructed in advance with the Hash coding to be retrieved
Difference meets the target image of preset condition are as follows:
It is searched in the Hash code database and is worth the smallest preceding T figure with the Hamming distance of the Hash coding to be retrieved
Picture;
T images are ranked up from small to large according to the Hamming distance difference with the Hash coding to be retrieved;
T images after output sequence.
On the other hand the embodiment of the present invention provides a kind of image retrieving apparatus, comprising:
Model training module, for, to for input, being obtained with described image to mapping with the image in image data base
The characteristic similarity of distance, label classification and described image pair between Hash coding pair is penalty values, is optimized using machine learning
Penalty values described in algorithm optimization obtain depth Hash mapping model with training;The label classification is for identifying described image centering
The similitude of two images;
Hash encodes generation module, to be retrieved for being mapped as image to be retrieved using the depth Hash mapping model
Hash coding;
Image retrieval module, for searching the Chinese with the Hash coding to be retrieved in the Hash code database constructed in advance
Prescribed distance difference meets the target image of preset condition, using as the image to be retrieved image data base search result;
The Hash code database be by every image in described image database through the depth Hash mapping model mapping after gained.
The embodiment of the invention also provides a kind of image retrieval apparatus, including processor, the processor is deposited for executing
It is realized when the computer program stored in reservoir such as the step of any one of preceding described image search method.
The embodiment of the present invention finally additionally provides a kind of computer readable storage medium, the computer readable storage medium
On be stored with image retrieval program, realized when described image search program is executed by processor such as any one of preceding described image retrieval
The step of method.
The advantages of technical solution provided by the present application, is, considers respectively to scheme in image data in training Hash mapping model
The similarity as between, more similar between image, the Hamming distance between Hash coding mapped is with regard to smaller, more not phase between image
Seemingly, the Hamming distance between Hash coding mapped is just slightly larger, between the Hash coding for promoting similar image to map
Hamming distance determined according to the similarity between image, efficiently solve in the related technology same category image Hash coding
The problem of excessively unanimously causing similar image that can not retrieve, to realize the accurate retrieval of same category image.
In addition, the embodiment of the present invention provides corresponding realization device, equipment and computer also directed to image search method
Readable storage medium storing program for executing, further such that the method has more practicability, described device, equipment and computer readable storage medium
Have the advantages that corresponding.
It should be understood that the above general description and the following detailed description are merely exemplary, this can not be limited
It is open.
Detailed description of the invention
It, below will be to embodiment or correlation for the clearer technical solution for illustrating the embodiment of the present invention or the relevant technologies
Attached drawing needed in technical description is briefly described, it should be apparent that, the accompanying drawings in the following description is only this hair
Bright some embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is a kind of flow diagram of image search method provided in an embodiment of the present invention;
Fig. 2 is a kind of disclosure generating process signal of depth Hash mapping model shown according to an exemplary embodiment
Figure;
Fig. 3 is a kind of disclosure VGG-16 schematic network structure shown according to an exemplary embodiment;
Fig. 4 is disclosure another kind VGG-16 schematic network structure shown according to an exemplary embodiment;
Fig. 5 is the generating process schematic diagram of disclosure similarity matrix shown according to an exemplary embodiment;
Fig. 6 is that the binary coding of disclosure technical scheme shown according to an exemplary embodiment is distributed signal
Figure;
Fig. 7 is the binary coding distribution schematic diagram of disclosure the relevant technologies shown according to an exemplary embodiment;
Fig. 8 is a kind of specific embodiment structure chart of image retrieving apparatus provided in an embodiment of the present invention.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.Obviously, described embodiments are only a part of the embodiments of the present invention, rather than
Whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
The description and claims of this application and term " first ", " second ", " third " " in above-mentioned attached drawing
Four " etc. be for distinguishing different objects, rather than for describing specific sequence.Furthermore term " includes " and " having " and
Their any deformations, it is intended that cover and non-exclusive include.Such as contain a series of steps or units process, method,
System, product or equipment are not limited to listed step or unit, but may include the step of not listing or unit.
After describing the technical solution of the embodiment of the present invention, the various non-limiting realities of detailed description below the application
Apply mode.
Referring first to Fig. 1, Fig. 1 is a kind of flow diagram of image search method provided in an embodiment of the present invention, this hair
Bright embodiment may include the following contents:
S101: in advance with the image in image data base between the Hash coding pair for input, penetrated with image map
Distance, label classification and image pair characteristic similarity be penalty values, use machine learning optimization algorithm optimization penalty values with
Training obtains depth Hash mapping model.
Building convolutional neural networks structure first is used for Hash mapping, is then based on image data base and image similarity square
Battle array study obtains the parameters weighting of the convolutional network structure, to complete the depth Hash mapping model structure based on image similarity
It builds, the training process of depth Hash mapping model can be as described below, it please refers to shown in Fig. 2:
Using the convolutional neural networks model that full articulamentum is Hash coding layer as the network knot of depth Hash mapping model
Any convolutional neural networks structure can be used in structure, convolutional neural networks structure, may be, for example, VGG-16 network shown in Fig. 3
Model, VGG-16 network model improve the final performance of entire convolutional neural networks by increasing network depth.Certainly, may be used
For other convolutional network structures, this not realization of image the application.It can be by adding before the last layer of convolutional neural networks
Add one layer of full articulamentum, nodal point number is Hash code length, and as Hash coding layer, the generating process of Hash coding layer can join
Any realization process recorded in the related technology is read just to repeat no more herein.
Image data base is the database of image retrieval to be retrieved and its similar or identical image, which includes a large amount of
Image.
Training depth Hash mapping model when, the comparison of image similarity can be added, so that different classes of image reflects
The Hamming distance between Hash coding penetrated is big as far as possible;Between the Hash coding that same category image maps
Hamming distance determined according to the similarity between image, be embodied as more similar between image, to map Hash coding
Between Hamming distance with regard to smaller;More dissimilar between image, the Hamming distance between Hash coding mapped is with regard to slightly larger.
In consideration of it, can be using the random image in image data base to the input as network structure, image data
Any two images constitute an image pair at random in library, and each image, to label classification, label classification is for identifying
The similitude of this two images of image pair is constituted, if this two images are same category of image, the tag class of image pair
It Wei not be similar;If this two images are not same category of image, the label classification of image pair is dissmilarity.For example,
Image in image data base is expressed as set X={ x1,x2,…,xn, that is, n images are shared in image data base, i-th
Image and the image of jth image construction are to being represented by (xi,xj), with symbol yi,jTo indicate the label of the image, yi,j=
1 indicates that this two images are similar, yi,j=0 indicates this two image dissmilarities.
The depth characteristic of every image is extracted, can judge two according to the depth characteristic for two images for constituting image pair
Whether image has similitude.When extracting the depth characteristic of image, any convolutional neural networks model can be used, such as
See the network structure that VGG-16 network shown in Fig. 4 is used as, by the feature that convolutional layer and pond layer obtain generally have compared with
High dimension, therefore usually choose the feature vector value of full articulamentum output as depth characteristic, fc7 layers of output is then in Fig. 4
For depth characteristic.
In a kind of specific embodiment, corresponding similarity matrix, image pair can be generated for image data base in advance
Similarity value can directly be obtained from the similarity matrix.The generating process of similarity matrix is seen shown in Fig. 5, and i-th
The depth characteristic similarity value of image and jth image can be obtained from the similarity matrix precalculated, similarity matrix
Calculating process can are as follows:
Using the depth characteristic of each image in the convolutional neural networks model extraction image data base constructed in advance, generate deep
Spend feature vector set.Convolutional neural networks may be based on deep learning and train gained using image in image data base, in order to
The depth characteristic that accuracy rate extracts image is improved, using the convolutional neural networks model of the image training in image data base.
With every image zooming-out depth characteristic that trained convolutional neural networks model is in image data base set X, i-th image
It extracts obtained feature vector and is denoted as fi, in image data base set X the depth characteristic vector of image constitute depth characteristic to
Duration set is represented by F={ f1,f2,…,fn}。
The distance between depth characteristic vector two-by-two is calculated separately in depth characteristic vector set, using as two images
Similarity value, such as the Euclidean distance between vector two-by-two can be calculated separately in depth characteristic vector set, generate image data
The similarity matrix in library, the matrix that similarity matrix can be as shown in Figure 5.
The depth characteristic similarity value of i-th image and jth image can be calculated using following formula:
In formula, fiFor the depth characteristic vector of i-th image, fjFor the depth characteristic vector of jth image, yi,j=1 table
Show that i-th image and jth image are same category image, yi,j=0 i-th image of expression is different classes of with jth image
Image, α, β are hyper parameter.
The penalty values of depth Hash mapping model can be by the classification, similarity and binary system of the image pair of input network structure
Change constraint to determine, the image data base of image is opened for sharing n, it can be using following formula as the penalty values of the network structure:
In formula, Loss is penalty values, and i-th image and jth image form image pair, and n is image in image data base
Total number, Si,jFor the depth characteristic similarity of i-th image and jth image, oiFor the Hash coding of i-th image, ojFor
The Hash coding of jth image, | | oi-oj||2For Hash coding pair between distance, α is hyper parameter, yi,jFor label classification,
yi,j=1 indicates that i-th image and jth image are same category image, yi,j=0 i-th image of expression opens image with jth
Different classes of image.
Then stochastic gradient descent optimization penalty values, which can be used, trains the network structure to obtain Model Weight, to complete depth
The training of Hash mapping model.Certainly, other optimization algorithms can also be used, the application does not do any restriction to this.
Depth Hash mapping model can be used in the output of Hash coding layerIt indicates, leads to
It crosses setting threshold value and maps it onto Hash coding.
S102: image to be retrieved is mapped as Hash to be retrieved using depth Hash mapping model and is encoded.
Image to be retrieved is input to depth Hash mapping model, the Hash coding layer of the model is exported and is reflected based on threshold value
The Hash coding for image to be retrieved is penetrated, i.e., Hash coding to be retrieved.
S103: it searches in the Hash code database constructed in advance and meets in advance with the Hamming distance difference of Hash to be retrieved coding
If the target image of condition, using as image to be retrieved image data base search result.
Hash code database is corresponding with image data base, each Hash coding and image data base that Hash code database includes
In image it is uniquely corresponding, every image in image data base is input to depth Hash mapping model, pass through setting threshold
The output of the Hash coding layer of depth Hash mapping model is mapped as corresponding Hash and encoded, then according to every image by value
Hash coding generate Hash code database.M codings of i-th image of image data baseIt can be such as following formula:
In formula,For the m output in Hash coding layer of i-th image, θ is threshold value.
Image similar with image to be retrieved namely target image are retrieved in image data base, it can be according to figure to be retrieved
The Hash of picture encode and image data base in each image Hash coding between Hamming distance determine, can be according to Hamming distance
Difference, the number of target image output and the total picture number of image data base are arranged preset condition, such as a kind of specific
In embodiment, it can be searched in Hash code database and be worth the smallest preceding T image, T with the Hamming distance of Hash to be retrieved coding
The size of value can mutually be determined that this does not influence the reality of the application by amount of images sum in image data base with user's actual need
It is existing.For example, 2 T calculate the Hash coding to be retrieved successively Hamming distance with the Hash of image each in image data base coding
Deviation value, then choosing minimal difference and the corresponding image of secondary small difference from 10 Hamming distance differences.
In order to facilitate the output of similar image, can according to the Hamming distance encoded with Hash to be retrieved difference from small to large
T images are ranked up, then T images after output sequence.It certainly can also be according to the Hamming encoded with Hash to be retrieved
The difference of distance is from big to small ranked up T images, this not realization of image the application.
In technical solution provided in an embodiment of the present invention, consider respectively to scheme in image data in training Hash mapping model
The similarity as between, more similar between image, the Hamming distance between Hash coding mapped is with regard to smaller, more not phase between image
Seemingly, the Hamming distance between Hash coding mapped is just slightly larger, between the Hash coding for promoting similar image to map
Hamming distance determined according to the similarity between image, efficiently solve in the related technology same category image Hash coding
The problem of excessively unanimously causing similar image that can not retrieve, to realize the accurate retrieval of same category image.
Technical scheme is more clearly understood for the ease of those skilled in the art, present invention also provides one to show
Meaning property example, is tested using CIFAR-10 data set as image data base.It altogether include 60000 in CIFAR-10 data set
The color image for opening 32*32*3, shares 10 classifications, it may include following the description:
Image in image data base is expressed as set X={ x1,x2,…,xn, n images are shared, take CIFAR- herein
First 50000 of 10 are used as training set, thus n=50000.Image random two are partnered, i-th image and jth
The image of image construction is to being expressed as (xi,xj), with symbol yi,jTo indicate the label of the image, yi,j=1 indicates this two images
It is similar, yi,j=0 indicates this two image dissmilarities.
As shown in Fig. 2, in this example for data sets CIFAR-10 to convolutional neural networks structure C NN1It is modified slightly, Quan Lian
It connects layer to be constituted by 2 layers, wherein fc6 is characterized extract layer.Use image data base set X as training set training CNN1, so that instruction
CNN after white silk1The feature extracted can preferably indicate image.
With trained CNN1For every image zooming-out depth characteristic in database collection X, i-th image zooming-out is obtained
Feature vector be denoted as fi, thus the set expression that the feature vector of every image of database collection X is constituted is F={ f1,
f2,…,fn}.As shown in Fig. 2, feature extraction layer fc6 is made of 512 nodes herein, thus feature vector fiDimension be (1,
512), the dimension of feature vector set F is (50000,512).
The distance between vector two-by-two in feature vector set F is calculated, similarity matrix between image is obtained, is denoted as S.For
Similarity matrix S between image, the i-th row, jth column indicate: image is to (xi,xj) corresponding feature vector fiAnd fjBetween
Distance value Si,j。
Herein, for CIFAR-10 data set, setup parameter β=30, α=30.
Depth Hash mapping model based on image similarity is expressed as CNNhash.In this example, the node of Hash coding layer
Number k=12, therefore the digit of Hash coding is also 12.So image xiIt is in the output of Hash coding layerTo indicate.
With random image to (xi,xj) it is used as CNNhashInput, the output of Hash coding layer is (fi,fj).Such as Fig. 2 institute
Show, loss function is determined by 3 kinds of factors: (1) classification yi,j;(2) similarity Si,j;(3) binarization constrains.Therefore, for total
Number is the database X of n, total penalty values are as follows:
On image data base X, CNN is obtained by gradient reduced minimum LosshashModel Weight w, will train
CNNhashIt is expressed as w-CNNhash.Herein, the learning rate of stochastic gradient descent may be configured as lr=0.00001.
Image xiAs the application depth Hash mapping model w-CNNhashInput, obtain output oi, this example threshold θ=0
Hash coding is mapped it onto, Hash coded representation isWithTo indicate i-th image
M codings, in which:
Thus, image data base X is through w-CNNhashHash code database is obtained after mapping, is expressed as H.
For image x to be retrievedquery, most like preceding T images are retrieved from image data base X.
With model w-CNN of the present inventionhashBy image x to be retrievedqueryIt is mapped to Hash coding hquery;
The smallest preceding T of Hamming distance images are searched in Hash code database H.
By this T images according to xqueryHamming distance resequence from small to large.
It is returned using the T after rearrangement images as search result.
From the foregoing, it will be observed that the embodiment of the present invention efficiently solves same category image Hash coding excessively one in the related technology
The problem of cause, to realize the accurate retrieval of same category image.
Further, in order to confirm that technical solution provided by the present application can realize the accurate retrieval to same category image,
Retrieval rate Precison can be used as evaluation criteria to measure image retrieval algorithm in the retrieval effectiveness of a retrieval image.
The calculation method of Precison are as follows:
Rel (i) indicates whether i-th image in image to be retrieved and image data base be similar, and value is if similar
1, dissimilar then value is 0.For retrieving image set Xtest, can be measured with the average retrieval accuracy rate MRP of all retrieval images
The retrieval performance of distinct methods.
By experiment, the MRP of the depth Hash mapping model based on image similarity is 83.42%, the MRP of the relevant technologies
It is 81.78%.It can be seen that the depth Hash mapping model based on image similarity has good retrieval rate.Such as Fig. 6 and
Shown in 7, Fig. 6 is to indicate that the binary system that CIFAR-10 maps on the depth Hash mapping model based on image similarity is compiled
Code distribution, shares 320 kinds;Fig. 7 show the binary coding distribution that image data base obtains after mapping, and shares 89 kinds.By
, as it can be seen that the Hash coding that the depth Hash mapping model based on image similarity maps has diversity, energy is preferable for this
The Hash coding for breaking off relations similar image is excessively consistent, thus similar image almost has no the problem of discrimination.
The embodiment of the present invention provides corresponding realization device also directed to image search method, further such that the method
With more practicability.Image retrieving apparatus provided in an embodiment of the present invention is introduced below, image retrieval described below
Device can correspond to each other reference with above-described image search method.
Referring to Fig. 8, Fig. 8 is a kind of structure of the image retrieving apparatus provided in an embodiment of the present invention under specific embodiment
Figure, the device can include:
Model training module 801, for the image in image data base to for input, the Kazakhstan penetrated with image map
The characteristic similarity of distance, label classification and image pair between uncommon coding pair is penalty values, excellent using machine learning optimization algorithm
Change penalty values and depth Hash mapping model is obtained with training;Label classification is for identifying the similitude that image pair two opens image.
Hash encodes generation module 802, to be retrieved for being mapped as image to be retrieved using depth Hash mapping model
Hash coding.
Image retrieval module 803, for searching the Chinese with Hash to be retrieved coding in the Hash code database constructed in advance
Prescribed distance difference meets the target image of preset condition, using as image to be retrieved image data base search result;Hash
Code database be by every image in image data base through depth Hash mapping model mapping after gained.
Optionally, in some embodiments of the present embodiment, the model training module 801 can also be used in full connection
Layer is network structure of the convolutional neural networks model of Hash coding layer as depth Hash mapping model;
Using the image in image data base to the input as network structure, using following formula as the loss of network structure
Value, using stochastic gradient descent optimization penalty values training network structure:
In formula, Loss is penalty values, and i-th image and jth image form image pair, and n is image in image data base
Total number, Si,jFor the depth characteristic similarity of i-th image and jth image, oiFor the Hash coding of i-th image, ojFor
The Hash coding of jth image, | | oi-oj||2For Hash coding pair between distance, α is hyper parameter, yi,jFor label classification,
yi,j=1 indicates that i-th image and jth image are same category image, yi,j=0 i-th image of expression opens image with jth
Different classes of image.
In some other embodiment of the embodiment of the present invention, the model training module 801 can also be used in using pre-
The depth characteristic of each image in the convolutional neural networks model extraction image data base first constructed generates depth characteristic vector set
It closes;
The Euclidean distance between vector two-by-two is calculated separately in depth characteristic vector set, the similar of image data base is generated
Spend matrix;
Wherein, convolutional neural networks are to utilize image training gained in image data base based on deep learning.
In the other embodiment of the embodiment of the present invention, the model training module 801 can also be used under utilization
State the depth characteristic similarity value that formula calculates i-th image and jth image:
In formula, fiFor the depth characteristic vector of i-th image, fjFor the depth characteristic vector of jth image, yi,j=1 table
Show that i-th image and jth image are same category image, yi,j=0 i-th image of expression is different classes of with jth image
Image, α, β are hyper parameter.
Optionally, in other embodiments of the present embodiment, described image retrieval module 803 for example can also be due to
It is searched in the Hash code database and is worth the smallest preceding T image with the Hamming distance of the Hash coding to be retrieved;According to
The Hamming distance of the Hash coding to be retrieved is from small to large ranked up T images;T images after output sequence.
The function that described image of the embodiment of the present invention retrieves each functional module of device can be according in above method embodiment
Method specific implementation, specific implementation process is referred to the associated description of above method embodiment, and details are not described herein again.
From the foregoing, it will be observed that the embodiment of the present invention efficiently solves same category image Hash coding excessively one in the related technology
The problem of cause, to realize the accurate retrieval of same category image.
The embodiment of the invention also provides a kind of image retrieval apparatus, specifically can include:
Memory, for storing computer program;
Processor realizes the step of any one embodiment described image search method as above for executing computer program
Suddenly.
The function of each functional module of described image retrieval facility of the embodiment of the present invention can be according in above method embodiment
Method specific implementation, specific implementation process is referred to the associated description of above method embodiment, and details are not described herein again.
From the foregoing, it will be observed that the embodiment of the present invention efficiently solves same category image Hash coding excessively one in the related technology
The problem of cause, to realize the accurate retrieval of same category image.
The embodiment of the invention also provides a kind of computer readable storage mediums, are stored with image retrieval program, the figure
When being executed by processor as search program as above any one embodiment described image search method the step of.
The function of each functional module of computer readable storage medium described in the embodiment of the present invention can be according to above method reality
The method specific implementation in example is applied, specific implementation process is referred to the associated description of above method embodiment, herein no longer
It repeats.
From the foregoing, it will be observed that the embodiment of the present invention efficiently solves same category image Hash coding excessively one in the related technology
The problem of cause, to realize the accurate retrieval of same category image.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with it is other
The difference of embodiment, same or similar part may refer to each other between each embodiment.For being filled disclosed in embodiment
For setting, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is referring to method part
Explanation.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Above to a kind of image search method provided by the present invention, device, equipment and computer readable storage medium into
It has gone and has been discussed in detail.Used herein a specific example illustrates the principle and implementation of the invention, the above implementation
The explanation of example is merely used to help understand method and its core concept of the invention.It should be pointed out that for the general of the art
, without departing from the principle of the present invention, can be with several improvements and modifications are made to the present invention for logical technical staff, this
A little improvement and modification are also fallen within the protection scope of the claims of the present invention.
Claims (10)
1. a kind of image search method characterized by comprising
Image to be retrieved Hash to be retrieved is mapped as using the depth Hash mapping model constructed in advance to encode;
It is searched in the Hash code database constructed in advance and meets default item with the Hamming distance difference of the Hash coding to be retrieved
The target image of part, using as the image to be retrieved image data base search result;
Wherein, the depth Hash mapping model is with the image in described image database to for input, with described image pair
The characteristic similarity of distance, label classification and described image pair between the Hash coding pair that mapping obtains is penalty values, using machine
Device learns optimization algorithm and optimizes the penalty values training gained;The label classification opens image for identifying described image centering two
Similitude;The Hash code database is to map every image in described image database through the depth Hash mapping model
Gained afterwards.
2. image search method according to claim 1, which is characterized in that the depth Hash mapping model was trained
Journey includes:
Using the convolutional neural networks model that full articulamentum is Hash coding layer as the network knot of the depth Hash mapping model
Structure;
Using the image in described image database to the input as the network structure, using following formula as the network knot
The penalty values of structure optimize the penalty values training network structure using stochastic gradient descent:
In formula, Loss is the penalty values, and i-th image and jth image form described image pair, and n is described image data
Image total number in library, Si,jFor the depth characteristic similarity of i-th image and jth image, oiFor the Hash of i-th image
Coding, ojHash for jth image encodes, | | oi-oj||2For Hash coding pair between distance, α is hyper parameter, yi,jFor
Label classification, yi,j=1 indicates that i-th image and jth image are same category image, yi,j=0 indicates i-th image and the
J images are different classes of image.
3. image search method according to claim 2, which is characterized in that the depth of i-th image and jth image is special
Sign similarity value can be obtained from the similarity matrix precalculated, the similarity matrix calculating process are as follows:
Using the depth characteristic of each image in the convolutional neural networks model extraction described image database constructed in advance, generate deep
Spend feature vector set;
The Euclidean distance between vector two-by-two is calculated separately in the depth characteristic vector set, described image database is generated
Similarity matrix;
Wherein, the convolutional neural networks are to utilize image training gained in described image database based on deep learning.
4. image search method according to claim 2, which is characterized in that the depth of i-th image and jth image
Spend the calculating process of characteristic similarity value are as follows:
The depth characteristic similarity value of i-th image and jth image is calculated using following formula:
In formula, fiFor the depth characteristic vector of i-th image, fjFor the depth characteristic vector of jth image, yi,j=1 indicates i-th
Opening image and jth image is same category image, yi,j=0 indicates that i-th image and jth image are different classes of image,
α, β are hyper parameter.
5. image search method according to claim 3, which is characterized in that the convolutional neural networks model is VGG-16
Network model.
6. according to claim 1 to image search method described in 5 any one, which is characterized in that the Hash code database
Generating process are as follows:
Every image in described image database is input to the depth Hash mapping model, by the way that threshold value is arranged, by institute
The output for stating the Hash coding layer of depth Hash mapping model is mapped as Hash coding;
The Hash code database is generated according to the Hash of every image coding;
Wherein, m codings of i-th image of described image databaseSuch as following formula:
In formula,For the m output in Hash coding layer of i-th image, θ is the threshold value.
7. according to claim 1 to image search method described in 5 any one, which is characterized in that described to construct in advance
The target image for meeting preset condition with the Hamming distance difference of the Hash coding to be retrieved is searched in Hash code database are as follows:
It is searched in the Hash code database and is worth the smallest preceding T image with the Hamming distance of the Hash coding to be retrieved;
T images are ranked up from small to large according to the Hamming distance difference with the Hash coding to be retrieved;
T images after output sequence.
8. a kind of image retrieving apparatus characterized by comprising
Model training module, for the image in image data base to for input, with described image to the obtained Hash of mapping
The characteristic similarity of distance, label classification and described image pair between coding pair is penalty values, using machine learning optimization algorithm
Optimize the penalty values and depth Hash mapping model is obtained with training;The label classification is opened for identifying described image centering two
The similitude of image;
Hash encodes generation module, for image to be retrieved to be mapped as Hash to be retrieved using the depth Hash mapping model
Coding;
Image retrieval module, for searching the Hamming distance with the Hash coding to be retrieved in the Hash code database constructed in advance
Deviation value meets the target image of preset condition, using as the image to be retrieved image data base search result;It is described
Hash code database be by every image in described image database through the depth Hash mapping model mapping after gained.
9. a kind of image retrieval apparatus, which is characterized in that including processor, the processor is used to execute to store in memory
It is realized when computer program such as the step of any one of claim 1 to 7 described image search method.
10. a kind of computer readable storage medium, which is characterized in that be stored with image inspection on the computer readable storage medium
Suo Chengxu is realized when described image search program is executed by processor such as any one of claim 1 to 7 described image search method
The step of.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910175051.0A CN109918532B (en) | 2019-03-08 | 2019-03-08 | Image retrieval method, device, equipment and computer readable storage medium |
PCT/CN2020/077586 WO2020182019A1 (en) | 2019-03-08 | 2020-03-03 | Image search method, apparatus, device, and computer-readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910175051.0A CN109918532B (en) | 2019-03-08 | 2019-03-08 | Image retrieval method, device, equipment and computer readable storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109918532A true CN109918532A (en) | 2019-06-21 |
CN109918532B CN109918532B (en) | 2023-08-18 |
Family
ID=66963962
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910175051.0A Active CN109918532B (en) | 2019-03-08 | 2019-03-08 | Image retrieval method, device, equipment and computer readable storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109918532B (en) |
WO (1) | WO2020182019A1 (en) |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111079828A (en) * | 2019-12-13 | 2020-04-28 | 智慧神州(北京)科技有限公司 | Image clustering method and device based on neural network and storage medium |
CN111581421A (en) * | 2020-04-30 | 2020-08-25 | 京东方科技集团股份有限公司 | Image retrieval method, image retrieval device, and image retrieval system |
CN111612080A (en) * | 2020-05-22 | 2020-09-01 | 深圳前海微众银行股份有限公司 | Model interpretation method, device and readable storage medium |
CN111626408A (en) * | 2020-05-22 | 2020-09-04 | 深圳前海微众银行股份有限公司 | Hash coding method, device and equipment and readable storage medium |
WO2020182019A1 (en) * | 2019-03-08 | 2020-09-17 | 苏州大学 | Image search method, apparatus, device, and computer-readable storage medium |
CN111813975A (en) * | 2020-07-09 | 2020-10-23 | 国网电子商务有限公司 | Image retrieval method and device and electronic equipment |
CN111967609A (en) * | 2020-08-14 | 2020-11-20 | 深圳前海微众银行股份有限公司 | Model parameter verification method, device and readable storage medium |
CN112131421A (en) * | 2020-09-23 | 2020-12-25 | 平安科技(深圳)有限公司 | Medical image classification method, device, equipment and storage medium |
CN112132099A (en) * | 2020-09-30 | 2020-12-25 | 腾讯科技(深圳)有限公司 | Identity recognition method, palm print key point detection model training method and device |
CN112182262A (en) * | 2020-11-30 | 2021-01-05 | 江西师范大学 | Image query method based on feature classification |
CN112215302A (en) * | 2020-10-30 | 2021-01-12 | Oppo广东移动通信有限公司 | Image identification method and device and terminal equipment |
CN112561053A (en) * | 2020-11-10 | 2021-03-26 | 北京百度网讯科技有限公司 | Image processing method, training method and device of pre-training model and electronic equipment |
CN112925940A (en) * | 2021-03-04 | 2021-06-08 | 浙江中设天合科技有限公司 | Similar image retrieval method and device, computer equipment and storage medium |
CN113094531A (en) * | 2021-03-22 | 2021-07-09 | 华中科技大学 | In-memory image retrieval method and retrieval system |
CN113111953A (en) * | 2021-04-20 | 2021-07-13 | 深圳吉祥星科技股份有限公司 | Projection image quality processing device |
CN113177130A (en) * | 2021-06-09 | 2021-07-27 | 山东科技大学 | Image retrieval and identification method and device based on binary semantic embedding |
WO2021159898A1 (en) * | 2020-02-12 | 2021-08-19 | 深圳壹账通智能科技有限公司 | Privacy protection-based deep learning method, system and server, and storage medium |
CN113282777A (en) * | 2021-04-20 | 2021-08-20 | 北京沃东天骏信息技术有限公司 | Model training method and device, electronic equipment and storage medium |
CN113326393A (en) * | 2021-05-31 | 2021-08-31 | 西安理工大学 | Image retrieval method based on deep hash feature and heterogeneous parallel processing |
CN113918753A (en) * | 2021-07-23 | 2022-01-11 | 腾讯科技(深圳)有限公司 | Image retrieval method based on artificial intelligence and related equipment |
CN114446407A (en) * | 2022-03-03 | 2022-05-06 | 冰洲石生物科技(上海)有限公司 | Method, system, medium, and electronic device for extracting reaction template for chemical reaction |
CN117156138A (en) * | 2023-10-31 | 2023-12-01 | 季华实验室 | Method, device, equipment and storage medium for variable-length depth hash coding of image |
CN118365904A (en) * | 2024-06-20 | 2024-07-19 | 浙江方圆检测集团股份有限公司 | Trademark infringement detection and identification method and system |
Families Citing this family (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112527855B (en) * | 2020-09-23 | 2024-05-03 | 广东协城信息科技有限公司 | Face vector quick comparison technology |
CN112256895B (en) * | 2020-10-16 | 2024-07-19 | 江南大学 | Fabric image retrieval method based on multitask learning |
CN112231514B (en) * | 2020-10-19 | 2024-01-05 | 腾讯科技(深圳)有限公司 | Data deduplication method and device, storage medium and server |
CN112270760B (en) * | 2020-11-03 | 2023-07-25 | 北京百度网讯科技有限公司 | Positioning method, positioning device, electronic equipment and storage medium |
CN112307248B (en) * | 2020-11-26 | 2023-11-03 | 国网数字科技控股有限公司 | Image retrieval method and device |
CN112801867B (en) * | 2020-12-30 | 2023-10-31 | 杭州趣链科技有限公司 | Satellite remote sensing image splicing method, device and equipment based on block chain |
CN112907712A (en) * | 2021-01-22 | 2021-06-04 | 杭州电子科技大学 | Three-dimensional model feature representation method based on multi-view hash enhanced hash |
CN112766288B (en) * | 2021-03-03 | 2024-01-23 | 重庆赛迪奇智人工智能科技有限公司 | Image processing model construction method, device, electronic equipment and readable storage medium |
CN112966755A (en) * | 2021-03-10 | 2021-06-15 | 深圳市固电电子有限公司 | Inductance defect detection method and device and readable storage medium |
CN113270199B (en) * | 2021-04-30 | 2024-04-26 | 贵州师范大学 | Medical cross-mode multi-scale fusion class guide hash method and system thereof |
CN113536012A (en) * | 2021-05-26 | 2021-10-22 | 浙江大华技术股份有限公司 | Image retrieval method, image retrieval device, electronic equipment and computer-readable storage medium |
CN113542750B (en) * | 2021-05-27 | 2024-06-25 | 绍兴市北大信息技术科创中心 | Data coding method for searching by adopting more than two sets of hash tables |
CN113239226A (en) * | 2021-06-01 | 2021-08-10 | 平安科技(深圳)有限公司 | Image retrieval method, device, equipment and storage medium |
CN113344826B (en) * | 2021-07-06 | 2023-12-19 | 北京锐安科技有限公司 | Image processing method, device, electronic equipment and storage medium |
CN113688261B (en) * | 2021-08-25 | 2023-10-13 | 山东极视角科技股份有限公司 | Image data cleaning method and device, electronic equipment and readable storage medium |
CN113886628A (en) * | 2021-10-09 | 2022-01-04 | 上海识装信息科技有限公司 | Image retrieval method, device and storage medium |
CN114417251A (en) * | 2022-01-25 | 2022-04-29 | 腾讯科技(深圳)有限公司 | Retrieval method, device, equipment and storage medium based on hash code |
CN114547354B (en) * | 2022-02-15 | 2023-04-07 | 华南师范大学 | Deep hash method based on function adaptive mapping |
CN114780763B (en) * | 2022-03-28 | 2024-09-06 | 东南大学 | Efficient retrieval method for indoor design furniture model |
CN114743035A (en) * | 2022-04-12 | 2022-07-12 | 上海极豆科技有限公司 | Image screening method and device, electronic equipment and readable storage medium |
CN115964527B (en) * | 2023-01-05 | 2023-09-26 | 北京东方通网信科技有限公司 | Label characterization construction method for single-label image retrieval |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104834748A (en) * | 2015-05-25 | 2015-08-12 | 中国科学院自动化研究所 | Image retrieval method utilizing deep semantic to rank hash codes |
CN106407352A (en) * | 2016-09-06 | 2017-02-15 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Traffic image retrieval method based on depth learning |
CN107423306A (en) * | 2016-05-24 | 2017-12-01 | 华为技术有限公司 | A kind of image search method and device |
CN108491528A (en) * | 2018-03-28 | 2018-09-04 | 苏州大学 | A kind of image search method, system and device |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11275747B2 (en) * | 2015-03-12 | 2022-03-15 | Yahoo Assets Llc | System and method for improved server performance for a deep feature based coarse-to-fine fast search |
CN108399211B (en) * | 2018-02-02 | 2020-11-24 | 清华大学 | Large-scale image retrieval algorithm based on binary characteristics |
CN108427738B (en) * | 2018-03-01 | 2022-03-25 | 中山大学 | Rapid image retrieval method based on deep learning |
CN108932314A (en) * | 2018-06-21 | 2018-12-04 | 南京农业大学 | A kind of chrysanthemum image content retrieval method based on the study of depth Hash |
CN109918532B (en) * | 2019-03-08 | 2023-08-18 | 苏州大学 | Image retrieval method, device, equipment and computer readable storage medium |
-
2019
- 2019-03-08 CN CN201910175051.0A patent/CN109918532B/en active Active
-
2020
- 2020-03-03 WO PCT/CN2020/077586 patent/WO2020182019A1/en active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104834748A (en) * | 2015-05-25 | 2015-08-12 | 中国科学院自动化研究所 | Image retrieval method utilizing deep semantic to rank hash codes |
CN107423306A (en) * | 2016-05-24 | 2017-12-01 | 华为技术有限公司 | A kind of image search method and device |
CN106407352A (en) * | 2016-09-06 | 2017-02-15 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Traffic image retrieval method based on depth learning |
CN108491528A (en) * | 2018-03-28 | 2018-09-04 | 苏州大学 | A kind of image search method, system and device |
Non-Patent Citations (2)
Title |
---|
JUN LU等: "Cascaded Deep Hashing for Large-Scale Image Retrieval", 《HTTPS://LINK.SPRINGER.COM/CHAPTER/10.1007/978-3-030-04224-0_36》 * |
LI MENGTING等: "Deep Hashing for Large-scale Image Retrieval", 《PROCEEDINGS OF THE 36TH CHINESE CONTROL CONFERENCE》 * |
Cited By (39)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020182019A1 (en) * | 2019-03-08 | 2020-09-17 | 苏州大学 | Image search method, apparatus, device, and computer-readable storage medium |
CN111079828B (en) * | 2019-12-13 | 2023-10-20 | 北京因特睿软件有限公司 | Image clustering method, device and storage medium based on neural network |
CN111079828A (en) * | 2019-12-13 | 2020-04-28 | 智慧神州(北京)科技有限公司 | Image clustering method and device based on neural network and storage medium |
WO2021159898A1 (en) * | 2020-02-12 | 2021-08-19 | 深圳壹账通智能科技有限公司 | Privacy protection-based deep learning method, system and server, and storage medium |
CN111581421B (en) * | 2020-04-30 | 2024-06-04 | 京东方科技集团股份有限公司 | Image retrieval method, image retrieval device, and image retrieval system |
CN111581421A (en) * | 2020-04-30 | 2020-08-25 | 京东方科技集团股份有限公司 | Image retrieval method, image retrieval device, and image retrieval system |
CN111612080B (en) * | 2020-05-22 | 2021-11-09 | 深圳前海微众银行股份有限公司 | Model interpretation method, device and readable storage medium |
CN111612080A (en) * | 2020-05-22 | 2020-09-01 | 深圳前海微众银行股份有限公司 | Model interpretation method, device and readable storage medium |
CN111626408A (en) * | 2020-05-22 | 2020-09-04 | 深圳前海微众银行股份有限公司 | Hash coding method, device and equipment and readable storage medium |
CN111626408B (en) * | 2020-05-22 | 2021-08-06 | 深圳前海微众银行股份有限公司 | Hash coding method, device and equipment and readable storage medium |
WO2021232752A1 (en) * | 2020-05-22 | 2021-11-25 | 深圳前海微众银行股份有限公司 | Hash encoding method, apparatus and device, and readable storage medium |
CN111813975A (en) * | 2020-07-09 | 2020-10-23 | 国网电子商务有限公司 | Image retrieval method and device and electronic equipment |
CN111967609A (en) * | 2020-08-14 | 2020-11-20 | 深圳前海微众银行股份有限公司 | Model parameter verification method, device and readable storage medium |
CN111967609B (en) * | 2020-08-14 | 2021-08-06 | 深圳前海微众银行股份有限公司 | Model parameter verification method, device and readable storage medium |
CN112131421A (en) * | 2020-09-23 | 2020-12-25 | 平安科技(深圳)有限公司 | Medical image classification method, device, equipment and storage medium |
CN112131421B (en) * | 2020-09-23 | 2023-09-15 | 平安科技(深圳)有限公司 | Medical image classification method, device, equipment and storage medium |
CN112132099A (en) * | 2020-09-30 | 2020-12-25 | 腾讯科技(深圳)有限公司 | Identity recognition method, palm print key point detection model training method and device |
CN112215302A (en) * | 2020-10-30 | 2021-01-12 | Oppo广东移动通信有限公司 | Image identification method and device and terminal equipment |
CN112561053B (en) * | 2020-11-10 | 2023-09-08 | 北京百度网讯科技有限公司 | Image processing method, training method and device of pre-training model and electronic equipment |
CN112561053A (en) * | 2020-11-10 | 2021-03-26 | 北京百度网讯科技有限公司 | Image processing method, training method and device of pre-training model and electronic equipment |
CN112182262A (en) * | 2020-11-30 | 2021-01-05 | 江西师范大学 | Image query method based on feature classification |
CN112182262B (en) * | 2020-11-30 | 2021-03-19 | 江西师范大学 | Image query method based on feature classification |
CN112925940A (en) * | 2021-03-04 | 2021-06-08 | 浙江中设天合科技有限公司 | Similar image retrieval method and device, computer equipment and storage medium |
CN112925940B (en) * | 2021-03-04 | 2022-07-01 | 浙江中设天合科技有限公司 | Similar image retrieval method and device, computer equipment and storage medium |
CN113094531A (en) * | 2021-03-22 | 2021-07-09 | 华中科技大学 | In-memory image retrieval method and retrieval system |
CN113111953A (en) * | 2021-04-20 | 2021-07-13 | 深圳吉祥星科技股份有限公司 | Projection image quality processing device |
CN113282777A (en) * | 2021-04-20 | 2021-08-20 | 北京沃东天骏信息技术有限公司 | Model training method and device, electronic equipment and storage medium |
CN113111953B (en) * | 2021-04-20 | 2022-08-26 | 深圳吉祥星科技股份有限公司 | Projection image quality processing device |
CN113326393A (en) * | 2021-05-31 | 2021-08-31 | 西安理工大学 | Image retrieval method based on deep hash feature and heterogeneous parallel processing |
CN113326393B (en) * | 2021-05-31 | 2023-04-07 | 深圳前瞻资讯股份有限公司 | Image retrieval method based on deep hash feature and heterogeneous parallel processing |
CN113177130B (en) * | 2021-06-09 | 2022-04-08 | 山东科技大学 | Image retrieval and identification method and device based on binary semantic embedding |
CN113177130A (en) * | 2021-06-09 | 2021-07-27 | 山东科技大学 | Image retrieval and identification method and device based on binary semantic embedding |
CN113918753A (en) * | 2021-07-23 | 2022-01-11 | 腾讯科技(深圳)有限公司 | Image retrieval method based on artificial intelligence and related equipment |
CN113918753B (en) * | 2021-07-23 | 2024-05-28 | 腾讯科技(深圳)有限公司 | Image retrieval method based on artificial intelligence and related equipment |
CN114446407A (en) * | 2022-03-03 | 2022-05-06 | 冰洲石生物科技(上海)有限公司 | Method, system, medium, and electronic device for extracting reaction template for chemical reaction |
CN117156138A (en) * | 2023-10-31 | 2023-12-01 | 季华实验室 | Method, device, equipment and storage medium for variable-length depth hash coding of image |
CN117156138B (en) * | 2023-10-31 | 2024-02-09 | 季华实验室 | Method, device, equipment and storage medium for variable-length depth hash coding of image |
CN118365904A (en) * | 2024-06-20 | 2024-07-19 | 浙江方圆检测集团股份有限公司 | Trademark infringement detection and identification method and system |
CN118365904B (en) * | 2024-06-20 | 2024-09-17 | 浙江方圆检测集团股份有限公司 | Trademark infringement detection and identification method and system |
Also Published As
Publication number | Publication date |
---|---|
WO2020182019A1 (en) | 2020-09-17 |
CN109918532B (en) | 2023-08-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109918532A (en) | Image search method, device, equipment and computer readable storage medium | |
Tang et al. | Tri-clustered tensor completion for social-aware image tag refinement | |
Karatzoglou et al. | A Seq2Seq learning approach for modeling semantic trajectories and predicting the next location | |
CN110032632A (en) | Intelligent customer service answering method, device and storage medium based on text similarity | |
CN102693311B (en) | Target retrieval method based on group of randomized visual vocabularies and context semantic information | |
CN113095415B (en) | Cross-modal hashing method and system based on multi-modal attention mechanism | |
CN109829065A (en) | Image search method, device, equipment and computer readable storage medium | |
CN110059807A (en) | Image processing method, device and storage medium | |
CN110647904B (en) | Cross-modal retrieval method and system based on unmarked data migration | |
CN109426781A (en) | Construction method, face identification method, device and the equipment of face recognition database | |
CN105589938A (en) | Image retrieval system and retrieval method based on FPGA | |
CN113177141B (en) | Multi-label video hash retrieval method and device based on semantic embedded soft similarity | |
WO2014160282A1 (en) | Classifying resources using a deep network | |
CN110059206A (en) | A kind of extensive hashing image search method based on depth representative learning | |
CN105849720A (en) | Visual semantic complex network and method for forming network | |
Wei et al. | Scalable heterogeneous translated hashing | |
CN104391908B (en) | Multiple key indexing means based on local sensitivity Hash on a kind of figure | |
CN110119477A (en) | A kind of information-pushing method, device and storage medium | |
CN111461175B (en) | Label recommendation model construction method and device of self-attention and cooperative attention mechanism | |
KR20180129001A (en) | Method and System for Entity summarization based on multilingual projected entity space | |
US12032605B2 (en) | Searchable data structure for electronic documents | |
Yan et al. | Image retrieval for structure-from-motion via graph convolutional network | |
Huang et al. | Deep multimodal embedding model for fine-grained sketch-based image retrieval | |
CN113591881B (en) | Intention recognition method and device based on model fusion, electronic equipment and medium | |
CN110502613A (en) | A kind of model training method, intelligent search method, device 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 |