CN101950400B - Picture retrieving method of network shopping guiding method - Google Patents

Picture retrieving method of network shopping guiding method Download PDF

Info

Publication number
CN101950400B
CN101950400B CN201010501027.0A CN201010501027A CN101950400B CN 101950400 B CN101950400 B CN 101950400B CN 201010501027 A CN201010501027 A CN 201010501027A CN 101950400 B CN101950400 B CN 101950400B
Authority
CN
China
Prior art keywords
picture
commodity
described
feature
client
Prior art date
Application number
CN201010501027.0A
Other languages
Chinese (zh)
Other versions
CN101950400A (en
Inventor
姚建
Original Assignee
姚建
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 姚建 filed Critical 姚建
Priority to CN201010501027.0A priority Critical patent/CN101950400B/en
Publication of CN101950400A publication Critical patent/CN101950400A/en
Application granted granted Critical
Publication of CN101950400B publication Critical patent/CN101950400B/en

Links

Abstract

The invention discloses a network shopping guiding method. In the method, a client searches for the commodity picture display of the association group of the pictures of solid commodities on a server through a picture searching engine. The method comprises that: the server receives commodity pictures and establishes an association group of the similarity of all commodity pictures; the display interface of the client sends a clicked commodity picture retrieving instruction through the image searching engine; and the server sends the clicked commodity picture in the association group of the commodity pictures to the client to display the picture on a display interface. The network shopping guiding method of the invention guides users in commodity purchasing by using the association group of the commodity pictures, allows the users to find target commodities with reduced click times and improves network shopping efficiency.

Description

The picture retrieval method in network shopping guiding method field

Technical field

The invention belongs to shopping at network technical field, relate in particular to the picture retrieval method in a kind of network shopping guiding method field.

Background technology

Along with the development of infotech, the appearance of online shopping mall's picture as emerging rapidly in large numbersBamboo shoots after a spring rain, the appearance of magnanimity commodity makes how to organize these information to become a new difficult problem.Traditional method by brand and pattern classification cannot meet the growing needs of shopping at network.Consumer is generally can search for for the demand of oneself carrying out shopping at network, to dwindle the commodity amount that need to browse.Nonetheless, in some large-scale online shopping malls, still have a large amount of commodity and enter Search Results.For example, in some large-scale shopping at network platform, it is two that men and women's formula footwear that businessman sells reach hundreds of thousands, even if the kind to footwear, pattern, size, price etc. limit, the quantity of the footwear that search obtains often still has thousands of even up to ten thousand two, the footwear that here face is looked for a pair of special style just as look for a needle in a haystack, experience and allow consumer be difficult to accept by the shopping of formula of browsing page by page.

The search technique for particular commodity of having applied is at present all the search of the Word message based on commodity, specifically, the supplier of information first inputs the various information of commodity with written form, and consumer carrys out the content of limit search and selects specific commodity to carry out detailed consulting in the result of search in the process of shopping by word.The shortcoming of this searching method is to have ignored completely the information such as a large amount of shape that the picture of commodity comprises, texture, color, and these intuitively information in net purchase process, to consumer, provide huge convenient and client is finally made a decision to buy the considerable effect that plays.

Therefore, there is defect in prior art, is further improved and develops.

Summary of the invention

The object of the present invention is to provide the picture retrieval method in a kind of network shopping guiding method field, make user in the magnanimity commodity of network, can find easily own needed commodity.

Technical scheme of the present invention is as follows:

The picture retrieval method in network shopping guiding method field, client is searched the associated group's of physical commodity picture commodity picture presentation at server end by photographic search engine, comprise the following steps:

A, described server end receive commodity picture, set up the associated group of all commodity picture analogies degree;

The display interface of B, described client sends the search instruction of clicked commodity picture by described photographic search engine;

C, described server end send to described client to show at display interface by search engine in the associated group's of clicked commodity picture commodity picture.

Described method, wherein, described steps A also comprises the pick up and store of picture product features, the calculating of the distance between product features and storage, the associated group's of the calculating of feature weight and storage and commodity picture calculating and storage.

Described method, wherein, the extraction of the product features in described steps A also comprises: the normal distribution Regularization process of feature, obtains the relative value of feature.

Described method, wherein, adopts the associated group of similarity between K-means algorithm or intensified learning classifier algorithm counting yield feature in described steps A.

Described method, wherein, the calculating of described feature weight comprises initial weight and the calculating of weight in real time.

Described method, wherein, in described step C, client is shown multiple commodity features in the mode of various dimensions.

Described method, wherein, in described step C, client is shown the feature of various dimensions with the mode projection of two-dimentional Cartesian coordinates.

Described method, wherein, commodity picture in described step C in associated group is filled full described two-dimentional Cartesian coordinates according to the distance between two-dimentional Cartesian coordinates, and the two-dimentional cartesian coordinate map of the commodity picture of filling full associated group is shown at described display interface.

Described method, wherein, the distance between described Cartesian coordinates is Euclidean distance or mahalanobis distance.

Compared with prior art, the invention provides the picture retrieval method in a kind of network shopping guiding method field, client is crossed photographic search engine by the information exchange of clicked commodity picture and is sent to server end, server end sends the associated group's of clicked commodity picture commodity picture to return to client by photographic search engine, and the display interface of client shows the associated group's of clicked commodity picture commodity picture.The picture retrieval method in network shopping guiding method of the present invention field, in the associated group's of commodity picture mode as user shopping guide commodity, makes user in the situation that reducing number of clicks, find rapidly target commodity, improves the efficiency of shopping at network.

Accompanying drawing explanation

Fig. 1 is the structural drawing of network shopping guidance system of the present invention;

Fig. 2 is the structural drawing of server end of the present invention.

Fig. 3 is the sparse distribution figure of toe-cap feature of the present invention;

Fig. 4 is the normal distribution of toe-cap feature of the present invention;

Fig. 5 is the schematic diagram of display of commodity display on client display interface;

Fig. 6 is the schematic diagram of the proper vector that represents of client multi-dimensional degree module;

Fig. 7 is the schematic diagram that client two dimension is shown the two-dimentional Cartesian coordinates that projection module represents;

Fig. 8 is client display interface search instruction schematic diagram.

Embodiment

Below in conjunction with accompanying drawing, preferred embodiment of the present invention is described in further detail.

Network shopping guidance system provided by the invention, can, for any physical commodity shopping navigation on network, take footwear as example, be described in detail network shopping guidance system now.

Described network shopping guidance system is shopping at network navigation by picture presentation, picture searching, as shown in Figure 1, has:

Server end, comprises data server end and calculation server end, the data such as distance between feature, the product features of described data server end storing commodity picture, commodity picture, feature weight; Described calculation server is for distance, feature weight between feature, the product features of analytical calculation commodity picture;

Photographic search engine, according to the search instruction of client, at the picture of described server end inquiry command adapted thereto, and returns to described client;

Client, for connecting with server end, and sends the search instruction of doing shopping.

The peripheral profile of footwear, color, toe-cap, upper of a shoe, sole, texture play considerable effect to client's the purchase of finally making a decision, and traditional text search can only be carried out index to relevant word, the information such as the peripheral profile, color, toe-cap, upper of a shoe, sole, texture of the footwear that product picture comprises have been ignored completely.

According to the commodity picture of footwear, utilize image analysis technology to extract the feature of picture, comprise the information such as peripheral profile, color, toe-cap, upper of a shoe, sole, texture.Described calculation server comprises pretreatment unit, can process the irregular picture receiving, to obtain for showing and the material of picture analyzing.

Calculation server of the present invention comprises product input block, for the input of commodity picture; Feature extraction unit, for the extraction of the value of commodity picture feature.Described product input block can be also a client separation with described server.

For this commodity of footwear, described feature extraction unit can comprise peripheral profile extraction module, color extraction module, toe-cap extraction module, upper of a shoe extraction module, sole extraction module and texture extraction module.

Described peripheral profile extraction module can utilize the profile of footwear in commodity picture to calculate the peripheral contour feature of footwear.Color extraction module can be used the color histogram of HSV to obtain the color characteristic of footwear according to the picture of footwear.Toe-cap extraction module can do by opening to remaining silent, by point to blunt analysis according to the feature of toe-cap in picture, thereby obtains toe-cap feature.Upper of a shoe extraction module is exactly the upper of a shoe feature that obtains footwear according to the specificity analysis of upper of a shoe relevant position in picture.Sole extraction module is that edge detection method adopts Prewitt/canny algorithm according to the sole feature of the shape of sole in picture being done rim detection and obtained footwear.Texture extraction module is the textural characteristics that obtains footwear by the image of footwear being carried out to carry out wavelet transform again after gray scale processing

The peripheral contour feature of described footwear, color characteristic, toe-cap feature, upper of a shoe feature, sole feature, these six features of textural characteristics, can use proper vector S(s 1, s 2, s 3, s 4, s 5, s 6), and the weight of each eigenwert in S vector is T (t 1, t 2, t 3, t 4, t 5, t 6), the population characteristic value of each like this shoes is exactly K=S*T, and every two couples of shoes (K 1, K 2) between similarity just can use | K 1– K 2| represent.

Use the feature of these vector representations do between shoes similarity relatively before, we will first carry out the relative value processing of feature because the absolute value of feature is nonsensical, just relative value is just meaningful.

Such as the toe-cap feature s to every pair of footwear 3, there are the footwear of 20,000 pairs, just, with 20,000 discrete points, these 20,000 points may be equally distributed, possibility is but more uneven distribution, very intensive of certain region, and very as shown in Figure 3 sparse of other region.

In order better to reflect similarity, the Regularization that the data of feature is carried out to normal distribution, makes the curve that reflects these 20,000 pairs of footwear meet normal distribution, as shown in Figure 4.

Server end of the present invention arranges Regularization module, for the feature that feature extraction unit is extracted, does normal distribution Regularization, obtains the relative value of feature.The relative value of feature can be stored in the characteristic storage unit of described database server.

The weight that how to calculate each feature is key of the present invention, and solution of the present invention is as follows:

The weight of feature comprises initial weight and real-time weight.

Initial weight is that network shopping guidance system is initially the weight that each feature arranges, the quantity of for example getting sample set is 100, be exactly 100 pairs of shoes, each feature of these 100 pairs of footwear is queued up by artificial method, guarantee that the similarity between adjacent shoes is the highest, distance between any three couples of shoes A B C that sorted like this, | A-B|<|A-C| and | B-C|<|A-C|, so just can find and adapt to the weighted value T of all these 100 orders.

Weight is take initial weight as basis in real time, according to client's click and selection, does machine learning, then weight is adjusted.Certainly to dissimilar footwear, all will have a set of different weighted value, the weighted value of for example sandals is different with the weighted value of sport footwear.Concrete, described calculation server comprises weight unit, the click according to user to feature and selection constantly regulate the weight calculation of individual features to go out real-time weight, for example:

Client has selected successively A, B, tri-pairs of shoes of C in the process of selecting footwear, and the weight of the feature of A is less than B, and the weight of the feature of B is less than C, and described weight unit is adjusted T vector according to user's selection.

The adjustment of weight is mainly to calculate when variety classes shoes are adjusted in real time, the interval of weight, or the step-length of weight.And the calculating of weight step-length is by the training set of a group, by the method for intensified learning, progressively approach acquisition with the observation of human eye.Such as eigenwert overall for sport footwear is 1 to 100, the present invention supposes that the step-length of weight is 5, the training set of this group selected, and such as the training set of this group finds the clicks of target shoes, be (10,8,2,15,7,4,10), and the desirable clicks that finds target is: 1, and that tries one's best lacks.2, the difference of each shoes clicks is also little as far as possible.There have been this two standards, just can have ceaselessly changed step-length, found the optimal step-length of approaching.

The weight storage unit of database server side comprises having and the corresponding weight memory module in real time of client, can store the real-time weight vectors of the specific client that weight unit calculates.

The present invention is according to the similarity between feature, i.e. distance between feature, sets up the associated group between commodity, is analyzing after good these characteristic distances, utilizes the similarity between commodity to set up associated group.Described calculation server comprises associated computing unit, for realizing the associated group of similarity between commodity.

A kind of interrelational form of described associated computing unit is, according to the kind of footwear, footwear are divided into footwear that N kind is different as starting condition; Concrete associated account form is as follows:

First, footwear are divided into the variety classeses such as sport footwear, playshoes, sandals, take the associated group that sets up sport footwear as example, introduce the computing method of associated computing unit.

Secondly, the vectorial S of feature that chooses a pair of shoes in sport footwear is as initial characteristics vector.

Three, the proper vector S that calculates successively every pair of trainers, to the distance between initial characteristics vector, according to the distance between sport footwear proper vector, sets up associated group.Distance of the present invention is exactly similarity, but the standard that similarity is eyes to be found out, and distance is the numerical value that utilizes eigenwert to calculate, the present invention is exactly the difference of utilizing range marker eyes similarity.

Described associated computing unit can adopt K-means algorithm or intensified learning classifier algorithm to carry out the associated group of similarity between counting yield feature.K-means algorithm is accepted input quantity k; Then n data object is divided into k cluster to obtained cluster is met: the object similarity in same cluster is higher; And object similarity in different clusters is less.

Client of the present invention comprises user's display interface, and described user's display interface comprises search instruction display unit, as shown in Figure 8, comprises the kind of commodity, for example: clothes, footwear, bag etc.; Comprise the type of commodity, such as sport footwear, leather shoes, sandals etc.

The described type of merchandise can display with picture, and for example Fig. 5 displays out the picture of dissimilar footwear.Because every pictures all has associated group, user clicks any one in picture, and the associated group's of clicked commodity picture commodity figure sector-meeting is presented at display interface.Every kind of picture of display interface can be connected by the association store unit of photographic search engine and server end.

Described every pictures has information display unit, and described information display unit can be by photographic search engine from the commodity price link storage unit query of server end to data.Described information display unit is linked at display interface by the commodity price inquiring and shows, for user provides approach and the information of purchase.

Described client also comprises merchandise display unit, and described merchandise display unit comprises various dimensions module and two dimension displaying projection module.

Described various dimensions module is for representing a kind of proper vector of commodity, and each feature of commodity picture occupies a dimension, as shown in Figure 6.

Described two dimension is shown projection module, by two-dimentional Cartesian coordinates, represent the proper vector of commodity, a concrete axle using a feature in proper vector as two-dimentional Cartesian coordinates, other the another one axle of the two-dimentional Cartesian coordinates of comprehensive conduct of proper vector of commodity of commodity.

For example, the two dimension displaying of footwear as shown in Figure 7, transverse axis is shape facility, and transverse axis combines the 5 large characteristics such as peripheral profile, toe-cap, upper of a shoe, sole, texture, and ordinate is the color characteristic of footwear.The two dimension of the feature of product as shown in Figure 7 shows, take initial point (upper left) as starting point, lower right is terminal, from the point of starting point near distance, more similar, and from the point away from starting point, similarity is just poorer.The present invention can use Euclidean distance (Euclidean distance) or mahalanobis distance (Mahalanobis) distance as eigen vector.

Two dimension of the present invention is shown projection module, be exactly the eigenvector projection of multidimensional in two-dimensional coordinate, by reducing dimension, solve the complexity of problem.Be exactly briefly that various dimensions module is set up the vector that peripheral profile, color, toe-cap, upper of a shoe, sole, texture 6 are tieed up.Two dimension show projection module by 6 vector projections that are in two-dimentional coordinate system, the extension of each dimension is exactly Euclidean or the mahalanobis distance of this vector value, and part between any two axles is exactly the two comprehensive distance.

Described two dimension displaying projection module is set up after two-dimentional Cartesian coordinates, and described client also comprises product packing module.Described packing module is filled the picture of footwear to show according to the distance between two-dimensional coordinate as shown in Figure 7.

Described packing module has different filling algorithms, and principle is, the point from initial point close to more is first filled, by initial point, set out and can draw countless concentric circless like this, and on each specific circumference, the direction of filling can be clockwise, or counterclockwise.Such filling effect just, as radar scanning, is from the close-by examples to those far off filled two-dimensional coordinate full successively.At the sector region of each fritter, owing to being between two features, can calculate his distance to adjacent two feature axis (reference), by the distance of these two features, add like this weight of distance, just can be by populated middle sector region.

In the two-dimentional Cartesian coordinates of projection, certain feature is successively decreased along increasing progressively of a direction, the contour similarity of for example picture successively decreases along increasing progressively of X-axis, the similarity of texture is along X, Y-axis 45 is spent increasing progressively of angle and is successively decreased, and the similarity of color is successively decreased along increasing progressively of Y-axis.Fan-shaped to external radiation along with initial point like this, picture analogies degree is to stretch out according to certain rule, is exactly in some sense to have simulated the progressive law of human eye to commodity similarity.

The picture retrieval method in network shopping guiding method provided by the invention field, by the associated group of picture product features similarity, in the mode of photographic search engine, by product, according to the associated group's of similarity mode, provide the displaying of commodity picture for user, make website product picture rule present to client, very convenient client selects product.

Take network, purchase footwear as example, the picture retrieval method in network shopping guiding method field comprises the following steps:

Step 1, the server end of navigational system receives the picture of footwear, sets up the associated group of the similarity of footwear on all pictures, specifically comprises:

Step 101, server end extracts the feature of picture, comprises peripheral contour feature, color characteristic, toe-cap feature, upper of a shoe feature, sole feature, the textural characteristics of footwear, and by the vectorial S(s after normal distribution Regularization for these features 1, s 2, s 3, s 4, s 5, s 6) represent, and by the eigenwert after normal distribution Regularization or vector value storage at server end.

Step 102, server end arranges the initial weight of the each feature of footwear, and by initial weight T (t 1, t 2, t 3, t 4, t 5, t 6) storage.

Step 103, the general characteristic that server end calculates each shoes is exactly K=S*T, and every two couples of shoes (K 1, K 2) between similarity just can use | K 1– K 2| represent.The picture retrieval method in network shopping guiding method of the present invention field is differentiated the difference of product similarity with the numerical value sign product naked eyes of similarity.

Step 104, server end adopts K-means algorithm or intensified learning classifier algorithm to carry out the association of similarity between counting yield, builds the association group of similarity.

Step 2, client comprises the display interface for display of commodity, display interface arranges search instruction; Also comprise the various dimensions module of representation feature vector, and proper vector is shown to projection module with the two dimension of two-dimentional Cartesian coordinates sign.The two-dimentional Cartesian coordinates of display interface after according to eigenvector projection, shows in two-dimentional mode the picture of commodity at display interface.Client, by search engine, sends to server end by search instruction.

Step 3, any commodity picture of client display interface is clicked, and this clicked picture, by photographic search engine, inquires the associated group of similarity of this clicked picture at server end, and associated group's picture is shown according to associated group's two-dimentional Cartesian coordinates, as shown in Figure 7.

When user clicks picture, photographic search engine sends to server end by the record of the picture of click, and server end, according to clicking record, calculates the real-time weight of commodity.

Owing to clicking every commodity picture, client can show the associated group's of similarity of clicked picture photo, for example 50 commodity of display interface, the associated group of similarity of each commodity has 50 similar commodity, 2500 commodity have just been contained in the click of such 2 times, click the commodity that can cover 12.5 10,000 three times, just click for four times and can arrive millions of commodity, consider that can there be the commodity of repetition the inside, like this 7, after the click of 8 times, most clients can find target commodity by network picture searching fast.Although the present invention is with the embodiment that is retrieved as of footwear, any physical commodities such as clothes, case and bag can arrive with picture form quick-searching by the present invention on network, have improved effectiveness of retrieval.

Should be understood that, the above-mentioned statement for preferred embodiment of the present invention is comparatively detailed, can not therefore think the restriction to scope of patent protection of the present invention, and scope of patent protection of the present invention should be as the criterion with claims.

Claims (3)

1. the picture retrieval method in network shopping guiding method field, client is searched the associated group's of physical commodity picture commodity picture presentation at server end by photographic search engine, comprise the following steps:
A, described server end receive commodity picture, the product features of pick up and store picture, and each trade mark feature represents by proper vector;
The distance of calculating between storing commodity feature comprises: using the proper vector of arbitrary trade mark as initial characteristics vector, the proper vector of other trade mark is two similarities between trade mark same characteristic features with the distance between initial characteristics vector;
Calculate and storage feature weight, comprise initial weight and the calculating of weight in real time;
Adopt the associated group of similarity between K-means algorithm or intensified learning classifier algorithm counting yield feature;
And the associated group of the similarity of calculating and storing commodity picture;
The display interface of B, described client sends the search instruction of clicked commodity picture by described photographic search engine;
C, described server end send to described client by the associated group's of clicked commodity picture commodity picture by search engine;
The merchandise display unit of described client comprises various dimensions module and two-dimentional display module; Commodity picture in associated server end group is filled to full described two-dimentional Cartesian coordinates according to the distance between two-dimentional Cartesian coordinates, and the two-dimentional cartesian coordinate map of the commodity picture of filling full associated group is shown and shown at described display interface.
2. method according to claim 1, is characterized in that, the extraction of the product features in described steps A also comprises: the normal distribution Regularization process of feature, obtains the relative value of feature.
3. method according to claim 1, is characterized in that, the distance between described Cartesian coordinates is Euclidean distance or mahalanobis distance.
CN201010501027.0A 2010-10-09 2010-10-09 Picture retrieving method of network shopping guiding method CN101950400B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201010501027.0A CN101950400B (en) 2010-10-09 2010-10-09 Picture retrieving method of network shopping guiding method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201010501027.0A CN101950400B (en) 2010-10-09 2010-10-09 Picture retrieving method of network shopping guiding method

Publications (2)

Publication Number Publication Date
CN101950400A CN101950400A (en) 2011-01-19
CN101950400B true CN101950400B (en) 2014-04-16

Family

ID=43453887

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201010501027.0A CN101950400B (en) 2010-10-09 2010-10-09 Picture retrieving method of network shopping guiding method

Country Status (1)

Country Link
CN (1) CN101950400B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120173390A1 (en) * 2010-12-29 2012-07-05 Microsoft Corporation Single, mixed-view presentation of related products
BR112013002095A2 (en) 2011-03-30 2016-05-24 Rakuten Inc device and method of providing information, recording medium, device and method of displaying information, and information search system
TWI505108B (en) * 2011-03-31 2015-10-21 Rakuten Inc Information providing apparatus, information providing method, information display apparatus, information display method, information retrieval system, program product, and recording medium
WO2013029234A1 (en) * 2011-08-30 2013-03-07 Nokia Corporation Method and apparatus for providing deal combinations
CN103186616A (en) * 2011-12-30 2013-07-03 上海博泰悦臻电子设备制造有限公司 Vehicle-mounted information-processing system, cloud service center and vehicle-mounted equipment
CN103412938B (en) * 2013-08-22 2016-06-29 成都数之联科技有限公司 A kind of commodity price-comparing method extracted based on picture interactive multiobjective
CN104424230B (en) * 2013-08-26 2019-10-29 阿里巴巴集团控股有限公司 A kind of cyber recommended method and device
CN103473279A (en) * 2013-08-28 2013-12-25 上海合合信息科技发展有限公司 Query method, device, system and client for product descriptions
CN105608459B (en) 2014-10-29 2018-09-14 阿里巴巴集团控股有限公司 The dividing method and its device of commodity picture
CN105843816A (en) * 2015-01-15 2016-08-10 阿里巴巴集团控股有限公司 Method and device for determining display information of picture

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101004748A (en) * 2006-10-27 2007-07-25 北京航空航天大学 Method for searching 3D model based on 2D sketch

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101004748A (en) * 2006-10-27 2007-07-25 北京航空航天大学 Method for searching 3D model based on 2D sketch

Also Published As

Publication number Publication date
CN101950400A (en) 2011-01-19

Similar Documents

Publication Publication Date Title
Yu et al. Semisupervised multiview distance metric learning for cartoon synthesis
Zhu et al. Deep learning in remote sensing: A comprehensive review and list of resources
Wang et al. Sketch-based 3d shape retrieval using convolutional neural networks
Akgül et al. 3D model retrieval using probability density-based shape descriptors
Hu et al. A performance evaluation of gradient field hog descriptor for sketch based image retrieval
US8861844B2 (en) Pre-computing digests for image similarity searching of image-based listings in a network-based publication system
US8949252B2 (en) Product category optimization for image similarity searching of image-based listings in a network-based publication system
US20140114977A1 (en) System and method for document analysis, processing and information extraction
US8787679B1 (en) Shape-based search of a collection of content
Li et al. A comparison of 3D shape retrieval methods based on a large-scale benchmark supporting multimodal queries
US7774227B2 (en) Method and system utilizing online analytical processing (OLAP) for making predictions about business locations
EP2612263B1 (en) Sketch-based image search
CN103207879B (en) The generation method and apparatus of image index
Xiang et al. Objectnet3d: A large scale database for 3d object recognition
AU2011269050B2 (en) Method and system for fast and robust identification of specific products in images
CN100578508C (en) Interactive type image search system and method
KR20040005895A (en) Image retrieval using distance measure
US8589457B1 (en) Training scoring models optimized for highly-ranked results
US9405773B2 (en) Searching for more products like a specified product
CN103329126B (en) Utilize the search of joint image-audio query
US8898169B2 (en) Automated product attribute selection
Velmurugan et al. Content-based image retrieval using SURF and colour moments
US8977629B2 (en) Image-based popularity prediction
Akata et al. Non-negative matrix factorization in multimodality data for segmentation and label prediction
US8682071B1 (en) Contour detection and image classification

Legal Events

Date Code Title Description
PB01 Publication
C06 Publication
SE01 Entry into force of request for substantive examination
C10 Entry into substantive examination
GR01 Patent grant
C14 Grant of patent or utility model
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20140416

Termination date: 20151009

EXPY Termination of patent right or utility model