CN108197180A - A kind of method of the editable image of clothing retrieval of clothes attribute - Google Patents
A kind of method of the editable image of clothing retrieval of clothes attribute Download PDFInfo
- Publication number
- CN108197180A CN108197180A CN201711422438.9A CN201711422438A CN108197180A CN 108197180 A CN108197180 A CN 108197180A CN 201711422438 A CN201711422438 A CN 201711422438A CN 108197180 A CN108197180 A CN 108197180A
- Authority
- CN
- China
- Prior art keywords
- clothes
- image
- clothing
- attribute
- user
- 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
- 238000000034 method Methods 0.000 title claims abstract description 72
- 239000013598 vector Substances 0.000 claims abstract description 57
- 239000000284 extract Substances 0.000 claims abstract description 9
- 230000004048 modification Effects 0.000 claims abstract description 8
- 238000012986 modification Methods 0.000 claims abstract description 8
- 238000009826 distribution Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 230000015572 biosynthetic process Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 5
- 238000013480 data collection Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 description 11
- 210000003739 neck Anatomy 0.000 description 11
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 239000000463 material Substances 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000004069 differentiation Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 230000000454 anti-cipatory effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000000205 computational method Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000035755 proliferation Effects 0.000 description 1
- 238000005086 pumping Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000004919 topotaxy Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Library & Information Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention discloses a kind of methods of the editable image of clothing retrieval of clothes attribute.This method includes：Extract the feature vector of the image of clothing in existing garment data；Obtain the image of clothing input by user for search；Extract feature vector of the user for the image of clothing of search；The clothes attribute of image of clothing is obtained with clothes attribute method of discrimination；Clothes attribute and optional modification option are showed into user, and obtain the clothes attribute item that user changed；With the feature vector of image of clothing and the modified clothes attribute item of user of original search, new garment feature vector is formed；Calculate the similarity between clothes attributive character in the clothes attributive character newly formed and garment data；Show the highest preceding k parts image of clothing of similarity.Implement the embodiment of the present invention, user can be caused to carry out the costume retrieval of image content-based again according to the attribute of itself hobby modification image of clothing to be retrieved, obtain satisfied query result.
Description
Technical field
The present invention relates to image technique fields, and in particular to a kind of side of the editable image of clothing retrieval of clothes attribute
Method.
Background technology
With the rapid proliferation of internet and the rapid development of e-commerce, the behavior of people's online shopping is also more universal
With it is frequent, in addition the category of commodity is also more and more, online commercial products retrieval system must not when becoming user's on-line purchase commodity
The tool that can lack.Meanwhile in numerous commodity, clothes are one of commodity that people often buy.So efficient, function perfects
Costume retrieval system have great importance to online shopping platform, particularly, for the electronics of special clothes on-line purchase
Business platform is even more indispensable function.Traditional image of clothing search method usually first passes through artificial mode to clothes
It fills image and carries out labeling, recycle keyword to retrieve corresponding image of clothing, it is this to provide inspection according to image tag matching
The method of hitch fruit, i.e. text based image of clothing search method not only need to take a substantial amount of time, and seriously by people
For subjective factor influence.Image of clothing search method based on content overcomes text based image of clothing search method
Deficiency, directly from Image Visual Feature to be found, found out in image of clothing library (seeking scope) similar
Image of clothing is a kind of mode that image searching result is provided according to vision similarity degree, generally can be divided into three levels, respectively
It is retrieved for one, according to low-level features such as the color of image, texture, edges；2nd, the low-level feature based on image passes through knowledge
The spatial topotaxy between object type and object in other image is retrieved；3rd, pushing away based on image high-level semantic
Neo-Confucianism is practised and being retrieved.
Search method general Study based on image of clothing content is mainly concentrated in that suitable global characteristics is selected to go
It describes image of clothing content and which type of method for measuring similarity image of clothing matching is carried out using.Either based on region
Image of clothing search method, main thought is that image of clothing cutting techniques extract object in image of clothing, then to every
A region is described using local feature, and comprehensive each provincial characteristics can obtain the feature description of image of clothing.The two grind
It is all image-centric to study carefully direction, although compensating for a large amount of manual tags of needs of text based image of clothing search method
The shortcomings that changing, and being influenced by artificial subjective factor, but can be further perfect in terms of user function, such as can increase
Clothes attribute editable function so that user can replace the association attributes of clothes according to the hobby of oneself.
Prior art is typically based on the image of clothing search method of text and the costume retrieval of image content-based
Method.The costume retrieval method and step of wherein image content-based is as follows：1) to automatically extract user using feature extracting method defeated
The feature of the image of clothing entered；2) phase is carried out using characteristic matching technology and the feature vector in existing garment image data library
It is calculated like property；3) the corresponding image of most like preceding k feature is returned into user.
Wherein common image of clothing feature has color characteristic, textural characteristics and shape feature.Color characteristic is clothes figure
As, using very extensive feature, can intuitively describe the color that image of clothing includes object, to image in itself in retrieval
Size is big, the view angle dependency of direction and clothes object is smaller.Textural characteristics are that have not depending on color or brightness reflection
The visual signature of image homogeneity phenomenon, can be from surrounding and watching different object in upper differentiation image.The description method of usual textural characteristics
There are statistic law, Spectrum Method, Structure Method and modelling.Shape feature is common including the description to profile and region and one kind
Feature, big effect, the extraction algorithm of shape feature can be divided into based on profile and be based in being played in image of clothing retrieval
Two kinds of region.Due to the rapid development of deep learning method, there is the performance for making us being pleasantly surprised in image processing field, based on
The method of deep learning is also introduced into the costume retrieval problem of picture material to extract the visual signature of image, by depth nerve net
CNN (convolutional neural networks, CNN, Convolutional Neural Network) feature of network extraction has the pumping of higher
As semanteme, it can preferably meet the Search Requirement of user.
Functionally, the costume retrieval method of image content-based can only be examined according to the picture material that user inquires
Rope, user can not be according to the hobby editors of itself for the association attributes of the image of clothing of inquiry, to obtain satisfied inquiry knot
Fruit.
Invention content
The purpose of the present invention is overcome the shortcomings of existing method, it is proposed that a kind of editable image of clothing inspection of clothes attribute
The method of rope.Clothes attributive character is generated using the method for clothes attributive character generation, is then melted again with former image of clothing feature
It closes, forms the new garment feature for meeting user intent, retrieved in conjunction with the costume retrieval method of image content-based, solve
Determined original image content-based costume retrieval method can only be retrieved according to the picture material that user inquires, user can not
The problem of being used for the association attributes for the image of clothing inquired according to the hobby editor of itself, can cause user to be liked according to itself
The attribute of image of clothing to be retrieved is changed, obtains satisfied query result.
To solve the above-mentioned problems, the method that the present invention proposes a kind of editable image of clothing retrieval of clothes attribute,
The method includes：
Extract the feature vector of the image of clothing in existing garment data；
Obtain the image of clothing input by user for search；
Extract feature vector of the user for the image of clothing of search；
The clothes attribute of image of clothing is obtained with clothes attribute method of discrimination；
Clothes attribute and optional modification option are showed into user, and obtain the clothes attribute item that user changed；
With the feature vector of image of clothing and the modified clothes attribute item of user of original search, new garment feature is formed
Vector；
Calculate the similarity between clothes attributive character in the clothes attributive character newly formed and garment data；
Show the highest preceding k parts image of clothing of similarity.
Preferably, the feature vector of the image of clothing in the existing garment data of extraction, specially：
Calculate the N-dimensional feature vector for obtaining the image of clothing in garment data；The feature vector of extraction is saved in clothes
It fills in database, subsequently to calculate inquiry.
Preferably, it is described to obtain the clothes attribute of image of clothing with clothes attribute method of discrimination, specially：
Clothes property distribution probability is obtained with clothes attribute method of discrimination；Clothes category is obtained according to clothes property distribution probability
Property label；The description of clothes attribute is obtained according to clothes attributive character.
Preferably, the feature vector of the image of clothing with original search and the modified clothes attribute item of user are formed newly
Garment feature vector, specially：
Confrontation network is generated with clothes attribute data collection pre-training condition；The generation item of acquisition condition generation confrontation network
Part；Generate new clothes attributive character；Generate new image of clothing feature.
In embodiments of the present invention, it is proposed that a kind of method of the editable image of clothing retrieval of clothes attribute, this method
Clothes attributive character is generated using the method for clothes attributive character generation, is then formed new with former image of clothing Fusion Features again
The garment feature for meeting user intent, retrieved, solved original in conjunction with the costume retrieval method of image content-based
The costume retrieval method of image content-based can only according to user inquire picture material be retrieved, user can not be according to itself
Hobby editor for inquiry image of clothing association attributes the problem of, can cause user according to itself hobby modification it is to be checked
The attribute of rope image of clothing obtains satisfied query result.
Description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention, for those of ordinary skill in the art, without creative efforts, can be with
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is the flow chart of the embodiment of the present invention；
Fig. 2 is the system schematic of the embodiment of the present invention.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present invention, the technical solution in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other without creative efforts
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is the flow chart of the embodiment of the present invention, as shown in Figure 1, this method includes：
S1 extracts the feature vector of the image of clothing in existing garment data；
S2 obtains the image of clothing input by user for search；
S3, extraction user is for the feature vector for the image of clothing searched for；
S4 obtains the clothes attribute of image of clothing with clothes attribute method of discrimination；
Clothes attribute and optional modification option are showed user, and obtain the clothes attribute that user changed by S5
；
S6 with the feature vector of image of clothing and the modified clothes attribute item of user of original search, forms new clothes
Feature vector；
S7 calculates the similarity between clothes attributive character in the clothes attributive character newly formed and garment data；
S8 shows the highest preceding k parts image of clothing of similarity.
Step S1, it is specific as follows：
S11 carries out feature to existing toggery database using the method that feature extraction is carried out to image of clothing and carries
It takes, specially using deep neural network VGG16, by every image input network model of toggery database, preserves
Feature of the output of first full articulamentum FC6 as image, the main function of FC6 is the figure for extracting front convolution layer network
Vector form is turned to as block is smooth, the vector dimension size after planarizing is 4096 dimensions, is represented with following formula：
F(Is)=Xs=(xs1,xs2,xs3,...,xsN)
Wherein：F is the net model methodology using VGG16, with a certain image of clothing in existing garment data
IsFor, XsIt is to I by Fs4096 dimensional feature vectors of extraction, (xs1,xs2,xs3,...,xsN) it is XsValue in each dimension, institute
Using N values therein as 4096, X is representedsDimension values.
S12, by all feature X of extractionsIt is saved in garment data, subsequently to calculate inquiry.
Step S1 is pretreatment, does not need to all implement in each retrieval.
Step S2, it is specific as follows：
The image of clothing input by user for search is obtained, is generally obtained and used using currently a popular human-computer interaction form
Family upload image of clothing, the method with respect to user's operation simply and also user's unaware.Specifically, when user clicks search work(
Can when, need user select an anticipatory remark or online image, after selection, front end will transmit automatically user selection
Image to retrieval server on.The image of clothing of user search is the clothes that user search is used in the embodiment of the present invention in Fig. 2
Fill image.
Step S3, it is specific as follows：
The image of clothing feature extraction uploaded using the method for carrying out feature extraction to image of clothing to user, is specially adopted
With deep neural network VGG16, method is consistent with the method for the image of clothing feature of step S1 extraction toggery databases.
It is calculated according to following formula and obtains N-dimensional feature vector of the user for the image of clothing of search：
F(Ic)=Xc=(xc1,xc2,xc3,...,xcN)
Wherein：F is the net model methodology of VGG16, IcIt is that user is used for the image of clothing searched for, XcIt is to be extracted by F
4096 dimensional feature vectors, (Xc1,Xc2,Xc3,...,XcN) it is XcValue in each dimension, so N values therein are 4096, table
Show XcDimension values.
Step S4, it is specific as follows：
Clothes attribute tags are represented with vector M, each m in MijIt is { 0, a 1 } value, represents the i-th kind of clothes
Property, j is j-th of value of the i-th attribute, and no this property value is represented for 0, and 1 represents there is this property value.
Here illustrate by taking four type of clothes attribute, collar, sleeve length and color attributes as an example, wherein type has T-shirt, lining
3 kinds of shirt, vest values, use m11m12m13It represents, value represents shirt when being 010；Collar whether there is 3 kinds of neck, V necks, crew neck values,
Use m21m22m23It represents, value represents crew neck when being 001；Sleeve length has 3 kinds of sleeveless, cotta, long sleeves values, uses m31m32m33It represents,
Value represents cotta when being 010；Color has 3 kinds of white, black, grey values, uses m41m42m43It represents, value represents when being 100
White.Attribute tags vector M is 12 dimensions.
S41 differentiates that network obtains clothes property distribution probability with clothes attribute：
Differentiate that network obtains clothes property distribution probability with clothes attribute, refer to using clothes attribute method of discrimination to generation
Garment feature carry out property distribution probabilistic forecasting.Clothes attribute differentiates that network is mainly three full articulamentum compositions, complete to connect
Using Relu activation primitives between layer, the feature vector, X for the image of clothing of search for the user obtained in S3 is inputtedc, it is defeated
Go out the feature P for that can represent clothes attribute probability distribution after the conversion of sigmoid functions.
S42 obtains clothes attribute tags according to clothes property distribution probability：
Clothes attribute tags vector M is obtained with the feature P of clothes attribute probability distribution：The dimension of P and M is equal, P tables
Show the probability of j-th of value of the ith attribute of M, the corresponding m in the position of maximum probability during this is groupedijValue for 1,
Remaining is set to 0.Such as P for [0.35,0.55,0.77, | 0.73,0.35,0.57, | 0.45,0.15,0.77, | 0.26,0.85,
0.47 ,], then M for [001 | 100 | 001 | 010].
S43 obtains the description of clothes attribute according to clothes attribute tags：
Clothes can be easily obtained by the clothes property distribution probability characteristics of S41 or the clothes attribute tags of S42
Attribute description A1,A2,A3,….Illustrate by taking four type of clothes attribute, collar, sleeve length and color attributes as an example, wherein type
There are 3 kinds of T-shirt, shirt, vest values, m11m12m13Value represents shirt when being 010；Collar whether there is 3 kinds of neck, V necks, crew neck values,
m21m22m23Value represents crew neck when being 001；Sleeve length has 3 kinds of sleeveless, cotta, long sleeves values, m31m32m33Value represents when being 010
Cotta；Color has 3 kinds of white, black, grey values, m41m42m43White is represented when value is 100.Four attribute are then shared to retouch
State A1,A2,A3,A4For shirt, crew neck, cotta, white.
S41, S42, S43 are summarized, the attribute for obtaining user's clothes can be calculated according to following formula：
J(Xc)=A1,A2,A3,...
Wherein：J is according to the overall model of the method for garment feature vector differentiation clothes attribute, XcIt is the use obtained in S3
Family is for the feature vector A for the image of clothing searched for1,A2,A3... it is each attribute description of clothes.
Clothes attribute method of discrimination is to the result of the determined property of user search image in Fig. 2, it can be seen that clothes attribute
Method of discrimination has been accurately judged to type, collar, sleeve length and the color attribute of the image of clothing of user search, their value difference
For T-shirt, crew neck, cotta and white.
Step S5, it is specific as follows：
Clothes attribute and optional modification option are showed user, refer to use currently a popular human-computer interaction by S51
Clothes property value and revisable option are presented to user by form in the form of user-friendly.It clothes attribute and can be repaiied in Fig. 2
It is that clothes attribute differentiates network to the determined property of user search image as a result, for type in the embodiment of the present invention to reelect item：T
Sympathize, collar：Crew neck, sleeve length：Cotta, color：White.
S52 obtains clothes attribute A of the user to former image of clothing1,A2,A3... the new clothes property value after modifying
A′1,A'2,A'3..., as shown in the clothes attribute that user changed in Fig. 2, user changes clothes color from original white
For black.
Step S6, it is specific as follows：
By on user newly modified attribute migration to the feature of user search picture, new retrieval character is formed.
S61 generates new clothes attributive character
Original generation confrontation network has generator and determining device two parts, and generator is mainly generated from training data
The sample of same distribution, and arbiter is then to judge that input is truthful data or generation data.Generator can pass through clothes
Attribute data collection is trained to generate the network of random clothes attributive character, but this method is it is desirable that the clothes attribute specified
Feature so generate confrontation network using condition, can replace original random noise input to allow generation according to the condition provided
The specified clothes attributive character of device generation.
S611 generates confrontation network with clothes attribute data collection pre-training condition：
Clothes attributive character is extracted to Stanford Clothing Attributes Dataset (SCAD) using VGG16
Collection (each attributive character dimension is 4096, identical with image of clothing characteristic dimension) generates the true number of confrontation network as condition
The form of clothes attribute tags in S42 is converted into as condition entry according to collection, and using the attribute tags of SCAD.
After generator receives the forms of attribute tags as condition entry, by the vector that it is extended to 4096 dimensions with 0, then
With the attribute feature vector that truthful data is concentrated by the way of stacking, i.e., using the square of two 1 × 4096 formation 2 × 4096
Battle array generates new attributive character using three full articulamentums, Relu activation primitives is used between full articulamentum.Arbiter receives
For the attributive character of one 4096 dimension as input, it is the attribute spy of the attributive character that truthful data is concentrated or generation to judge it
Sign.For the feature of vector form, two layers of full articulamentum may be used in arbiter, and centre uses the form of Relu activation primitives,
The result of output passes through simoid function transition probability forms, and it is the probability for the feature that truthful data is concentrated to represent input feature vector.
S612 obtains the formation condition of condition generation confrontation network：
Using user in S52 to the modification of the clothes attribute of former image of clothing as a result, being translated into clothes attribute in S42
The form of label generates the formation condition of confrontation network as condition.
S613 generates new clothes attributive character：
Clothes category is generated using the formation condition in the generator and S612 of the condition confrontation generation network of S611 pre-training
Property feature, for promoted generation attributive character stability, using multiple clothes attributive character are first generated, then take its average value make
For new clothes attributive character.
S62 generates new image of clothing feature：
The user obtained in step S3 for search image of clothing feature vector and S613 in generate new clothes category
Property Fusion Features generate new image of clothing feature vector.Since the dimension of the two vectors is identical, it is possible to directly
By the way of stacking, i.e., using the matrix of two 1 × 4096 formation 2 × 4096, using the transformation being made of full articulamentum
Network forms new image of clothing feature vector.Converting network is made of three full articulamentums, and Relu is used between full articulamentum
Activation primitive inputs the matrix for 2 × 4096, exports the vector for 1 × 4096.S61, S62 are summarized, it can be according to following formula
It calculates and obtains new garment feature vector：
G(Xc,A′1,A′2,A′3...) and=X'c=(x'c1,x'c2,x'c3,...,x'cN)
Wherein：G is garment feature generation method, i.e., after being changed using the feature vector and user of the image of clothing of former search
The corresponding new attributive character of clothes attribute generate the method for new garment feature vector, XcIt is that the user obtained in S3 is used for
The feature vector of the image of clothing of search, Xc' it is the new garment feature vector generated by G, (x'c1,x'c2,x'c3,...,x
'cN) it is X 'cProperty value.
Step S7, it is specific as follows：
Using characteristic distance computational methods, it is special to calculate clothes in the new garment feature vector sum garment data that S6 is obtained
The distance between vector is levied, then characteristic distance is converted into similarity.
The characteristic distance obtained between two clothes is calculated according to following formula：
Wherein, d (Xs,X'c) represent two clothes between characteristic distance, L2(Xs,X′c) represent Euclidean distance calculation formula.
The characteristic similarity obtained between two garment features is calculated according to following formula：
s(Xs,X′c)=max (0,1-d (Xs,X′c)/d)
Wherein s (Xs,X′c) representing similarity between two clothes, max takes higher value in two values, and d is one preselected
Distance constant, twice of the general feature average distance for choosing the image of clothing in garment data.
Step S8, it is specific as follows：
It shows the highest preceding k parts image of clothing of similarity, refers to the similarity for calculating S7, arranged in the way of descending
Row, then k similarity is highest as a result, corresponding commodity image of clothing is showed user before selecting, k here can basis
Actual needs is chosen.Highest 6 retrieval results of similarity are that user examines after changing color attribute in the embodiment of the present invention in Fig. 2
Rope as a result, the k of use be equal to 6.
One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage
Medium can include：Read-only memory (ROM, Read Only Memory), random access memory (RAM, Random
Access Memory), disk or CD etc..
In addition, a kind of method of the editable image of clothing retrieval of clothes attribute provided above the embodiment of the present invention
Be described in detail, specific case used herein is expounded the principle of the present invention and embodiment, more than it is real
The explanation for applying example is merely used to help understand the method and its core concept of the present invention；Meanwhile for the general technology of this field
Personnel, thought according to the present invention, there will be changes in specific embodiments and applications, in conclusion this theory
Bright book content should not be construed as limiting the invention.
Claims (5)
- A kind of 1. method of the editable image of clothing retrieval of clothes attribute, which is characterized in that the method includes：Extract the feature vector of the image of clothing in existing garment data；Obtain the image of clothing input by user for search；Extract feature vector of the user for the image of clothing of search；The clothes attribute of image of clothing is obtained with clothes attribute method of discrimination；Clothes attribute and optional modification option are showed into user, and obtain the clothes attribute item that user changed；With original search the feature vector of image of clothing and the modified clothes attribute item of user, formed new garment feature to Amount；Calculate the similarity between clothes attributive character in the clothes attributive character newly formed and garment data；Show the highest preceding k parts image of clothing of similarity.
- 2. the method for the editable image of clothing retrieval of clothes attribute as described in claim 1, which is characterized in that the clothes Dress attribute method of discrimination obtains the clothes attribute step of image of clothing, specially：Clothes property distribution probability is obtained with clothes attribute method of discrimination；Clothes attribute mark is obtained according to clothes property distribution probability Label；The description of clothes attribute is obtained according to clothes attributive character.
- 3. the method for the editable image of clothing retrieval of clothes attribute as claimed in claim 2, which is characterized in that the clothes Dress attribute method of discrimination obtains clothes property distribution probability, specially：Property distribution probabilistic forecasting is carried out to the garment feature of generation using clothes attribute method of discrimination.Clothes attribute differentiates network Mainly three full articulamentums form, and using Relu activation primitives between full articulamentum, input the clothes for search for user The feature vector of image exports the feature vector for that can represent clothes attribute probability distribution after the conversion of sigmoid functions.
- 4. the method for the editable image of clothing retrieval of clothes attribute as described in claim 1, which is characterized in that described with former The feature vector of the image of clothing of search and the modified clothes attribute item of user form new garment feature vector, specially：Confrontation network is generated with clothes attribute data collection pre-training condition；The formation condition of acquisition condition generation confrontation network；It is raw The clothes attributive character of Cheng Xin；Generate new image of clothing feature.
- 5. the method for the editable image of clothing retrieval of clothes attribute as claimed in claim 4, which is characterized in that the generation New image of clothing feature, specially：User generates new image of clothing spy for the feature vector of the image of clothing of search and the fusion of new clothes attributive character Sign vector.Due to the two vector dimensions be identical, it is possible to directly using stacking by the way of, i.e., using two 1 × 4096 form 2 × 4096 matrix, and new image of clothing feature vector is formed using the converting network being made of full articulamentum. Converting network is made of three full articulamentums, using Relu activation primitives between full articulamentum, inputs the matrix for 2 × 4096, Export the vector for 1 × 4096.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711422438.9A CN108197180A (en) | 2017-12-25 | 2017-12-25 | A kind of method of the editable image of clothing retrieval of clothes attribute |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711422438.9A CN108197180A (en) | 2017-12-25 | 2017-12-25 | A kind of method of the editable image of clothing retrieval of clothes attribute |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108197180A true CN108197180A (en) | 2018-06-22 |
Family
ID=62583725
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711422438.9A Pending CN108197180A (en) | 2017-12-25 | 2017-12-25 | A kind of method of the editable image of clothing retrieval of clothes attribute |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108197180A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109064249A (en) * | 2018-06-28 | 2018-12-21 | 中山大学 | A kind of clothes recommendation optimization method and its system based on feature personalization modification |
| CN111784787A (en) * | 2019-07-17 | 2020-10-16 | 北京沃东天骏信息技术有限公司 | Image generation method and device |
| CN112598806A (en) * | 2020-12-28 | 2021-04-02 | 深延科技(北京)有限公司 | Virtual fitting method and device based on artificial intelligence, computer equipment and medium |
| CN112699261A (en) * | 2020-12-28 | 2021-04-23 | 大连工业大学 | Automatic clothing image generation system and method |
| CN113297475A (en) * | 2021-03-26 | 2021-08-24 | 阿里巴巴新加坡控股有限公司 | Commodity object information searching method and device and electronic equipment |
| CN114556435A (en) * | 2021-05-28 | 2022-05-27 | 株式会社威亚视 | Information processing apparatus, information processing method, and program |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104933697A (en) * | 2014-03-20 | 2015-09-23 | 株式会社东芝 | Image Processing Apparatus, Image Processing System, And Image Processing Method |
| CN107203548A (en) * | 2016-03-17 | 2017-09-26 | 阿里巴巴集团控股有限公司 | Attribute acquisition methods and device |
-
2017
- 2017-12-25 CN CN201711422438.9A patent/CN108197180A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104933697A (en) * | 2014-03-20 | 2015-09-23 | 株式会社东芝 | Image Processing Apparatus, Image Processing System, And Image Processing Method |
| CN107203548A (en) * | 2016-03-17 | 2017-09-26 | 阿里巴巴集团控股有限公司 | Attribute acquisition methods and device |
Non-Patent Citations (1)
| Title |
|---|
| ZHENGZHONG ZHOU ET AL.: "Interactive Image Search for Clothing Recommendation", 《PROCEEDINGS OF THE 24TH ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109064249A (en) * | 2018-06-28 | 2018-12-21 | 中山大学 | A kind of clothes recommendation optimization method and its system based on feature personalization modification |
| CN111784787A (en) * | 2019-07-17 | 2020-10-16 | 北京沃东天骏信息技术有限公司 | Image generation method and device |
| CN111784787B (en) * | 2019-07-17 | 2024-04-09 | 北京沃东天骏信息技术有限公司 | Image generation method and device |
| CN112598806A (en) * | 2020-12-28 | 2021-04-02 | 深延科技(北京)有限公司 | Virtual fitting method and device based on artificial intelligence, computer equipment and medium |
| CN112699261A (en) * | 2020-12-28 | 2021-04-23 | 大连工业大学 | Automatic clothing image generation system and method |
| CN113297475A (en) * | 2021-03-26 | 2021-08-24 | 阿里巴巴新加坡控股有限公司 | Commodity object information searching method and device and electronic equipment |
| CN113297475B (en) * | 2021-03-26 | 2024-10-22 | 淘宝（中国）软件有限公司 | Commodity object information searching method and device and electronic equipment |
| CN114556435A (en) * | 2021-05-28 | 2022-05-27 | 株式会社威亚视 | Information processing apparatus, information processing method, and program |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108197180A (en) | A kind of method of the editable image of clothing retrieval of clothes attribute | |
| CN113795834B (en) | Techniques for modifying query images | |
| CN107861972B (en) | Method and equipment for displaying full commodity result after user inputs commodity information | |
| CN102232220B (en) | Method and system for extracting and correlating video interested objects | |
| CN104424296B (en) | Query word sorting technique and device | |
| CN108109055B (en) | Cross-scene clothing retrieval method based on image rendering | |
| Laenen et al. | Web search of fashion items with multimodal querying | |
| Mortara et al. | Semantics-driven best view of 3D shapes | |
| US20220245391A1 (en) | Text-conditioned image search based on transformation, aggregation, and composition of visio-linguistic features | |
| CN105787490A (en) | Commodity fashion identification method and device based on deep learning | |
| Vempati et al. | Enabling hyper-personalisation: Automated ad creative generation and ranking for fashion e-commerce | |
| CN101950400A (en) | Network shopping guiding method | |
| CN108984577A (en) | Commodity patent acquisition methods and system on a kind of line based on machine recognition | |
| Afifi et al. | Image Recoloring Based on Object Color Distributions. | |
| Chun et al. | A novel clothing attribute representation network-based self-attention mechanism | |
| KR102301663B1 (en) | Identifying physical objects using visual search query | |
| Gong et al. | Aesthetics, personalization and recommendation: A survey on deep learning in fashion | |
| Li et al. | Retrieval of clothing images based on relevance feedback with focus on collar designs | |
| Tautkute et al. | What looks good with my sofa: Multimodal search engine for interior design | |
| Misraa et al. | Multi-modal retrieval using graph neural networks | |
| CN109325529A (en) | Sketch identification method and application of sketch identification method in commodity retrieval | |
| Fuentes et al. | Sketch-qnet: A quadruplet convnet for color sketch-based image retrieval | |
| CN107622071A (en) | By indirect correlation feedback without clothes image searching system and the method looked under source | |
| Liu et al. | FRSFN: A semantic fusion network for practical fashion retrieval | |
| Chi et al. | UbiShop: Commercial item recommendation using visual part-based object representation |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180622 |