CN109241318A - Picture recommendation method, device, computer equipment and storage medium - Google Patents
Picture recommendation method, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN109241318A CN109241318A CN201811106466.4A CN201811106466A CN109241318A CN 109241318 A CN109241318 A CN 109241318A CN 201811106466 A CN201811106466 A CN 201811106466A CN 109241318 A CN109241318 A CN 109241318A
- Authority
- CN
- China
- Prior art keywords
- picture
- candidate
- pictures
- training
- preset
- 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses picture recommendation method, device, computer equipment and storage mediums.This method comprises: being classified to pre-stored candidate picture according to preset picture classification rule to obtain multiple candidate pictures;Picture recommended models are established according to preset characteristic matching formula and multiple candidate pictures;Picture recommended models are trained by multiple groups training parameter and multiple training pictures;If receiving the picture to be matched that user is inputted, picture to be matched is matched with multiple candidate pictures according to the picture recommended models after training, and the matching probability of corresponding candidate picture is calculated;The matching probability of obtained all candidate pictures is ranked up, Target Photo is obtained according to preset Target Photo quantity.The present invention is based on intelligent Decision Technology, the picture that can be efficiently inputted to user carries out accurate matching to obtain similar pictures, significantly reduces the error during picture match, save match time.
Description
Technical field
The present invention relates to field of computer technology more particularly to a kind of picture recommendation method, device, computer equipment and deposit
Storage media.
Background technique
It, need to be by multiple in the picture inputted and picture library in the development process for carrying out computer programming project
Picture is matched, to obtain Target Photo similar with the picture inputted in picture library.In existing picture match method,
Carrying out matching to the global feature of picture, to will lead to matching error rate larger, and is matched to each pixel in picture and then can
Because time-consuming for matching caused by huge calculation amount.Thus existing picture match method exists and can not efficiently and accurately match to obtain
The problem of similar pictures.
Summary of the invention
The embodiment of the invention provides a kind of picture recommendation method, device, computer equipment and storage mediums, it is intended to solve
The problem of obtaining similar pictures can not efficiently and accurately be matched by existing in the prior art.
In a first aspect, the embodiment of the invention provides a kind of picture recommendation methods comprising:
Multiple pre-stored candidate pictures are obtained, the candidate picture is divided according to preset picture classification rule
Class is to obtain multiple candidate pictures;
Picture recommended models are established according to preset characteristic matching formula and obtained multiple candidate pictures;
The picture recommended models established are carried out by preset multiple groups training parameter and multiple preset training pictures
Picture recommended models after being trained;
If receiving the picture to be matched that user is inputted, according to the picture recommended models after training to picture to be matched with
Multiple candidate's pictures are matched, and the matching probability of corresponding candidate picture is calculated;
The matching probability of obtained all candidate pictures is ranked up, mesh is obtained according to preset Target Photo quantity
It marks on a map piece.
Second aspect, the embodiment of the invention provides a kind of picture recommendation apparatus comprising:
Picture classification unit, it is right according to preset picture classification rule for obtaining multiple pre-stored candidate pictures
Candidate's picture is classified to obtain multiple candidate pictures;
Picture recommended models construction unit, for according to preset characteristic matching formula and obtained multiple candidate pictures
Collection establishes picture recommended models;
Picture recommended models training unit, for passing through preset multiple groups training parameter and multiple preset training pictures pair
The picture recommended models established are trained the picture recommended models after being trained;
Matching probability computing unit, if the picture to be matched inputted for receiving user, according to the picture after training
Recommended models match picture to be matched with multiple candidate pictures, and the matching that corresponding candidate picture is calculated is general
Rate;
Target Photo acquiring unit is ranked up, according to pre- for the matching probability to obtained all candidate pictures
If Target Photo quantity obtain Target Photo.
The third aspect, the embodiment of the present invention provide a kind of computer equipment again comprising memory, processor and storage
On the memory and the computer program that can run on the processor, the processor execute the computer program
Picture recommendation method described in the above-mentioned first aspect of Shi Shixian.
Fourth aspect, the embodiment of the invention also provides a kind of computer readable storage mediums, wherein the computer can
It reads storage medium and is stored with computer program, it is above-mentioned that the computer program when being executed by a processor executes the processor
Picture recommendation method described in first aspect.
The embodiment of the invention provides a kind of picture recommendation method, device, computer equipment and storage mediums.By that will wait
It selects picture classification to multiple candidate pictures, and establishes picture recommended models according to candidate pictures, pass through preset training ginseng
Several and training picture is trained picture recommended models, and matches to obtain matching probability according to the picture recommended models after training
Highest Target Photo is exported, and the picture that can be efficiently inputted to user carries out accurate matching to obtain similar diagram
Piece significantly reduces the error during picture match, saves match time.
Detailed description of the invention
Technical solution in order to illustrate the embodiments of the present invention more clearly, below will be to needed in embodiment description
Attached drawing is briefly described, it should be apparent that, drawings in the following description are some embodiments of the invention, general for this field
For logical technical staff, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is the flow diagram of picture recommendation method provided in an embodiment of the present invention;
Fig. 2 is another flow diagram of picture recommendation method provided in an embodiment of the present invention;
Fig. 3 is the sub-process schematic diagram of picture recommendation method provided in an embodiment of the present invention;
Fig. 4 is another sub-process schematic diagram of picture recommendation method provided in an embodiment of the present invention;
Fig. 5 is another sub-process schematic diagram of picture recommendation method provided in an embodiment of the present invention;
Fig. 6 is the schematic block diagram of picture recommendation apparatus provided in an embodiment of the present invention;
Fig. 7 is another schematic block diagram of picture recommendation apparatus provided in an embodiment of the present invention;
Fig. 8 is the subelement schematic block diagram of picture recommendation apparatus provided in an embodiment of the present invention;
Fig. 9 is another subelement schematic block diagram of picture recommendation apparatus provided in an embodiment of the present invention;
Figure 10 is another subelement schematic block diagram of picture recommendation apparatus provided in an embodiment of the present invention;
Figure 11 is the schematic block diagram of computer equipment provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
It should be appreciated that ought use in this specification and in the appended claims, term " includes " and "comprising" instruction
Described feature, entirety, step, operation, the presence of element and/or component, but one or more of the other feature, whole is not precluded
Body, step, operation, the presence or addition of element, component and/or its set.
It is also understood that mesh of the term used in this description of the invention merely for the sake of description specific embodiment
And be not intended to limit the present invention.As description of the invention and it is used in the attached claims, unless on
Other situations are hereafter clearly indicated, otherwise " one " of singular, "one" and "the" are intended to include plural form.
It will be further appreciated that the term "and/or" used in description of the invention and the appended claims is
Refer to any combination and all possible combinations of one or more of associated item listed, and including these combinations.
Referring to Fig. 1, Fig. 1 is the flow diagram of picture recommendation method provided in an embodiment of the present invention.The picture is recommended
Method is applied to server end, and this method is executed by the application software being installed in server end, wherein server end
It is the enterprise terminal for executing picture recommendation method to recommend similar pictures.
As shown in Figure 1, the method comprising the steps of S101~S105.
S101, multiple pre-stored candidate pictures are obtained, according to preset picture classification rule to the candidate picture
Classify to obtain multiple candidate pictures.
Server end is previously stored with multiple candidate pictures, according to preset picture classification rule to the candidate picture into
Row classification is to obtain multiple candidate pictures.Server end is the enterprise for picture recommended models to be constructed and used
Terminal.Classification processing is carried out to all candidate pictures according to preset picture classification rule, to obtain multiple candidate pictures.Figure
Piece classifying rules is the Rule Information for classifying to candidate picture, wherein picture classification rule includes feature extraction
Formula, K-means clustering algorithm.
In one embodiment, as shown in Fig. 2, further including step S101A before step S101.
S101A, according to format of the preset Format adjusting rule to pre-stored all candidate pictures be adjusted with
Obtain the candidate picture of uniform format.
Since pre-stored candidate picture is unable to ensure format complete unity, before classifying to candidate picture,
Unified adjustment can be carried out according to format of the preset Format adjusting rule to pre-stored all candidate pictures.For example, can be right
The size of all candidate's pictures is adjusted, and the resolution adjustment by all candidate pictures is 200 × 200, and lattice can be obtained
The unified candidate picture of formula.
In one embodiment, as shown in figure 3, step S101 includes sub-step S1011, S1012 and S1013.
S1011, feature extraction is carried out to the candidate picture to obtain all candidates according to preset feature extraction formula
The characteristic variable of picture.
It is extracted according to characteristic variable of the preset feature extraction formula to all candidate pictures.Wherein, preset spy
Sign is extracted formula and is constructed based on convolutional neural networks.Preset feature extraction formula includes the first convolution calculation formula, pond
Change calculation formula, the second convolution calculation formula, the first full connection calculation formula, the second full connection calculation formula.It is mentioned by feature
Take the i.e. extractable characteristic variable for obtaining all candidate pictures of formula.The characteristic variable of candidate picture is by carrying out to picture
It is obtained for reflecting that the multi-C vector of picture characteristic, computer program can not carry out different pictures after feature extraction
Identification, but after multiple pictures are converted to characteristic variable, computer program is analyzed by the characteristic variable to multiple pictures
Different pictures can be identified.
Specifically, if the resolution ratio of the candidate picture of input is 200 × 200, according to the first convolution calculation formula, to differentiate
Rate 10*10 carries out convolution operation as window, step-length 1, to obtain the vector matrix that size is 190 × 190, that is to say time
Select the shallow-layer feature of picture;According to pond calculation formula, using resolution ratio 10 × 10 as window, step-length 10, progress is down-sampled,
To obtain the vector matrix that size is 18 × 18, the profound feature of candidate picture that is to say;According to the second convolution calculation formula,
Using resolution ratio 2 × 2 as window, the carry out convolution operation that step-length is 2, to obtain the vector matrix that size is 9 × 9.Pass through
One full connection calculation formula, calculates obtained 9 × 9 vector matrix, altogether includes five in the first full connection formula
Node, each node is associated with 9 × 9 vector matrix, that is to say and is calculated and 9 by five calculation formula respectively
The value of associated five nodes of × 9 vector matrix, first calculation formula are represented by C1=W1 × x1+B1, wherein C1
For the calculated value of first node, x1 is the value of candidate's picture vector matrix, and W1 and B1 are first node and the vector matrix
The default parameter value into associated first calculation formula can be calculated with the vector matrix by five calculation formula to pass
The value of five nodes of connection;The value of five nodes is calculated by the second full connection calculation formula to obtain the final candidate
The characteristic variable of picture, calculation formula are F1=a1 × C1+a2 × C2+a3 × C3+a4 × C4+a5 × C5;Wherein C1, C2, C3,
C4, C5 be five nodes associated with the vector matrix of the candidate picture value, a1, a2, a3, a4, a5 for five nodes extremely
The preset parameter value of final output node, the characteristic variable for finally obtaining candidate's picture is the moment of a vector of one 1 × 128 dimension
Battle array can be indicated using F1=(f1, f2 ... f128).
Feature extraction can be carried out to all candidate pictures to obtain all candidate pictures by extracting formula according to features described above
Characteristic variable: F1, F2, F3 ... Fn.
S1012, it is clustered according to characteristic variable of the preset K-means clustering algorithm to the candidate picture to obtain
Multiple monoids comprising mass center.
It is clustered according to characteristic variable of the preset K-means clustering algorithm to the candidate picture comprising multiple classes
The cluster result of group.Specifically, the quantity K of final required monoid is arranged in K-means clustering algorithm, according to the numerical value pair of K
The characteristic variable of all candidate's pictures is clustered, and the characteristic variable value of each monoid mass center in K monoid is obtained.For example,
K=5 is set in K-means clustering algorithm, then finally obtains the characteristic variable value of 5 monoids and each monoid mass center, mass center
It is the central point in the monoid, the characteristic variable value of mass center is the mean value of the quasi-group characteristic variable.
S1013, classified all candidate pictures to obtain comprising candidate picture according to obtained multiple monoids
Multiple candidate's pictures.
Classified according to obtained multiple monoids to all candidate pictures, to obtain multiple times comprising candidate picture
Select pictures.Since each monoid after cluster includes mass center of the characteristic variable value as the monoid, pass through meter
Candidate the distance between a picture and all mass centers are calculated, a mass center nearest with candidate's picture distance can be obtained, and
It will be in monoid corresponding to candidate's picture classification to the mass center.All candidate pictures are carried out classifying it according to the above method
Afterwards, multiple candidate pictures comprising corresponding candidate picture can be obtained.
S102, picture recommended models are established according to preset characteristic matching formula and obtained multiple candidate pictures.
Picture recommended models are established according to preset characteristic matching formula and obtained multiple candidate pictures.To realize
The picture inputted to user identifies, and calculates the inputted picture of user and the matching probability of candidate picture, needs
Picture recommended models are established by preset characteristic matching formula and obtained multiple candidate pictures.Specifically, described pre-
If characteristic matching formula be S=1/ (((f1-t1)2+(f2-t2)2+……+(fn-tn)2)-1+ 1), wherein f1, f2 ... fn
For the characteristic variable value of a certain candidate picture, t1, t2 ... tn are inputted the characteristic variable value of picture by user, due to candidate
The characteristic variable of picture is the vector matrix of one 1 × 128 dimension, then n=128 in above formula.
S103, by preset multiple groups training parameter and it is preset multiple training pictures to the picture recommended models established
It is trained the picture recommended models after being trained.
By preset multiple groups training parameter and it is preset multiple training photos to the picture recommended models established into
Row training, with the picture recommended models after being trained.Specifically, including learning rate, training time in each group of training parameter
Several and training termination condition, learning rate is the amplitude being adjusted to the feature extraction Parameters in Formula of picture recommended models
And the direction of adjustment, frequency of training are the number that picture recommended models are carried out with maximum training, training termination condition is pair
The conditional information that the training of picture recommended models is terminated, if reaching trained termination condition or reaching preset frequency of training,
Then terminate the training to picture recommended models.
For example, learning rate is adjustment amplitude 2% in preset one group of training parameter, adjustment direction is amplification adjustment, training
Number is 15 times, and training termination condition is that front and back trains the difference of obtained matching probability less than 3% twice.
In one embodiment, as shown in figure 4, step S103 includes sub-step S1031, S1032 and S1033.
S1031, feature extraction is carried out to multiple training pictures according to the feature extraction formula to obtain all training and scheme
The characteristic variable of piece.
Feature extraction is carried out to obtain all trained pictures to a variety of trained pictures according to preset feature extraction formula
Characteristic variable.During being trained to picture recommended models, by all times in all trained pictures and candidate pictures
Select picture as model training picture, all trained pictures are used as the positive sample in model training picture, in candidate pictures
Picture be used as the negative sample in model training picture.
The characteristic variable of S1032, the preset one group of training parameter of acquisition and all trained pictures push away the picture established
Model is recommended repeatedly to be trained, the accuracy rate of training picture in last time training is accurate as the model of this group of training parameter
Rate.
The characteristic variable of preset one group of training parameter and all trained pictures is obtained to the picture recommended models established
It is repeatedly trained, using the accuracy rate of training picture in last time training as the model accuracy rate of this group of training parameter.Tool
Body, it obtains preset one group of training parameter and feature extraction Parameters in Formula in picture recommended models is adjusted, according to tune
Feature extraction formula after whole extracts the characteristic variable of all trained pictures again.By being preset in picture recommended models
Characteristic matching formula the matching probability of a certain Zhang Xunlian picture and all model training pictures is calculated, selection with the instruction
Practice the highest model training picture of picture match probability and be used as training objective picture, to whether matching obtained training objective picture
Judged for the training picture itself, if the training objective picture that matching obtains is the training picture itself, the training figure
The matching result of piece is successful match;If matching obtained training objective picture is not the training picture itself, the training figure
The matching result of piece is that it fails to match, counts the matching result of all trained pictures, training picture match success in matching result
Probability be this training in training picture accuracy rate.Wherein, model training picture includes all trained pictures and candidate
All candidate pictures in pictures.
For example, preset trained picture is 100, all trained pictures and all models are instructed by picture recommended models
Practice picture to be matched, it is 78 that its own is matched in 100 trained pictures, and 22 trained pictures are not matched to it certainly
Body, then the accuracy rate of training picture is 78% in this training.
According to the accuracy rate of the obtained trained picture of first time training and this group of training parameter in picture recommended models
The parameter of feature extraction formula is adjusted again, and according to picture recommended models adjusted again to all trained pictures with
It is matched with all model training pictures, executes above-mentioned training process repeatedly, until reaching trained termination condition or reaching pre-
If frequency of training, then terminate training to picture recommended models, obtain the accurate of training picture in last time training process
Model accuracy rate of the rate as this group of training parameter.
Default multiple groups training parameter is sequentially input by the above method, picture recommended models are trained, obtained
The model accuracy rate of all training parameters.
S1033, select one group of optimal training parameter to figure according to the model accuracy rate of obtained multiple groups training parameter
The parameter of feature extraction formula is configured with the picture recommended models after being trained in piece recommended models.
One group of optimal training parameter is selected according to the model accuracy rate of obtained multiple groups training parameter, that is to say selection
The highest one group of training parameter of model accuracy rate, is configured the parameter of feature extraction formula in picture recommended models
Picture recommended models after being trained.Specifically, the highest one group of training parameter of preference pattern accuracy rate, group training is joined
The parameter of number obtained feature extraction formula in last time training process when being trained to picture recommended models, as
The parameter of feature extraction formula is configured in picture recommended models after training.
If S104, the picture to be matched that user is inputted is received, according to the picture recommended models after training to be matched
Picture is matched with multiple candidate pictures, and the matching probability of corresponding candidate picture is calculated.
If receiving the picture to be matched that user is inputted, according to the picture recommended models after training to picture to be matched and more
It is matched in a candidate's pictures, and the matching probability of corresponding candidate picture is calculated by characteristic matching formula.
In one embodiment, as shown in figure 5, step S104 includes sub-step S1041, S1042, S1043 and S1044.
S1041, feature extraction is carried out to picture to be matched to obtain picture to be matched according to the feature extraction formula
Characteristic variable.
Specifically, feature extraction is carried out to picture to be matched according to the feature extraction formula of picture recommended models after training,
To obtain the characteristic variable of picture to be matched.
S1042, it is carried out according to the characteristic variable value of mass center in the characteristic variable of picture to be matched and multiple candidate pictures
It calculates to obtain target candidate pictures.
According to the characteristic variable value of mass center in the characteristic variable of picture to be matched and multiple candidate pictures calculated with
Obtain target candidate pictures.Each candidate corresponding specific monoid of pictures, each monoid includes a matter
The heart, specifically, by the distance between the characteristic variable value of mass center in the characteristic variable of picture to be matched and each candidate pictures
It is calculated, candidate pictures corresponding to the mass center obtained in all mass centers closest to the picture to be matched according to calculated result
As target candidate pictures.
S1043, feature extraction is carried out to the candidate picture in target candidate pictures to obtain according to feature extraction formula
The characteristic variable of candidate picture in the target candidate pictures.
The candidate picture in target candidate pictures is carried out according to the feature extraction formula of picture recommended models after training
Feature extraction is to obtain the characteristic variable of all candidate's pictures in the target candidate pictures.
S1044, according to the characteristic matching formula to of picture and picture to be matched candidate in target candidate pictures
It is calculated with probability to obtain the matching probability of all candidate pictures.
According to the characteristic matching formula in target candidate pictures it is all candidate pictures characteristic variable with it is to be matched
The matching probability of the characteristic variable of picture is calculated, to obtain the matching probability of all candidate's pictures in candidate's pictures.
S105, the matching probability of obtained all candidate pictures is ranked up, according to preset Target Photo quantity
Obtain Target Photo.
The matching probability of obtained all candidate pictures is ranked up, is obtained according to preset Target Photo quantity
To Target Photo to recommend to user.Preset Target Photo quantity, which is that user is default, obtains Target Photo
Quantity information, be ranked up according to the matching probability of candidate's pictures all in target candidate pictures, and pre- according to user institute
If Target Photo quantity obtain Target Photo.For example, the default Target Photo quantity of user is 10, then waited according to target
The ranking results for selecting candidate picture match probability in pictures select highest 10 candidates of matching probability in all candidate pictures
Picture is exported as Target Photo.Specifically, also exportable Target Photo during being exported to Target Photo
Essential information, the time of Target Photo publication, the network address for obtaining Target Photo, mesh are contained in the essential information of Target Photo
It marks on a map the title of piece and the label information of Target Photo etc..
By by candidate picture classification to multiple candidate pictures, and picture recommended models are established according to candidate pictures,
Picture recommended models are trained by preset training parameter and training picture, and according to the picture recommended models after training
Matching obtains the highest Target Photo of matching probability and is exported, and the picture that can be efficiently inputted to user carries out accurate
It is equipped with to obtain similar pictures, significantly reduces the error during picture match, save match time.
The embodiment of the present invention also provides a kind of picture recommendation apparatus, which recommends for executing aforementioned picture
Any embodiment of method.Specifically, referring to Fig. 6, Fig. 6 is the schematic of picture recommendation apparatus provided in an embodiment of the present invention
Block diagram.The picture recommendation apparatus 100 can be configured in management server.
As shown in fig. 6, picture recommendation apparatus 100 include picture classification unit 101, picture recommended models construction unit 102,
Picture recommended models training unit 103, matching probability computing unit 104 and Target Photo acquiring unit 105.
Picture classification unit 101, for obtaining multiple pre-stored candidate pictures, according to preset picture classification rule
Classify to the candidate picture to obtain multiple candidate pictures.
Server end is previously stored with multiple candidate pictures, according to preset picture classification rule to the candidate picture into
Row classification is to obtain multiple candidate pictures.Server end is the enterprise for picture recommended models to be constructed and used
Terminal.Classification processing is carried out to all candidate pictures according to preset picture classification rule, to obtain multiple candidate pictures.Figure
Piece classifying rules is the Rule Information for classifying to candidate picture, wherein picture classification rule includes feature extraction
Formula, K-means clustering algorithm.
In other inventive embodiments, as shown in fig. 7, described, picture recommendation apparatus 100 further includes subelement: candidate picture
Format adjusting unit 101A.
Candidate picture format adjustment unit 101A is used for according to preset Format adjusting rule to pre-stored all times
The format of picture is selected to be adjusted to obtain the candidate picture of uniform format.
Since pre-stored candidate picture is unable to ensure format complete unity, before classifying to candidate picture,
Unified adjustment can be carried out according to format of the preset Format adjusting rule to pre-stored all candidate pictures.
In other inventive embodiments, as shown in figure 8, the picture classification unit 101 includes subelement: candidate picture feature
Variable extraction unit 1011, cluster cell 1012 and candidate picture classification unit 1013.
Candidate picture feature variable extraction unit 1011 is used for according to preset feature extraction formula to the candidate picture
Feature extraction is carried out to obtain the characteristic variable of all candidate pictures.
It is extracted according to characteristic variable of the preset feature extraction formula to all candidate pictures.Wherein, preset spy
Sign is extracted formula and is constructed based on convolutional neural networks.Preset feature extraction formula includes the first convolution calculation formula, pond
Change calculation formula, the second convolution calculation formula, the first full connection calculation formula, the second full connection calculation formula.It is mentioned by feature
Take the i.e. extractable characteristic variable for obtaining all candidate pictures of formula.The characteristic variable of candidate picture is by carrying out to picture
It is obtained for reflecting that the multi-C vector of picture characteristic, computer program can not carry out different pictures after feature extraction
Identification, but after multiple pictures are converted to characteristic variable, computer program is analyzed by the characteristic variable to multiple pictures
Different pictures can be identified.
Specifically, if the resolution ratio of the candidate picture of input is 200 × 200, according to the first convolution calculation formula, to differentiate
Rate 10*10 carries out convolution operation as window, step-length 1, to obtain the vector matrix that size is 190 × 190, that is to say time
Select the shallow-layer feature of picture;According to pond calculation formula, using resolution ratio 10 × 10 as window, step-length 10, progress is down-sampled,
To obtain the vector matrix that size is 18 × 18, the profound feature of candidate picture that is to say;According to the second convolution calculation formula,
Using resolution ratio 2 × 2 as window, the carry out convolution operation that step-length is 2, to obtain the vector matrix that size is 9 × 9.Pass through
One full connection calculation formula, calculates obtained 9 × 9 vector matrix, altogether includes five in the first full connection formula
Node, each node is associated with 9 × 9 vector matrix, that is to say and is calculated and 9 by five calculation formula respectively
The value of associated five nodes of × 9 vector matrix, first calculation formula are represented by C1=W1 × x1+B1, wherein C1
For the calculated value of first node, x1 is the value of candidate's picture vector matrix, and W1 and B1 are first node and the vector matrix
The default parameter value into associated first calculation formula can be calculated with the vector matrix by five calculation formula to pass
The value of five nodes of connection;The value of five nodes is calculated by the second full connection calculation formula to obtain the final candidate
The characteristic variable of picture, calculation formula are F1=a1 × C1+a2 × C2+a3 × C3+a4 × C4+a5 × C5;Wherein C1, C2, C3,
C4, C5 be five nodes associated with the vector matrix of the candidate picture value, a1, a2, a3, a4, a5 for five nodes extremely
The preset parameter value of final output node, the characteristic variable for finally obtaining candidate's picture is the moment of a vector of one 1 × 128 dimension
Battle array can be indicated using F1=(f1, f2 ... f128).
Feature extraction can be carried out to all candidate pictures to obtain all candidate pictures by extracting formula according to features described above
Characteristic variable: F1, F2, F3 ... Fn.
Cluster cell 1012, for according to preset K-means clustering algorithm to the characteristic variable of the candidate picture into
Row cluster is to obtain multiple monoids comprising mass center.
It is clustered according to characteristic variable of the preset K-means clustering algorithm to the candidate picture comprising multiple classes
The cluster result of group.Specifically, the quantity K of final required monoid is arranged in K-means clustering algorithm, according to the numerical value pair of K
The characteristic variable of all candidate's pictures is clustered, and the characteristic variable value of each monoid mass center in K monoid is obtained.Mass center is
Central point in the monoid, the characteristic variable value of mass center are the mean value of the quasi-group characteristic variable.
Candidate picture classification unit 1013, for according to obtained multiple monoids to all candidate pictures classify with
Obtain multiple candidate pictures comprising candidate picture.
Classified according to obtained multiple monoids to all candidate pictures, to obtain multiple times comprising candidate picture
Select pictures.Since each monoid after cluster includes mass center of the characteristic variable value as the monoid, pass through meter
Candidate the distance between a picture and all mass centers are calculated, a mass center nearest with candidate's picture distance can be obtained, and
It will be in monoid corresponding to candidate's picture classification to the mass center.All candidate pictures are carried out classifying it according to the above method
Afterwards, multiple candidate pictures comprising corresponding candidate picture can be obtained.
Picture recommended models construction unit 102, for according to preset characteristic matching formula and obtained multiple candidates
Pictures establish picture recommended models.
Picture recommended models are established according to preset characteristic matching formula and obtained multiple candidate pictures.To realize
The picture inputted to user identifies, and calculates the inputted picture of user and the matching probability of candidate picture, needs
Picture recommended models are established by preset characteristic matching formula and obtained multiple candidate pictures.Specifically, described pre-
If characteristic matching formula be S=1/ (((f1-t1) 2+ (f2-t2) 2+ ...+(fn-tn) 2) -1+1), wherein f1, f2 ...
Fn is the characteristic variable value of a certain candidate picture, and t1, t2 ... tn are inputted the characteristic variable value of picture by user, due to waiting
The characteristic variable for selecting picture is one 1 × 128 vector matrix tieed up, then n=128 in above formula.
Picture recommended models training unit 103, for being schemed by preset multiple groups training parameter and multiple preset training
Piece is trained the picture recommended models after being trained to the picture recommended models established.
By preset multiple groups training parameter and it is preset multiple training photos to the picture recommended models established into
Row training, with the picture recommended models after being trained.Specifically, including learning rate, training time in each group of training parameter
Several and training termination condition, learning rate is the amplitude being adjusted to the feature extraction Parameters in Formula of picture recommended models
And the direction of adjustment, frequency of training are the number that picture recommended models are carried out with maximum training, training termination condition is pair
The conditional information that the training of picture recommended models is terminated, if reaching trained termination condition or reaching preset frequency of training,
Then terminate the training to picture recommended models.
In other inventive embodiments, as shown in figure 9, the picture recommended models training unit 103 includes subelement: training
Picture feature variable extraction unit 1031, model accuracy rate acquiring unit 1032 and parameter set unit 1033.
Training picture feature variable extraction unit 1031, for according to the feature extraction formula to multiple training pictures into
Row feature extraction is to obtain the characteristic variable of all trained pictures.
Feature extraction is carried out to obtain all trained pictures to a variety of trained pictures according to preset feature extraction formula
Characteristic variable.During being trained to picture recommended models, by all times in all trained pictures and candidate pictures
Select picture as model training picture, all trained pictures are used as the positive sample in model training picture, in candidate pictures
Picture be used as the negative sample in model training picture.
Model accuracy rate acquiring unit 1032, for obtaining the feature of preset one group of training parameter and all trained pictures
Variable repeatedly trains the picture recommended models established, will last time training in training picture accuracy rate as be somebody's turn to do
The model accuracy rate of group training parameter.
The characteristic variable of preset one group of training parameter and all trained pictures is obtained to the picture recommended models established
It is repeatedly trained, using the accuracy rate of training picture in last time training as the model accuracy rate of this group of training parameter.Tool
Body, it obtains preset one group of training parameter and feature extraction Parameters in Formula in picture recommended models is adjusted, according to tune
Feature extraction formula after whole extracts the characteristic variable of all trained pictures again.By being preset in picture recommended models
Characteristic matching formula the matching probability of a certain Zhang Xunlian picture and all model training pictures is calculated, selection with the instruction
Practice the highest model training picture of picture match probability and be used as training objective picture, to whether matching obtained training objective picture
Judged for the training picture itself, if the training objective picture that matching obtains is the training picture itself, the training figure
The matching result of piece is successful match;If matching obtained training objective picture is not the training picture itself, the training figure
The matching result of piece is that it fails to match, counts the matching result of all trained pictures, training picture match success in matching result
Probability be this training in training picture accuracy rate.Wherein, model training picture includes all trained pictures and candidate
All candidate pictures in pictures.
According to the accuracy rate of the obtained trained picture of first time training and this group of training parameter in picture recommended models
The parameter of feature extraction formula is adjusted again, and according to picture recommended models adjusted again to all trained pictures with
It is matched with all model training pictures, executes above-mentioned training process repeatedly, until reaching trained termination condition or reaching pre-
If frequency of training, then terminate training to picture recommended models, obtain the accurate of training picture in last time training process
Model accuracy rate of the rate as this group of training parameter.
Default multiple groups training parameter is sequentially input by the above method, picture recommended models are trained, obtained
The model accuracy rate of all training parameters.
Parameter set unit 1033, for selecting optimal one according to the model accuracy rate of obtained multiple groups training parameter
Group training parameter, which is configured the parameter of feature extraction formula in picture recommended models, recommends mould with the picture after being trained
Type.
One group of optimal training parameter is selected according to the model accuracy rate of obtained multiple groups training parameter, that is to say selection
The highest one group of training parameter of model accuracy rate, is configured the parameter of feature extraction formula in picture recommended models
Picture recommended models after being trained.Specifically, the highest one group of training parameter of preference pattern accuracy rate, group training is joined
The parameter of number obtained feature extraction formula in last time training process when being trained to picture recommended models, as
The parameter of feature extraction formula is configured in picture recommended models after training.
Matching probability computing unit 104, if the picture to be matched inputted for receiving user, according to the figure after training
Piece recommended models match picture to be matched with multiple candidate pictures, and the matching that corresponding candidate picture is calculated is general
Rate.
If receiving the picture to be matched that user is inputted, according to the picture recommended models after training to picture to be matched and more
It is matched in a candidate's pictures, and the matching probability of corresponding candidate picture is calculated by characteristic matching formula.
In other inventive embodiments, as shown in Figure 10, the matching probability computing unit 104 includes subelement: to be matched
Picture feature extraction unit 1041, candidate pictures acquiring unit 1042, characteristic variable extraction unit 1043 and candidate picture
With probability calculation unit 1044.
Picture feature extraction unit 1041 to be matched, it is special for being carried out according to the feature extraction formula to picture to be matched
Sign is extracted to obtain the characteristic variable of picture to be matched.
Specifically, feature extraction is carried out to picture to be matched according to the feature extraction formula of picture recommended models after training,
To obtain the characteristic variable of picture to be matched.
Candidate pictures acquiring unit 1042, in the characteristic variable and multiple candidate pictures according to picture to be matched
The characteristic variable value of mass center is calculated to obtain target candidate pictures.
According to the characteristic variable value of mass center in the characteristic variable of picture to be matched and multiple candidate pictures calculated with
Obtain target candidate pictures.Each candidate corresponding specific monoid of pictures, each monoid includes a matter
The heart, specifically, by the distance between the characteristic variable value of mass center in the characteristic variable of picture to be matched and each candidate pictures
It is calculated, candidate pictures corresponding to the mass center obtained in all mass centers closest to the picture to be matched according to calculated result
As target candidate pictures.
Characteristic variable extraction unit 1043, for according to feature extraction formula to the candidate picture in target candidate pictures
Feature extraction is carried out to obtain the characteristic variable of candidate picture in the target candidate pictures.
The candidate picture in target candidate pictures is carried out according to the feature extraction formula of picture recommended models after training
Feature extraction is to obtain the characteristic variable of all candidate's pictures in the target candidate pictures.
Candidate picture match probability calculation unit 1044 is used for according to the characteristic matching formula to target candidate pictures
The matching probability of middle candidate's picture and picture to be matched is calculated to obtain the matching probability of all candidate pictures.
According to the characteristic matching formula in target candidate pictures it is all candidate pictures characteristic variable with it is to be matched
The matching probability of the characteristic variable of picture is calculated, to obtain the matching probability of all candidate's pictures in candidate's pictures.
Target Photo acquiring unit 105 is ranked up for the matching probability to obtained all candidate pictures, according to
Preset Target Photo quantity obtains Target Photo.
The matching probability of obtained all candidate pictures is ranked up, is obtained according to preset Target Photo quantity
To Target Photo to recommend to user.Preset Target Photo quantity, which is that user is default, obtains Target Photo
Quantity information, be ranked up according to the matching probability of candidate's pictures all in target candidate pictures, and pre- according to user institute
If Target Photo quantity obtain Target Photo.Specifically, also exportable target during being exported to Target Photo
The essential information of picture contains the time of Target Photo publication in the essential information of Target Photo, obtains the net of Target Photo
Location, the title of Target Photo and label information of Target Photo etc..
By by candidate picture classification to multiple candidate pictures, and picture recommended models are established according to candidate pictures,
Picture recommended models are trained by preset training parameter and training picture, and according to the picture recommended models after training
Matching obtains the highest Target Photo of matching probability and is exported, and the picture that can be efficiently inputted to user carries out accurate
It is equipped with to obtain similar pictures, significantly reduces the error during picture match, save match time.
Above-mentioned picture recommendation apparatus can be implemented as the form of computer program, which can be in such as Figure 11 institute
It is run in the computer equipment shown.
Figure 11 is please referred to, Figure 11 is the schematic block diagram of computer equipment provided in an embodiment of the present invention.
Refering to fig. 11, which includes processor 502, memory and the net connected by system bus 501
Network interface 505, wherein memory may include non-volatile memory medium 503 and built-in storage 504.
The non-volatile memory medium 503 can storage program area 5031 and computer program 5032.The computer program
5032 are performed, and processor 502 may make to execute picture recommendation method.
The processor 502 supports the operation of entire computer equipment 500 for providing calculating and control ability.
The built-in storage 504 provides environment for the operation of the computer program 5032 in non-volatile memory medium 503, should
When computer program 5032 is executed by processor 502, processor 502 may make to execute picture recommendation method.
The network interface 505 is for carrying out network communication, such as the transmission of offer data information.Those skilled in the art can
To understand, structure shown in Figure 11, only the block diagram of part-structure relevant to the present invention program, is not constituted to this hair
The restriction for the computer equipment 500 that bright scheme is applied thereon, specific computer equipment 500 may include than as shown in the figure
More or fewer components perhaps combine certain components or with different component layouts.
Wherein, the processor 502 is for running computer program 5032 stored in memory, to realize following function
Can: obtain multiple pre-stored candidate pictures, according to preset picture classification rule to the candidate picture classify with
Obtain multiple candidate pictures;It establishes picture according to preset characteristic matching formula and obtained multiple candidate pictures and recommends
Model;The picture recommended models established are trained by preset multiple groups training parameter and multiple preset training pictures
Picture recommended models after being trained;If receiving the picture to be matched that user is inputted, recommended according to the picture after training
Model matches picture to be matched with multiple candidate pictures, and the matching probability of corresponding candidate picture is calculated;It is right
The matching probability of obtained all candidate pictures is ranked up, and obtains Target Photo according to preset Target Photo quantity.
In one embodiment, processor 502 is executing multiple pre-stored candidate pictures of acquisition, according to preset picture
When classifying rules classifies to the candidate picture to obtain the step of multiple candidate pictures, perform the following operations: according to
Preset feature extraction formula carries out feature extraction to the candidate picture to obtain the characteristic variable of all candidate pictures;According to
Preset K-means clustering algorithm clusters to obtain multiple classes comprising mass center the characteristic variable of the candidate picture
Group;Classified to all candidate pictures according to obtained multiple monoids to obtain multiple candidate pictures comprising candidate picture
Collection.
In one embodiment, processor 502 is being executed through preset multiple groups training parameter and multiple preset training figures
When piece is trained the step of the picture recommended models after being trained to the picture recommended models established, following behaviour is executed
Make: feature extraction being carried out to multiple training pictures according to the feature extraction formula to obtain the feature of all trained pictures and become
Amount;The characteristic variable for obtaining preset one group of training parameter and all trained pictures is more to the picture recommended models progress established
Secondary training, using the accuracy rate of training picture in last time training as the model accuracy rate of this group of training parameter;According to gained
To the model accuracy rate of multiple groups training parameter select one group of optimal training parameter public to feature extraction in picture recommended models
The parameter of formula is configured with the picture recommended models after being trained.
In one embodiment, if processor 502 is executing the picture to be matched for receiving user and being inputted, after training
Picture recommended models picture to be matched is matched with multiple candidate pictures, and of corresponding candidate picture is calculated
When step with probability, perform the following operations: according to the feature extraction formula to picture to be matched carry out feature extraction with
To the characteristic variable of picture to be matched;Become according to the feature of mass center in the characteristic variable of picture to be matched and multiple candidate pictures
Magnitude is calculated to obtain target candidate pictures;According to feature extraction formula to the candidate picture in target candidate pictures
Feature extraction is carried out to obtain the characteristic variable of candidate picture in the target candidate pictures;According to the characteristic matching formula pair
The matching probability of candidate picture and picture to be matched is calculated to obtain of all candidate pictures in target candidate pictures
With probability.
In one embodiment, processor 502 is executing multiple pre-stored candidate pictures of acquisition, according to preset picture
Before the step of classifying rules classifies to obtain multiple candidate pictures to the candidate picture, also perform the following operations:
It is adjusted according to format of the preset Format adjusting rule to pre-stored all candidate pictures to obtain uniform format
Candidate picture.
It will be understood by those skilled in the art that the embodiment of computer equipment shown in Figure 11 is not constituted to computer
The restriction of equipment specific composition, in other embodiments, computer equipment may include components more more or fewer than diagram, or
Person combines certain components or different component layouts.For example, in some embodiments, computer equipment can only include depositing
Reservoir and processor, in such embodiments, the structure and function of memory and processor are consistent with embodiment illustrated in fig. 11,
Details are not described herein.
It should be appreciated that in embodiments of the present invention, processor 502 can be central processing unit
(CentralProcessing Unit, CPU), which can also be other general processors, digital signal processor
(Digital Signal Processor, DSP), specific integrated circuit (Application Specific
IntegratedCircuit, ASIC), ready-made programmable gate array (Field-Programmable GateArray, FPGA) or
Person other programmable logic device, discrete gate or transistor logic, discrete hardware components etc..Wherein, general processor
It can be microprocessor or the processor be also possible to any conventional processor etc..
Computer readable storage medium is provided in another embodiment of the invention.The computer readable storage medium can be with
For non-volatile storage medium.The computer-readable recording medium storage has computer program, and wherein computer program is located
Reason device performs the steps of when executing obtains multiple pre-stored candidate pictures, according to preset picture classification rule to institute
Candidate picture is stated to classify to obtain multiple candidate pictures;According to preset characteristic matching formula and obtained multiple times
Pictures are selected to establish picture recommended models;By preset multiple groups training parameter and multiple preset training pictures to being established
Picture recommended models are trained the picture recommended models after being trained;If receiving the picture to be matched that user is inputted,
Picture to be matched is matched with multiple candidate pictures according to the picture recommended models after training, and corresponding time is calculated
Select the matching probability of picture;The matching probability of obtained all candidate pictures is ranked up, according to preset Target Photo
Quantity obtains Target Photo.
In one embodiment, multiple pre-stored candidate pictures of the acquisition, it is right according to preset picture classification rule
The step of candidate's picture is classified to obtain multiple candidate pictures, comprising: according to preset feature extraction formula pair
Candidate's picture carries out feature extraction to obtain the characteristic variable of all candidate pictures;It is clustered and is calculated according to preset K-means
Method clusters to obtain multiple monoids comprising mass center the characteristic variable of the candidate picture;According to obtained multiple classes
Group classifies to all candidate pictures to obtain multiple candidate pictures comprising candidate picture.
In one embodiment, described by preset multiple groups training parameter and multiple preset training pictures are to being established
Picture recommended models are trained the step of picture recommended models after being trained, comprising: according to the feature extraction formula
Feature extraction is carried out to obtain the characteristic variable of all trained pictures to multiple training pictures;Obtain preset one group of training parameter
And the characteristic variable of all trained pictures repeatedly trains the picture recommended models established, and will instruct in last time training
Practice model accuracy rate of the accuracy rate of picture as this group of training parameter;Model according to obtained multiple groups training parameter is accurate
Rate selects one group of optimal training parameter to be configured to be trained the parameter of feature extraction formula in picture recommended models
Picture recommended models afterwards.
In one embodiment, if the picture to be matched for receiving user and being inputted, recommended according to the picture after training
Model matches picture to be matched with multiple candidate pictures, and the step of the matching probability of corresponding candidate picture is calculated
Suddenly, comprising: feature extraction is carried out to picture to be matched according to the feature extraction formula to obtain the feature of picture to be matched and become
Amount;It is calculated according to the characteristic variable value of mass center in the characteristic variable of picture to be matched and multiple candidate pictures to obtain mesh
Mark candidate pictures;Feature extraction is carried out to be somebody's turn to do to the candidate picture in target candidate pictures according to feature extraction formula
The characteristic variable of candidate picture in target candidate pictures;According to the characteristic matching formula to candidate in target candidate pictures
The matching probability of picture and picture to be matched is calculated to obtain the matching probability of all candidate pictures.
In one embodiment, multiple pre-stored candidate pictures of the acquisition, it is right according to preset picture classification rule
Before the step of candidate's picture is classified to obtain multiple candidate pictures, further includes: according to preset Format adjusting
Rule is adjusted to obtain the candidate picture of uniform format the format of pre-stored all candidate pictures.
It is apparent to those skilled in the art that for convenience of description and succinctly, foregoing description is set
The specific work process of standby, device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
Those of ordinary skill in the art may be aware that unit described in conjunction with the examples disclosed in the embodiments of the present disclosure and algorithm
Step can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and software
Interchangeability generally describes each exemplary composition and step according to function in the above description.These functions are studied carefully
Unexpectedly the specific application and design constraint depending on technical solution are implemented in hardware or software.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
The scope of the present invention.
In several embodiments provided by the present invention, it should be understood that disclosed unit and method, it 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 logical function partition, there may be another division manner in actual implementation, can also will be with the same function
Unit set is at a unit, such as multiple units or components can be combined or can be integrated into another system or some
Feature can be ignored, or not execute.In addition, shown or discussed mutual coupling, direct-coupling or communication connection can
Be through some interfaces, the indirect coupling or communication connection of device or unit, be also possible to electricity, mechanical or other shapes
Formula connection.
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.Some or all of unit therein can be selected to realize the embodiment of the present invention according to the actual needs
Purpose.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, is also possible to two or more units and is integrated in one unit.It is above-mentioned integrated
Unit 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 one storage medium.Based on this understanding, technical solution of the present invention is substantially in other words to existing
The all or part of part or the technical solution that technology contributes can be embodied in the form of software products, should
Computer software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be
Personal computer, server or network equipment etc.) execute all or part of step of each embodiment the method for the present invention
Suddenly.And storage medium above-mentioned include: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), magnetic disk or
The various media that can store program code such as person's CD.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can readily occur in various equivalent modifications or replace
It changes, these modifications or substitutions should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with right
It is required that protection scope subject to.
Claims (10)
1. a kind of picture recommendation method characterized by comprising
Obtain multiple pre-stored candidate pictures, according to preset picture classification rule to the candidate picture classify with
Obtain multiple candidate pictures;
Picture recommended models are established according to preset characteristic matching formula and obtained multiple candidate pictures;
The picture recommended models established are trained by preset multiple groups training parameter and multiple preset training pictures
Picture recommended models after being trained;
If receiving the picture to be matched that user is inputted, according to the picture recommended models after training to picture to be matched with it is multiple
Candidate pictures are matched, and the matching probability of corresponding candidate picture is calculated;
The matching probability of obtained all candidate pictures is ranked up, target figure is obtained according to preset Target Photo quantity
Piece.
2. picture recommendation method according to claim 1, which is characterized in that described right according to preset picture classification rule
Candidate's picture is classified to obtain multiple candidate pictures, comprising:
Feature extraction is carried out to the candidate picture according to preset feature extraction formula to obtain the feature of all candidate pictures
Variable;
It is clustered according to characteristic variable of the preset K-means clustering algorithm to the candidate picture to obtain comprising mass center
Multiple monoids;
Classified to all candidate pictures according to obtained multiple monoids to obtain multiple candidate figures comprising candidate picture
Piece collection.
3. picture recommendation method according to claim 2, which is characterized in that it is described by preset multiple groups training parameter and
Multiple preset training pictures are trained the picture recommended models after being trained to the picture recommended models established, and wrap
It includes:
Feature extraction is carried out to multiple training pictures according to the feature extraction formula to obtain the feature of all trained pictures and become
Amount;
The characteristic variable for obtaining preset one group of training parameter and all trained pictures carries out the picture recommended models established
Repeatedly training, using the accuracy rate of training picture in last time training as the model accuracy rate of this group of training parameter;
Select one group of optimal training parameter to picture recommended models according to the model accuracy rate of obtained multiple groups training parameter
The parameter of middle feature extraction formula is configured with the picture recommended models after being trained.
4. picture recommendation method according to claim 2, which is characterized in that the picture recommended models according to after training
Picture to be matched is matched with multiple candidate pictures to obtain the matching probability of corresponding candidate picture, comprising:
Feature extraction is carried out to obtain the characteristic variable of picture to be matched to picture to be matched according to the feature extraction formula;
It is calculated according to the characteristic variable of picture to be matched and the characteristic variable value of mass center in multiple candidate pictures to obtain
Target candidate pictures;
Feature extraction is carried out to obtain the target candidate to the candidate picture in target candidate pictures according to feature extraction formula
The characteristic variable of candidate picture in pictures;
The matching probability of candidate picture and picture to be matched in target candidate pictures is carried out according to the characteristic matching formula
It calculates to obtain the matching probability of all candidate pictures.
5. picture recommendation method according to claim 1, which is characterized in that described to obtain multiple pre-stored candidate's figures
Piece also wraps before being classified the candidate picture to obtain multiple candidate pictures according to preset picture classification rule
It includes:
It is adjusted according to format of the preset Format adjusting rule to pre-stored all candidate pictures to obtain format system
One candidate picture.
6. a kind of picture recommendation apparatus characterized by comprising
Picture classification unit, for obtaining multiple pre-stored candidate pictures, according to preset picture classification rule to described
Candidate picture is classified to obtain multiple candidate pictures;
Picture recommended models construction unit, for being built according to preset characteristic matching formula and obtained multiple candidate pictures
Vertical picture recommended models;
Picture recommended models training unit, for by preset multiple groups training parameter and multiple preset training pictures to being built
Vertical picture recommended models are trained the picture recommended models after being trained;
Matching probability computing unit is recommended if the picture to be matched inputted for receiving user according to the picture after training
Model matches picture to be matched with multiple candidate pictures, and the matching probability of corresponding candidate picture is calculated;
Target Photo acquiring unit is ranked up, according to preset for the matching probability to obtained all candidate pictures
Target Photo quantity obtains Target Photo.
7. picture recommendation apparatus according to claim 6, which is characterized in that the picture classification unit, comprising:
Candidate picture feature variable extraction unit, for carrying out feature to the candidate picture according to preset feature extraction formula
It extracts to obtain the characteristic variable of all candidate pictures;
Cluster cell, for according to characteristic variable of the preset K-means clustering algorithm to the candidate picture clustered with
Obtain multiple monoids comprising mass center;
Candidate picture classification unit, for being classified all candidate pictures to be included according to obtained multiple monoids
Multiple candidate pictures of candidate picture.
8. picture recommendation apparatus according to claim 6, which is characterized in that the picture recommended models training unit, packet
It includes:
Training picture feature variable extraction unit is mentioned for carrying out feature to multiple training pictures according to the feature extraction formula
It takes to obtain the characteristic variable of all trained pictures;
Model accuracy rate acquiring unit, for obtaining the characteristic variable of preset one group of training parameter and all trained pictures to institute
The picture recommended models of foundation are repeatedly trained, and are joined the accuracy rate of training picture in last time training as group training
Several model accuracys rate;
Parameter set unit, for selecting one group of optimal training ginseng according to the model accuracy rate of obtained multiple groups training parameter
The parameter of feature extraction formula is configured with the picture recommended models after being trained in several pairs of picture recommended models.
9. a kind of computer equipment, including memory, processor and it is stored on the memory and can be on the processor
The computer program of operation, which is characterized in that the processor realizes such as claim 1 to 5 when executing the computer program
Any one of described in picture recommendation method.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer journey
Sequence, the computer program make the processor execute such as figure described in any one of claim 1 to 5 when being executed by a processor
Piece recommended method.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811106466.4A CN109241318B (en) | 2018-09-21 | 2018-09-21 | Picture recommendation method and device, computer equipment and storage medium |
PCT/CN2018/124798 WO2020056999A1 (en) | 2018-09-21 | 2018-12-28 | Picture recommendation method and apparatus, computer device, and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811106466.4A CN109241318B (en) | 2018-09-21 | 2018-09-21 | Picture recommendation method and device, computer equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109241318A true CN109241318A (en) | 2019-01-18 |
CN109241318B CN109241318B (en) | 2023-06-13 |
Family
ID=65056387
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811106466.4A Active CN109241318B (en) | 2018-09-21 | 2018-09-21 | Picture recommendation method and device, computer equipment and storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109241318B (en) |
WO (1) | WO2020056999A1 (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110232401A (en) * | 2019-05-05 | 2019-09-13 | 平安科技(深圳)有限公司 | Lesion judgment method, device, computer equipment based on picture conversion |
CN110457469A (en) * | 2019-07-05 | 2019-11-15 | 中国平安财产保险股份有限公司 | Information classification approach, device based on shot and long term memory network, computer equipment |
CN110533050A (en) * | 2019-07-22 | 2019-12-03 | 平安科技(深圳)有限公司 | Picture geography information acquisition methods, device, computer equipment and storage medium |
CN111461228A (en) * | 2020-04-01 | 2020-07-28 | 腾讯科技(深圳)有限公司 | Image recommendation method and device and storage medium |
CN112069337A (en) * | 2020-09-03 | 2020-12-11 | Oppo广东移动通信有限公司 | Picture processing method and device, electronic equipment and storage medium |
CN113792202A (en) * | 2021-08-31 | 2021-12-14 | 中国电子科技集团公司第三十研究所 | Screening method for user classification |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112558958B (en) * | 2020-12-16 | 2023-07-25 | 中国平安人寿保险股份有限公司 | Template-based push content generation method and device and computer equipment |
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 |
CN113769387A (en) * | 2021-09-18 | 2021-12-10 | 网易(杭州)网络有限公司 | Game graphic parameter configuration method and device and terminal equipment |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8429173B1 (en) * | 2009-04-20 | 2013-04-23 | Google Inc. | Method, system, and computer readable medium for identifying result images based on an image query |
CN103914527A (en) * | 2014-03-28 | 2014-07-09 | 西安电子科技大学 | Graphic image recognition and matching method based on genetic programming algorithms of novel coding modes |
CN105045818A (en) * | 2015-06-26 | 2015-11-11 | 腾讯科技(深圳)有限公司 | Picture recommending method, apparatus and system |
US9373057B1 (en) * | 2013-11-01 | 2016-06-21 | Google Inc. | Training a neural network to detect objects in images |
CN106202362A (en) * | 2016-07-07 | 2016-12-07 | Tcl集团股份有限公司 | Image recommendation method and image recommendation device |
US20170329804A1 (en) * | 2016-05-10 | 2017-11-16 | Libo Fu | Method And Apparatus Of Generating Image Characteristic Representation Of Query, And Image Search Method And Apparatus |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101551809B (en) * | 2009-05-13 | 2011-04-06 | 西安电子科技大学 | Search method of SAR images classified based on Gauss hybrid model |
US9373033B2 (en) * | 2012-03-13 | 2016-06-21 | Massachusetts Institute Of Technology | Assisted surveillance of vehicles-of-interest |
CN106383912B (en) * | 2016-10-14 | 2019-09-03 | 北京字节跳动网络技术有限公司 | A kind of picture retrieval method and device |
CN108509466A (en) * | 2017-04-14 | 2018-09-07 | 腾讯科技(深圳)有限公司 | A kind of information recommendation method and device |
-
2018
- 2018-09-21 CN CN201811106466.4A patent/CN109241318B/en active Active
- 2018-12-28 WO PCT/CN2018/124798 patent/WO2020056999A1/en active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8429173B1 (en) * | 2009-04-20 | 2013-04-23 | Google Inc. | Method, system, and computer readable medium for identifying result images based on an image query |
US9373057B1 (en) * | 2013-11-01 | 2016-06-21 | Google Inc. | Training a neural network to detect objects in images |
CN103914527A (en) * | 2014-03-28 | 2014-07-09 | 西安电子科技大学 | Graphic image recognition and matching method based on genetic programming algorithms of novel coding modes |
CN105045818A (en) * | 2015-06-26 | 2015-11-11 | 腾讯科技(深圳)有限公司 | Picture recommending method, apparatus and system |
US20170329804A1 (en) * | 2016-05-10 | 2017-11-16 | Libo Fu | Method And Apparatus Of Generating Image Characteristic Representation Of Query, And Image Search Method And Apparatus |
CN106202362A (en) * | 2016-07-07 | 2016-12-07 | Tcl集团股份有限公司 | Image recommendation method and image recommendation device |
Non-Patent Citations (1)
Title |
---|
李宇等, pages 200 - 207 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110232401A (en) * | 2019-05-05 | 2019-09-13 | 平安科技(深圳)有限公司 | Lesion judgment method, device, computer equipment based on picture conversion |
CN110232401B (en) * | 2019-05-05 | 2023-08-04 | 平安科技(深圳)有限公司 | Focus judging method, device and computer equipment based on picture conversion |
CN110457469A (en) * | 2019-07-05 | 2019-11-15 | 中国平安财产保险股份有限公司 | Information classification approach, device based on shot and long term memory network, computer equipment |
CN110533050A (en) * | 2019-07-22 | 2019-12-03 | 平安科技(深圳)有限公司 | Picture geography information acquisition methods, device, computer equipment and storage medium |
CN110533050B (en) * | 2019-07-22 | 2023-11-24 | 平安科技(深圳)有限公司 | Picture geographic information acquisition method and device, computer equipment and storage medium |
CN111461228A (en) * | 2020-04-01 | 2020-07-28 | 腾讯科技(深圳)有限公司 | Image recommendation method and device and storage medium |
CN111461228B (en) * | 2020-04-01 | 2024-03-15 | 腾讯科技(深圳)有限公司 | Image recommendation method and device and storage medium |
CN112069337A (en) * | 2020-09-03 | 2020-12-11 | Oppo广东移动通信有限公司 | Picture processing method and device, electronic equipment and storage medium |
CN113792202A (en) * | 2021-08-31 | 2021-12-14 | 中国电子科技集团公司第三十研究所 | Screening method for user classification |
CN113792202B (en) * | 2021-08-31 | 2023-05-05 | 中国电子科技集团公司第三十研究所 | User classification screening method |
Also Published As
Publication number | Publication date |
---|---|
WO2020056999A1 (en) | 2020-03-26 |
CN109241318B (en) | 2023-06-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109241318A (en) | Picture recommendation method, device, computer equipment and storage medium | |
US10515104B2 (en) | Updating natural language interfaces by processing usage data | |
CN107103057B (en) | A kind of resource supplying method and device | |
CN108804641A (en) | A kind of computational methods of text similarity, device, equipment and storage medium | |
CN109948641A (en) | Anomaly groups recognition methods and device | |
CN111797320B (en) | Data processing method, device, equipment and storage medium | |
CN105118518B (en) | A kind of semantic analysis and device of sound | |
CN110413780A (en) | Text emotion analysis method, device, storage medium and electronic equipment | |
CN109242135A (en) | A kind of model method for running, device and service server | |
CN108205570A (en) | A kind of data detection method and device | |
CN108536595A (en) | Test case intelligence matching process, device, computer equipment and storage medium | |
CN108108743A (en) | Abnormal user recognition methods and the device for identifying abnormal user | |
US10162879B2 (en) | Label filters for large scale multi-label classification | |
JP2018514857A (en) | Data processing method and apparatus | |
CN109214772A (en) | Item recommendation method, device, computer equipment and storage medium | |
US12026592B2 (en) | Machine learning model training method and machine learning model training device | |
US20200278960A1 (en) | Method and system for deriving a procedure for querying data | |
CN112579891A (en) | Cloud resource recommendation method and device, electronic terminal and storage medium | |
CN109859035A (en) | A kind of data processing method of application product functional test, apparatus and system | |
CN116306866A (en) | Deployment method of network model, computer equipment and storage medium | |
CN115982323A (en) | Big data analysis method and artificial intelligence system applied to cloud online service | |
CN114610965A (en) | Data processing method, device, equipment and storage medium | |
CN110427213A (en) | Method, device, equipment and storage medium for realizing intelligent configuration software | |
CN110134701A (en) | A kind of buffering benefit account method and device | |
CN110309335A (en) | A kind of picture match method, apparatus, equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |