CN105956878A - Network advertisement pushing method and network advertisement pushing device - Google Patents
Network advertisement pushing method and network advertisement pushing device Download PDFInfo
- Publication number
- CN105956878A CN105956878A CN201610260604.9A CN201610260604A CN105956878A CN 105956878 A CN105956878 A CN 105956878A CN 201610260604 A CN201610260604 A CN 201610260604A CN 105956878 A CN105956878 A CN 105956878A
- Authority
- CN
- China
- Prior art keywords
- picture
- similarity
- target photo
- reference picture
- advertisement
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
Landscapes
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a network advertisement pushing method and a network advertisement pushing device, and relates to the Internet advertisement putting technology field. The network advertisement pushing method is characterized in that a target picture and a reference picture disposed in a to-be-put webpage are acquired; the similarity between the reference picture and the target picture is calculated; when the picture similarity satisfies a preset requirement, the advertisement information corresponding to the target picture is pushed to the preset position of the to-be-put website; the network advertisement is pushed by aiming at the picture content according to the picture similarity matching, and the pushing of the irrelevant advertisement information according to the key word/label data matching is prevented, and therefore the pushing precision is high, and network user experience is improved at the same time of reducing input cost of merchants.
Description
Technical field
The present invention relates to Internet advertising and throw in technical field, in particular to a kind of web advertisement
The method and device pushed.
Background technology
Along with the fast development of the Internet, the network platform has provided the user increasing facility, as
At enterprising product, the service transacting etc. of doing business of network, and how to enable to commodity, service wide by other people
For knowing, it is a good way of promotion by throwing in advertisement on webpage.
The most common web advertisement is thrown in, is pushed scheme, mainly according to keyword/label data
Association or matching degree, same or similar relation, be i.e. the keyword/label data by advertisement and advertisement
Medium (as web page contents, network picture or about current web page viewer) keyword/label data
Between association coupling, similarity carry out web advertisement input, propelling movement.But, due to above-mentioned advertisement
Put-on method itself uses fuzzy match mode, and it is precisely retrieved, the difficulty of input is relatively big, again
In view of there is stronger subjectivity or deliberately property during arranging above-mentioned advertisement keywords, and lead
The information causing to push may not be the content that user is interested, it addition, commercial audience, page browsing
Keyword/the label data of person and the keyword/label data of current page often differ, such as, close
" viewer of the work transaction such as cloud computing and content browses " house fitting-up " to note " patent application "
Or during the web page contents of the leisure such as " travel in holiday " pastime, the most also can be appreciated that " patent application " is acted on behalf of
The web advertisement that the relatednesss such as service or " Cloud Server " advertising campaign are low.
Inventor finds under study for action, and existing putting method of network advertisement utilizes advertisement and advertising media
The association coupling of keyword/label data by advertisement pushing to corresponding advertising media, owing to it cannot
Throw in accurately and advertisement, thus cause that advertising input is relatively costly, economic well-being of workers and staff is the lowest.
Summary of the invention
In view of this, it is an object of the invention to provide the method and device that a kind of web advertisement pushes,
Avoid and push incoherent advertising message by keyword/label data coupling, improve advertisement
The precision of information.
First aspect, embodiments provides a kind of method that web advertisement pushes, described method
Including:
Obtain Target Photo and the reference picture being positioned in webpage to be put；
Calculate the picture similarity of described reference picture and described Target Photo；
If described picture similarity meets preset requirement, then the advertisement corresponding to described Target Photo is believed
Breath pushes to the predeterminated position on described webpage to be put.
In conjunction with first aspect, embodiments provide the first possible embodiment party of first aspect
Formula, wherein, the described reference picture of described calculating and the picture similarity of described Target Photo, including:
Fisrt feature information and second feature is extracted respectively from described reference picture and described Target Photo
Information；
According to described fisrt feature information and described second feature information, calculate described reference picture and institute
State the picture similarity of Target Photo.
In conjunction with the first possible embodiment of first aspect, embodiments provide first party
The embodiment that the second in face is possible, wherein, described respectively from described reference picture and described target
Picture extracts fisrt feature information and second feature information, including:
Use the object module obtained in advance, respectively described reference picture and described Target Photo are carried out
Detection, second target area in first object region and described Target Photo to obtain described reference picture
Territory；
Extract the fisrt feature information in described first object region；And, extract described second target area
The second feature information in territory.
In conjunction with the embodiment that the second of first aspect is possible, embodiments provide first party
The third possible embodiment in face, wherein, described according to described fisrt feature information with described
Two characteristic informations, calculate the picture similarity of described reference picture and described Target Photo, including:
Use the similarity model obtained in advance, calculate described fisrt feature information and described second feature
The characteristic similarity of information；
According to described characteristic similarity, calculate described reference picture similar with the picture of described Target Photo
Degree.
In conjunction with first aspect, embodiments provide the 4th kind of possible embodiment party of first aspect
Formula, wherein, described by the transmitting advertisement information corresponding to described Target Photo to described webpage to be put
On predeterminated position, including:
Search picture similarity higher than the Target Photo presetting similarity threshold；
By picture similarity higher than the transmitting advertisement information corresponding to the Target Photo of default similarity threshold
Predeterminated position to described webpage to be put.
In conjunction with first aspect, embodiments provide the 5th kind of possible embodiment party of first aspect
Formula, wherein, described by the transmitting advertisement information corresponding to described Target Photo to described webpage to be put
On predeterminated position, including:
According to described picture similarity clooating sequence from high to low, described Target Photo is carried out ranking,
To generate similarity ranking；
In described similarity ranking, search ranking exceed the Target Photo of predetermined ranking；
Ranking is exceeded the transmitting advertisement information corresponding to Target Photo of predetermined ranking to described to be put
Predeterminated position on webpage.
In conjunction with the embodiment that the second of first aspect is possible, embodiments provide first party
The 6th kind of possible embodiment in face, wherein, described object module obtains as follows:
Limit the Weakly supervised learning method of frame based on object, set up and comprise target area in the range of advertisement putting
First group of samples pictures in territory；
Utilize computer vision subject detection algorithm that described first group of samples pictures is carried out pixel scale
Labelling, to obtain the first group echo picture；
Described first group of samples pictures and described first group echo picture are carried out the study of the full convolution degree of depth,
Obtain the characteristic information of target area corresponding to described first group of samples pictures；
Described characteristic information is trained by the method and the hole wave filter that utilize densification sampling, and it is right to obtain
The object module of target area described in Ying Yu.
In conjunction with the third possible embodiment of first aspect, embodiments provide first party
The 7th kind of possible embodiment in face, wherein, described similarity model obtains as follows:
Obtain second group of samples pictures and the 3rd group of samples pictures；Wherein, described second group of samples pictures
Between there is similar Design；Design between described 3rd group of samples pictures and differ greatly；
By the symmetrical network learning method of shared weight to described second group of samples pictures and the described 3rd
Group samples pictures is trained, to obtain described second group of samples pictures and/or described 3rd group of sample graph
The similarity model of the target area of sheet.
In conjunction with first aspect, embodiments provide the 8th kind of possible embodiment party of first aspect
Formula, wherein, described method also includes:
If described picture similarity meets preset requirement, search picture similarity higher than presetting similarity threshold
Value reference picture or according to described picture similarity clooating sequence from high to low to described with reference to figure
Sheet carries out ranking, to generate similarity ranking；In described similarity ranking, search ranking exceed predetermined
The reference picture of ranking；
Reference picture lookup obtained carries out the matching treatment of filter attribute, to obtain object reference figure
Sheet；
Treat described in corresponding for the transmitting advertisement information corresponding to Target Photo to described object reference picture
Throw in the predeterminated position on webpage.
Second aspect, the embodiment of the present invention additionally provides the device that a kind of web advertisement pushes, described dress
Put and include:
Acquisition module, for obtaining Target Photo and the reference picture being positioned in webpage to be put；
Computing module, for calculating described reference picture and the described target figure that described acquisition module obtains
The picture similarity of sheet；
Pushing module, for judging that the described picture similarity that described computing module calculates meets pre-
If require, by the transmitting advertisement information corresponding to described Target Photo to described webpage to be put
Predeterminated position.
The method and device that the web advertisement that the embodiment of the present invention provides pushes, it is possible to according to the ginseng calculated
Examine the picture similarity of picture and Target Photo, by the transmitting advertisement information corresponding to Target Photo to ginseng
Examine the predeterminated position on the webpage to be put that picture is corresponding, with prior art in based on keyword/label
The advertisement placement method of Data Matching pushes incoherent advertising message, essence owing to using fuzzy matching
Accuracy is relatively low to be compared, and first it obtain Target Photo and the reference picture being positioned in webpage to be put, so
Rear calculating Target Photo and the picture similarity of reference picture, and meet preset requirement in picture similarity
Time, by the predeterminated position on the transmitting advertisement information corresponding to Target Photo to webpage to be put, pass through
The coupling of above-mentioned picture similarity carries out the mode of transmitting advertisement information makes the web advertisement can be for picture
Content pushes, and not only pushes precision higher, and can promote network user's Experience Degree
Reduce the input cost of businessman simultaneously.
For making the above-mentioned purpose of the present invention, feature and advantage to become apparent, preferable reality cited below particularly
Execute example, and coordinate appended accompanying drawing, be described in detail below.
Accompanying drawing explanation
In order to be illustrated more clearly that the technical scheme of the embodiment of the present invention, below will be to required in embodiment
Accompanying drawing to be used is briefly described, it will be appreciated that the following drawings illustrate only some of the present invention
Embodiment, is therefore not construed as the restriction to scope, for those of ordinary skill in the art,
On the premise of not paying creative work, it is also possible to obtain other relevant accompanying drawings according to these accompanying drawings.
Fig. 1 shows the method flow diagram that a kind of web advertisement that the embodiment of the present invention is provided pushes；
Fig. 2 shows the method flow diagram that the another kind of web advertisement that the embodiment of the present invention is provided pushes；
Fig. 3 shows the method flow diagram that the another kind of web advertisement that the embodiment of the present invention is provided pushes；
Fig. 4 shows the method flow diagram that the another kind of web advertisement that the embodiment of the present invention is provided pushes；
Fig. 5 shows the method flow diagram that the another kind of web advertisement that the embodiment of the present invention is provided pushes；
Fig. 6 shows the method flow diagram that the another kind of web advertisement that the embodiment of the present invention is provided pushes；
Fig. 7 shows the method flow diagram that the another kind of web advertisement that the embodiment of the present invention is provided pushes；
Fig. 8 shows the method flow diagram that the another kind of web advertisement that the embodiment of the present invention is provided pushes；
Fig. 9 shows the method flow diagram that the another kind of web advertisement that the embodiment of the present invention is provided pushes；
Figure 10 shows that the structure of the device that a kind of web advertisement that the embodiment of the present invention provided pushes is shown
It is intended to.
Main element symbol description:
11, acquisition module；22, computing module；33, pushing module.
Detailed description of the invention
For making the purpose of the embodiment of the present invention, technical scheme and advantage clearer, below in conjunction with this
Accompanying drawing in inventive embodiments, is clearly and completely described the technical scheme in the embodiment of the present invention,
Obviously, described embodiment is only a part of embodiment of the present invention rather than whole embodiments.
Generally herein described in accompanying drawing and the assembly of the embodiment of the present invention that illustrates can be with various different joining
Put and arrange and design.Therefore, retouching in detail the embodiments of the invention provided in the accompanying drawings below
State the scope being not intended to limit claimed invention, but be merely representative of the selected reality of the present invention
Execute example.Based on embodiments of the invention, those skilled in the art are not before making creative work
Put the every other embodiment obtained, broadly fall into the scope of protection of the invention.
In view of putting method of network advertisement of the prior art, utilize the key of advertisement and advertising media
The association coupling of word/label data is by the advertising media of advertisement pushing to correspondence, owing to it cannot be accurately
Throw in and advertisement, thus cause that advertising input is relatively costly, economic well-being of workers and staff is the lowest.Based on this,
Embodiments providing the method and device that a kind of web advertisement pushes, it passes through picture similarity
Coupling be capable of the web advertisement and can push for image content, push precision higher, make
Obtain network user's Experience Degree and also reduce the input cost of businessman the most simultaneously.
The flow chart of the method that the web advertisement that the embodiment of the present invention shown in Figure 1 provides pushes,
Described method specifically includes following steps:
S101, obtain Target Photo and the reference picture being positioned in webpage to be put；
Concrete, it is contemplated that specifically should of the method that the web advertisement that the embodiment of the present invention is provided pushes
By scene, the method that the web advertisement that the embodiment of the present invention provides pushes needs Target Photo and reference
Picture obtains respectively, wherein, Target Photo and reference picture all can by data-interface or
The mode of web crawlers obtains, and from the point of view of data-interface aspect, both obtain the data of picture and connect
Mouth also differs, and Target Photo is the most open from internet site (such as sky cat, Amazon etc.)
Data-interface obtains, and reference picture is then the most open from internet site (such as Tengxun)
The data-interface that can throw in advertisement obtains, it is seen then that obtains the carrier of two pictures and differs.
Additionally, Target Photo and reference picture the most all can obtain by the way of web crawlers, both
Web crawlers technology all can be used, as python realizes the function of reptile, in the source generation wanting to obtain
Picture in Ma crawls this locality.Although both of which can be obtained by reptile, but, sum
Identical according to interface, above-mentioned two picture is by the picture number on the different internet site of reptile
According to and obtain.Additionally, above-mentioned Target Photo can also is that the advertisement figure of the traditional industries of certain forms
Sheet.
Wherein, during obtaining Target Photo, believe obtaining corresponding to the advertisement of this Target Photo
Breath, this advertising message at least includes: advertising pictures, brand/trade mark, trade name, introduction describe,
Price etc. and can point to, jump to show the link/hyperlink of the page of above-mentioned Target Photo.
What deserves to be explained is, when obtaining reference picture and Target Photo, not only by PC computer
The corresponding data interface at the page/interface inside application software or this page/interface is carried out network climb
The mode of worm carries out picture acquisition；Can also be by the page/interface inside APP application program of mobile phone
Corresponding data interface or this page/interface is carried out the mode of web crawlers carry out picture acquisition, this
Bright embodiment does not do concrete restriction.
After getting Target Photo and reference picture by the operation of above-mentioned S101, by following S102
Operation Target Photo and reference picture are carried out the calculating of picture similarity.
S102, calculating reference picture and the picture similarity of Target Photo；
Concrete, above-mentioned reference picture and Target Photo to obtaining carry out intellectual analysis, excavation, inspection
Survey, identify reference picture and the target area of Target Photo, by reference picture and the mesh of Target Photo
Characteristic similarity between mark region, calculates reference picture and the picture similarity of Target Photo.
Wherein, it is contemplated that the embodiment of the present invention is based on the detection identification to object, therefore, above-mentioned reference
Picture refers to object in picture or picture region corresponding to article with the target area of Target Photo, and
Characteristic similarity then refers to the characteristic similarity between the characteristic of correspondence information of target area.Features described above
Information at least includes visual signature information, then, by the spy of the visual signature information of object or article
Levy similarity and understand object or the similarity relation of article visual signature, by this feature similarity relation, can
Know reference picture and the picture similarity of Target Photo of correspondence.
If S103 picture similarity meets preset requirement, then by the advertising message corresponding to Target Photo
Push to the predeterminated position on webpage to be put.
Concrete, on the premise of obtaining above-mentioned picture similarity, the net that the embodiment of the present invention is provided
This picture similarity will be analyzed processing by the method for network advertisement pushing, obtain current with user terminal
The picture similarity of the reference picture on webpage to be put meets the Target Photo of preset requirement, and will symbol
Close the transmitting advertisement information of the Target Photo required presetting to webpage to be put corresponding to reference picture
Position.
Wherein, above-mentioned propelling movement can be to embed display type, it is also possible to is to implant display type, and considers
The experience effect of user, above-mentioned Target Photo is preferably shown at webpage to be put by the embodiment of the present invention
Predetermined position, and above-mentioned propelling movement is at the active operation received on user terminal interface or quilt
Carry out under the dynamic control instruction triggered.
Wherein, above-mentioned predeterminated position can be the viewing area of the reference picture on webpage to be put, also
Can be the side, edge of reference picture on webpage to be put, the embodiment of the present invention do concrete limit
System.
The embodiment of the present invention provide the web advertisement push method, with prior art in based on keyword/
The advertisement placement method of label data coupling pushes incoherent advertisement letter owing to using fuzzy matching
Breath, precision is relatively low compares, its first obtain Target Photo and be positioned in webpage to be put with reference to figure
Sheet, then calculates Target Photo and the picture similarity of reference picture, and meets pre-in picture similarity
If require, by the predeterminated position on the transmitting advertisement information corresponding to Target Photo to webpage to be put,
Make the web advertisement can be for by the way of the coupling of above-mentioned picture similarity carries out transmitting advertisement information
Image content pushes, and not only pushes precision higher, and can experience promoting the network user
The input cost of businessman is reduced while degree.
In order to preferably calculate above-mentioned picture similarity, the calculating process of above-mentioned S102, especially by such as
Lower step realizes, flow chart shown in Figure 2, and described method also includes:
S201, respectively extraction fisrt feature information and second feature letter from reference picture and Target Photo
Breath；
In view of when reference picture and Target Photo are carried out feature extraction, it is necessary first to reference to figure
Sheet and Target Photo carry out semantic segmentation, then by the result of semantic segmentation to reference picture and target
Target area in picture is identified and feature extraction, sees Fig. 3, the concrete mistake that features described above is extracted
Journey includes:
The object module that S2011, use obtain in advance, examines reference picture and Target Photo respectively
Survey, second target area in first object region and Target Photo to obtain reference picture；
Concrete, along with the arrival of big data age, the most more complicated model, express in other words
The model that ability is strong, could fully excavate the abundant information contained in mass data, so, the present invention
Object module in embodiment uses more powerful depth model, so that we can be from big number
According to excavating more valuable information and knowledge, and this object module is as a grader, permissible
The picture of input is classified.The above-mentioned object module obtained by training in advance is respectively to reference to figure
Sheet and Target Photo carry out semantic segmentation, i.e. detect, identify the object included in picture, are joined
Examine the first object region of picture and the second target area of Target Photo.
S2012, the fisrt feature information in extraction first object region；And, extract the second target area
Second feature information.
Concrete, that the first object region obtained and the second target area are comprised picture pixels
Cut and extract, to obtain the fisrt feature information corresponding to first object region and corresponding to the
The second feature information of two target areas.
S202, according to fisrt feature information and second feature information, calculate reference picture and Target Photo
Picture similarity.
In order to preferably calculate picture similarity, the embodiment of the present invention first pass through training obtain similar
Degree model, carries out the calculating of characteristic similarity between above-mentioned first object region and the second target area,
The most again by characteristic similarity and the corresponding relation of picture similarity, obtain reference picture and target figure
Picture similarity between sheet, sees Fig. 4, and the calculating process of above-mentioned picture similarity, especially by such as
Lower step realizes:
The similarity model that S2021, use obtain in advance, calculates fisrt feature information and believes with second feature
The characteristic similarity of breath；
Concrete, utilize big data and degree of depth learning art, training sample is simultaneously entered shared weights
Symmetrical network in, training obtains calculating the similarity model of different visual signature similarity, and this is similar
Degree model is also depth model, can be to the mesh of any two picture of input according to this similarity model
The characteristic similarity in mark region calculates.
S2022, according to characteristic similarity, calculate reference picture and the picture similarity of Target Photo.
Concrete, according to fisrt feature information and second target area in the first object region of above-mentioned calculating
The characteristic similarity of the second feature information in territory, it is known that, first object region and the second target area
Characteristic similarity is the highest, illustrates that the picture similarity of the reference picture of its correspondence and Target Photo is the highest,
Be i.e. picture similarity be feature based similarity, refer to include the target of the second target area
Picture and include the similarity degree of reference picture content in first object region, it is possible to according to this phase
The propelling movement of Target Photo is carried out like degree.
In order to realize the propelling movement of Target Photo extremely webpage to be put, above-mentioned propelling movement process, especially by such as
Lower step realizes, flow chart shown in Figure 5, and described method also includes:
S301, lookup picture similarity are higher than the Target Photo presetting similarity threshold；
S302, by picture similarity higher than preset similarity threshold Target Photo corresponding to advertisement letter
Breath pushes to the predeterminated position on webpage to be put.
Concrete, after being calculated the picture similarity of reference picture and Target Photo, first from mesh
Mark on a map and sheet is searched higher than the Target Photo corresponding to picture similarity presetting similarity threshold, then
The letter of the advertisement corresponding to Target Photo that the picture similarity meeting preset requirement that lookup obtained is corresponding
Breath pushes to the predeterminated position on webpage to be put.
It addition, the method that the web advertisement that the embodiment of the present invention is provided pushes can also be similar to picture
Degree carries out ranking, carries out the propelling movement of above-mentioned Target Photo extremely webpage to be put according to ranking result, sees
Flow chart shown in Fig. 6, described method also includes:
S401, according to picture similarity clooating sequence from high to low, Target Photo is carried out ranking, with
Generate similarity ranking；
S402, similarity ranking is searched ranking exceed the Target Photo of predetermined ranking；
S403, ranking is exceeded the transmitting advertisement information corresponding to Target Photo of predetermined ranking throw to waiting
Put the predeterminated position on webpage.
Concrete, (it is i.e. reference according to the picture similarity of above-mentioned calculating clooating sequence from high to low
Picture and the similarity degree of Target Photo) Target Photo is carried out ranking, to generate similarity ranking,
Then in similarity ranking, search ranking exceed the Target Photo of predetermined ranking and wait to open up as First ray
Show advertising pictures, using to be presented as first for First ray advertising pictures to be presented the highest for similarity degree
Advertising pictures, to ensure first to push according to similarity ranking during the web advertisement pushes
The highest Target Photo of degree of correlation, on webpage to be put corresponding to reference picture, improves wide further
Accuse the precision pushed.
When reference picture and Target Photo are carried out feature information extraction, it is necessary first to obtain and instruct in advance
The object module perfected, to realize above-mentioned reference picture and the semantic segmentation of Target Photo, sees Fig. 7,
The acquisition process of above-mentioned object module specifically includes:
S501, Weakly supervised learning method based on object restriction frame, set up in the range of comprising advertisement putting
First group of samples pictures of target area；
S502, utilize computer vision subject detection algorithm that first group of samples pictures is carried out pixel scale
Labelling, to obtain the first group echo picture；
S503, first group of samples pictures and the first group echo picture are carried out the full convolution degree of depth study,
Characteristic information to the target area corresponding to first group of samples pictures；
Concrete, it is primarily based on object and limits the Weakly supervised learning method of frame, set up and comprise on a large scale
First group of samples pictures of target area in the range of advertisement putting；Furthermore utilize computer vision main body to examine
Method of determining and calculating makes data base obtain the labelling of pixel scale, to obtain the first group echo picture；Furthermore by
One group of samples pictures and labelling picture input the degree of depth learning network of full convolution and learn, wherein, and this
The convolutional layer using full convolutional neural networks in inventive embodiments replaces the full connection of traditional neutral net
Layer, wherein, full convolutional neural networks is shared by weight and the image of input is entered by different convolution kernels
Row convolution, can obtain the characteristic information of the target area of a sheet by a sheet first group of samples pictures semantic segmentation.
Wherein, it is contemplated that image content rich, a certain and in the picture often that user pays close attention to
Individual key area, so the target area in the embodiment of the present invention refers to object or article in picture
Corresponding part picture region, and the target area obtained is split according to above-mentioned full convolutional neural networks,
Being to support any regular or irregular picture region, the accuracy rate that segmentation identifies is higher.
Characteristic information is trained by S504, the method for utilization densification sampling and hole wave filter, obtains
Object module corresponding to target area.
Concrete, utilize the method for densification sampling and hole wave filter that characteristic information is trained, obtain
Corresponding to the object module of target area, it uses said method to be possible not only to semantic label and causes
Close prediction, it is also possible to keep the down-sampled constant rate between network two-layer.This mainly considers
By 2*2 in tradition CNN, step-length be also 2 maximum Pooling (pond) to change step-length into be 1, often
Secondary such operation makes Pooling output image size identical with input, uses in a network twice
Step-length is the Pooling of 1 so that down-sampled rate is reduced to 8 times from 32 times, remains artwork more
Details, but, consequent problem is that the input pixel quantity corresponding to each output reduces, then
After using wave filter (as a example by the wave filter of 3*3), the input of convolution is subtracted by 6 pixels originally
Being 4 less, the artwork part causing output corresponding may only fall and produce error in commodity local.
In order to solve the problems referred to above, the method that the web advertisement that the embodiment of the present invention is provided pushes uses hole
Wave filter, interleaves the null value not updating weights in 3*3 wave filter so that wave filter size becomes 5*5,
And do not increase the number of parameters needing study, it is ensured that Receptive Field (acceptance region) is constant,
Characteristic spectrum input is risen sample level, utilizes Deconvolution Technique, initial with bilinear interpolation for liter sampling
Changing, training obtains the weight of each pixel correspondence underlying pixels so that network output photo resolution with
Original image keeps consistent, and the ginseng that the Softmax loss function of employing pixel scale is to each convolutional layer
Number is optimized.The loss function of the most each pixel is L (i)=-log (p (g)), and wherein g is labelling
GroundTruth class, p (g) is the probability being predicted as g class.
Fisrt feature information and the second mesh of Target Photo in the first object region of calculating reference picture
When marking the characteristic similarity between the second feature information in region, it is necessary first to obtain training in advance good
Similarity model to realize calculating to the picture similarity between above-mentioned reference picture and Target Photo,
Seeing Fig. 8, the acquisition process of above-mentioned similarity model specifically includes:
S601, obtain second group of samples pictures and the 3rd group of samples pictures；Wherein, second group of sample graph
There is between sheet similar Design；Design between 3rd group of samples pictures and differ greatly；
S602, by the symmetrical network learning method of shared weight to second group of samples pictures with the 3rd group
Samples pictures is trained, to obtain second group of samples pictures and/or the target area of the 3rd group of samples pictures
The similarity model in territory.
Concrete, obtain the existing second group of samples pictures carrying cohort labelling data and do not contain
3rd group of samples pictures of cohort labelling information, i.e. refers to have similar setting between second group of samples pictures
Meter, belongs to similar design；Design between 3rd group of samples pictures and differ greatly, and with second group of sample
This picture belongs to non-similar design.By using the symmetrical network method sharing weight to second group of sample
This picture carries out the similar or training of Similarity Class with the 3rd group of samples pictures.Wherein, the portion of weight is shared
It is divided into VGG network, extracts 4096 dimensional features, 8 groups of similar pictures are simultaneously entered network every time,
Similar for positive sample with labelling, randomly draws extra 2 non-similar pictures as negative sample, is i.e.
Part picture in second group of samples pictures is as positive sample, the part picture in the 3rd group of samples pictures
As negative sample, then carry out above-mentioned positive sample and negative sample by Contrastive loss function
Training, to obtain the similarity mould of the target area of second group of samples pictures and/or the 3rd group of samples pictures
Type, can be to the feature similarity of the target area of any two picture of input according to this similarity model
Degree calculates.
In order to save the input cost of businessman, the side that the web advertisement that the embodiment of the present invention is provided pushes
Reference picture on webpage to be put can also be sieved by method according to the calculating of above-mentioned picture similarity
Choosing processes, and to realize shooting the arrow at the target of targeted advertisements information, thus the input reducing businessman further becomes
This.Seeing Fig. 9, the propelling movement process in reference picture after above-mentioned Target Photo to screening, by as follows
Step implements:
If S701 picture similarity meets preset requirement, search picture similarity higher than presetting similarity
The reference picture of threshold value or reference picture is carried out according to picture similarity clooating sequence from high to low
Ranking, to generate similarity ranking；In similarity ranking, search ranking exceed the reference of predetermined ranking
Picture；
S702, reference picture lookup obtained carry out the matching treatment of filter attribute, to obtain target
Reference picture；
S703, corresponding for the transmitting advertisement information corresponding to Target Photo to object reference picture waiting is thrown
Put the predeterminated position on webpage.
Concrete, after being calculated the picture similarity of reference picture and Target Photo, first pass through
Following two mode chooses the ginseng of the candidate corresponding to picture similarity meeting preset requirement from reference picture
Examining picture, one is that the picture directly can searched from reference picture and be higher than default similarity threshold is similar
Candidate reference picture corresponding to degree, two be can first according to the picture similarity of above-mentioned calculating by up to
Reference picture is arranged by low clooating sequence (being i.e. the similarity degree of reference picture and Target Photo)
Name, to generate similarity ranking, then searches ranking in similarity ranking and exceedes the ginseng of predetermined ranking
Examine picture as candidate reference picture, then by filter attribute, above-mentioned candidate reference picture is chosen
Choosing, labelling and determining, thus obtain object reference picture, be finally wide by corresponding to Target Photo
Accuse the information pushing predeterminated position to webpage to be put corresponding to this object reference picture, according to difference
Gray different demand, enters with the webpage to be put that the part reference picture higher to similarity is corresponding
The propelling movement of row advertisement, thus provide cost savings.
Wherein, above-mentioned filter attribute can be the ground at the commercial audience place that candidate reference picture is covered
Reason area attribute and/or age-sex's attribute, it is also possible to be that other are subject to about candidate reference picture or advertisement
Many characteristic attributes.
The embodiment of the present invention provide the web advertisement push method, with prior art in based on keyword/
The advertisement placement method of label data coupling pushes incoherent advertisement letter owing to using fuzzy matching
Breath, precision is relatively low compares, its first obtain Target Photo and be positioned in webpage to be put with reference to figure
Sheet, then calculates Target Photo and the picture similarity of reference picture, and meets pre-in picture similarity
If require, by the predeterminated position on the transmitting advertisement information corresponding to Target Photo to webpage to be put,
Make the web advertisement can be for by the way of the coupling of above-mentioned picture similarity carries out transmitting advertisement information
Image content pushes, and not only pushes precision higher, and can experience promoting the network user
The input cost of businessman is reduced while degree.
The embodiment of the present invention additionally provides the device that a kind of web advertisement pushes, and device is used for performing above-mentioned
The method that the web advertisement pushes, sees Figure 10, and described device includes:
Acquisition module 11, for obtaining Target Photo and the reference picture being positioned in webpage to be put；
Computing module 22, for calculating reference picture and the picture of Target Photo that acquisition module 11 obtains
Similarity；
Pushing module 33, for judging that the picture similarity that computing module 22 calculates meets default wanting
When asking, by the predeterminated position on the transmitting advertisement information corresponding to Target Photo to webpage to be put.
In order to preferably calculate above-mentioned picture similarity, above-mentioned computing module 22 includes: extraction unit and
Computing unit, wherein,
Extraction unit, extracts for the reference picture obtained from acquisition module 11 respectively and Target Photo
Fisrt feature information and second feature information；
Computing unit, for the fisrt feature information extracted according to extraction unit and second feature information,
Calculate reference picture and the picture similarity of Target Photo.
In view of when reference picture and Target Photo are carried out feature extraction, it is necessary first to reference to figure
Sheet and Target Photo carry out semantic segmentation, then by the result of semantic segmentation to reference picture and target
Target area in picture is identified and feature extraction, and said extracted unit includes: detection sub-unit
With extract subelement, wherein,
Detection sub-unit, for using the object module obtained in advance, obtains acquisition module 11 respectively
Reference picture and Target Photo detect, to obtain first object region and the target of reference picture
Second target area of picture；
Extract subelement, for extracting the fisrt feature in the first object region that detection sub-unit detects
Information；And, extract the second feature information of the second target area that detection sub-unit detects.
In order to preferably calculate picture similarity, the embodiment of the present invention first pass through training obtain similar
Degree model, carries out the calculating of characteristic similarity between above-mentioned first object region and the second target area,
The most again by characteristic similarity and the corresponding relation of picture similarity, obtain reference picture and target figure
Picture similarity between sheet, above-mentioned computing unit includes: computation subunit and acquisition subelement, its
In,
Computation subunit, for using the similarity model obtained in advance, calculates what extraction unit extracted
Fisrt feature information and the characteristic similarity of second feature information；
Generate subelement, for the characteristic similarity calculated according to computation subunit, calculate reference picture
Picture similarity with Target Photo.
In order to realize the propelling movement of Target Photo extremely webpage to be put, above-mentioned pushing module 33 includes: first
Lookup unit and the first push unit, wherein,
First searches unit, and the picture similarity calculated for searching computing module 22 is higher than default similar
The Target Photo of degree threshold value；
First push unit, wide for search first that unit searches corresponding to the Target Photo that obtains
Accuse the information pushing predeterminated position to webpage to be put.
It addition, the device that the web advertisement that the embodiment of the present invention is provided pushes can also be similar to picture
Degree carries out ranking, carries out the propelling movement of above-mentioned Target Photo extremely webpage to be put according to ranking result, so,
Above-mentioned pushing module 33 also includes: ranking unit, the second lookup unit and the second push unit, wherein,
Ranking unit, suitable for the sequence from high to low of the picture similarity according to computing module 22 calculating
Ordered pair Target Photo carries out ranking, to generate similarity ranking；
Second searches unit, surpasses for searching ranking in the similarity ranking that ranking unit ranking obtains
Cross the Target Photo of predetermined ranking；
Second push unit, wide for what the second lookup unit was found corresponding to the Target Photo obtained
Accuse the information pushing predeterminated position to webpage to be put.
When reference picture and Target Photo are carried out feature information extraction, it is necessary first to obtain and instruct in advance
The object module perfected is to realize above-mentioned reference picture and the semantic segmentation of Target Photo, therefore, and this
The device that the web advertisement that inventive embodiments provides pushes also includes the first training mould obtaining object module
Block, this first training module includes: set up unit, indexing unit, unit and training unit,
Wherein,
Set up unit, for limiting the Weakly supervised learning method of frame based on object, set up and comprise advertisement throwing
First group of samples pictures of target area in the range of putting；
Indexing unit, for utilizing computer vision subject detection algorithm first to setting up that unit obtains
Group samples pictures carries out the labelling of pixel scale, to obtain the first group echo picture；
Unit, for setting up first group of samples pictures that unit obtains and indexing unit obtains
First group echo picture carries out the study of the full convolution degree of depth, obtains the target corresponding to first group of samples pictures
The characteristic information in region；
First training unit, for utilizing the fine and close method sampled and hole wave filter to unit
Acquistion to characteristic information be trained, obtain the object module corresponding to target area.
Fisrt feature information and the second mesh of Target Photo in the first object region of calculating reference picture
When marking the characteristic similarity between the second feature information in region, it is necessary first to obtain training in advance good
Similarity model to realize calculating to the picture similarity between above-mentioned reference picture and Target Photo,
Therefore, the device that the web advertisement that the embodiment of the present invention provides pushes also includes obtaining similarity model
Second training module, this second training module includes: acquiring unit and the second training unit, wherein,
Acquiring unit, for obtaining second group of samples pictures and the 3rd group of samples pictures；Wherein, second
Between group samples pictures, there is similar Design；Design between 3rd group of samples pictures and differ greatly；
Second training unit, for obtaining acquiring unit by the symmetrical network learning method of shared weight
The second group of samples pictures taken is trained with the 3rd group of samples pictures, to obtain second group of samples pictures
And/or the similarity model of the target area of the 3rd group of samples pictures.
In order to save the input cost of businessman, the dress that the web advertisement that the embodiment of the present invention is provided pushes
Put and according to the calculating of above-mentioned picture similarity, the reference picture on webpage to be put can also be sieved
Choosing processes, and to realize shooting the arrow at the target of targeted advertisements information, thus the input reducing businessman further becomes
This.The device that the web advertisement that the embodiment of the present invention is provided pushes also includes screening module, this screening
Module includes: screening unit and matching unit；
Screening unit, for judging that the picture similarity that computing module 22 calculates meets preset requirement
Time, search picture similarity higher than presetting the reference picture of similarity threshold or according to picture similarity
Clooating sequence from high to low carries out ranking to reference picture, to generate similarity ranking；In similarity
Ranking is searched ranking and exceedes the reference picture of predetermined ranking；
Matching unit, carries out the coupling of filter attribute for screening unit is searched the reference picture obtained
Process, to obtain object reference picture；
Push unit, pushes away for matching unit mates the advertising message corresponding to the Target Photo obtained
Deliver to the predeterminated position on the webpage to be put that object reference picture is corresponding.
The device that the web advertisement that the embodiment of the present invention provides pushes, it obtains mesh by acquisition module 11
The reference picture that sheet marking on a map and is positioned in webpage to be put, computing module 22 calculates Target Photo and reference
The picture similarity of picture, and when picture similarity meets preset requirement, will by pushing module 33
Predeterminated position on transmitting advertisement information corresponding to Target Photo extremely webpage to be put, by above-mentioned picture
The coupling of face similarity carries out the mode of transmitting advertisement information makes the web advertisement can enter for image content
Row pushes, and not only pushes precision higher, and can drop while promoting network user's Experience Degree
The input cost of low businessman.
The computer program of the method carrying out web advertisement propelling movement that the embodiment of the present invention is provided,
Including storing the computer-readable recording medium of program code, the instruction that described program code includes can
For performing the method described in previous methods embodiment, implement and can be found in embodiment of the method,
This repeats no more.
The device that the web advertisement that the embodiment of the present invention is provided pushes can be the specific hardware on equipment
Or the software being installed on equipment or firmware etc..The device that the embodiment of the present invention is provided, it realizes
The technique effect of principle and generation is identical with preceding method embodiment, for briefly describing, and device embodiment
The not mentioned part of part, refers to corresponding contents in preceding method embodiment.The technology people of art
Member is it can be understood that arrive, for convenience and simplicity of description, and system, device and list described above
The specific works process of unit, is all referred to the corresponding process in said method embodiment, at this no longer
Repeat.
In embodiment provided by the present invention, it should be understood that disclosed apparatus and method, permissible
Realize by another way.Device embodiment described above is only schematically, such as,
The division of described unit, is only a kind of logic function and divides, and actual can have other drawing when realizing
Point mode, the most such as, multiple unit or assembly can in conjunction with or be desirably integrated into another system,
Or some features can ignore, or do not perform.Another point, shown or discussed coupling each other
Conjunction or direct-coupling or communication connection can be the indirect couplings by some communication interfaces, device or unit
Close or communication connection, can be electrical, machinery or other form.
The described unit illustrated as separating component can be or may not be physically separate,
The parts shown as unit can be or may not be physical location, i.e. may be located at a ground
Side, or can also be distributed on multiple NE.Can select therein according to the actual needs
Some or all of unit realizes the purpose of the present embodiment scheme.
Process single it addition, each functional unit in the embodiment that the present invention provides can be integrated in one
In unit, it is also possible to be that unit is individually physically present, it is also possible to two or more unit are integrated
In a unit.
If described function realizes using the form of SFU software functional unit and as independent production marketing or make
Used time, can be stored in a computer read/write memory medium.Based on such understanding, this
Part that prior art is contributed by bright technical scheme the most in other words or this technical scheme
Part can embody with the form of software product, and this computer software product is stored in a storage
In medium, including some instructions with so that a computer equipment (can be personal computer, take
Business device, or the network equipment etc.) perform completely or partially walking of method described in each embodiment of the present invention
Suddenly.And aforesaid storage medium includes: USB flash disk, portable hard drive, read only memory (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic disc or CD
Etc. the various media that can store program code.
It should also be noted that similar label and letter represent similar terms, therefore, one in following accompanying drawing
A certain Xiang Yi the accompanying drawing of denier is defined, then need not it is carried out further in accompanying drawing subsequently
Definition and explanation, describe additionally, term " first ", " second ", " the 3rd " etc. are only used for distinguishing,
And it is not intended that indicate or hint relative importance.
Last it is noted that the detailed description of the invention of embodiment described above, the only present invention, use
So that technical scheme to be described, being not intended to limit, protection scope of the present invention is not limited to
This, is although the present invention being described in detail with reference to previous embodiment, the ordinary skill of this area
Personnel are it is understood that any those familiar with the art is at the technical scope that the invention discloses
In, the technical scheme described in previous embodiment still can be modified by it maybe can readily occur in change
Change, or wherein portion of techniques feature is carried out equivalent；And these are revised, change or replace,
The essence not making appropriate technical solution departs from the spirit and scope of embodiment of the present invention technical scheme.All
Should contain within protection scope of the present invention.Therefore, protection scope of the present invention should be described with right
The protection domain required is as the criterion.
Claims (10)
1. the method that a web advertisement pushes, it is characterised in that described method includes:
Obtain Target Photo and the reference picture being positioned in webpage to be put；
Calculate the picture similarity of described reference picture and described Target Photo；
If described picture similarity meets preset requirement, then the advertisement corresponding to described Target Photo is believed
Breath pushes to the predeterminated position on described webpage to be put.
The method that the web advertisement the most according to claim 1 pushes, it is characterised in that described meter
Calculate the picture similarity of described reference picture and described Target Photo, including:
Fisrt feature information and second feature is extracted respectively from described reference picture and described Target Photo
Information；
According to described fisrt feature information and described second feature information, calculate described reference picture and institute
State the picture similarity of Target Photo.
The method that the web advertisement the most according to claim 2 pushes, it is characterised in that described point
From described reference picture and described Target Photo, do not extract fisrt feature information and second feature information,
Including:
Use the object module obtained in advance, respectively described reference picture and described Target Photo are carried out
Detection, second target area in first object region and described Target Photo to obtain described reference picture
Territory；
Extract the fisrt feature information in described first object region；And, extract described second target area
The second feature information in territory.
The method that the web advertisement the most according to claim 3 pushes, it is characterised in that described
According to described fisrt feature information and described second feature information, calculate described reference picture and described target
The picture similarity of picture, including:
Use the similarity model obtained in advance, calculate described fisrt feature information and described second feature
The characteristic similarity of information；
According to described characteristic similarity, calculate described reference picture similar with the picture of described Target Photo
Degree.
The method that the web advertisement the most according to claim 1 pushes, it is characterised in that described general
Predeterminated position on transmitting advertisement information corresponding to described Target Photo extremely described webpage to be put, bag
Include:
Search picture similarity higher than the Target Photo presetting similarity threshold；
By picture similarity higher than the transmitting advertisement information corresponding to the Target Photo of default similarity threshold
Predeterminated position to described webpage to be put.
The method that the web advertisement the most according to claim 1 pushes, it is characterised in that described general
Predeterminated position on transmitting advertisement information corresponding to described Target Photo extremely described webpage to be put, bag
Include:
According to described picture similarity clooating sequence from high to low, described Target Photo is carried out ranking,
To generate similarity ranking；
In described similarity ranking, search ranking exceed the Target Photo of predetermined ranking；
Ranking is exceeded the transmitting advertisement information corresponding to Target Photo of predetermined ranking to described to be put
Predeterminated position on webpage.
The method that the web advertisement the most according to claim 3 pushes, it is characterised in that described mesh
Mark model obtains as follows:
Limit the Weakly supervised learning method of frame based on object, set up and comprise target area in the range of advertisement putting
First group of samples pictures in territory；
Utilize computer vision subject detection algorithm that described first group of samples pictures is carried out pixel scale
Labelling, to obtain the first group echo picture；
Described first group of samples pictures and described first group echo picture are carried out the study of the full convolution degree of depth,
Obtain the characteristic information of target area corresponding to described first group of samples pictures；
Described characteristic information is trained by the method and the hole wave filter that utilize densification sampling, and it is right to obtain
The object module of target area described in Ying Yu.
The method that the web advertisement the most according to claim 4 pushes, it is characterised in that described phase
Obtain as follows like degree model:
Obtain second group of samples pictures and the 3rd group of samples pictures；Wherein, described second group of samples pictures
Between there is similar Design；Design between described 3rd group of samples pictures and differ greatly；
By the symmetrical network learning method of shared weight to described second group of samples pictures and the described 3rd
Group samples pictures is trained, to obtain described second group of samples pictures and/or described 3rd group of sample
The similarity model of the target area of picture.
The method that the web advertisement the most according to claim 1 pushes, it is characterised in that described side
Method also includes:
If described picture similarity meets preset requirement, search picture similarity higher than presetting similarity threshold
Value reference picture or according to described picture similarity clooating sequence from high to low to described reference
Picture carries out ranking, to generate similarity ranking；In described similarity ranking, search ranking exceed pre-
Name secondary reference picture；
Reference picture lookup obtained carries out the matching treatment of filter attribute, to obtain object reference figure
Sheet；
Treat described in corresponding for the transmitting advertisement information corresponding to Target Photo to described object reference picture
Throw in the predeterminated position on webpage.
10. the device that a web advertisement pushes, it is characterised in that including:
Acquisition module, for obtaining Target Photo and the reference picture being positioned in webpage to be put；
Computing module, for calculating described reference picture and the described target figure that described acquisition module obtains
The picture similarity of sheet；
Pushing module, for judging that the described picture similarity that described computing module calculates meets pre-
If require, by the transmitting advertisement information corresponding to described Target Photo to described webpage to be put
Predeterminated position.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610260604.9A CN105956878A (en) | 2016-04-25 | 2016-04-25 | Network advertisement pushing method and network advertisement pushing device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610260604.9A CN105956878A (en) | 2016-04-25 | 2016-04-25 | Network advertisement pushing method and network advertisement pushing device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105956878A true CN105956878A (en) | 2016-09-21 |
Family
ID=56916849
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610260604.9A Pending CN105956878A (en) | 2016-04-25 | 2016-04-25 | Network advertisement pushing method and network advertisement pushing device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105956878A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106454442A (en) * | 2016-11-03 | 2017-02-22 | Tcl集团股份有限公司 | Advertisement putting method and advertisement receiving end |
| CN106529996A (en) * | 2016-10-24 | 2017-03-22 | 北京百度网讯科技有限公司 | Deep learning-based advertisement display method and device |
| CN106686404A (en) * | 2016-12-16 | 2017-05-17 | 中兴通讯股份有限公司 | Video analysis platform, matching method, accurate advertisement delivery method and system |
| CN107066916A (en) * | 2016-10-26 | 2017-08-18 | 中国科学院自动化研究所 | Scene Semantics dividing method based on deconvolution neutral net |
| CN107330750A (en) * | 2017-05-26 | 2017-11-07 | 北京三快在线科技有限公司 | A kind of recommended products figure method and device, electronic equipment |
| CN107578300A (en) * | 2017-10-24 | 2018-01-12 | 济南浪潮高新科技投资发展有限公司 | A kind of elevator card launches the method and device that work is audited automatically |
| CN108322824A (en) * | 2018-02-27 | 2018-07-24 | 四川长虹电器股份有限公司 | A kind of method and system that television image is carried out to scene displacement |
| CN108717587A (en) * | 2018-05-25 | 2018-10-30 | 杭州知智能科技有限公司 | A method of text prediction forwarding task is pushed away based on the solution of multi-panel sorting network |
| CN109118275A (en) * | 2018-07-26 | 2019-01-01 | 王振 | A kind of accurate distribution method of advertisement position and system |
| CN109543051A (en) * | 2018-10-25 | 2019-03-29 | 北京奇虎科技有限公司 | A kind of advertisement placement method and device |
| WO2019076294A1 (en) * | 2017-10-19 | 2019-04-25 | 腾讯科技（深圳）有限公司 | Promotion information delivery method and apparatus, and medium and device |
| CN110489685A (en) * | 2019-08-19 | 2019-11-22 | 腾讯科技（武汉）有限公司 | A kind of Webpage display process, system and relevant apparatus and storage medium |
| CN112150174A (en) * | 2019-06-27 | 2020-12-29 | 百度在线网络技术（北京）有限公司 | Advertisement matching method and device and electronic equipment |
| CN112446720A (en) * | 2019-08-29 | 2021-03-05 | 北京搜狗科技发展有限公司 | Advertisement display method and device |
| CN112702260A (en) * | 2020-12-23 | 2021-04-23 | 维沃移动通信（杭州）有限公司 | Image sending method and device and electronic equipment |
| CN112766997A (en) * | 2019-11-01 | 2021-05-07 | 百度在线网络技术（北京）有限公司 | Picture delivery method, device, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102103641A (en) * | 2011-03-08 | 2011-06-22 | 西安交通大学 | Method for adding banner advertisement into user-browsed network image |
| CN102486793A (en) * | 2010-12-06 | 2012-06-06 | 武汉若鱼网络科技有限公司 | Method and system for searching target user |
| CN102693311A (en) * | 2012-05-28 | 2012-09-26 | 中国人民解放军信息工程大学 | Target retrieval method based on group of randomized visual vocabularies and context semantic information |
| CN103218734A (en) * | 2013-04-01 | 2013-07-24 | 天脉聚源(北京)传媒科技有限公司 | Advertisement information pushing method and device |
| CN105160561A (en) * | 2015-09-30 | 2015-12-16 | 北京中搜网络技术股份有限公司 | Situational picture advertising method and apparatus |
-
2016
- 2016-04-25 CN CN201610260604.9A patent/CN105956878A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102486793A (en) * | 2010-12-06 | 2012-06-06 | 武汉若鱼网络科技有限公司 | Method and system for searching target user |
| CN102103641A (en) * | 2011-03-08 | 2011-06-22 | 西安交通大学 | Method for adding banner advertisement into user-browsed network image |
| CN102693311A (en) * | 2012-05-28 | 2012-09-26 | 中国人民解放军信息工程大学 | Target retrieval method based on group of randomized visual vocabularies and context semantic information |
| CN103218734A (en) * | 2013-04-01 | 2013-07-24 | 天脉聚源(北京)传媒科技有限公司 | Advertisement information pushing method and device |
| CN105160561A (en) * | 2015-09-30 | 2015-12-16 | 北京中搜网络技术股份有限公司 | Situational picture advertising method and apparatus |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529996A (en) * | 2016-10-24 | 2017-03-22 | 北京百度网讯科技有限公司 | Deep learning-based advertisement display method and device |
| CN107066916B (en) * | 2016-10-26 | 2020-02-07 | 中国科学院自动化研究所 | Scene semantic segmentation method based on deconvolution neural network |
| CN107066916A (en) * | 2016-10-26 | 2017-08-18 | 中国科学院自动化研究所 | Scene Semantics dividing method based on deconvolution neutral net |
| CN106454442A (en) * | 2016-11-03 | 2017-02-22 | Tcl集团股份有限公司 | Advertisement putting method and advertisement receiving end |
| CN106686404B (en) * | 2016-12-16 | 2021-02-02 | 中兴通讯股份有限公司 | Video analysis platform, matching method, and method and system for accurately delivering advertisements |
| CN106686404A (en) * | 2016-12-16 | 2017-05-17 | 中兴通讯股份有限公司 | Video analysis platform, matching method, accurate advertisement delivery method and system |
| WO2018107914A1 (en) * | 2016-12-16 | 2018-06-21 | 中兴通讯股份有限公司 | Video analysis platform, matching method, and accurate advertisement push method and system |
| CN107330750A (en) * | 2017-05-26 | 2017-11-07 | 北京三快在线科技有限公司 | A kind of recommended products figure method and device, electronic equipment |
| CN107330750B (en) * | 2017-05-26 | 2019-03-08 | 北京三快在线科技有限公司 | A kind of recommended products figure method and device, electronic equipment |
| CN110020123A (en) * | 2017-10-19 | 2019-07-16 | 腾讯科技（深圳）有限公司 | A kind of promotion message put-on method, device, medium and equipment |
| CN110020123B (en) * | 2017-10-19 | 2023-05-12 | 腾讯科技（深圳）有限公司 | Popularization information delivery method, device, medium and equipment |
| WO2019076294A1 (en) * | 2017-10-19 | 2019-04-25 | 腾讯科技（深圳）有限公司 | Promotion information delivery method and apparatus, and medium and device |
| CN107578300A (en) * | 2017-10-24 | 2018-01-12 | 济南浪潮高新科技投资发展有限公司 | A kind of elevator card launches the method and device that work is audited automatically |
| CN108322824B (en) * | 2018-02-27 | 2020-11-03 | 四川长虹电器股份有限公司 | Method and system for carrying out scene replacement on television picture |
| CN108322824A (en) * | 2018-02-27 | 2018-07-24 | 四川长虹电器股份有限公司 | A kind of method and system that television image is carried out to scene displacement |
| CN108717587A (en) * | 2018-05-25 | 2018-10-30 | 杭州知智能科技有限公司 | A method of text prediction forwarding task is pushed away based on the solution of multi-panel sorting network |
| CN108717587B (en) * | 2018-05-25 | 2022-03-15 | 杭州一知智能科技有限公司 | Method for solving tweet prediction forwarding task based on multi-face sequencing network |
| CN109118275A (en) * | 2018-07-26 | 2019-01-01 | 王振 | A kind of accurate distribution method of advertisement position and system |
| CN109543051A (en) * | 2018-10-25 | 2019-03-29 | 北京奇虎科技有限公司 | A kind of advertisement placement method and device |
| CN112150174A (en) * | 2019-06-27 | 2020-12-29 | 百度在线网络技术（北京）有限公司 | Advertisement matching method and device and electronic equipment |
| CN112150174B (en) * | 2019-06-27 | 2024-04-02 | 百度在线网络技术（北京）有限公司 | Advertisement picture allocation method and device and electronic equipment |
| CN110489685A (en) * | 2019-08-19 | 2019-11-22 | 腾讯科技（武汉）有限公司 | A kind of Webpage display process, system and relevant apparatus and storage medium |
| CN112446720A (en) * | 2019-08-29 | 2021-03-05 | 北京搜狗科技发展有限公司 | Advertisement display method and device |
| CN112766997A (en) * | 2019-11-01 | 2021-05-07 | 百度在线网络技术（北京）有限公司 | Picture delivery method, device, equipment and storage medium |
| CN112702260A (en) * | 2020-12-23 | 2021-04-23 | 维沃移动通信（杭州）有限公司 | Image sending method and device and electronic equipment |
| CN112702260B (en) * | 2020-12-23 | 2022-08-05 | 维沃移动通信（杭州）有限公司 | Image sending method and device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105956878A (en) | Network advertisement pushing method and network advertisement pushing device | |
| US11537719B2 (en) | Deep neural network system for similarity-based graph representations | |
| CN106204165A (en) | A kind of advertisement placement method and device | |
| US10666784B2 (en) | Intuitive computing methods and systems | |
| US10713794B1 (en) | Method and system for using machine-learning for object instance segmentation | |
| US11055557B2 (en) | Automated extraction of product attributes from images | |
| CN107690657B (en) | Trade company is found according to image | |
| US20170109615A1 (en) | Systems and Methods for Automatically Classifying Businesses from Images | |
| US10198635B2 (en) | Systems and methods for associating an image with a business venue by using visually-relevant and business-aware semantics | |
| JP5843207B2 (en) | Intuitive computing method and system | |
| JP6011938B2 (en) | Sensor-based mobile search, related methods and systems | |
| CN110334272B (en) | Intelligent question-answering method and device based on knowledge graph and computer storage medium | |
| CN106202362A (en) | Image recommendation method and image recommendation device | |
| EP3493105A1 (en) | Optimizations for dynamic object instance detection, segmentation, and structure mapping | |
| EP3493106B1 (en) | Optimizations for dynamic object instance detection, segmentation, and structure mapping | |
| US10769428B2 (en) | On-device image recognition | |
| CN110737783A (en) | method, device and computing equipment for recommending multimedia content | |
| EP3493104A1 (en) | Optimizations for dynamic object instance detection, segmentation, and structure mapping | |
| Vargas Munoz et al. | Deploying machine learning to assist digital humanitarians: making image annotation in OpenStreetMap more efficient | |
| CN111191127B (en) | Travel recommendation method and system based on correlation analysis algorithm | |
| KR20190124436A (en) | Method for searching building based on image and apparatus for the same | |
| CN103399900A (en) | Image recommending method based on location service | |
| CN112711645B (en) | Method and device for expanding position point information, storage medium and electronic equipment | |
| WO2014159111A2 (en) | Clustering of ads with organic map content | |
| CN114972947B (en) | Depth scene text detection method and device based on fuzzy semantic modeling |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160921 |