CN110472090A - Image search method and relevant apparatus, storage medium based on semantic label - Google Patents
Image search method and relevant apparatus, storage medium based on semantic label Download PDFInfo
- Publication number
- CN110472090A CN110472090A CN201910770152.2A CN201910770152A CN110472090A CN 110472090 A CN110472090 A CN 110472090A CN 201910770152 A CN201910770152 A CN 201910770152A CN 110472090 A CN110472090 A CN 110472090A
- Authority
- CN
- China
- Prior art keywords
- label
- image
- node
- preset
- term vector
- 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
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Library & Information Science (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application discloses a kind of image search method and relevant apparatus based on semantic label, storage medium, pass through the training for image classification model, to carry out label association to all images in searching database, when user inputs retrieval information, carrying out similarity-rough set according to the label of the label of retrieval information instruction and image can be obtained the image of Search Requirement, to realize the image retrieval procedure based on semantic label, the high efficiency and accuracy of image retrieval procedure can be improved in this method, and due to the operability of label, further improve convenience and the flexibility of image retrieval procedure.
Description
Technical field
This application involves field of computer technology more particularly to a kind of image search methods and phase based on semantic label
Close device, storage medium.
Background technique
Computer vision technique (computer vision, CV) is one and studies the science of machine " seeing " of how making, more into
As soon as step is said, refers to and the machine vision such as replace human eye to be identified, tracked to target with video camera and computer and measured, go forward side by side
One step does graphics process, and computer is made to be treated as the image for being more suitable for eye-observation or sending instrument detection to.As a section
Learn subject, the relevant theory and technology of computer vision research, it is intended to which foundation can obtain letter from image or multidimensional data
The artificial intelligence system of breath.
Image retrieval can be carried out based on computer vision technique, present image searching system is mainly the base for utilizing image
Plinth feature to carry out similarity calculation to image, for example the model resnet that pre-training is good on ImageNet is utilized to extract data
All characteristics of image in library, then similarity calculation is carried out to image zooming-out feature to be retrieved, then to the two, it is high to return to similarity
Image.
But such method have the defects that in the retrieval on image, semantic it is certain, such as user want inspection
The image of rope cake needs first to find a cake image zooming-out feature, under the scene of great amount of samples, the process be not easy into
Row, and influence convenience and the flexibility of image retrieval procedure.
Summary of the invention
In view of this, the application first aspect provides a kind of image search method based on semantic label, figure can be applied to
As specifically including in searching system or program process: determining that the corresponding relationship of the first image Yu N number of label, N number of label are used
In the display content of instruction the first image, N is positive integer；
The characteristic information of the first image is extracted by the first preset algorithm, with based on default dimension generate feature to
Amount, first preset algorithm includes convolutional neural networks；
N number of label is converted into N number of term vector according to the second preset algorithm；
Preset model is trained according to described eigenvector and the corresponding relationship of N number of term vector, to obtain figure
As disaggregated model, described image disaggregated model is used to generate corresponding M label according to the second image, and M≤N, M are positive integer；
If the similarity of the M label and R label meets preset condition, it is determined that second image is retrieval letter
Corresponding search result is ceased, the R label is the label of the retrieval information instruction, and R is positive integer.
Preferably, in some possible implementations of the application, it is described according to described eigenvector and N number of word to
The corresponding relationship of amount is trained preset model, to obtain image classification model, comprising:
A node is determined according to N number of term vector, and to construct Adjacency matrix, the Adjacency matrix is used to indicate the A
The incidence relation of any two node in a node, A≤N, A are positive integer；
Network topology structure is determined according to the incidence relation of any two node in the A node；
The network topology structure is trained by the second preset algorithm, to obtain image classification model, described
Two preset algorithms are used to indicate A node described in the network topology structure and carry out feature vector exchange, and described second is default
Algorithm includes figure convolutional neural networks.
Preferably, in some possible implementations of the application, N number of term vector includes the first label and the second mark
Label, it is described that A node is determined according to N number of term vector, to construct Adjacency matrix, comprising:
N number of term vector is handled according to preset rules, to construct objective matrix, the preset rules are based on institute
The cooccurrence relation setting of N number of term vector is stated, the objective matrix is used to indicate first label and second label is common
The ratio of appearance；
If the objective matrix meets preset condition, N number of term vector is converted into the A node, so that
The objective matrix is converted to the Adjacency matrix, and the preset condition is based on first label and second label is common
The ratio set of appearance.
Preferably, described according to any two node in the A node in some possible implementations of the application
Incidence relation determine network topology structure, comprising:
Determine the first node in the A node；
Calculate separately the first node according to judgment rule is with the node for removing the first node in the A node
It is no to there is connection side, to obtain judging result；
Network topology structure is determined according to the corresponding relationship of the A node and the judging result.
Preferably, in some possible implementations of the application, it is described according to the second preset algorithm by N number of label
Be converted to N number of term vector, comprising:
Obtain the vocabulary of each label in N number of label；
If described have a large vocabulary in preset threshold, the average value of the vocabulary is calculated；
The average value of the vocabulary is converted into N number of term vector according to according to the second preset algorithm.
Preferably, described that corresponding M mark is generated according to the second image in some possible implementations of the application
Label, comprising:
Judge that N number of label corresponds to the probability of second image according to described image disaggregated model；
Determine that probability in N number of label meets M label of class condition.
Preferably, described that the N is judged according to described image disaggregated model in some possible implementations of the application
A label corresponds to the probability of second image, comprising:
The feature vector of second image is exported according to described image disaggregated model；
According to the feature vector of second image, knot vector corresponding with the N number of label passes through sigmoid respectively
Function is calculated, to obtain tag parameter；
The tag parameter is normalized, to obtain N number of label corresponding to the general of second image
Rate.
The application second aspect provides the device of another image retrieval, comprising: determination unit, for determining the first image
With the corresponding relationship of N number of label, N number of label is used to indicate the display content of the first image, and N is positive integer；
Extraction unit, for extracting the characteristic information of the first image by the first preset algorithm, based on default dimension
Degree generates feature vector, and first preset algorithm includes convolutional neural networks；
Converting unit, for N number of label to be converted to N number of term vector according to the second preset algorithm；
Training unit, for being carried out according to the corresponding relationship of described eigenvector and N number of term vector to preset model
Training, to obtain image classification model, described image disaggregated model is used to generate corresponding M label according to the second image, and M≤
N, M are positive integer；
Retrieval unit, if the similarity for the M label and R label meets preset condition, it is determined that described the
Two images are the corresponding search result of retrieval information, and the R label is the label of the retrieval information instruction, and R is positive integer.
Preferably, in some possible implementations of the application,
The training unit, it is described to construct Adjacency matrix specifically for determining A node according to N number of term vector
Adjacency matrix is used to indicate the incidence relation of any two node in the A node, and A≤N, A are positive integer；
The training unit, specifically for determining network according to the incidence relation of any two node in the A node
Topological structure；
The training unit, specifically for being trained by the second preset algorithm to the network topology structure, with
To image classification model, second preset algorithm is used to indicate A node described in the network topology structure and carries out feature
Vector exchange, second preset algorithm includes figure convolutional neural networks.
Preferably, in some possible implementations of the application, N number of term vector includes the first label and the second mark
Label,
The training unit, specifically for being handled according to preset rules N number of term vector, to construct target square
Battle array, the preset rules are set based on the cooccurrence relation of N number of term vector, and the objective matrix is used to indicate first mark
The ratio that label and second label occur jointly；
The training unit converts N number of term vector if meeting preset condition specifically for the objective matrix
For the A node, so that the objective matrix is converted to the Adjacency matrix, the preset condition is based on first mark
The ratio set that label and second label occur jointly.
Preferably, in some possible implementations of the application,
The training unit, specifically for the first node in the determination A node；
The training unit, specifically for being calculated separately in the first node and the A node according to judgment rule
Except the node of the first node is with the presence or absence of connection side, to obtain judging result；
The training unit, specifically for determining network according to the corresponding relationship of the A node and the judging result
Topological structure.
Preferably, in some possible implementations of the application,
The converting unit, specifically for obtaining the vocabulary of each label in N number of label；
The converting unit calculates being averaged for the vocabulary if having a large vocabulary specifically for described in preset threshold
Value；
The converting unit, specifically for the average value of the vocabulary is converted to N according to according to the second preset algorithm
A term vector.
Preferably, in some possible implementations of the application,
The training unit, for judging that N number of label corresponds to second figure according to described image disaggregated model
The probability of picture；
The training unit, for determining that probability in N number of label meets M label of class condition.
Preferably, in some possible implementations of the application,
The training unit, for exporting the feature vector of second image according to described image disaggregated model；
The training unit, for according to the feature vector of second image section corresponding with the N number of label respectively
Point vector is calculated by sigmoid function, to obtain tag parameter；
The training unit is corresponded to for being normalized the tag parameter with obtaining N number of label
The probability of second image.
The application third aspect provides a kind of computer equipment, comprising: memory, processor and bus system；It is described
Memory is for storing program code；The processor is used for according to the above-mentioned first aspect of instruction execution in said program code
Or the described in any item image search methods based on semantic label of first aspect.
The application fourth aspect provides a kind of computer readable storage medium, stores in the computer readable storage medium
There is instruction, when run on a computer, so that computer executes above-mentioned first aspect or first aspect is described in any item
Image search method based on semantic label.
As can be seen from the above technical solutions, the embodiment of the present application has the advantage that
By determining the corresponding relationship of the first image and multiple labels, then using convolutional neural networks to the first image
Characteristic information extracts, to generate feature vector based on default dimension, further according to the second preset algorithm by the multiple label
Be converted to multiple term vectors；Preset model is instructed according to described eigenvector and the corresponding relationship of the multiple term vector
Practice, to obtain image classification model, to carry out label association to all images in searching database, when user inputs retrieval letter
When breath, carrying out similarity-rough set according to the label of the label of retrieval information instruction and image can be obtained the image of Search Requirement,
To realize the image retrieval procedure based on semantic label, the high efficiency of image retrieval procedure and accurate is can be improved in this method
Property, and due to the operability of label, further improve convenience and the flexibility of image retrieval procedure.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of application for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is the network architecture diagram of image indexing system operation；
Fig. 2 is a kind of image retrieval process frame diagram；
Fig. 3 is a kind of flow chart of the image search method based on semantic label provided by the embodiments of the present application；
Fig. 4 is the flow chart of another image search method based on semantic label provided by the embodiments of the present application；
Fig. 5 is a kind of schematic diagram of interface display of image retrieval provided by the embodiments of the present application；
Fig. 6 is the structural schematic diagram of image retrieving apparatus provided by the embodiments of the present application；
Fig. 7 is the structural schematic diagram of another image retrieving apparatus provided by the embodiments of the present application.
Specific embodiment
The embodiment of the present application provides a kind of image search method based on semantic label and relevant apparatus, storage are situated between
Matter during can be applied to image retrieval, especially by the corresponding relationship for determining the first image and multiple labels, then makes
It is extracted with characteristic information of the convolutional neural networks to the first image, to generate feature vector based on default dimension, further according to
The multiple label is converted to multiple term vectors by the second preset algorithm；According to described eigenvector and the multiple term vector
Corresponding relationship is trained preset model, to obtain image classification model, to carry out to all images in searching database
Label association carries out similarity ratio according to the label of the label of retrieval information instruction and image when user inputs retrieval information
The image of Search Requirement relatively can be obtained, to realize the image retrieval procedure based on semantic label, this method be can be improved
The high efficiency and accuracy of image retrieval procedure, and due to the operability of label, further improve image retrieval procedure
Convenience and flexibility.
The description and claims of this application and term " first ", " second ", " third ", " in above-mentioned attached drawing
The (if present)s such as four " are to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should manage
The data that solution uses in this way are interchangeable under appropriate circumstances, so that embodiments herein described herein for example can be to remove
Sequence other than those of illustrating or describe herein is implemented.In addition, term " includes " and " corresponding to " and their times
What is deformed, it is intended that cover it is non-exclusive include, for example, contain the process, method of a series of steps or units, system,
Product or equipment those of are not necessarily limited to be clearly listed step or unit, but may include be not clearly listed or for
The intrinsic other step or units of these process, methods, product or equipment.
It should be understood that image search method provided by the present application can be applied to can be applied to the operation of image indexing system
Cheng Zhong, specifically, image indexing system can be run in the network architecture as shown in Figure 1, as shown in Figure 1, being image retrieval
The network architecture diagram of system operation, as figure shows, image indexing system can obtain Search Requirement by multiple terminals, pass through figure
As database acquisition image data, analyzing and training is carried out to above-mentioned image according to default rule, generates corresponding multiple labels,
It is understood that showing five terminals in Fig. 1, there can be more or fewer terminal devices to participate in actual scene
In experiment test, depending on particular number is because of actual scene, herein without limitation；In addition, showing an image data in Fig. 1
Library, but in actual scene, there can also be the participation of multiple images database, especially in the field of more application image data interactions
Jing Zhong, depending on specific image data base quantity is because of actual scene.
It is understood that above-mentioned image indexing system can run on individual mobile terminal, service can also be run on
Device is also used as running on third party device to provide the image retrieval service of client, to obtain search report；Specifically
Image indexing system can be to be run in above equipment in the form of a kind of program, can also be used as the system in above equipment
Component is run, and is also used as one kind of cloud service program and is not done herein depending on specific operating mode is because of actual scene
It limits.
Computer vision technique (computer vision, CV) is one and studies the science of machine " seeing " of how making, more into
As soon as step is said, refers to and the machine vision such as replace human eye to be identified, tracked to target with video camera and computer and measured, go forward side by side
One step does graphics process, and computer is made to be treated as the image for being more suitable for eye-observation or sending instrument detection to.As a section
Learn subject, the relevant theory and technology of computer vision research, it is intended to which foundation can obtain letter from image or multidimensional data
The artificial intelligence system of breath.
Image retrieval can be carried out based on computer vision technique, present image searching system is mainly the base for utilizing image
Plinth feature to carry out similarity calculation to image, for example the model resnet that pre-training is good on ImageNet is utilized to extract data
All characteristics of image in library, then similarity calculation is carried out to image zooming-out feature to be retrieved, then to the two, it is high to return to similarity
Image.
But such method have the defects that in the retrieval on image, semantic it is certain, such as user want inspection
The image of rope cake needs first to find a cake image zooming-out feature, under the scene of great amount of samples, the process be not easy into
Row, and influence convenience and the flexibility of image retrieval procedure.
To solve the above-mentioned problems, present applicant proposes a kind of image search method based on semantic label, this method is answered
For image retrieval process frame shown in Fig. 2, as shown in Fig. 2, being a kind of image retrieval process frame diagram, user passes through in figure
Input retrieval text, obtains multiple labels；Image data base side is by scheming the image input picture searching system in database
As searching system makes the image in database all generate corresponding multiple labels by model training method provided by the present application,
Similarity measures, available corresponding retrieval knot are carried out by the resulting label of label and model training for retrieving text generation
Fruit realizes the image retrieval procedure based on semantic label.
It is understood that showing three labels in figure, in actual scene, there can be more or fewer labels
Number, depending on particular number is because of actual scene, herein without limitation.
In conjunction with above-mentioned process frame, the image search method in the application based on semantic label will be introduced below,
Referring to Fig. 3, Fig. 3 is a kind of flow chart of the image search method based on semantic label provided by the embodiments of the present application, this Shen
Please embodiment at least include the following steps:
301, the corresponding relationship of the first image Yu N number of label is determined
In the present embodiment, N number of label is used to indicate the display content of the first image, and N is positive integer.First figure
As can be any one image in image data base, N number of label, which can be, perhaps amplifies content according in the first image
Multiple labels indicated by corresponding relationship i.e. the first picture material of a series of label of setting, the first image and N number of label
Set.
302, the characteristic information of the first image is extracted, by the first preset algorithm to generate feature based on default dimension
Vector.
In the present embodiment, it may include convolutional neural networks that first preset algorithm is set based on neural network algorithm；
Such as: feature is carried out to image image data set using the good model resnet50 convolutional network of the pre-training on ImageNet and is mentioned
It takes, obtains the feature vector of 2048 dimensions.
It is understood that preset dimension can be set according to the related needs of user, it can also be according to image retrieval
The setting range setting that dimension is preset in the historical record of system, depending on specific method is because of actual scene.
303, N number of label is converted to by N number of term vector according to the second preset algorithm；
In the present embodiment, the second preset algorithm is used to determine N number of label vector between such label and label
The connection of amount gone to measure the relationship between them, excavate between label.
In a kind of possible scene, N number of label can be converted to as mark by N number of term vector by word2vec
The character representation of label.
It is understood that the term vector of multiple words in label can be averaged, so if including multiple words in label
Next calculating process is carried out in the term vector for being denoted as corresponding label afterwards.
304, preset model is trained according to described eigenvector and the corresponding relationship of N number of term vector, with
To image classification model.
In the present embodiment, described image disaggregated model is used to generate corresponding M label according to the second image, and M≤N, M are
Positive integer.
It is understood that in order to which all images in image data base are determined corresponding label, it can be according to
The data of one image are trained, and can also be chosen multiple images and be repeated the above steps the process of 301-304, obtain feature vector
It with the corresponding relationship of term vector, is trained with inputting preset model, and further increases accuracy.
Optionally, feature vector and the determination process of N number of term vector corresponding relationship can be based on feature vector
Blocking processing, i.e., determine A node according to N number of term vector, to construct Adjacency matrix, the Adjacency matrix is used to indicate
The incidence relation of any two node, A≤N, A are positive integer in the A node；Then according to any in the A node
The incidence relation of two nodes determines network topology structure；The network topology structure is instructed by the second preset algorithm
Practice, to obtain image classification model, second preset algorithm be used to indicate A node described in the network topology structure into
The exchange of row feature vector, second preset algorithm includes figure convolutional neural networks.
Wherein, according to N number of term vector determine A node process can by according to preset rules to described N number of
Term vector is handled, and to construct objective matrix, the preset rules are set based on the cooccurrence relation of N number of term vector, institute
It states objective matrix and is used to indicate the ratio that first label and second label occur jointly；If the objective matrix meets
N number of term vector is then converted to the A node by preset condition, is connect so that the objective matrix is converted to the neck
Matrix, the ratio set that the preset condition is occurred jointly based on first label and second label, such as mark
Label i, j calculating m (i, j)=p (i | j)=nij/nj, nijFor the number that label i, j occur jointly, njThe number occurred for label j.
If 305, the similarity of the M label and R label meets preset condition, it is determined that second image is inspection
The corresponding search result of rope information.
In the present embodiment, the R label is the label of the retrieval information instruction, and R is positive integer.
It is understood that retrieval information is the relevant textual information of user's input, can be determined by text information
Corresponding R label can determine that the second image is if the similarity of the M label and R label meets preset condition
Retrieve the corresponding search result of information.Wherein, the second image can be the highest figure of label similarity corresponding with retrieval information
Picture, such as the similarity of the second image reach 90%；It is also possible to label similarity corresponding with retrieval information and is greater than certain threshold
The image of value, for example, image of the similarity greater than 80% is classified as search result；Specifically, retrieving output can be one
The highest image of similarity is opened, is also possible to the set of multiple images with similitude, and be ranked up in sequence,
In order to which user selects.
In conjunction with above-described embodiment it is found that passing through the corresponding relationship for determining the first image and multiple labels, convolution is then used
Neural network extracts the characteristic information of the first image, pre- further according to second to generate feature vector based on default dimension
The multiple label is converted to multiple term vectors by imputation method；It is closed according to described eigenvector is corresponding with the multiple term vector
System is trained preset model, to obtain image classification model, to carry out label pass to all images in searching database
Connection carries out similarity-rough set according to the label of the label of retrieval information instruction and image when user inputs retrieval information
The image of Search Requirement is obtained, to realize the image retrieval procedure based on semantic label, image inspection is can be improved in this method
The high efficiency and accuracy of rope process, and due to the operability of label, further improve the convenience of image retrieval procedure
Property and flexibility.
In a kind of possible scene, the association process of feature vector and multiple label term vectors for image be can be
The judgement on the connection side based on term vector and generate；In the following, the scene is described with reference to the drawings, as shown in figure 4, Fig. 4
For the flow chart of another image search method based on semantic label provided by the embodiments of the present application, the embodiment of the present application is at least
The following steps are included:
401, the corresponding relationship of the first image Yu N number of label is determined
In the present embodiment, N number of label is used to indicate the display content of the first image, and N is positive integer.First figure
As can be any one image in image data base, N number of label, which can be, perhaps amplifies content according in the first image
Multiple labels indicated by corresponding relationship i.e. the first picture material of a series of label of setting, the first image and N number of label
Set.
402, the characteristic information of the first image is extracted, by the first preset algorithm to generate feature based on default dimension
Vector.
In the present embodiment, it may include convolutional neural networks that first preset algorithm is set based on neural network algorithm；
Such as: feature is carried out to image image data set using the good model resnet50 convolutional network of the pre-training on ImageNet and is mentioned
It takes, obtains the feature vector of 2048 dimensions.
It is understood that preset dimension can be set according to the related needs of user, it can also be according to image retrieval
The setting range setting that dimension is preset in the historical record of system, depending on specific method is because of actual scene.
403, N number of label is converted to by N number of term vector according to the second preset algorithm；
In the present embodiment, the second preset algorithm is used to determine N number of label vector between such label and label
The connection of amount gone to measure the relationship between them, excavate between label.
In a kind of possible scene, N number of label can be converted to as mark by N number of term vector by word2vec
The character representation of label.
It is understood that the term vector of multiple words in label can be averaged, so if including multiple words in label
Next calculating process is carried out in the term vector for being denoted as corresponding label afterwards.
404, the incidence relation between N number of term vector is determined, whether to judge the corresponding node of image feature vector
In the presence of connection side.
In the present embodiment, the process of the incidence relation between N number of term vector is determined, first according to preset rules to institute
It states N number of term vector to be handled, to construct objective matrix, the preset rules are set based on the cooccurrence relation of N number of term vector
Fixed, the objective matrix is used to indicate the ratio that first label and second label occur jointly；If the target square
Battle array meets preset condition, then N number of term vector is converted to the A node, so that the objective matrix is converted to institute
State Adjacency matrix, the ratio set that the preset condition is occurred jointly based on first label and second label.Such as:
According in training dataset label cooccurrence relation construct n × n matrix m, wherein n be label number, wherein m (i, j)=p (i |
J)=nij/nj, nijFor the number that label i, j occur jointly, njThe number occurred for label j.As m (i, j) > τ, a is enabledij=1,
Adjacency matrix A (n, n) is obtained to construct, wherein A (i, j)=aij。
Then, it is determined that the first node in the A node；The first node and institute are calculated separately according to judgment rule
It states except the node of the first node is with the presence or absence of connection side in A node, to obtain judging result；According to the A node
Network topology structure is determined with the corresponding relationship of the judging result, it is to be appreciated that for each section in A node
Point can carry out above-mentioned processing analysis, such as: tectonic network topological structure G=(V, E), wherein V is node, that is, corresponds to N
Whether a label, connection side of the E between node have connection side between two nodes, according to the member in the adjacency matrix A of node
Depending on prime number value, work as aijThere is connection side when=1, between node i and node j, otherwise then there is no connection sides.
405, model is trained by figure convolutional neural networks, to obtain image classification model.
In the present embodiment, the network topology structure constructed in 404 through the above steps, then pass through training picture scroll product nerve net
Network model is trained network topology structure, and when training, the feature vector of each node can be traveled in adjacent node, because
This each node can absorb information from adjacent knot vector, and nodal information has text label to convert by word2vec
Study obtains, and can retain the semantic neighbor information on certain, therefore still can between adjacent node after network transmission
It is enough to keep certain semantic neighbor information, after training in the case where keeping semanteme centainly adjacent, obtain image point
Class model.
406, label is carried out to image in database according to image classification model to determine.
In the present embodiment, the process that label determines, which can be, exports second image according to described image disaggregated model
Feature vector；According to the feature vector of second image, knot vector corresponding with the N number of label passes through sigmoid respectively
Function is calculated, to obtain tag parameter；The tag parameter is normalized, to obtain N number of label pair
The probability of second image described in Ying Yu.
Specifically, in order to be matched with characteristics of image dimension, can be set since each node obtains knot vector
The output dimension of each node is preset 2048, then each node can be used as classifier wi, knot vector and figure to be sorted
Score σ (the w being multiplied as vector xiX), wherein image to be classified vector x is the corresponding image data base of the first image (comprising the
Two images) in not set corresponding label each image characteristics of image vector indicate, σ be sigmoid function, will output point
Number normalizes between [0,1], this score is then used as text content to belong to the probability of the label, passes through setting for predetermined probabilities
It is fixed, such as: Probability p > 0.5, then explanation includes the label, if to realize the purpose of multi-tag classification.
407, according to the similitude of the retrieval information of user's input and associated picture label.
In the present embodiment, retrieval information is the relevant textual information of user's input, can determine phase by text information
The R label answered can determine the second image for inspection if the similarity of the M label and R label meets preset condition
The corresponding search result of rope information.
408, search result is determined.
In the present embodiment, search result is the second image, it is to be understood that the second image can be and retrieval information pair
The highest image of the label similarity answered, such as the similarity of the second image reach 90%；It is also possible to corresponding with retrieval information
Label similarity be greater than the image of certain threshold value, for example, similarity is classified as search result greater than 80% image；Specifically,
Retrieving output can be the highest image of similarity, be also possible to the set of multiple images with similitude,
And be ranked up in sequence, in order to which user selects.
In conjunction with above-described embodiment it is found that passing through the corresponding relationship for determining the first image and multiple labels, convolution is then used
Neural network extracts the characteristic information of the first image, pre- further according to second to generate feature vector based on default dimension
The multiple label is converted to multiple term vectors by imputation method；It is closed according to described eigenvector is corresponding with the multiple term vector
System is trained preset model, to obtain image classification model, to carry out label pass to all images in searching database
Connection carries out similarity-rough set according to the label of the label of retrieval information instruction and image when user inputs retrieval information
The image of Search Requirement is obtained, to realize the image retrieval procedure based on semantic label, image inspection is can be improved in this method
The high efficiency and accuracy of rope process, and due to the operability of label, further improve the convenience of image retrieval procedure
Property and flexibility.
In a kind of possible display mode, display mode as described in Figure 5 can be used, Fig. 5 is the embodiment of the present application
A kind of schematic diagram of interface display of the image retrieval provided.The interface may include label information and the output of user's input
As a result.When user needs to retrieve associated picture, corresponding multiple labels are generated according to the text information of user's input, such as:
Landscape, massif, it can obtain corresponding retrieving image, if user needs to know specific retrieving, can click in detail
Feelings button, can show the image similarity list retrieved, and user is also possible to others in selective listing and meets itself
The image of demand.
It is understood that the coherent element in step corresponding in above-mentioned Fig. 3 and Fig. 4 embodiment may be displayed on this
In interface, depending on particular content answers actual scene, herein without limitation.
For the above scheme of better implementation the embodiment of the present application, correlation for implementing the above scheme is also provided below
Device.Referring to Fig. 6, Fig. 6 is the structural schematic diagram of image retrieving apparatus provided by the embodiments of the present application, image retrieving apparatus
600 include:
Determination unit 601, for determining that the corresponding relationship of the first image Yu N number of label, N number of label are used to indicate institute
The display content of the first image is stated, N is positive integer；
Extraction unit 602, for extracting the characteristic information of the first image by the first preset algorithm, based on default
Dimension generates feature vector, and first preset algorithm includes convolutional neural networks；
Converting unit 603, for N number of label to be converted to N number of term vector according to the second preset algorithm；
Training unit 604, for according to the corresponding relationship of described eigenvector and N number of term vector to preset model into
Row training, to obtain image classification model, described image disaggregated model is used to generate corresponding M label, M according to the second image
≤ N, M are positive integer；
Retrieval unit 605, if the similarity for the M label and R label meets preset condition, it is determined that described
Second image is the corresponding search result of retrieval information, and the R label is the label of the retrieval information instruction, and R is positive whole
Number.
Preferably, in some possible implementations of the application,
The training unit 604, specifically for determining A node according to N number of term vector, to construct Adjacency matrix,
The Adjacency matrix is used to indicate the incidence relation of any two node in the A node, and A≤N, A are positive integer；
The training unit 604, specifically for determining net according to the incidence relation of any two node in the A node
Network topological structure；
The training unit 604, specifically for being trained by the second preset algorithm to the network topology structure, with
Image classification model is obtained, second preset algorithm is used to indicate A node described in the network topology structure and carries out spy
Vector exchange is levied, second preset algorithm includes figure convolutional neural networks.
Preferably, in some possible implementations of the application, N number of term vector includes the first label and the second mark
Label,
The training unit 604, specifically for being handled according to preset rules N number of term vector, to construct mesh
Matrix is marked, the preset rules are set based on the cooccurrence relation of N number of term vector, and the objective matrix is used to indicate described the
The ratio that one label and second label occur jointly；
The training unit 604, if meeting preset condition specifically for the objective matrix, by N number of term vector
The A node is converted to, so that the objective matrix is converted to the Adjacency matrix, the preset condition is based on described the
The ratio set that one label and second label occur jointly.
Preferably, in some possible implementations of the application,
The training unit 604, specifically for the first node in the determination A node；
The training unit 604, specifically for calculating separately the first node and the A node according to judgment rule
In except the first node node with the presence or absence of connection side, to obtain judging result；
The training unit 604, specifically for determining net according to the corresponding relationship of the A node and the judging result
Network topological structure.
Preferably, in some possible implementations of the application,
The converting unit 603, specifically for obtaining the vocabulary of each label in N number of label；
The converting unit 603 calculates the flat of the vocabulary if having a large vocabulary specifically for described in preset threshold
Mean value；
The converting unit 603, specifically for converting the average value of the vocabulary according to according to the second preset algorithm
For N number of term vector.
Preferably, in some possible implementations of the application,
The training unit 604, for judging that N number of label corresponds to described second according to described image disaggregated model
The probability of image；
The training unit 604, for determining that probability in N number of label meets M label of class condition.
Preferably, in some possible implementations of the application,
The training unit 604, for exporting the feature vector of second image according to described image disaggregated model；
The training unit 604, for corresponding with N number of label respectively according to the feature vector of second image
Knot vector is calculated by sigmoid function, to obtain tag parameter；
The training unit 604, for the tag parameter to be normalized, to obtain N number of label pair
The probability of second image described in Ying Yu.
By determining the corresponding relationship of the first image and multiple labels, then using convolutional neural networks to the first image
Characteristic information extracts, to generate feature vector based on default dimension, further according to the second preset algorithm by the multiple label
Be converted to multiple term vectors；Preset model is instructed according to described eigenvector and the corresponding relationship of the multiple term vector
Practice, to obtain image classification model, to carry out label association to all images in searching database, when user inputs retrieval letter
When breath, carrying out similarity-rough set according to the label of the label of retrieval information instruction and image can be obtained the image of Search Requirement,
To realize the image retrieval procedure based on semantic label, the high efficiency of image retrieval procedure and accurate is can be improved in this method
Property, and due to the operability of label, further improve convenience and the flexibility of image retrieval procedure.
The embodiment of the present application also provides a kind of image retrieving apparatus, referring to Fig. 7, Fig. 7 is that the embodiment of the present application provides
Another image retrieving apparatus structural schematic diagram, which can generate ratio because configuration or performance are different
Biggish difference may include one or more central processing units (central processing units, CPU) 722
(for example, one or more processors) and memory 732, one or more storage application programs 742 or data 744
Storage medium 730 (such as one or more mass memory units).Wherein, memory 732 and storage medium 730 can be with
It is of short duration storage or persistent storage.The program for being stored in storage medium 730 may include that (diagram does not have one or more modules
Mark), each module may include to the series of instructions operation in image retrieving apparatus.Further, central processing unit
722 can be set to communicate with storage medium 730, and a series of fingers in storage medium 730 are executed on image retrieving apparatus 700
Enable operation.
Image retrieving apparatus 700 can also include one or more power supplys 726, one or more wired or nothings
Wired network interface 750, one or more input/output interfaces 758, and/or, one or more operating systems 741,
Such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM etc..
The step as performed by image retrieving apparatus can be based on the image retrieving apparatus shown in Fig. 7 in above-described embodiment
Structure.
A kind of computer readable storage medium is also provided in the embodiment of the present application, is stored in the computer readable storage medium
There is image retrieval instruction, when run on a computer, so that computer is executed as earlier figures 2 to embodiment illustrated in fig. 5 is retouched
Step performed by image retrieving apparatus in the method stated.
A kind of computer program product including image retrieval instruction is also provided in the embodiment of the present application, when it is in computer
When upper operation, so that computer is executed as performed by the image retrieving apparatus into the method described in embodiment illustrated in fig. 5 of earlier figures 2
The step of.
The embodiment of the present application also provides a kind of image indexing system, described image searching system may include Fig. 6 and be retouched
State image retrieving apparatus described in the image retrieving apparatus or Fig. 7 in embodiment.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In several embodiments provided herein, it should be understood that disclosed system, device and method can be with
It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit
It divides, only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components
It can be combined or can be integrated into another system, or some features can be ignored or not executed.Another point, it is shown or
The mutual coupling, direct-coupling or communication connection discussed can be through some interfaces, the indirect coupling of device or unit
It closes or communicates to connect, can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, each functional unit in each embodiment of the application can integrate in one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage medium.Based on this understanding, the technical solution of the application is substantially
The all or part of the part that contributes to existing technology or the technical solution can be in the form of software products in other words
It embodies, which is stored in a storage medium, including some instructions are used so that a computer
Equipment (can be personal computer, image retrieving apparatus or the network equipment etc.) executes side described in each embodiment of the application
The all or part of the steps of method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (read-only
Memory, ROM), random access memory (random access memory, RAM), magnetic or disk etc. is various to deposit
Store up the medium of program code.
The above, above embodiments are only to illustrate the technical solution of the application, rather than its limitations；Although referring to before
Embodiment is stated the application is described in detail, those skilled in the art should understand that: it still can be to preceding
Technical solution documented by each embodiment is stated to modify or equivalent replacement of some of the technical features；And these
It modifies or replaces, the spirit and scope of each embodiment technical solution of the application that it does not separate the essence of the corresponding technical solution.
Claims (10)
1. a kind of image search method based on semantic label characterized by comprising
Determine that the corresponding relationship of the first image Yu N number of label, N number of label are used to indicate in the display of the first image
Hold, N is positive integer；
The characteristic information of the first image is extracted, by the first preset algorithm to generate feature vector, institute based on default dimension
Stating the first preset algorithm includes convolutional neural networks；
N number of label is converted into N number of term vector according to the second preset algorithm；
Preset model is trained according to described eigenvector and the corresponding relationship of N number of term vector, to obtain image point
Class model, described image disaggregated model are used to generate corresponding M label according to the second image, and M≤N, M are positive integer；
If the similarity of the M label and R label meets preset condition, it is determined that second image is retrieval information pair
The search result answered, the R label are the label of the retrieval information instruction, and R is positive integer.
2. the method according to claim 1, wherein described according to described eigenvector and N number of term vector
Corresponding relationship preset model is trained, to obtain image classification model, comprising:
A node is determined according to N number of term vector, and to construct Adjacency matrix, the Adjacency matrix is used to indicate the A section
The incidence relation of any two node in point, A≤N, A are positive integer；
Network topology structure is determined according to the incidence relation of any two node in the A node；
The network topology structure is trained by the second preset algorithm, to obtain image classification model, described second is pre-
Imputation method is used to indicate A node described in the network topology structure and carries out feature vector exchange, second preset algorithm
Including figure convolutional neural networks.
3. according to the method described in claim 2, it is characterized in that, N number of term vector include the first label and the second label,
It is described that A node is determined according to N number of term vector, to construct Adjacency matrix, comprising:
N number of term vector is handled according to preset rules, to construct objective matrix, the preset rules are based on the N
The cooccurrence relation of a term vector is set, and the objective matrix is used to indicate first label and second label occurs jointly
Ratio；
If the objective matrix meets preset condition, N number of term vector is converted into the A node, so that described
Objective matrix is converted to the Adjacency matrix, and the preset condition is based on first label and second label occurs jointly
Ratio set.
4. according to the method described in claim 2, it is characterized in that, described according to any two node in the A node
Incidence relation determines network topology structure, comprising:
Determine the first node in the A node；
The first node is calculated separately according to judgment rule whether to deposit with the node for removing the first node in the A node
On connection side, to obtain judging result；
Network topology structure is determined according to the corresponding relationship of the A node and the judging result.
5. method according to claim 1-4, which is characterized in that it is described according to the second preset algorithm by the N
A label is converted to N number of term vector, comprising:
Obtain the vocabulary of each label in N number of label；
If described have a large vocabulary in preset threshold, the average value of the vocabulary is calculated；
The average value of the vocabulary is converted into N number of term vector according to according to the second preset algorithm.
6. method according to claim 1-4, which is characterized in that described to generate corresponding M according to the second image
A label, comprising:
Judge that N number of label corresponds to the probability of second image according to described image disaggregated model；
Determine that probability in N number of label meets M label of class condition.
7. according to the method described in claim 6, it is characterized in that, it is described judged according to described image disaggregated model it is described N number of
Label corresponds to the probability of second image, comprising:
The feature vector of second image is exported according to described image disaggregated model；
According to the feature vector of second image, knot vector corresponding with the N number of label passes through sigmoid function respectively
It is calculated, to obtain tag parameter；
The tag parameter is normalized, to obtain the probability that N number of label corresponds to second image.
8. a kind of image retrieving apparatus based on semantic label characterized by comprising
Determination unit, for determining that the corresponding relationship of the first image Yu N number of label, N number of label are used to indicate described first
The display content of image, N are positive integer；
Extraction unit, for extracting the characteristic information of the first image by the first preset algorithm, with raw based on default dimension
At feature vector, first preset algorithm includes convolutional neural networks；
Converting unit, for N number of label to be converted to N number of term vector according to the second preset algorithm；
Training unit, for being trained according to described eigenvector and the corresponding relationship of N number of term vector to preset model,
To obtain image classification model, described image disaggregated model is used to generate corresponding M label according to the second image, and M≤N, M are
Positive integer；
Retrieval unit, if the similarity for the M label and R label meets preset condition, it is determined that second figure
As to retrieve the corresponding search result of information, the R label is the label of the retrieval information instruction, and R is positive integer.
9. a kind of computer equipment, which is characterized in that the computer equipment includes processor and memory:
The memory is for storing program code；The processor is used for according to the instruction execution right in said program code
It is required that 1 to 7 described in any item image search methods based on semantic label.
10. a kind of computer readable storage medium, it is stored with instruction in the computer readable storage medium, when it is in computer
When upper operation, so that computer executes the image search method based on semantic label described in the claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910770152.2A CN110472090B (en) | 2019-08-20 | 2019-08-20 | Image retrieval method based on semantic tags, related device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910770152.2A CN110472090B (en) | 2019-08-20 | 2019-08-20 | Image retrieval method based on semantic tags, related device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110472090A true CN110472090A (en) | 2019-11-19 |
| CN110472090B CN110472090B (en) | 2023-10-27 |
Family
ID=68512976
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910770152.2A Active CN110472090B (en) | 2019-08-20 | 2019-08-20 | Image retrieval method based on semantic tags, related device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110472090B (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111291212A (en) * | 2020-01-24 | 2020-06-16 | 复旦大学 | Zero sample sketch image retrieval method and system based on graph convolution neural network |
| CN111368934A (en) * | 2020-03-17 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Image recognition model training method, image recognition method and related device |
| CN111597375A (en) * | 2020-05-19 | 2020-08-28 | 清华大学 | Picture retrieval method based on similar picture group representative feature vector and related equipment |
| CN111613299A (en) * | 2020-06-15 | 2020-09-01 | 山东搜搜中医信息科技有限公司 | Multi-label analysis technology of traditional Chinese medicine data |
| CN111813967A (en) * | 2020-07-14 | 2020-10-23 | 中国科学技术信息研究所 | Retrieval method, retrieval device, computer equipment and storage medium |
| CN112084359A (en) * | 2020-09-18 | 2020-12-15 | 维沃移动通信有限公司 | Picture retrieval method and device and electronic equipment |
| CN112434722A (en) * | 2020-10-23 | 2021-03-02 | 浙江智慧视频安防创新中心有限公司 | Label smooth calculation method and device based on category similarity, electronic equipment and medium |
| CN112732968A (en) * | 2021-01-12 | 2021-04-30 | 特赞（上海）信息科技有限公司 | Case material image retrieval method, device, equipment and storage medium |
| CN113204659A (en) * | 2021-03-26 | 2021-08-03 | 北京达佳互联信息技术有限公司 | Label classification method and device for multimedia resources, electronic equipment and storage medium |
| CN113343013A (en) * | 2019-12-24 | 2021-09-03 | 北京旷视科技有限公司 | Target object determination method and device and electronic equipment |
| CN113836933A (en) * | 2021-07-27 | 2021-12-24 | 腾讯科技（深圳）有限公司 | Method and device for generating graphic mark, electronic equipment and storage medium |
| CN113868447A (en) * | 2021-09-27 | 2021-12-31 | 新智认知数据服务有限公司 | Picture retrieval method, electronic device and computer-readable storage medium |
| CN115146103A (en) * | 2022-09-01 | 2022-10-04 | 太平金融科技服务(上海)有限公司深圳分公司 | Image retrieval method, image retrieval apparatus, computer device, storage medium, and program product |
| CN115238081A (en) * | 2022-06-14 | 2022-10-25 | 杭州原数科技有限公司 | Cultural relic intelligent identification method and system and readable storage medium |
| CN117194698A (en) * | 2023-11-07 | 2023-12-08 | 清华大学 | Task processing system and method based on OAR semantic knowledge base |
| WO2024104438A1 (en) * | 2022-11-16 | 2024-05-23 | 中移(苏州)软件技术有限公司 | Multimedia retrieval method and apparatus, and device, medium and program product |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120209833A1 (en) * | 2011-02-11 | 2012-08-16 | Siemens Aktiengesellschaft | Methods and devices for data retrieval |
| CN109271546A (en) * | 2018-07-25 | 2019-01-25 | 西北大学 | The foundation of image retrieval Feature Selection Model, Database and search method |
-
2019
- 2019-08-20 CN CN201910770152.2A patent/CN110472090B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120209833A1 (en) * | 2011-02-11 | 2012-08-16 | Siemens Aktiengesellschaft | Methods and devices for data retrieval |
| CN109271546A (en) * | 2018-07-25 | 2019-01-25 | 西北大学 | The foundation of image retrieval Feature Selection Model, Database and search method |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113343013A (en) * | 2019-12-24 | 2021-09-03 | 北京旷视科技有限公司 | Target object determination method and device and electronic equipment |
| CN111291212A (en) * | 2020-01-24 | 2020-06-16 | 复旦大学 | Zero sample sketch image retrieval method and system based on graph convolution neural network |
| CN111291212B (en) * | 2020-01-24 | 2022-10-11 | 复旦大学 | Zero sample sketch image retrieval method and system based on graph convolution neural network |
| CN111368934A (en) * | 2020-03-17 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Image recognition model training method, image recognition method and related device |
| CN111368934B (en) * | 2020-03-17 | 2023-09-19 | 腾讯科技（深圳）有限公司 | Image recognition model training method, image recognition method and related device |
| CN111597375A (en) * | 2020-05-19 | 2020-08-28 | 清华大学 | Picture retrieval method based on similar picture group representative feature vector and related equipment |
| CN111613299A (en) * | 2020-06-15 | 2020-09-01 | 山东搜搜中医信息科技有限公司 | Multi-label analysis technology of traditional Chinese medicine data |
| CN111813967A (en) * | 2020-07-14 | 2020-10-23 | 中国科学技术信息研究所 | Retrieval method, retrieval device, computer equipment and storage medium |
| CN111813967B (en) * | 2020-07-14 | 2024-01-30 | 中国科学技术信息研究所 | Retrieval method, retrieval device, computer equipment and storage medium |
| CN112084359A (en) * | 2020-09-18 | 2020-12-15 | 维沃移动通信有限公司 | Picture retrieval method and device and electronic equipment |
| CN112434722A (en) * | 2020-10-23 | 2021-03-02 | 浙江智慧视频安防创新中心有限公司 | Label smooth calculation method and device based on category similarity, electronic equipment and medium |
| CN112434722B (en) * | 2020-10-23 | 2024-03-19 | 浙江智慧视频安防创新中心有限公司 | Label smooth calculation method and device based on category similarity, electronic equipment and medium |
| CN112732968A (en) * | 2021-01-12 | 2021-04-30 | 特赞（上海）信息科技有限公司 | Case material image retrieval method, device, equipment and storage medium |
| CN113204659A (en) * | 2021-03-26 | 2021-08-03 | 北京达佳互联信息技术有限公司 | Label classification method and device for multimedia resources, electronic equipment and storage medium |
| CN113204659B (en) * | 2021-03-26 | 2024-01-19 | 北京达佳互联信息技术有限公司 | Label classification method and device for multimedia resources, electronic equipment and storage medium |
| CN113836933A (en) * | 2021-07-27 | 2021-12-24 | 腾讯科技（深圳）有限公司 | Method and device for generating graphic mark, electronic equipment and storage medium |
| CN113868447A (en) * | 2021-09-27 | 2021-12-31 | 新智认知数据服务有限公司 | Picture retrieval method, electronic device and computer-readable storage medium |
| CN115238081B (en) * | 2022-06-14 | 2024-04-30 | 杭州原数科技有限公司 | Intelligent cultural relic identification method, system and readable storage medium |
| CN115238081A (en) * | 2022-06-14 | 2022-10-25 | 杭州原数科技有限公司 | Cultural relic intelligent identification method and system and readable storage medium |
| CN115146103A (en) * | 2022-09-01 | 2022-10-04 | 太平金融科技服务(上海)有限公司深圳分公司 | Image retrieval method, image retrieval apparatus, computer device, storage medium, and program product |
| WO2024104438A1 (en) * | 2022-11-16 | 2024-05-23 | 中移(苏州)软件技术有限公司 | Multimedia retrieval method and apparatus, and device, medium and program product |
| CN117194698A (en) * | 2023-11-07 | 2023-12-08 | 清华大学 | Task processing system and method based on OAR semantic knowledge base |
| CN117194698B (en) * | 2023-11-07 | 2024-02-06 | 清华大学 | Task processing system and method based on OAR semantic knowledge base |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110472090B (en) | 2023-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110472090A (en) | Image search method and relevant apparatus, storage medium based on semantic label | |
| CN110866140B (en) | Image feature extraction model training method, image searching method and computer equipment | |
| CN111523621B (en) | Image recognition method and device, computer equipment and storage medium | |
| WO2020192442A1 (en) | Method for generating classifier using a small number of annotated images | |
| CN110647802A (en) | Remote sensing image ship target detection method based on deep learning | |
| CN108229347A (en) | For the method and apparatus of the deep layer displacement of the plan gibbs structure sampling of people's identification | |
| Adedoja et al. | Intelligent mobile plant disease diagnostic system using NASNet-mobile deep learning | |
| Shoohi et al. | DCGAN for Handling Imbalanced Malaria Dataset based on Over-Sampling Technique and using CNN. | |
| CN114818734B (en) | Method and device for analyzing antagonism scene semantics based on target-attribute-relation | |
| CN111340213B (en) | Neural network training method, electronic device, and storage medium | |
| US8150212B2 (en) | System and method for automatic digital image orientation detection | |
| Banerjee et al. | Cnn-svm model for accurate detection of bacterial diseases in cucumber leaves | |
| CN112052816A (en) | Human behavior prediction method and system based on adaptive graph convolution countermeasure network | |
| CN112418256A (en) | Classification, model training and information searching method, system and equipment | |
| CN111767985B (en) | Neural network training method, video identification method and device | |
| Fujii et al. | Hierarchical group-level emotion recognition in the wild | |
| US20210292805A1 (en) | Semi-supervised classification of microorganism | |
| Liao et al. | A flower classification method combining DenseNet architecture with SVM | |
| Yu et al. | Avocado ripeness classification using graph neural network | |
| CN110457387A (en) | A kind of method and relevant apparatus determining applied to user tag in network | |
| CN109460485A (en) | A kind of image library method for building up, device and storage medium | |
| CN112929380B (en) | Trojan horse communication detection method and system combining meta-learning and spatiotemporal feature fusion | |
| CN115048504A (en) | Information pushing method and device, computer equipment and computer readable storage medium | |
| CN114022698A (en) | Multi-tag behavior identification method and device based on binary tree structure | |
| Bahrami et al. | Image concept detection in imbalanced datasets with ensemble of convolutional neural networks |
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 |