CN104376052B - A kind of same money commodity merging method based on commodity image - Google Patents
A kind of same money commodity merging method based on commodity image Download PDFInfo
- Publication number
- CN104376052B CN104376052B CN201410607607.6A CN201410607607A CN104376052B CN 104376052 B CN104376052 B CN 104376052B CN 201410607607 A CN201410607607 A CN 201410607607A CN 104376052 B CN104376052 B CN 104376052B
- Authority
- CN
- China
- Prior art keywords
- image
- money
- machine node
- node
- commodity
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
Landscapes
- Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of same money commodity merging method based on commodity image, this method utilizes global and local visual signature, using distributed storage and calculating match pattern, first approximate multiimage is searched in efficiency relatively higher global characteristics, if global search fails, approximate multiimage is searched in local feature;The approximate multiimage of most like one is retrieved from image library, commodity are belonged into the same same money group belonging to approximate multiimage, commodity are belonged into a new same money group if without approximate multiimage.The index structure of global and local feature can be supported concurrently to access, read-write, distributed storage and calculating in real time, quickly realize that the real-time online of mass image data is calculated with money.
Description
Technical field
It is more particularly to a kind of special based on global and local the present invention relates to computer vision and technical field of information retrieval
The same money commodity merging method levied.
Background technology
Approximate multiimage(Image Near-Duplicate)This concept is existed by Yan Ke et al. earliest
《Efficient Near-duplicate Detection and Sub-image Retrieval》Middle proposition.It is approximate to repeat
Image refers to same object or scene in different shooting situations(Illumination, yardstick, angle, block)The image of lower acquisition.
Approximate multiimage retrieval has wide application scenarios as a branch of image retrieval.For example, image infringement detection, figure
As shopping search, link related web page, video frequency searching etc..Blocking in approximate multiimage, displacement, yardstick, light change etc.
All challenge is brought for the retrieval of automatic approximation multiimage.
David G.Lowe exist《Distinctive Image Features from Scale-Invariant
Keypoints》In propose it is a kind of it is based on metric space, image scaling, rotation even affine transformation are maintained the invariance
Image local feature description -- SIFT description, its full name is Scale Invariant Feature Transform, i.e. chi
Spend invariant features conversion.Practice have shown that SIFT description not only have good yardstick and brightness consistency, while becoming to affine
Shape, visual angle change and noise also has certain robustness.
Robustness, locality and the strong characteristic of resolving ability make local feature turn into image retrieval, video copy detection
The feature favored etc. multiple fields.However, its dimension is high, the characteristic more than quantity is to require high to Time & Space Complexity
Searching field brings index and the big challenge of inquiry burden.To make full use of local feature and reducing amount of calculation, researcher
Use for reference the bag of words method in text retrieval field(Bag of Words, BoW), i.e., the frequency histogram representative occurred using keyword
The method of one document, it is proposed that visual dictionary method(Bag of Visual Words).Algorithm is first from a large amount of training sample figures
Local feature is extracted as in, then these characteristic points are clustered, obtained in a vision code book, i.e. visual dictionary, code book
Each cluster centre represent a vision word;Local feature region is all mapped to vision list finally for every image
Word, the matching of two images changes into the matching similar to two text documents, and the index of image, which can also be used, is similar to text
The inverted index of shelves.
The content of the invention
It is an object of the invention to provide a kind of same money commodity merging method based on commodity image.
The purpose of the present invention is achieved through the following technical solutions:A kind of same money commodity merging side based on commodity image
Method, comprises the following steps:
(1)The extraction of characteristics of image:The overall situation of each image in extraction image library, local two kinds of visual signatures, wherein
Global characteristics describe integral color distribution and the grain details of image, the gradient letter of the crucial regional area of one group of local feature description
Breath;
(2)Global characteristics are matched, and find approximate repeated matching image:Using distributed storage and calculating match pattern, profit
Feature database is managed with N number of machine node, is divided into following steps:
(2.1)Determine the retrieval machine node of match query:Utilize step(1)Obtained global image characteristic vector is right
Global characteristics data carry out two differentiation and combined to obtain a hash value, by carrying out grid section division to hash values, set up
Corresponding relation of the hash values with each machine node;Specially:CLD and EHD characteristic vectors are selected from global characteristics vector V,
Wherein for this two category features data, the magnitude range per category feature is CLD:0 ~ 63, EHD:0 ~ 7, each characteristic is entered
Row two breaks up, and output 0/1 is encoded, and multiple codings are joined together to constitute a hash value;Evenly distributed to each machine node
A part of hash values, multiple combination of nodes are entirety hash tables together;When needing to inquire about approximate multiimage, by image meter
Calculate feature and hash values are generated by binaryzation, affiliated machine node is then found by the hash values;
(2.2 )Global characteristics are matched:Use the picture in the global data base in the machine node determined in step 2.1
Global characteristics with retrieval image carry out matching comparison, analyse whether there is same money, and carry out data renewal;Specially:This is looked into
The global characteristics for asking image are matched with the set of image characteristics of all images in the image library in the machine node;If similar
Degree distance is less than 0.01, then it is assumed that is approximate repeated matching image, i.e., with money commodity, returns to the approximate repeated matching image ID,
As do not inquired identical money, then the retrieval image ID and characteristics of image are written to the global image feature set of the machine node
In, into step 3;When writing characteristic vector to machine node, write-protect is carried out using locking;
(3)Approximate repeated matching image is looked in local feature matching choosing:Using distributed storage and calculating match pattern, use
N number of machine node is managed and inquired about to local property data base, is divided into following steps:
(3.1)The retrieval and inquisition of individual node:For query image, using step 1, local feature is calculated first, is obtained
A series of vision word, different with global registration, the local feature vectors of query image will be sent to N number of storage simultaneously and calculate
Node is inquired about, and the local feature vectors of query image are with the figure in the feature database in { the Local Data } in each node
As being matched, if phase knowledge and magnanimity are more than 0.9, then it is assumed that the machine node has the approximate multiimage of query image, i.e., same to money
Commodity, return to the ID of the approximate multiimage, otherwise without with money commodity in the node diagnostic storehouse;
(3.2)Merge the retrieval result of each node:The Query Result of N number of calculating storage node is merged, if depositing
In same money, then into step(4)If not retrieving approximate repeated matching image, into step 3.3 characteristics of image stored record
Link,
(3.3)Characteristic storage is recorded:The local feature vectors of query image are with the spy in write buffer WriteCache first
Levy data to be compared, inquiry whether there is with amount of money evidence, enter step if data cached middle presence is with money(4), characteristic
According to not preserving;If data cached middle nothing is with amount of money evidence, into step(4), while the local feature vectors write-in of query image
In WriteCache, and analyze the cache size threshold value whether WriteCache property data base size exceedes setting;If slow
Deposit full, then the characteristic in WriteCache is written to the machine node to machine node using repeating query hit mode
In local feature storehouse, the machine node being written into is locked in write-in and protected;
(4)Obtain with money group number:If finding approximate multiimage, by step(2)Or step(3)The approximate weight matched
The same money group number of complex pattern distributes to new images, if step(2)And step(3)Do not find approximate multiimage, use input
The image ID of figure is used as the same money group number for inputting figure;New result recorded in same money database.
The beneficial effects of the invention are as follows:Using global and local visual signature, stored using Distributed Calculation and calculating
With pattern, approximate multiimage is searched in efficiency relatively higher global characteristics first, if global search fails, relative
Approximate multiimage is searched in slower local feature;The approximate multiimage of most like one is retrieved from image library, will
Commodity belong to the same same money group belonging to approximate multiimage, and commodity are belonged into one if without approximate multiimage
New same money group.Local inquiry mode, on the premise of recall ratio is not influenceed, improves the inquiry of system after first global
Can, whole system can be supported concurrently to access, read-write, distributed storage and calculating in real time, be rapidly completed mass image data
Real-time online is calculated with money.
Brief description of the drawings
Fig. 1 is that input picture is calculated with money overall flow figure.
Embodiment
The present invention is described in detail below in conjunction with the accompanying drawings, the purpose of the present invention and effect will be apparent.
As shown in figure 1, a kind of same money commodity merging method based on commodity image comprises the following steps:
Step 1:The overall situation of image, local visual feature extraction.
Following several visual signatures are specifically used in the present invention:
Global color feature employs MPEG-7 color layout descriptors(Color layout Descriptor,
CLD), express distributed intelligence of the color in space.The present invention uses partition strategy to the operator(It is divided into full figure block and image
Central block, totally 2 block images difference extraction feature), totally 12 tie up(2*6)Color layout feature.
Global textural characteristics employ MPEG-7 edge histogram descriptor(Edge Histogram Descriptor,
EHD), it is extracted 80 dimension datas of the description from 16 sub-picture contents.
Vision word of the local feature based on SIFT feature;First by the SIFT feature cluster generation one of samples pictures
The individual visual dictionary for including million rank vision word quantity, each SIFT key point of image is described as a vision
Word simultaneously includes the auxiliary matched information such as space coordinate, yardstick, direction.SIFT key points are cut by conspicuousness, a member-retaining portion
Remarkable characteristic, to improve the accuracy and speed of matching.
3 kinds of features carry out feature extraction to commodity image present invention employs more than, and global characteristics are retouched comprising color layout
Symbol CLD and edge histogram EHD is stated, local feature is the vision word that SIFT feature is generated.
Step 2:Global characteristics are matched, and find approximate reproducible results;
Same money under large data sets is calculated, and searches approximate repeated matching image in more than one hundred million width commodity images from commodity storehouse,
In order to improve systematic function and efficiency, because global characteristics characteristic matching speed is fast, searches gold first by global characteristics and repeat
Matching result, in order to improve the inquiry velocity of system, using distributed storage computation schema, N is used to image overall feature database
The individual storage machine rows of nodes that calculates is calculated and managed, and specific meter search procedure is as follows:
2.1 determine the retrieval machine node of match query, and any commodity image in image library is chosen first, is carried
Overall Vision characteristic vector V is taken, due to using distributed storage and calculating, there is N number of storage and calculate node, in order to accelerate
With calculating speed, we produce a series of hash values, according to Hash by the way of grid section division using characteristic vector
Value, sets up one-to-one relationship between hash value and machine node i (i=0...N), and storage and calculated value are only in a machine every time
Device section is carried out.
N represents N number of distributed machines node of the presence of system;
Specific features vector sets up as follows with machine node relationships:
CLD and EHD characteristic vectors are selected from global characteristics vector V, wherein for this two category features data, per category feature
Magnitude range be CLD:0 ~ 63, EHD:0 ~ 7, two are carried out to each characteristic and is broken up, output 0/1 is encoded, by multiple codings
Join together to constitute a hash value;A part of hash values are evenly distributed to each machine node, multiple combination of nodes are together
For overall hash tables;
When needing to inquire about approximate multiimage, feature is calculated by binaryzation generation hash values by image, then passed through
The hash values find affiliated machine node.,
2.2 global characteristics are matched:
Carried out using the picture in the global data base in the machine node determined in 2.1 with the global characteristics of query image
Matching is compared, and analyses whether there is same money, and carries out data renewal, specific as follows:
The set of image characteristics of the global characteristics of the query image and all images in the image library in the machine node enters
Row matching.If similarity distance is less than 0.01, then it is assumed that is approximate repeated matching image, is same money commodity, returns to the approximate weight
Image ID is matched again, does not such as inquire identical money, then the retrieval image id and characteristics of image are written to the machine node, write-in
Characteristics of image to the machine node is concentrated, and in order to ensure the security and integrality of data, is writing characteristic vector to machine section
During point, write-protect is carried out, a machine node synchronization does not allow the characteristic vector for writing multiple images.
Step 3 local feature is matched
The local feature of image is one group of vision word, and local feature sets up inverted index using vision word(Foundation is fallen
Row's indexing means are general conventional method, can check prior art document);
In order to improve the speed of retrieval, local feature matching uses distributed many machine sections with global characteristics matching, system
Point calculates match pattern, and image local feature storehouse is calculated and managed using N number of storage machine rows of nodes that calculates, each node
Characteristic data set is recorded as { Local Data },
In order to ensure in the matching analysis that any two images are completed on single machine, the local feature of each image
Multiple vision words are stored in same machine node data, rather than are evenly distributed according to single vision word many
In individual machine node.
In order to improve the concurrency performance of system, a write buffer data space Write is set to characteristics of image storage
Cache, the data of write buffer can be written in the local feature storehouse on different machine nodes { Local Data } in turn, so that
The characteristic of distributed machines is uniformly distributed.
Specific matching process is as follows:
The retrieval and inquisition of 3.1 individual nodes:
For query image, first choice calculates local feature, obtains a series of vision word, different with global registration, looks into
Ask the local feature vectors of image and will be sent to N number of storage calculate node simultaneously and inquired about, the local feature of query image to
Amount is matched with the image in the feature database in { the Local Data } in each node, if phase knowledge and magnanimity are more than 0.9, then it is assumed that
The machine node exists for approximate multiimage, is same money commodity, returns to the ID of the approximate multiimage, and otherwise the node is special
Levy in storehouse without with money commodity.
3.2 merge the retrieval result of each node:
The Query Result of N number of calculating storage node is merged, if there is same money, into step 4, if not retrieving
To approximate repeated matching image, into 3.3 characteristics of image stored record links.
3.3 features storage record:
The local feature vectors of query image are compared with the characteristic in write buffer WriteCache first, inquiry
With the presence or absence of same amount of money evidence, enter step 4 if data cached middle presence is with money, characteristic is not preserved, if in data cached
Without with amount of money evidence, then in the local feature vectors write-in WriteCache of query image, and WriteCache characteristic is analyzed
Whether exceed the cache size threshold value of setting according to storehouse size.If caching is full, the characteristic in WriteCache is write
Into one of machine node, the selection of the machine node is by the way of repeating query, to ensure that characteristic is evenly distributed on
In each machine node.To prevent from writing the data characteristics that two are same money image simultaneously, the loss of data is prevented, is written into
Machine node write-in when lock protected, to ensure the recall ratio of system.
Step 4:Obtain with money group number.
The image ID of the approximate multiimage obtained according to step 2 and step 3 inquires affiliated same money in database
Group number, regard this as the same money group number for inputting figure with money group number;
When step 2 and step 3 all do not find approximate multiimage, the image ID using input figure is used as input figure
With money group number;New same money result recorded in database, be same money with all images of money group number identical.
The present invention, using Distributed Calculation storage mode, is existed first by extracting global and local feature to input picture
Searched in efficiency higher global characteristics relatively, global search searches approximate multiimage in local feature again when failing;Entirely
The index structure of office and local feature can be supported concurrently to access, read-write, distributed storage and calculating in real time, realize magnanimity figure
As the real-time online of data is calculated with money.
Claims (1)
1. a kind of same money commodity merging method based on commodity image, it is characterised in that comprise the following steps:
(1)The extraction of characteristics of image:The overall situation of each image in extraction image library, local two kinds of visual feature vectors, wherein
Global characteristics describe integral color distribution and the grain details of image, the gradient letter of the crucial regional area of one group of local feature description
Breath;
(2)Global characteristics are matched, and find approximate repeated matching image:Using distributed storage and match pattern is calculated, using N number of
Machine node is managed to feature database, is divided into following steps:
(2.1)Determine the retrieval machine node of match query:Utilize step(1)Obtained image overall characteristic vector, to the overall situation
Characteristic carries out binaryzation and combined to obtain a hash value, by carrying out grid section division to hash values, sets up hash
It is worth the corresponding relation with each machine node;Specially:Select color layout descriptors and edge straight from global characteristics vector V
Square figure descriptor characteristic vector, wherein for this two category features data, the magnitude range per category feature is color layout descriptors:
0 ~ 63, edge histogram descriptor:0 ~ 7, binaryzation is carried out to each characteristic, output 0/1 is encoded, multiple codings are combined
Get up to constitute a hash value;A part of hash values are evenly distributed to each machine node, multiple combination of nodes are together to be whole
Body hash tables;When needing to inquire about approximate repeated matching image, hash values are generated by binaryzation by image calculating feature, then
Affiliated machine node is found by the hash values;
(2.2)Global characteristics are matched:Use step(2.1)The picture in global data base in the machine node of middle determination is looked into together
The global characteristics for asking image carry out matching comparison, analyse whether there is same money, and carry out data renewal;Specially:The query graph
The global characteristics of picture are matched with the set of image characteristics of all images in the image library in the machine node;If similarity away from
From less than 0.01, then it is assumed that be approximate repeated matching image, i.e., with money commodity, return to the approximate repeated matching image ID, do not have such as
Identical money is inquired, then query image ID and characteristics of image are written in the image overall feature set of the machine node, are entered
Enter step(3.1);When writing characteristic vector to machine node, write-protect is carried out using locking;
(3)Approximate repeated matching image is found in local feature matching:Using distributed storage and calculating match pattern, using N number of
Machine node is managed and inquired about to local property data base, is divided into following steps:
(3.1)The retrieval and inquisition of individual node:For query image, step is utilized(1), local feature is calculated first, obtains one
The vision word of series, difference is matched with global characteristics, and the local feature vectors of query image will be sent to N number of machine section simultaneously
Point is inquired about, and the local feature vectors of query image are carried out with the image in the feature database in the local data in each node
Matching, if similarity is more than 0.9, then it is assumed that the machine node has the approximate repeated matching image of query image, i.e., with money business
Product, return to the ID of the approximate repeated matching image, otherwise without with money commodity in the node diagnostic storehouse;
(3.2)Merge the retrieval result of each node:The Query Result of N number of machine node is merged, if there is same money,
Into step(4)If approximate repeated matching image is not retrieved, into step(3.3)Characteristics of image stored record link;
(3.3)Characteristics of image stored record:The local feature vectors of query image are with the spy in write buffer WriteCache first
Levy data to be compared, inquiry whether there is with amount of money evidence, enter step if data cached middle presence is with money(4), characteristic
According to not preserving;If data cached middle nothing is with amount of money evidence, into step(4), while the local feature vectors write-in of query image
In WriteCache, and analyze the cache size threshold value whether WriteCache property data base size exceedes setting;If slow
Deposit full, then the characteristic in WriteCache is written to the machine node to machine node using repeating query hit mode
In local feature storehouse, the machine node being written into is locked in write-in and protected;
(4)Obtain with money group number:If approximate repeated matching image is found, by step(2)Or step(3)The approximate weight matched
The same money group number of complex pattern distributes to new images, if step(2)And step(3)Do not find approximate multiimage, use input
The image ID of figure is used as the same money group number for inputting figure;New result recorded in same money database.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410607607.6A CN104376052B (en) | 2014-11-03 | 2014-11-03 | A kind of same money commodity merging method based on commodity image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410607607.6A CN104376052B (en) | 2014-11-03 | 2014-11-03 | A kind of same money commodity merging method based on commodity image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104376052A CN104376052A (en) | 2015-02-25 |
CN104376052B true CN104376052B (en) | 2017-07-14 |
Family
ID=52554959
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410607607.6A Expired - Fee Related CN104376052B (en) | 2014-11-03 | 2014-11-03 | A kind of same money commodity merging method based on commodity image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104376052B (en) |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106844381B (en) * | 2015-12-04 | 2020-06-30 | 富士通株式会社 | Image processing apparatus and method |
CN107533566A (en) * | 2016-02-25 | 2018-01-02 | 华为技术有限公司 | Method, portable electric appts and the graphic user interface retrieved to the content of picture |
US10558702B2 (en) * | 2016-04-06 | 2020-02-11 | Baidu Usa Llc | Unified storage system for online image searching and offline image analytics |
CN106777167B (en) * | 2016-12-21 | 2020-05-12 | 中国科学院上海高等研究院 | Massive human face image retrieval system and retrieval method based on Spark framework |
CN106951551B (en) * | 2017-03-28 | 2020-03-31 | 西安理工大学 | Multi-index image retrieval method combining GIST characteristics |
US10885118B2 (en) * | 2017-05-12 | 2021-01-05 | Futurewei Technologies, Inc. | Incremental graph computations for querying large graphs |
CN110675207A (en) * | 2018-07-03 | 2020-01-10 | 阿里巴巴集团控股有限公司 | Image display combination recommendation method, device and equipment |
CN110119460A (en) * | 2019-05-16 | 2019-08-13 | 广东三维家信息科技有限公司 | Image search method, device and electronic equipment |
CN111008210B (en) * | 2019-11-18 | 2023-08-11 | 浙江大华技术股份有限公司 | Commodity identification method, commodity identification device, codec and storage device |
CN113128923B (en) * | 2020-01-15 | 2024-05-21 | 北京京东乾石科技有限公司 | Storage recommendation method and device |
CN111552829B (en) * | 2020-05-07 | 2023-06-27 | 京东科技信息技术有限公司 | Method and apparatus for analyzing image material |
CN112182264B (en) * | 2020-10-10 | 2024-05-10 | 书行科技(北京)有限公司 | Method, device and equipment for determining landmark information and readable storage medium |
CN113763211A (en) * | 2021-09-23 | 2021-12-07 | 支付宝(杭州)信息技术有限公司 | Infringement detection method and device based on block chain and electronic equipment |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102339438A (en) * | 2010-07-22 | 2012-02-01 | 阿里巴巴集团控股有限公司 | Commodity information website publishing method, system and device |
CN102567543A (en) * | 2012-01-12 | 2012-07-11 | 北京搜狗信息服务有限公司 | Clothing picture search method and clothing picture search device |
CN102760144A (en) * | 2011-04-26 | 2012-10-31 | 乐活在线(北京)网络技术有限公司 | Information search method and system |
CN102890686A (en) * | 2011-07-21 | 2013-01-23 | 腾讯科技(深圳)有限公司 | Method and system for showing commodity search result |
CN103092861A (en) * | 2011-11-02 | 2013-05-08 | 阿里巴巴集团控股有限公司 | Method and system for selecting commodity representative picture |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140032359A1 (en) * | 2012-07-30 | 2014-01-30 | Infosys Limited | System and method for providing intelligent recommendations |
US20140052580A1 (en) * | 2012-08-17 | 2014-02-20 | Kallidus, Inc. | Product explorer page for use with interactive digital catalogs and touch-screen devices |
-
2014
- 2014-11-03 CN CN201410607607.6A patent/CN104376052B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102339438A (en) * | 2010-07-22 | 2012-02-01 | 阿里巴巴集团控股有限公司 | Commodity information website publishing method, system and device |
CN102760144A (en) * | 2011-04-26 | 2012-10-31 | 乐活在线(北京)网络技术有限公司 | Information search method and system |
CN102890686A (en) * | 2011-07-21 | 2013-01-23 | 腾讯科技(深圳)有限公司 | Method and system for showing commodity search result |
CN103092861A (en) * | 2011-11-02 | 2013-05-08 | 阿里巴巴集团控股有限公司 | Method and system for selecting commodity representative picture |
CN102567543A (en) * | 2012-01-12 | 2012-07-11 | 北京搜狗信息服务有限公司 | Clothing picture search method and clothing picture search device |
Also Published As
Publication number | Publication date |
---|---|
CN104376052A (en) | 2015-02-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104376052B (en) | A kind of same money commodity merging method based on commodity image | |
Schönberger et al. | A vote-and-verify strategy for fast spatial verification in image retrieval | |
Turcot et al. | Better matching with fewer features: The selection of useful features in large database recognition problems | |
Gao et al. | Database saliency for fast image retrieval | |
Liu et al. | Indexing of the CNN features for the large scale image search | |
Chen et al. | Ranking consistency for image matching and object retrieval | |
Yang et al. | An improved Bag-of-Words framework for remote sensing image retrieval in large-scale image databases | |
Xu et al. | Near duplicate identification with spatially aligned pyramid matching | |
Iakovidou et al. | Localizing global descriptors for content-based image retrieval | |
Sitaula et al. | Content and context features for scene image representation | |
Dharani et al. | Content based image retrieval system using feature classification with modified KNN algorithm | |
Feng et al. | An efficient indexing method for content-based image retrieval | |
Romberg et al. | Bundle min-Hashing: Speeded-up object retrieval | |
Yeh et al. | Fast visual retrieval using accelerated sequence matching | |
Hong et al. | Realtime and robust object matching with a large number of templates | |
Grana et al. | UNIMORE at ImageCLEF 2013: Scalable Concept Image Annotation. | |
Chen et al. | Mobile visual search from dynamic image databases | |
Ufer et al. | Object retrieval and localization in large art collections using deep multi-style feature fusion and iterative voting | |
Belarbi et al. | Web-based multimedia research and indexation for big data databases | |
Doulamis et al. | 3D modelling of cultural heritage objects from photos posted over the Twitter | |
Li et al. | A Transformer-Based Adaptive Semantic Aggregation Method for UAV Visual Geo-Localization | |
Staszewski et al. | Query-by-example image retrieval in microsoft SQL server | |
Brogan et al. | Dynamic spatial verification for large-scale object-level image retrieval | |
Bommisetty et al. | Content based video retrieval—methods, techniques and applications | |
Romberg | From local features to local regions |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170714 Termination date: 20201103 |
|
CF01 | Termination of patent right due to non-payment of annual fee |